Kaplan_Kapor-1990-Agenda by xiaoyounan


									        5. Jerrold         Kaplan,     Mitchell                  D. KapOr.  Edward     J. Belove,
                      Richard      A. LandSman,                     and Todd   R. Drake

                               A PERSONAL
The free-form,           evolving,                                                      needs, such as preparing                memos,
                                                                                        dcvclopiiig linancial project ions, and
personal           information                                                          planning prqjects.
that people deal with in                                                                    ‘Iho of the three ma.jor personal
                                                                                        computer          applications-spread-
the course of their daily                                                               sheets and word processing-are
activities       requires      more                                                     priularily     used    as pi:rsonaJ procluc-
                                                                                        tivily tools 1.1~individuals.         7%~: third
flexible       data structures                                                          major application-database                 man-
and data management                                                                     agemcnt systems-is             cliffcrent. It is
 systems         than      tabular                                                      used almost exclusively as a scalecl-
                                                                                        down version            of its mainframe
data structures           provide.                                                      counterpart.        Mainframe          database
A tool for managing              per-                                                   management          systems do ml typi-
                                                                                        cally serve the individual’s needs, but
 sonal information              must                                                    rather the organization’s            needs.
 conveniently          handle free-                                                         Dots this mean that individuals do
                                                                                        not neecl to store, organize, r&eve,
 textual       data; allow for                                                          and otherwise manage information?
 structure to evolve grace-                                                             No, they certainly do. But their needs
                                                                                        -and       the character of their data-
 fully      as the database                                                             arc: quite cliffcrcnt from CJrgaIliZa-
 grows; represent            unnor-                                                     tional needs and data.
 malized data; and sup-                                                                 Personal       lniocmation
 port data entry through                                                                In the course of their activities, pro-
                                                                                        fessionals and nianagcrs encounter a
 database views. We have                                                                variety of information     that can be
 designed         a new type of                                                         conveniently    expressed as a short
                                                                                        piece of test, that we call an &7fL An

 database that serves these                          IIC interactive,   single-use
                                                                                        item might represent (for example)
                                                     format of personal c:ompui
 needs-        called an “item/                    ‘ers has spawnecl applica
                                                                                        an idea, task, reminder, or fact. ‘ryp-
                                                                                        ically, an item consisls of a single
  category”           database-                      lions quite difl’erent         i   phrase or a scntencc and has little OI
                                                     character loom mainfram
  and realized this design                   m       applications.    Like main
                                                                                        no formal internal structure.
                                                                                           Some examples of items are:
  in a commercial          personal       franics, personal computers can selv
                                          the needs of an organization       in area            Call Fred on Tuesday about staff
  computer        software prod-          such as accounting, billing, or invcl

 uct named “Agenda.”
              -.                          tory. Unlike       typical mainframe:             l   Hotel Konadvcnture   guaranteed
        ;\c:.\q                   Sl..W however, they also serve person;
 Y;1!1!11l llll111-~lix1/!Il~/lli~lo-(1111’                                                     late arrival 2/31 confirniatioll
     #1487’L9. Mt*ct Mr. Silver andJoh                         When many categories are dc-                         magnetic   forI   (IIliiiIlly   tap)   Inorc
     I’dI 6:30 in lobby, dinner at Chez                   fined, the categories take on special                     coml)actly than in physical files.
     Albert 735.                                          signilicancc:       they provide the con-             l   Utility-l’he     data could be prc-
l    Consider new approach to pric-                       ccpls for orqanizing and uildcrstand-
                                                                          c                                         sented in different formats for dif-
     ing--“book      club”   model with                   irig tlic information al higher lcvcls ot                 ferciit purposcts.
     giveaways.                                           abstraction. ‘l’hc set IMI~~S          becomc
l    Expect sales rep011 liom Sally next                  the zewrd~ of‘ ;I language, which can                     ‘l’hcsc advantages of computerized
     Wed.                                                 describe the inli)rmntion          and its in-        databases over manual              processes
l   Clcancrs: pick up three shirts, drop                  terrelatiollships.      This ln~~gnagc~      can      prompted       both database theorists
     blue suit.                                           c:apturc salient fixtures ol’ the uni-                and practitioners        to focus on the cf-
l   *John Milton believed that li-cc will                 vcrsc of the inlhrmation,           reflecting        ficicnt, reliable manipulation         of large
    esisted so lhal loving God would be                   the “rral wxxld” from which they arc                  volumes of data. It was rccongized,
     meariingfiil.                                        drawn.                                                early on, that understanding                and
                                                               When itcrns/notcs can be ;lssigIlecl             dcscrihing the structure of the data in
    An individual      may handle several                                                                       advance (i.e., at database design
                                                          lo Iriultiple categories, an iudividual
items in a single day and have hun-                                                                             time) was a major advantage. Most
                                                          can alter his or her I’ocus to rcflcct dif-
dreds or even thousands of “active:”                                                                            ol’thc data being manipulated            had a
                                                          iixcnt perspectives by organizing             the
items that rlecd       lo tw stored, orga-
                                                                                                                repeating, record-oriented          structure
                                                          salnc infixma1ion         in multiple ways.
nized, ad sc:ariIiccl       rquhly           [8-i.                                                              that could bc used to improve pro-
                                                          We call CYKII such organization                   a
    While niucl~       of the information                                                                       cessing and storage etlicicncy. There-
                                                          oifnr. This is difficult     to accomplish
that individuals        nt~ecl to nlanagc            is                                                         I&e, relatively little attention            was
                                                          with physical storage melhods (such
shorl and self-generated,              sometimes                                                                paid to providing richer data descrip-
                                                          as paper in liles), whcrc: rach item 01
the granularity       is larger and ccmlcs                                                                      tion languages or designing systems
                                                          nole     exists in ii unique place.
from other sources. An illdiviclual                                                                             with more flexible data structures.
                                                               Most     pwple are unaware of the
may need to index or or~ganizc largeer                                                                              As a result, the dominant database
                                                          volume of information          they manage
bodies of test such as II~CIIIOS,                  re-                                                          management paradigms arc not well
                                                          011 a regular      basis, because they USC
ports, messages. or news slories. It is                                                                         suited for managing          personal data.
                                                          a variety of mechanisms for this pur-
useful to associate such ob.jects with                                                                          They are oriented toward the storage
                                                          pose. ‘The style of management                   is
items, whcrc: the item may be a topic,                                                                          and retrieval oflarge volumes of data
                                                          highly individual.       but often includes
summary. source. or htwllinr describ-                                                                           with a known, repetitive structure.
                                                          such aids as note cards, rolodcxcs,
ing the test. WC call such bodies of                                                                            These databases arc mainly used for
                                                          yellow Posted notes, while-you-wcrc-
text nofr:c.                                                                                                    keeping corporate records and com-
                                                          out messages, and lists on paper that
    To cope with the task ol’organizing                                                                         municating      within an organization.
                                                          are recopied on a regular basis. In
items arid notes, individuals               group
                                                          general, personal inl&mation              is too           Personal data has different charac-
them into sets--typically             in the form                                                               teristics: it is often of relatively lim-
                                                          ad hoc and poorly structured to war-
of an ordered list or file. The items/                                                                          ited volume;         its structure      is not
                                                          rant putting it into a record-oriented
noles then can he manipulated                   con-                                                            known in advance and evolves over
                                                          on-lint: database
vcnicntly as a unit. WC call these sets                                                                         time. It contains heterogeneous            data
cateprirs  and say that the items/notes                   Database           mmanamement                        types. This information         may be gcn-
are assigned to categories.                               Paradigms                                             erated and managed directly by an
    A representative       range of’catego-               In the lOGOs, mainframe computers
                                                                                                                individual     or may bc some body of
ries for one individual           is:                     were owned primarily by large orga-
                                                                                                                data that is indexed or acccsscd ac-
                                                          nizations and businesses ancl were
l   Phone calls                                                                                                 cording to some idiosyncratic           needs.
                                                          applied to their problems.            ‘l’hcsc
l   Things to read                                                                                              Therefore, mainframe databases and
                                                          consisted mainly of clerical data-
l   Errands                                                                                                     their personal         computer      counter-
                                                          base tasks. which usually        involved
l   Special dates and birthdays                                                                                 parts are not very useful for personal
                                                          filing, rctricving, sorting, processing,
l   Bring from home to of&x                                                                                     data, where discovering, presenting.
                                                          and reporting data. For these appli-
l   IEEE speech notes                                                                                           or modifying         the structure       of in-
                                                          cations, the computer        represented
l   Proposed picnic dates                                                                                       formation is more important.             A dif-
                                                          several advantages over the corre-
l   Program bugs                                                                                                fcrcnt data management             paradigm
                                                          sponding manual processes:
l   A~:coI~lplishn~crlts   for next per-                                                                        is required.
    formance appraisal                                    l   Speed-data        could be rctricvcd                   Many programs designed to man-
l   Major upcoming expenses                                   much faster than through nlimual                  age personal data have adopted a
l   High priority                                             nlcthods.                                         completely unstructured           approach,
l   Fred Smith                                            l   Accuracy-calculations        and 11Ia-            treating all data as ,@x fext that is
l   Pricing pO]iCy commitlee                                  nipulations   (such as sorting) could             searched and displayed in response to
l   Important      concepls of Western                        be perfbrmed ztm-c accurately.                    ad hoc keyword queries. This ap-
    philosophy                                            l   Density-data       could be stored in             proach confuses the lack of a fixed
structure with no slructure              at all.                      rile:              c:\yarm\P-2                                                                                                                   lQ/25/01                      02:36
Personal clara has structure, hi1 ils                                 View:              Initial   vhw
slruclure       is fluid,     cliangiiig      in                      VUIU
response to cvolviiig iiccds.                                             l       Junm            hu           w+&d                      thm          clue-r                  utiDf8ation                    8urvmY               for      th

    Hypertext          systems,      such as                                      2paYk2~dbhw8DhuuuIwilrprawrttbe                                                                                 rwult*                aut             fuaaday.
                                                                          l       spuk                 to      JOff           &out       inorum5ng              pwduotioa of tha sparklm
Hypercarcl for rhc Apple Macintosh,                                             diohwuho+                           by     the        and or u-r                       to 3000 unita          per -la.
provide a simple structuring             mcch-                                . Jtun     will                     rwt         with       odltou            of ‘9adocout0f”              uguina          on                                      lmdnud8y
                                                                                  to                                      dnlfot                   tha M       Spultl~       md op+rad        in M-t.
anism-links         bctwccn dala elements                                                 fiaop                        l

                                                                              l   *il             p=oduot~on                   mt.ho&                 of tkn ao8t        dryu        ha affaatd         if                                      lm
-which      allows the user to create a                                             l lb8tLtoto                        the UWaiiukl8                         l muabla         frtaAlAaa           io+tlw**
topological       spClce of adjact:nt clc-                                          l mumblo                          fnr        Yapa?
mcnls. Hpwevcr, snxcturc arises out
of the grouping of similar items and
relarionships        definccl aiiiong hose
groups, nol the simple linking                 of
individual        items togerhcr.         Most
liypcrtest systems assume that the in-
formation      of inleresl resides exlu-
sivcly in the data rather. than in the
slructure. They do not provirlc the
user with the incans lo view and                                      ICURE                       1 A Simple                      View          in Agenda
nianipulale      structure i&for         create
relationships       among groups of (lata
&111e11ts.                                                               Lttatdwad                     not6            for:       Call         Jeff         dut               th.         RCD    h++.                                            Note

                                                                                                                                                                                                                                  Line        o=ms:
                                                                                  It      would                .ppm.r             that         vm      l   xm     -r            mt                for   th             Dryor         ptoject.
Agenda:                   A Personal                                              also,apuk                            to     Jilna      - rho             may         h8rr         non         cutront               figurm*.
lniormation                     Manager
To study the requirenlcnts           of pcr-                                      Haa tha manporr porhapa undanotimat~d?
sonal rlalahnse        nianagcnicnt,     wc                                       Yrut justify upurs.s.
                                                                                  G+ misd      foncasta l xl chack out thoroughly....
constructed       several prototype     sys-
tems,  culniinaling     in a program suit-
able for release as a commercial
product.      This program,          named
 “Lotus Agenda," meets scvcral dc-
sign recluiremcnts fi.)r managing per-
sonal information.        These arc:
l           The user must bc able to easily
            enter, edit, and manipulate            fix!c
            teslual items without conccrri fix
            the underlying        slruclure    of the
                                                                  FIGURE                          2 Attached                          Note       for        an Item
    l       ‘T’he user must no1 be required to
            specili lhe structure oflhe       data    in
            advance and must be able to mod-
                                                                  described above. Items arc> the con-                                                                        natively, catcgorics can be under-
            ify the dalabase structure            as il
                                                                  lenl of the database and can consist                                                                        stood as one-place prcdicales defined
            cvolvcs without         losing data or
                                                                  of up to scvcral lines (350 charac-                                                                         over the universe of‘itcms. Given the
             reorganizing     the database.
                                                                  tars) of text. Items in Agenda arc                                                                          categories delined earlier, tlic useI
        l   The user must be able to define                       assigned to or filed in categories.                                                                         might assign an item “Call Fred nest
             rcporls in idiosyncratic       I’ormats.
                                                                  Categories provide the structure 01                                                                         Tuesday about pricing policy plans”
            ‘I’hrough     thcsc reports, the USCI
                                                                  the database and arc usctl to organize                                                                      to several categories: “Phone calls,”
             musl be able to create ant1 nmd-
                                                                  items and display        them convc-                                                                        “Fred Smith,” a11c1 “l’ricing     policy
             ify both database structure and                      nienlly. Like items, the categories                                                                         committee.”
                                                                  are not prc-dcfincd--the          user                                                                         Figure 1 shows a simple list of
        From the user’s perspective,                              crcatcs.    deletes ., and modifies                                                                         items in Agenda assigned fo a catc-
    Agenda      consists of a few simple                          categories as necessary in order to                                                                         gory callcrl “Tasks.” A bar cursol
    elements that are used lo incrmim-                            manage the items.                                                                                           scrvcx as a pointer, highlighting     the
    tally dcfinc and populate a database.                            Each category defines a subse( of                                                                        current itcni or category in reverse
        The first of these is the itcrn, as                       those items assigned to it. Altcr-                                                                          video. In the example, the cursor is

COMMUNICATIONS            OF   TREACM: J,,,, ,‘N,li,u,.:;3.N,,i                                                                                                                                                                                              107
 on item “Call Jeff about the K&D                  are implemented      in part by enforcing         hierarchy subtree of the selected
 budget.”                                          these relationships.                              category are then displayed in the
      Both items and categories can op-                                                              colu~m~.   Note that a column may
 tionally have an attachctl body of text                                                             contain multiple entries for a single
 called a clots Notes can be up to about           Items arc entered and displayed in                item, i.e., the information  need not
  10,000 characters in length and may              Agenda through          viezca. A view is         be normalized.      The columns also
 be internal to an Agcmda database or              analogous       to a report in a typical          c:an be different for each section in
 contained in a separate ASCII tile.               database manap;ement           system, ex-        a view.
 Notes arc easily accessible fi-om an              cept that it is dynamic: it is modified             Figure 4 is an example of a com-
 item or category by placing the bar               automatically      when items or catego-        plex view including          many sections
cursor on the item or category and                 ries change. ‘I’he user typically dc-           and columns. Iivo sections are visi-
pressing a function key.                           lines a number o:i‘dilYerent views and          ble, one headed by the category
     Figure ILillustrates a note attached          can quickly       switch among them.            “Gust      Clothes      Dryer”    and one
to item “Call Jeff about the K&D                     I’iezta allow the user to focus on dill       headed by the category “Sparklc
budget.” Typically, notes are used to              fercnt aspects ul’ the database for             Dishwasher”        which is incomplete on
capture some further detail or some                dit’lerent purposes by varying the              the screen. A column headed by
secondary information about an item                content or format of the display. For           “Priority”      is to the left of the item
or category. Another USC of an item                example, when deciding what to do               text in both sections, while a column
and its note is to make the note a                  next, the user might consult a view            headed by “Department”           is present
short document and use the item it-                listing items by priority; in cast a col-       only in the lirst section. The “Peo-
self as a title. Notes on categories are            league drops in, another view might            ple” column displays which category
often used to document the intent                   provide only the items that relate to          in the “People”        hierarchy the items
behind the category, i.e., what sorts of            that person, organized by topic.               are assigned to. Note that one item
items should be assigned there.                          The user constructs views visually        has multiple assignments within the
     The categories are organized into              by arranging categories into (possibly          “People”     hierarchy.
a tree structure called the category                irregular) row and column structures                Notic:c that all important     aspects
hierarchy. ‘I’he purpose of the hier-               on the display. Formally, views are            of a view arc defined by selecting
archy is to conveniently describe two               defined via three mechanisms:                  categories. In addition to providing
fundamental         structuring    principles:
                                                       Selection-A      query, called criteria,    a simple, visually oriented definition
subsumnption am1 n~utual e.&.sion. Items           l

                                                       caii be specified to select a subset of      method for the user, this allows the
that are assigned to a category in the
                                                       items to be displayed. The basis             program to modify the view struc-
hierarchy are also implicitly assigned
                                                       for the selection is assignment (or         tures gracefully as the database struc-
to the parent of that category in the
                                                       non-assignmenl)       of the item 10         ture evolves.
hierarchy, i.e., a parent category sub-
                                                       categories.                                      In summary, an Agenda database
sumes its children. A category in the
                                                                                                    consists of four basic elements. Items
hierarchy       can bc optionally         desig-   l   Demarcation--Once        a set of items
                                                                                                    and notes provide the content; cate-
nated as exclusive, indicating that no                 is selected, they are organized for
                                                                                                    gories organized         into a hierarchy
item can bc assigned to more than                      display into a collection        01’ lists
                                                                                                    provide the structure; and views pro-
one child in the category’s subtree.                   called sections. Each section is
                                                                                                    vide the data entry and reporting
     Figure 3 is a typical category hier-              headed by a category. The user
                                                                                                    (output) mechanism.
archy display. The brackets around                     defines the sections that are to ap-
the children of “Priority”             indicate        pear and indicates their order by          Updating            Databases
that Priority          is exclusive.      Since        selecting a category to head each          ThrOugh          VleWS
Priority is a parent of “High,”              any       section. Within a section, the items       The power of Agenda comes primar-
item assigned to High is implicitly                    appear as bullets and are wrapped          ily from the user’s ability to update
assigned to Priority as well.                          to fit onto the display as necessary.      the database implicitly       through the
     Subsumption         and mutual cxclu-         l   Annotation-Further        information      views. By manipulating           items and
 sion, defined through this tree struc-                about the items that appear in each        categories in a view, the user can
 ture, implement          a limited class of           section can be optionally displayed        make changes to the underlying
logical implications:           A implies B            in columm tither to the right or left      database       that are immediately
(subsumption);          and A implies not B            ofthe items in each section. The in- ~ reflected in other views, without con-
 (mutual exclusion).          Advanced I’ea-           formation displayed in the columns         cern for the internal organization          or
 turcs called conditions and actions allow             are other categories that the items        storage of the information.
 users to define more complex logical                  are assigned to. The user specifies            While this facility is a significant
 interrelationships         between catcgo-            a family of inibrmation       to be dis-   advantage to the user, it presents a
 ries, including       a form of declarative           played in a column simply by se-           complex implementation             problem.
 if-then rules. Automatic assignment                   lecting a category to head that            Theoretically,       updating   a database
 and other forms ofinte/ligmt behavior                 column;     only categories       in the   through     a view is not always well
defined IG, 71.Specifically, an aclion                                V
may be ambiguous       or evm contra-                                 I
dictory to interpret in the underlying                                                                          NAIN
database 141. Agenda LSCS a variety                                                                                    "Untitled-l"
of formal methods and heuristics to                                                                                    ovmrview
interpret these actions in a rcason-                                                                                   Dmpartmmnt:Dept
able way and affect reasonable                                                                                                Nsrluting
 changes to the database.                                                                                                     Production
    A normal suite ofscreen-oriented                                                                                          Sal*8
 commands is provided to the user in                                                                                   Projmts
 order to manipulate    the objects visi-                                                                                       Gust    Clothos       Dz'yW;Gust
ble in the view: insert, delete, move,                                                                                          Sparkle       Dishwashsr;Sparkl~

copy, edit, or replace. ‘I’he system in-
terprets these commands by making
 the simplest change to the underlying
database, which results in the literal
 change to the display indicated by the
 command. As a result, the actual in-
                                                                  d                                                                                                                                                   /
 tcrprctation  of a command is highly
 conte.xt sensitive.                                              FIN CURE             3 Typical               Category                  Hierarchy              Display
    Some simple examples are:
                                                                           Pila:     c:\AGumA\PRo.nCT2
l       Inserting a new item into a section                                                                                                                                               10/2J/Bf          02:31
                                                                           viw:      owta**        - uxgwt
        assigns that item to the category
        that heads the section.                                           'riority      Gust        Clothas         Dryer                                                                    People    Depwztaant
                                                                           High         Narksting              strstopy suggested
                                                                                                                                by                               Ml.ka      for     tha       -Xike    .Narkoting
    l   Deleting an item from a section                                                 Gust drysr               &a#           cot .ppasr                to be suppo*ed
        removes the item from the category                                              by tho,rasar&.                          IIe.t  with              him to diws88
                                                                                        this tomorrow.
        that heads the section. For conve-                                 nigh       .spOak          to Jaff       bout                 ccqdetinp              tha      r8vb.d               *Jaff    -Production
        nience, a separate command called                                                1988       production         plan                   for    Gustdryarm.               I will
                                                                                        Iwdit             bytheJan.                      l-ger~mntlng.
        discard removes an item or category                               Medium      *Have   Ni.h    mime                       tha          Gust     braak-             amlysim             *Nika    *Ilarkmting
        from the database completely.                                                   by next    weok    to                  tak.           the    stool      prim       drop
                                                                                         into       aooouut.
    l   Copying an item into a se&on                                       Low        *Look         at i&w              itea        0x1 a low-and
                                                                                                                                       Uikm               entry                               -N.ik*   ~nBrketing
        assigns the item to the category                                                iutI'mdryermark&,                      usingGusttwhr.olqy.
        that heads that section and re-                                               eNike       and DSW             must work     011 "sys       to up salm                                 .NikO    ' Sales
                                                                                        of      th   Gust         dryer    to 4OOO/month           by tha     and                              DAVO      Narksting
        moves the item from any category                                                of      Nww&o~.
        that is mutually exclusive with it.                               'riority      Sparkls    Diishrasher                                                                                             Paoplm
                                                                           iligh      l 8pe8k         Jeff
                                                                                                     to        about    iacr~sing      produotion   of                                                     'Joif
         Catcgorics can appear in a view in                                             thm Sparklm        dimhmahet          by thm aad of N-r
    three contexts-as      a section heading,                                           to 3000 wits           par month.
                                                                          Nadium      l ltmdw    tdavision           s&     frca   JUXW for thm n."    Sparkle.                                             . June
    a column heading, or in association                                                                                                                                                                               J
    with an item such as the contents of
    a column. By far the most common                                      CURE          4 Example                 of a complex                        View
    operations on categories in a view are
    insert and replace, which involve the
    selection of a category for display in                        der ;ired catc<gory.                                                                       biguity, Agenda examines assign-
    one of these three contexts.                                       The goal of thcsc methods is to                                                       ments of similar items and categories
         One approach to allowing the usei                        ma Ike the minimal       logically consis-                                                 in the vicinity of the user’s change.
     to select categories to define a view is                     ten II change to the database that will                                                         While these techniques       do not
     to present the category hierarchy and                        pr( educe the change to the view indi-                                                     always yield an optimal result, they
     allow the user to select the desired                         cat ed by the user. This often causes                                                      arc quite satisfactory in the majority
     categories. Experience        with users,                    Ott ier views to be modilied               im-                                             of cases. The residual cases are minor
     however, revealed that it was discon-                        plic citly as well. When the user makes                                                    annoyances, more than offset by the
     certing to keep switching back and                           aC hange to a view that is ambiguous                                                       convenience       of updating    through
     forth between       the partially    con-                    wil :h rcspcct to a database (i.e.,                                                        views.
     structcd view and a display of the                           thz tt has multiple     interpretations),                                                       This introduces some ambiguities
     category hierarchy. WC choose an                             AS   enda selects the interpretation      that                                              in interpreting    the user’s intent. Is
     alternative   approach,     allowing  the                    cat rses the minimal        disruption       to                                            the user creating a new category OI
    user to simply position the cursor on                         Ott ier views. If this heuristic    provides                                               trying to select an existing one? What
    the screen and type the name of the                           no guidance for resolving the am-                                                           if there are multiple categories with

COMMUNICATIONS     0FT”E   ACM/July   l!l’lO/V,,l.:33.   So   i                                                                                                                                                      109
                                                                                                                                                                            cessed in the background         when no
        1 Of     2 IMtChiXUJ                CAt9qOriaS.                 Current          choice:            Jdf                                        SXh
                                                                                                                                                                            other actions are pending. The pro-
                                                                                                                                                                            ccssing for each item takes place sc-
   Priority               Gust         Cl&has          Dryer                                                                      Peoplm        Departamnt
   . aiqll                Narkmtinq               l trateqy            l qqmtad            by llike  for                tha        .Nikm        *Mark&inq                   qucntially through rhe categories of
                       Gust  dryar     doaa not     appear                              to h aupportad                                                                      the hierarchy in depth-first order (the
                       by tha raaearch.        t&et     with                            him to diaoum
                       tbia  tomarror.                                                                                                                                      order in which the hierarchy              is
   . aiqa            z+Spmak to Jmff about          Complatinq                                   the   zmviamd                    . Dave        *Production                 displayed).
                       1988 production      plan      for    Gust                              4xymrs.      I will                               Salma
                       mad   it by thm Jan.         1 mumqmr.                                    mm&inq.
                                                                                                                                                                                 The user can associate with each
   ~bh3h18              *Ilava      Nika       ravine            tha          Gust    breakwan              analytic              *lU.ka        .Markdtinq                  category in the hierarchy a set ofcon-
                           by nrrt.      rsrlr        to       taka           thm    atmel       pticm        drop                                                          dilions and a sc‘l of actions, assem-
                          into         account.
   .LOW              *Look     at idmaa                 from           wilu     on      . low-and        entry                    *Nika         ~lhrketinq                  bled from a restricted \:ocabulary
                        in tha    dryer                market,            using         Gust    twbnoloqy.                                                                  available for this purpose. The con-
                     ‘Nikm     and Da4                  mat            work     on      “ay‘    to up *alma                       *Nile         . Sales
                          of     the      Gust       dzyaz        to          4000/month       by tha l nd
                                                                                                                                                                            ditions are evalualcrl each time an
                                                                                                                                    Dam          Marketinq
                          of     No-r.                                                                                                                                      item is chnng~l, and ifthe conditions
   Priority            Sparkle         Dishwasher                                                                                                     People                arc met, the changed item is auto-
   *Riqh             *Speak       to Jeff         about  increaainq                              production               of                           *Jeff                matically assigned to this category.
                       tba     Sparkla        dishwasher       by tha                          and of N-I:
                       to 3000         unita      per month.                                                                                                                Whcncvcr an assignmcnl is inatlc to
                                                                                                                                                                            a category (including         an explicit
                                                                                                                                                                            assignment by lhc user), any aclions
FIGURE                     5     Example             of    Category                  Matching                                                                               the user has dcfincd for that catcgor)
                                                                                                                                                                            are immediately        taken.
                                                                                                                                                                                 ‘l’hcrc arc three types of condi-
the siu~ic iiiunc iii diffcrcnt parts of                                                     technique that can force a new iter n                                          tions rhac can be tested f’or each
the hierarchy? ‘lb rcsolvc: thcsc. qucs-                                                     ro satisfy a given set ofcrirrria. Thu 5,                                      category. These are called string
[ions,         Agenda            incorporates                  a concep                      ciitcring a new item into a view th; it                                        conditions,     profile conditions.   and
c&d category matching, illustrated                                                           displays only items associated wit 11                                          date conditions.
in Figure 5. When rhe user begins to                                                         a palYlclrla              I projwl            or   chc          (Ihr   c? i-
                                                                                                                                                                            (1) String      conditions    perform        a
type,         a special           display           appears             at the               ariipk:) will assign that item to th; It
                                                                                                                                                                                match between the content ofrhe
top of the screen, which indicates                                                           project or dale.                                                                   item and the content of the cate-
how many catcgorics in the hierarchy                                                             In soiric casc:i, tlicrc arc: niultipl Ic
                                                                                                                                                                                gory (i.e., the category name),
llli~t(‘ll   the: string bciiig        critcrcd and                                          ways that a new item could satisfy th If2
                                                                                                                                                                                with the goa of computing                a
displays the lirst of thcsc (in hicrar-                                                      criteria. When such arribiguitics          0~ ._
                                                                                                                                                                                measure of semantic overlap be-
thy display order). While typing lhc                                                         cur, the context ofsurrounding         itclr IS
                                                                                                                                                                                twccn the two. This is done by
string, the user can use arrow keys lo                                                       in the view arc uscrl to sclcct a sing1 lc                                         comparing       the words appearing
display other matches, pop up ;I dis-                                                        interpretation.
                                                                                                                                                                                in each after stripping       suflises.
play box conraining the WlWilIlt                par-
                                                                                                                                                                                looking      for certain    common
Cons of the hierarchy, accept the                                                            Automatic                            Assignment                                    identifiable     data types (such as
currcritly displayed category, or crc-                                                       and Implicit         Actions
                                                                                                                                                                                proper names, dates, tirncs, and
ate a new category simply by typing                                                          Entering and categorizing        data in a
                                                                                                                                                                                numbers), and performing            lim-
a uriiquc string.                                                                            personal clalalxisc    can   Ix: ii chart :.
                                                                                                                                                                                ited syntactic analysis. (‘l’he de-
        This process has several advan-                                                      Ideally, a personal database systcl 11
                                                                                                                                                                                tails of this process are beyond
tagcs. In most contexts, the user only                                                       should aid the user in this proces: 5.
                                                                                                                                                                                the scope of this article.)
types a few characters to sclccl ii                                                          f\,genda uses 3 wide variety of cut 2s
category without sacrificing the abil-                                                       and contextual information          to hcl P                                   (2) Profile condilions    apply a set of
ity to make a sclcction horn the en-                                                         the user properly categorize          ilen- IS                                     criteria to the item that is iden-
tire hierarchy when desired. Also,                                                           ad    manage: the database. Updatin S                                              tical in form to the view criteria
new catcgorics can bc created with-                                                          the database through views is 011 le                                               describccl above. If the catego-
out special            c:oriiiriaritls-iricrc-                                               nspcct ofthis process. Another meet l-                                             rization   of the item meets the
mentally extending the structure of                                                          anism allows the user to pwgm ‘VI                                                  criteria, the condition succeeds.
the clatabase.                                                                               automnlir    assignments and implic il
        Perhaps the most powerful asp3                                                       act1ons.                                                                       (3)   Date conditions    test one of the
ol’updaling        through views is the role                                                     In addilion lo serving 3s the stru( ._                                           three intrinsic dates associated
oi‘critcria (qucrics). When an item is                                                       turc of the database, the categor Y                                                  with an item (entry date, complc-
entrrecl into a view for which a critc-                                                      hierarchy      embodies    a declaraliv ‘e                                           tion date, and user-assigned cal-
ria selection is specified, it follows                                                       program       that is executed again: jl                                             cndar date) for inclusion      in a
that the item is intended 10 satisfy the                                                     changed itcrns. I21c:h time a new iter                                               specified date range. Tfthe item’s
current criteria. Agxda iricorporatcs                                                        is cntcrcd or an old one is modifiec ;                                               date falls inside the range the
3 special rcvcrsc query evaluation                                                           it is placed on a C~LI~YK, which is pro-                                             condition   succeeds.
    Unless the user explicitly defines a             plementing ad hoc logical implica-            Ellmlnatlnm          FleldNalue
profile condition or a date condition,               tions.     For example,        any item       DIstInctIons
a category has only a string condition               assigned to both “John Smith” and             There are several structuring prin-
associated with it, since it exists im-              “policy committee”         should be as-      ciples that are inherent in tabular
plicitly by virtue of the category’s                 signed to “discuss with manager.”             database organization, but that seem
name. This is by far the most com-                   They are also handy for implement-            alien to personal data. Forcing users
mon situation, as it does not require                ing default assignments,          such as     to consider them inhibits their abil-
any special action on the part of the                classifying any “bug” not already             ity to organize their data.
user. Indeed, the user does not even                 assigned to a “classification”           as       The most important of these is a
need to know about the subtleties of                 “serious.”                                    fixed field/value     distinction.     In a
automatic assignment in order to                         Date conditions are mainly used to        tabular database, the fields (column
benefit from string conditions.                      define calendar views-views      in which     headings), and possibly the set of
    In the simplest case, items will get             the items are displayed in sections ac-       legal values that can appear in a field
assigned to categories that match one                cording to dates. Using date condi-           (data>, are normally           defined in
or more words in the item. For in-                   tions, it is possible to construct a daily    advance-when           the database is
stance, an item “call Mr. Smith                      calendar, weekly calendar, or any             designed. The organization              into
about the policy meeting”           might            other desired arrangement.           Date     fields and values, however, is not a
match the string conditions of cate-                 conditions are also useful for chang-         characteristic of the data itself, but a
gories “calls,” “John Smith,” “policy                ing an item’s assignments based on            representational      convenience;       the
committee,”        and “meetings”      and           some time-sensitive conditions. For           same information can be represented
hence be automatically        assigned to            example, an item that becomes over-           in different tabular forms.
them.                                                due can be automatically assigned to              For example, the “employment
    A set of special symbols can be                  a “late” category, or one due within          status” information in Figure 6a is
embedded into a category name to                     three days can be moved into a “high          represented in a different field/value
gain additional        control of string             priority”    category.                        format in 6b, by elevating the value
matching. These provide for alter-                       Date conditions work in conjunc-           “full-time” to a binary (yes/no) field.
native descriptions of the category                  tion with a natural language date             The same information appears in a
(called aliases); requiring        precise           parser that extracts English date ex-         different format in 6c, where the divi-
matches; accepting any substrings;                   pressions from items and assigns the          sions have been represented                as
and fixing the order of matched                      appropriate date. This grammar can            separate fields in the database, with
words or case sensitivity.                           correctly interpret expressions such          their departments as values. Explic-
    A global setting called 1&i&e                    as “two weeks from last Tuesday,”             itly representing the implication that
determines how strong a match must                    “the day after tomorrow,” and “the           the departments belong to the divi-
be in order for the string conditions                last week in June,” as well as standard       sions could be accomplished by nor-
to succeed. By varying the initiative,               American       date formats such as           malizing the information           into two
the user can control how freely the                  “May 25, 1989” and “2/5/87.”                  tables, such as in 6d.
program will accept matches be-                          Actions are somewhat simpler                  In general, the fields represent the
tween items and categories.                          than conditions in that they are taken        fixed information about what can be
     Another      global setting, called             immediately whenever assignments              in the domain, while the values in-
Authority,       determines      whether             occur. They can be used to date an            dicate the changing          information
 matches over the initiative threshold               item (date actions), pwh an item              about what Zs.The distinction is not
 are carried out immediately         or are          onto one or more other categories             inherent in the data, but rather
 put in a special queue of suggested                 (profile actions), or execute a limited       reflects how the data may change
 categorizations to be reviewed with the             set of special actions such as to dis-        over time.
 user. This technique,         sometimes             card the item from the database.                  While data from a traditional
 called mixed initiative [2,3,10], allows                Note that any change to an item           database system could be trans-
 the program to engage the user in                   that results from a successful condi-         formed and presented in any of these
 a clarification     dialog for weak or              tion or an action will place the item         formats, it would be quite difficult
 potentially    spurious automatic as-               back onto the queue of changed                to allow arbitrary       updates to the
 signments. When this queue is non-                  items, causing it to be processed             transformed data, as this may re-
 empty, a special ? symbol appears at                again through the hierarchy. Pro-             quire extensive internal reorganiza-
 the upper right of the screen. The                  grams of considerable complexity              tion. By not presuming a particular
 user begins the dialog by selecting a               can be constructed by cascading se-           field/value      orientation         at the
 menu item “questions,” after which                  quences of conditions and actions             database level, Agenda can present
 the program presents its pending                    throughout the hierarchy, In effect,          the data in any such format and sup-
 assignments and the strength of the                 the hierarchy is a declarative pro-           port updates in real time.
 evidence for them.                                  gram that is run against each                     In Agenda, both fields and values
     Profile conditions are used for im-             changed item.                                 are represented by the more general

Couw”“,C~r,C”sCFT”EliCY/July   199O/Vol.33,   No.7                                                                                        111
                                                                                                             concept of category. Whether               a
                                                                                                             category serves as a field or a value
      PERSON                DIVXSION
                            --                                   DEPARTMENT           FX?LOYlG?.NT STATUS    depends on its use in a given context
      Fred Smith                     - _
                            Test ECIU~D.                         Sales                part-time              -what role it plays in a view. For ex-
      Sally  Jones          Banking        Software              Development          full-time
      Bill  Harris          Banking        Software              cust.   support      full-time              ample, the structural information in
      ...                                                                                                    Figure 6 might be represented in
                                                                                                             Agenda as the category hierarchy in
FBGURE           6a                                                                                          Figure 7. This information could be
                                                                                                             presented in different forms in dif-
                                                                                                             ferent views, as shown in Figure 8a,
      PERSON            DIVXSION
                        _I-                                      DEPARTMENT              FULL-TIME           b, c, and d.
      Fred Smith        Test Equip.                              Sales                   NO                     Views a, b, and c in Figure 8
      Sally  Jones      Bankin&   Sbftware                       Development             Yes
      Bill  Harris      Banking    Software                      cust.   support         Yes                 display the data in a form similar to
      ...                                                                                                    the corresponding tables in Figure 6.
                                                                                                             The Agenda database is identical in
FlGURE           6 b                                                                                         each case, but the views are format-
                                                                                                             ted to reflect different field/value
                                                                                                             distinctions.    In this database, the
      PERSON                TES!C EQUIP.          DIV.          BANKING    SOETW. DIV.      FULL-TIME        people are treated as items.
      Fred Smith        Sales                                                                NO
                                                                                                                View 8a uses the same category
      Sally  Jones                                              Development                  Yes
      Bill  Harris                                              cust.   support              Yes             (Division/Dept)      to head two col-
      ...                                                                                                    umns. One column uses a child for-
                                                                                                             mat, however, to display only the
FBGURE           6C                                                                                          immediate children of the column
                                                                                                             heading (the divisions), while the
                                                                                                             other uses an actual format to display
      PERSON            DEPARTKENT
                        --                                PULL-TIME                                          the departments. View 8b shows a
      Fred     Smith    sales                             NO
                                                                                                             column headed by the “full-time”
      Sally     Jones   Development                       Yes
      Bill     Harris   cust.   support                   Yes                                                category instead of the “employment
      ...                                                                                                    status” category, with a “Y/N” for-
      DEPARTUENT              DXVISION
                              --                                                                             mat. View 8c selects “Test Equip.”
      Sales                   T(?st Equip.                                                                   and “Banking Software” as column
      Development             Banking   Software
      cust. support           Banking   Software
      ...                                                                                                       While it is possible in Agenda to
                                                                                                             create a multi-table      structure as in
FlCURE           6d                                                                                          Figure 6d by adding an additional
                                                                                                             section to the view, it is misleading to
                                                                                                             do so in this instance.           Agenda
                                                                  Vimv:Psrson      by Divisicm       3   :   databases     can contain        multiple
                                                                                                             record types, as in a multi-table
                                                                                                             database; however, only a limited
                                                                                                             form ofjoin is possible. In a typical
                                DivisiodRqat.                                                                tabular database, each table contains
                                       Test     lcquip.                                                      one or more distinguished            fields
                                  E Banking SoftwaEm                                                         constituting    a unique key for each
                                       Dstrr1oFalult                                                         entry. The corresponding concept in
                                        cust. support                                                        Agenda is the item itself, while the
                                aqrloynvnt     st*tus
                                    full-tinu                                                                categories correspond to fully in-
                                  c part-tims                                                                dexed non-key fields. Some relational
                                                                                                             database joins implicitly alter the in-
                                                                                                             terpretation of a given field from be-
                                                                                                             ing a key to a non-key field. In this
                                                                                                             example, one might expect to join the
                                                                                                             two tables in 6d on DEPARTMENT,

 ULIUL                                                                                                       yielding a table as in 6b. However,
                                                                                                             this implies
                                                                                                                                 a conversion
                                                                                                                                    from a key to a

FIGURE          7 Ca&gory             Hierarchy                                                              non-key field. The corresponding

112                                                                                                                      No
                                                                                                             JulymxxV”1.33, 7,COYY”WIC*TIONSOICT”E~~Y
FlCURE      8C

operation in Agenda requires con-        and the content varies over time.         volume of information that needs to
verting an item to a category, which     When dealing with personal data,          be handled. The amount of informa-
is not a view-level operation.           users must be able to reorganize their    tion can vary dramatically        as the
   Nonetheless, the referential in-      databases on-the-fly. The basic prob-     project actually unfolds over time
tegrity [l] implied by the decomposi-    lem is that the structure of a personal   -and      therefore the structural re-
tion of the data into two tables in      database may not be known in ad-          quirements change as well.
Figure 6 is enforced in Agenda.          vance and, indeed, may continuously           In this context, the ratio of items
Specifically, an individual could not    evolve over time as the nature of the     to categories becomes important. Ex-
be assigned to a department incon-       information or the context changes.       perience with Agenda suggests that
sistent with his or her division. In        For example, consider the problem      personal information       tends to be
Figure 8a, for example, the user         of organizing a new project to be         organized into a relatively        small
could position      the cursor over      undertaken. As the project begins to      number of categories, each of which
“Sales” and type “Development,”          take shape, it becomes necessary to       typically contains multiple items. If
thereby reassigning “Fred Smith” to      organize the mass of ideas, tasks, and    the items are too numerous, the user
“Development,”     as in 8d. Subsump-    information      into categories     or   tends to reline the categories; if they
tion in the category hierarchy insures   classifications so that the project can   are too few, the user tends to ag-
that “Fred Smith” is now assigned to     be understood and managed at an           gregate them. The purpose of this
“Banking Software” and, further, is      appropriate level of abstraction. A       process seems to be to maintain this
removed from “Test Equip.” through       limited amount of detail can be           ratio at a comfortable level. This re-
mutual exclusion.                        handled at a time, and the structure      quirement is in contrast to record-
                                         provides the means to focus on            oriented databases, where the ratio of
Modiiylng         Daeabase               manageable sub-tasks or aggregate         items to categories is typically either
EtrucCure        DynamIcally             the information     into a meaningful     one-to-one-such       as each person
In a conventional       database, the    big picture. The complexity of the        having a unique address-or          very-
tabular structure is fixed in advance    structure is often dictated by the        many-to-one-such       as all people be-
                                                                                                categories. This allows Agenda to

 A-IL *-

   ... +
                    rext conditions
                               Mgpr.   Current   Visr:Person   by   Division   3

                                                                                                modify the views as the database
                                                                                                structure changes. Since categories
                                                                                                are the vocabulary for defining the
                        Divisionbept          .                                                 elements of a view (criteria, sections,
                             zest Icquip.                                                       and columns), the program can re-
                                  Sale8                                                         spond intelligently to changes in the
                           [ Banking      software
                                  Lkmelopment                                                   category hierarchy.
                                  cust.   support                                                   For example, deleting a category
                        lmp1opent         status                                                need not make a view obsolete-it
                                  hourly                                                        may just remove a section or columns
                               c salaried                                                       from a view or modify the selection
                                                                                                criteria. Similarly, simple heuristics
                                                                                                can be used to augment a view when
                                                                                                new categories are added. For exam-
                                                                                                ple, adding a new category as a child
                                                                                                of the “Division/Dept”           category
                                                                                                would add a new section to a view in
                                                                                                which all the existing          divisions
FIGURE       s Refinement of “Full-Time Categ0I-V                                               already appear as sections.
                                                                                                    Perhaps the most complex adjust-
                                                                                                ment required when the category
ing assigned to either male or female.             gory hierarchies, as shown in Fii            hierarchy changes is the reinterpreta-
     We have also found that a single              ure 9. The role of “full-time”        as     tion of queries (criteria). In Agenda,
item tends to be filed in a relatively             binary-valued field in Figure 8c coul        the user does not directly specify a
small number of places. This is why                be easily modified to show the tw            boolean expression in order to select
filing cabinets-where       an item can            new children in the column. Altel            the items that will appear in a view.
be filed at most once--work accept-                natively, “hourly”        and “salaried      Rather,     the user identifies         the
ably, but do not provide the conve-                could be treated as a value of “En           categories of interest,       indicating
nience or flexibility    of a personal             ployment Status” in Figure 8a. C:,           whether items assigned to that cate-
information     manager. There is a course, existing items in the databar ;e gory are to be included or excluded,
definite departure point in power-a                would have to be recategorized fror n        and the system synthesizes an ap-
significant step up in utility-when                “full-time”      into “hourly”        an d propriate boolean expression.
items can be conveniently assigned to              “salaried.” But in the interim, the :Y           This expression is not a simple
multiple categories.                               can remain assigned directly to thte conjunction of the categories indi-
     The evolution of personal data-               non-terminal     “full-time”   category      cated-complex       expressions of logical
base structure often follows a par-                   Two design considerations al10 W          ANDs, ORs, and NOTs often result.
ticular pattern. Users seem to like                Agenda to provide this flexibilit Y. The heuristics for formulating these
regular, normalized      data. As they             First, the physical storage of the iten 1S expressions derive from a simple obser-
discover and refine the structure of               does not depend on the structure (,f         vation: the user would never intend
their data, however, it passes through             the database. Thus, no reformattinl 5        a query that would necessarily result
non-standard states and in fact may                reorganization,     or indexing again iFit in a null set of selected items [4].
be only partially      normalized           in     the physical level is required to implc ‘-       For example, if a group of cate-
steady state.                                      ment a change to the category hiera:         gories are selected that are exclu-
     To address the need for dynamic               thy. Second, no distinction is mad L         sive, i.e., an item can appear in one
modification    of the database struc-             internally between terminal and nor l-       of them, at most, these categories
ture, Agenda category hierarchies                  terminal categories in the hierarch Y will be ORed in the query, since con-
can be modified incrementally with-                Consequently,      an item can be a:S- joining them could never result in a
out making the database, defined                   signed to categories anywhere in thke non-empty set of items. While this
views, or other program structures                 hierarchy, including the root cats:-         approach limits the user’s flexibility
obsolete.                                          gory. To refine the categorization fcx- for specifying arbitrary queries, the
     As an example, suppose a user of a set of items, the user moves thbe query synthesis algorithm                                almost
 the database of Figure 8 decided it items’ assignments further down thle                       always produces the desired result in
 would be handy to know which full-                hierarchy;    to aggregate, the USC       :r practice. Most queries are actually
 time employees are paid on an hourly              moves the items’ assignments up thle quite simple and follow the heuristic
 basis and which are paid a salary.                hierarchy.                                   rules used in the program. When a
 This is easily accomplished by insert-               As noted earlier, most aspect ts category is deleted, queries in the sys-
 ing two child categories into the cate- of a view are defined by selecting                     tem are resynthesized when they are

                                                                                              July 199O/Vo1.33,
next needed, to reflect the change.                  comparison) and comparing           cate-    eoncluslons
                                                     gory hierarchy      structures in the        The development           of Agenda
Inter-Database                                       vicinity of the categories being com-        spanned approximately       two and a
communication                                        pared. When an item is exported from         half years from conception           to
Evolving databases create another                    me database to another, information          scheduled product release. During
challenge. How do you move data                      is carried along with it indicating not      this time, many novel and interesting
from one database to another? In a                   only the specific category assign-           concepts were developed that were
database where the structure is de-                  ments in the source database, but            not ultimately incorporated into the
fined in advance, it is possible                     certain information     about the con-       product and may form the basis for
(though a bit tedious) to write a pro-               text of those categories          in the     future versions or other products.
gram that picks up information from                  lategory hierarchy. In the destination          User and market testing occurred
one database, re-formats it, and                     database, the hierarchy is searched          throughout the final 12 months of
copies it into another database. In                  for matching categories, and if these        this process. The development team
an Agenda database, such a pro-                      are found, the item is assigned to           was often surprised by the range of
gram is in constant danger of be-                    them. If no matching categories are          novel applications     discovered by
coming obsolete, as the category                     found, an optional attempt is made           users. Several of these involved the
hierarchies of the source and desti-                 to locate an appropriate place in the        analysis of existing bodies of text
nation database continue to evolve.                  Sestination category hierarchy for the       or information    that was drawn in
    At the core of this problem is the               orphaned category, and a corre-              real-time    from external    sources.
indeterminacy of the translation [9].                sponding category is created.                These included:
From a linguistic         perspective, an                This approach to inter-database
                                                                                                  l       Tracking events around the world
item and a category assignment form                  communication       serves several pur-
                                                                                                          to help predict the availability and
a declarative proposition: the item                  poses in addition to the immediate
                                                                                                          price of oil.
is the subject, and the category as-                 requirement to move items around.
                                                                                                  l       Automatic     classification   of UP1
signment is a statement about that                   It allows databases with similar, but
                                                                                                          business stories according          to
subject. For example, an item “Write                 not necessarily identical, structures
                                                                                                          qualitative measures (i.e., whether
Agenda press release” may be as-                     to remain related, even though each
                                                                                                          the news was positive, negative,
signed to a category “Fred Smith,”                   may evolve differently over time. For
which has a parent category “Task                    example, many users maintain             a
                                                                                                  l       Managing      takeover rumors at a
assignments,” representing the prop-                  “shadow” database of completed ac-
                                                                                                          brokerage house.
osition that Fred Smith is responsible               tion items as a separate Agenda
                                                                                                  l       Planning office space and moves.
for writing a press release about                    database. By exporting items from
                                                                                                  l       Sharing experience and informa-
Agenda. Each database contains a                     one or more primary databases to
                                                                                                          tion among telephone product
vocabulary       of categories that, in              this database, the structure is in-
                                                                                                          support personnel.
their context, make assertions about                 crementally relined as necessary to
                                                                                                      l   Tabulating survey responses.
the items.                                           reflect changes in the category hierar-
                                                                                                      l   Selecting relevant precedents for
    While items can be transferred                   chies of the primary databases.
                                                                                                          legal cases.
easily, it is difficult to determine how                 A similar benefit accrues to shared
an item’s assignments in one data-                   databases. Members of a work group              Because Agenda         incorporates
base should be interpreted in another                may communicate           by exporting       a variety of heuristic methods for
database. When is a category in one                   items to each other or to a common          taking actions without an explicit
database the same as a category in                    shared database, which is consulted         command (such as automatic as-
another? What if there are multiple                  periodically. This can be used to ac-        signments),     it requires a subtle
categories in the destination database               cumulate changes to track progress           paradigm shift on the part of the user.
that match a category in the source                   on a project, or simply to send             Few computer users are accustomed
database?                                             messages back and forth.                    to programs that interpret intentions
    Our approach to this problem is to                   Transferring     items from one          in addition to simply executing com-
use partial match algorithms that at-                database to another is reminiscent of        mands. We found that people were
tempt to find similarities between the               the transfer      of DNA      in living      quite willing to accept this sharing
structure of two databases. Moving                   organisms. An agent carries with it a        of control, tolerating occasional mis-
data from one database to another                     genetic description that, in a suitably     takes in the program’s judgment
involves assigning the data as simi-                  related environment, can serve to in-       in return for the benefits of im-
larly as possible in the destination                  tegrate the agent into this new             proved personal organization         and
database to its category assignments                  environment.     Under some condi-          effectiveness. The results of cas-
in the source database.                               tions, the invading agent not only          cading program-initiated         actions
    Two basic measures are used in the                becomes integrated into the host, but       often surprise even experienced
partial match process-comparing                       transforms the basic structure of the       users, who continually fine-tune their
the category names (i.e., a string                    host to accommodate it.                      databases to more closely mirror

CCYY”WlCATlCYICFT”E   ACM/July   199O/Vo1.33, No.7                                                                                         115
                                                          their own judgment.                                                              [Information             Systems       Applications]:        Misccl-
                                                              From a user’s perspective, item/                                                 General        Terms:     Human        Factors
                                                          category databases are tools for con-                                                Additional         Key Words and Phrases: Automatic
                                                          trolling complexity. They make it                                                classification,          database       view     update,     flexible
                              staticand gkbal function                                                                                     databases,          personal      information         management,
                               pmtdype #include files
                                                          possible to organize volumes of detail                                           semantic        data models, textual          databases
Instantly recall ycur c
symbd definitions by           and a detailed log file    far greater than manual methods
hitting a hot-kzy              daumentln]?    all your
                               symbds plus where and      allow; quickly shift Derspective in 1                                            About   the Author:
any editod                                                                 *                            1       *

                               how the+      used. Uses   response to immediate        demands;                                            S. JERROLD      KAPLAN      is currently President and
Along with standard
CCBIDER features like          orpanded memay for                                                                                          CEO of GO Corporation.       His research interests are
Scarce view. Edit and         fastsou~e a-de analysis!    centralize    disparate notes, ideas,                                            in hardware    and software     research and develop-
~$~~;;;,?‘&~                  OukkHd/q, tool              reminders, lists, and tasks; and get                                             ment. Author’s    present address: GO Corporation.
                              CCSYM can also create *                                                                                      950 Tower Lane Suite 140, Foster City, CA 94404.
Scarce Browse mode that
                              helpdatabruforycw           the big picture by examining infor- I
lets you walk thru+  all
uslp d * symbd in your        grogrnntwhich   is          mation from multiple viewpoints and                                              MITCHELL          D. KAPOR       is currently      Chairman
                              ccmpattble with                                                                                              of the Board of ON Technology,           Inc. His current
code fu on-line hypcrtuf
cmss-nfmncing ycel
                              Miaosoh’s help sytem,       at varying levels of detail. Personal                                            research interests    are in software design. Author’s
                              QwLwldp. All symbds in
                              ycurappllcatico   a10       information      managers provide a                                              present  address:     ON Technology,          Incorporated,
                                                                                                                                           One Cambridge        Center,  Cambridge,         MA 02142.
FuncUon Prototupcr!           accessible from             means for coping with the hetero-
Build a database d us&d       zk/;$;zz;;;j                                                                                                 EDWARD       J. BELOVE         is currently     Vice President
infamatioo    for your C
                                                          geneous, evolving, free-form infor-
                              lurromaticnllyfnnl your                                                                                      of Corporate     Research and Development               at Lotus
cede with CCSYh4, a
                              original mnme code.         mation      that individuals     must                                            Development       Corporation.        His current       research
paverful bxwm co&
mtalyurwhich      generater                               manage on a continuing basis. (3                                                 interests include multimedia          “sage in productivity
a wealth d useful output,     wd3wJlAwc,*                                                                                                  application,    direct manipulation            user interfaces,
like fully cemented           MtmsoWZ~rlnd          C.
                                                          References                                                                       and new designs for analytical             software.   Author’s
                                 FREE DEMO
                                                            1. Date, CJ. Referential        integrity. In Proceedings                      present address: Lotus Development                Corporation,
                                                          of the &m&h   Inlernationnl     Conference on Very Large                         55 Cambridge        Parkway,     Cambridge,         MA 02142.
                                                          Data Barer (Cannes,       France,      Sept. g-11). IEEE,
                                                                                                                                           RICHARD          A. LANDSMAN             is currently     a System
                                                          New Jersey, 1981, pp. Z-12.
                                                                                                                                           Architect     at Lotus Development           Corporation.        His
                                                             2. Ehrenreich,    S.L. Query    languages:   Design                           current      research    interests      include       workgroup
                                                          recommendations       derived from the human    factors                          technologies      and their applicability        to spreadsheet
                                                          literature.    Human Factors 23, 6 (Dec. 1981).                                  products.     Author’s  present address: Lotus Develop-
                                                             3. Gershman,      A., and Wolf, T. The dynamics    of                         ment Corporation,         55 Cambridge          Parkway,     Cam-
                                                           man-machine      interaction  in a conversational  ad-                                _
                                                                                                                                           bridge, MA 02142.
                                                           visory system. In Pmceedingr ofthe Internalional  Con-
                                                                                                                                           TODD        R. DRAKE          is currently       a Senior Software
                                                          &xce    on Systems, Man, and Cybernelics. IEEE, New
                                                                                                                                           Engineer      at Lotus Development                Corporation.          His
                                                          Jersey, 1986.
Circle #21 on Reader Service                    Card                                                                                       current       research       interests         include        dynamic
                                                            4. Kaplan,  S.J. Cooperative              responses from a por-                hypermedia,          information           retrieval,       intelligent
                                                          table natural language query               system. Artif: In&l. 19,              interfaces,     and the cognitive           and social aspects of
                                                          2 (Oct. 1982), 165-187.                                                          computing.         Author’s         present        address:         Lotus
                                                            5. Kaplan,     SJ. and Davidson,       J.E. Interpreting                       Develop&nt         Corporation,        55 Cambridge            Parkway,
                                                          natural  language      database updates. In Proceedings o/                       Cambridge,        MA 02142.
                                                          the 19th Me&g          of Ihe Association for Computalion
                                                          Lingui&s     (Stanford,     Ca., June).    1981.
                                                            6. Keller, A.M.    Role of semantics                 in translating            Permission      to copy without         fee all or part of this
                                                          view updates.     IEEE    Compul. 19,                 1 uan.      1986),         material    is granted provided        that the copies are not
                                                          63-73.                                                                           made or distributed         for direct commercial         advan-
                                                                                                                                           tage, the ACM copyright           notice and the title ofthe
                                                            7. Larson,       P.A., and Yang,       J.Z.  Computing
                                                                                                                                           publication     and its date appear, and notice is given
                                                          queries from derived relations.        In Proceedings of Ihe
                                                                                                                                           that copying is by permission            ofthe Association     for
                                                          11th Inlernntional     Conference on Very Large Data Barer.
                                                                                                                                           Computing        Machinery.      To copy otherwise,         or to
                                                          IEEE, New Jersey, 1985, pp. 259-269.
                                                                                                                                           republish,    requires a fee and/or specific permission.
                                                           8. Lyngbaek,    P, and McLeod,     D. A personal data
                                                          manatp.      In Proceedings of Ihe 10th International
                                                          Conference on V5y Large Data Bases. IEEE, New Jersey,
                                                          1984, pp. 14-25.
                                                            9. Quine,      WV. Indeterminancy                   of translation
                                                          again.J.    Philos. 84, 1 (1987), 5-10.

                                                           10. Whalen,        T. A linguistic    inference     system
                                                          with mixed-initiative        dialog. In Proceeding5 of the
                                                          North American Fuzzy Information          Processing Society
                                                          (New Orleans,         La.). 1986.

                                                               CR Categories          and Subject          Descriptors:       D.4.3
                                                          [Operating        Systems]:        File Systems Management-
                                                          directory strulures;       distributedfile      systenu; E.5 [Data]:
                                                          Files-orsunitation/slruclure;                  F.2.2 [Analysis           of
                                                          Algorithms           and Problem             Complexity]:           Non-
                                                          numeical       Algorithms         and Problems--pa&m               match-
                                                          ing; H.1.2 [Models           and Principles]:         User/Machine
                                                          Systems-human             informationprocessing;         H.2.1 [Data-
                                                          base      Management]:                 Logical       Design;       H.2.5
                                                          [Information          Storage and Retriwal]:              Information
                                                          Search and Retrieval-clwleri                        queryjormula&m;
                                                          relrieual models; search process; selection process; H.4.m

  Circle #22 on Reader Service                    Card

                                                                                                                                        July 199O/Vol.33,      N~.~ICOMMUNICATIOWS                 OF TM ACM

To top