Visualizing web search results in 3d - Computer

Document Sample
Visualizing web search results in 3d - Computer Powered By Docstoc
					                                                                                     WEB TECHNOLOGIES

Visualizing Web
                                                                                         tion facilitates user navigation of
                                                                                         search results as well as interaction
                                                                                         with contents. For instance, translat-
                                                                                         ing and rotating objects in the 3D

Search Results
                                                                                         space might clarify their meaning,
                                                                                         while changing the viewpoint might
                                                                                         improve perception and understand-
                                                                                         ing of data. Spatial visualization also

in 3D
                                                                                         can represent information as various
                                                                                         shapes, colors, textures, positions,
                                                                                         sizes, orientations, and even behavior;
                                                                                         the only limiting factor in a 3D space
                                                                                         is user perception.
Wojciech Cellary, Wojciech Wiza, and Krzysztof Walczak
Poznan University of Economics                                                           ADAPTIVE VISUALIZATION
                                                                                           When the amount of presented data
                                                                                         differs significantly from one query to

          he exponential growth in Web
          sites is making it increasingly                                           Periscope presents a
          difficult to extract useful
          information on the Internet                                               holistic view of data
          using existing search engines.                                            related to a search
Despite a wide range of sophisticated
indexing and data retrieval features,                                               query.
search engines often deliver satisfac-
tory results only when users know
precisely what they are looking for.         documents. A user can interact with         another, Periscope’s user interface
   Traditional textual interfaces present    the scene to better understand the          adapts to the volume and characteris-
results as a list of links to Web pages.     nature of the search result and eventu-     tics of search results. The system per-
Because most users are unwilling to          ally refine the query.                       forms this adaptive visualization both
explore an extensive list, search engines       Periscope automatically selects the      externally, by selecting an appropriate
arbitrarily reduce the number of links       3D scene that best visualizes the result.   3D model from a number of available
returned, aiming also to provide quick       If the engine finds a large number of       ones, and internally, by applying the
response times. Moreover, their pro-         documents, Periscope classifies the         selected model’s visualization dimen-
prietary ranking algorithms often do         entire result according to one or more      sions to particular search result attrib-
not reflect individual user preferences.      criteria—for example, document lan-         utes.
   Those who need comprehensive gen-         guages, server hosts or domains, num-          To maximize 3D scene readability,
eral information about a topic or have       ber of matching keywords, or docu-          Periscope selects a visualization model
vague initial requirements instead want      ment types. A user can thus immedi-         that ensures that
a holistic presentation of data related to   ately focus on classes of documents
their queries. To address this need, we      rather than on individual documents.          • the user has a viewpoint to observe
have developed Periscope (periscope.            Automatic scene selection is based on        all presented objects,, a 3D search result        search result properties such as number       • the size of glyphs allows easy
visualization system that displays all the   of documents, document languages, file           manipulation and interaction,
Web pages found in a synthetic, yet          types, links to and from other docu-          • occlusions in the 3D scene do not
comprehensible format.                       ments, and matching keywords in each            prevent a user from interacting
                                             document title or body; the position of         with every glyph,
PERISCOPE OVERVIEW                           keywords within documents; and the            • the user can distinguish the dis-
  Periscope presents a comprehensive         most frequent phrases. User preferences         tance between subsequent visual-
view of the entire search result as a        and learned user behavior also can              ization dimension values, and
three-dimensional scene consisting of        influence the selection process.               • visible reference elements permit
virtual objects, called glyphs, that cor-       By reflecting the way humans per-            a user to properly interpret dis-
respond to documents or classes of           ceive the world, 3D spatial visualiza-          played data.

                                                                                                              May 2004               87
     W e b Te c h n o l o g i e s

                                                                                       Visualization models

                                         Search engine
                                             query                                                              User query

                                         Search engine                                                          Visualized
                                           response                                                              results
                                                                    Periscope visualization engine
           Indexing search engine                                                                                                 User

     Figure 1. Adaptive visualization in Periscope. The system selects an appropriate 3D visualization model according to an algorithm
     contained in a metamodel and applies the model’s visualization dimensions to particular search result attributes.

        Figure 1 illustrates adaptive visual-       The determinant consists of a size—the              For each triple, Periscope calculates
     ization in Periscope. After a user sub-        maximum number of glyphs in a                    a coefficient that expresses how well a
     mits a query, the system translates the        model—and a set of facets. Each facet            visualization model matches an aggre-
     query into a search engine’s proprietary       specifies a single visualization dimen-           gation table using a particular mapping.
     format and submits it for processing.          sion—a particular common property                If two or more triples have the same
     After retrieving the search result,            of all glyphs—and its capacity, the              maximum value of the coefficient, the
     Periscope selects a visualization model        maximum number of distinguishable                system arbitrarily selects one or lets the
     according to an algorithm contained in         values that may be assigned to the visu-         user decide. If no triple fulfills the
     a metamodel. The algorithm finds the            alization dimension in a particular              required conditions, Periscope modifies
     best match between quantitative prop-          visualization model.                             each table to reduce the number of clas-
     erties of a categorized search result and         The visualization engine’s logic tries        sification subdomains in the attribute
     visualization capabilities of available        to find the best match among all pos-             that has the most subdomains. For
     visualization models.                          sible triples consisting of an aggrega-          example, after one such grouping, the
        The quantitative properties of a            tion table, a visualization model, and           attribute “Document language” could
     search result are expressed by a set of        a mapping between aggregation table              consist of the values {English, Italian,
     multidimensional aggregation tables.           attributes and the model’s visualization         Polish, German, Other Languages}. For
     Each aggregation table is built on a dif-      dimensions. The engine matches each              modified tables, Periscope performs the
     ferent subset of document attributes           table against determinants of all avail-         matching procedure again.
     derived from the specific search               able visualization models, taking into              After selecting a visualization model,
     engine’s database. Each field of the           account that                                     Periscope instantiates it with data from
     aggregation table contains the set of                                                           the search result and displays it.
     documents that conforms to classifi-                • the number of interface facets
     cation constraints defined over each                  must be greater than or equal to          VISUALIZATION MODELS
     attribute’s domain. The system initially              the number of attributes forming            Each Periscope visualization model
     partitions quasicontinuous domains,                   the aggregation table;                    consists of three parts: a dynamic
     such as modification/creation date or                • the size of the model must be             model of the 3D scenes visualizing
     document size, into predefined subdo-                  greater than or equal to the total        search results, a model defining the 2D
     mains—for example, document sizes                     number of fields in the aggrega-          user interface controls, and the actions
     0-100 bytes, 100 bytes-0.5 Kbyte, 0.5-                tion table; and                           associated with objects in the 2D and
     1 Kbyte, and so on. Partitioning based              • a mapping must exist, such that           3D parts of the interface.
     on fixed ranges delivers richer seman-                 each visualization dimension                To focus on selected attributes of the
     tic information than dynamic subdo-                   capacity is greater than or equal to      retrieved documents, the user can
     mains containing a balanced number                    the number of different values or         change the attribute mapping to the
     of documents.                                         ranges used for classification of the      model’s visualization dimensions. For
        A visualization model’s capabilities               aggregation table attribute mapped        example, in the Periscope model in
     are expressed by a model determinant.                 onto this visualization dimension.        Figure 2, each glyph represents a sin-

88                Computer
                                                                                                        entire search result to precise visualiza-
                                                                                                        tion of information about particular
                                                                                                        documents of interest. Ongoing project
                                                                                                        work focuses on developing advanced
                                                                                                        visualization models based on analysis
                                                                                                        of existing models’ performance,
                                                                                                        including their ergonomics and the
                                                                                                        accuracy of applied visualization
                                                                                                        metaphors. We plan to test Periscope
                                                                                                        in a real scenario by connecting it to
                                                                                                        large search engine databases.I

                                                                                                        Wojciech Cellary heads the Depart-
                                                                                                        ment of Information Technology at the
                                                                                                        Poznan University of Economics. Con-
                                                                                                        tact him at

                                                                                                        Wojciech Wiza is a researcher in the
                                                                                                        Department of Information Technol-
                                                                                                        ogy at the Poznan University of Eco-
                                                                                                        nomics. Contact him at
Figure 2. Sample Periscope visualization model. Each glyph represents a single document,
and the user can assign different attributes to glyph colors, shapes, horizontal positions,
and levels. Here, the levels correspond to servers hosting documents: The user can
                                                                                                        Krzysztof Walczak is an assistant pro-
proceed to a specific level, switch the level indicators on and off, and visualize only
                                                                                                        fessor in the Department of Informa-
documents on a particular level.
                                                                                                        tion Technology at the Poznan Uni-
                                                                                                        versity of Economics. Contact him at
gle document. The user can assign                   y using multiple levels of abstrac-       
different attributes to glyph colors,
shapes, horizontal positions, and lev-
                                               B    tion and applying the most appro-
                                                    priate 3D environment at each
                                               step of the search process, Periscope
                                                                                                         Editor: Sumi Helal, Computer and Informa-
                                                                                                         tion Science and Engineering Dept., Univer-
   Additional interactive elements sup-        greatly improves searching efficiency.                     sity of Florida, P.O. Box 116125, Gainesville,
port exploration of search results. In         Users can navigate from a high-level                      FL 32611-6120;
Figure 2, the levels correspond to             aggregated and categorized view of the
servers hosting documents; the user
can proceed to a specific level, switch
the level indicators on and off, and
visualize only documents on a particu-
                                                                        I E E E
                                                                    C o m p u t e r
lar level.
   A user who is not satisfied with an
automatically selected visualization
model can select another model man-
ually. In addition, a user can reformu-
                                                                     S o c i e t y
                                                                     m e m b e r s                         25                           %
                                                                     Join online today!

late queries by either specifying new
                                                                                                 on all conferences
                                                    Not a member?

keywords or changing query con-
straints. Such query refinement can be
achieved by interacting with 2D inter-                                                           sponsored by the
face controls or with 3D scene objects.
The parallel processes of interface                                                           IEEE Computer Society
adaptation and query refinement effi-
ciently narrow results to a manageable
set of documents, even for complex                                                        w w w. c o m p u t e r. o r g / j o i n

                                                                                                                                  May 2004                89

Shared By:
Description: Visualizing web search results in 3d - Computer