Docstoc

zheng_iconference

Document Sample
zheng_iconference Powered By Docstoc
					                                     Exploring Hidden Connections among Historic Images
                                                                                                      Wu Zheng
                                                                                              wuzheng2@illinois.edu
                                                                                Graduate School of Library and Information Science
                                                                                    University of Illinois at Urbana-Champaign

                        ABSTRACT                                                                                                                                 Image 5: George Frederick                      Image 6: James Murray
Among historical images, there are numerous implicit pair-wise                                                                                                     Frye, Seattle, ca. 1888                     Colman, Seattle, ca. 1885
                                                                                              EXAMPLES OF FINDINGS
connections that cannot be found using traditional keyword
searches. Two seemingly unrelated images, when put together,
might reveal some interesting patterns that were overlooked before.
The background information in the text description provides clues to
these connections. An experiment was conducted to explore the             Image 1: Ladies' Aid posing                                Image 2: Juanita,
metadata of historical images using Literature-based Discovery            outside Grace Presbyterian                               Eleanor, and E. Martin                                           Henry
(LBD) methods. Several implicit relations among the images were              Church, Seattle, 1922                                Lewis, Seattle, ca. 1930                                          Yesler
identified. The result of the experiment shows that LBD can help
users better navigate and use historical image records.

                                                                                                                      Grace
                             DATA                                                                                  Presbyterian                                  Born in Germany, George                        …… In 1872 he moved to
The data used for experiment was selected from the Institute of                                                      Church
                                     Frederick Frye came to Seattle                 Seattle, where he operated
Museum and Library Services (IMLS) Digital Collection and Content                                                                                                in 1853, where he worked with                  Henry Yesler's mill. ……
(DCC) project. A single collection, the King County Snapshots, was                                                                                               Henry Yesler at his lumber mill.
chosen because of the high quality metadata provided. This is a           …… Ladies' Aid posing outside                                                          ……
historical image collection with more than 12,000 items portraying        Grace Presbyterian Church ……
people, places and events in the county.                                  The Ladies' Aid was formed in                           …… He was ordained by the                                   DISCUSSION
                                                                          1912 for the purpose of raising                         Presbytery of Seattle and
                                                                          funds for the church. ……                                                              Influence of Metadata Quality
                          METHOD                                                                                                  served as minister of Grace
                                                                                                                                                                The method used in the experiment relied on the human-assigned
                                                                                                                                  Presbyterian Church from
1. A-term identification. For each record, the subject of the image                                                                                             metadata to explore the implicit connections, thus the performance of the
                                                                                                                                  1922 to 1929. ……
(A-term) was extracted from the title field.                                                                                                                    method would be influenced by the quality of the metadata. For images
2. B-term identification. The title and description field of the record                                                                                         with little or no description, especially background information on the
were processed using the Stanford Named Entity Recognizer. All                                                                                                  subject, the connections are difficult to identified.
the named entities (person, organization and location names)
captured were potential B-terms that connected the A-record with                                                                                                Evaluation of Relatedness
another set of C-records. C-records contained images that had               Image 3: Anna Louise Strong,                       Image 4: John Henry              The relatedness in social science and humanities is very subjective, and
implicit relations to the topic represented by A-records.                                                                                                       depends heavily on the background and information need of the users.
                                                                              possibly in Seattle, 1939                       Smith, Seattle, ca. 1921
3. Searching for C-records. New searches were conducted for                                                                                                     How to automatically evaluate and rank the connections will be focus of
each B-term identified. For each record retrieved, if it was not in the                                                                                         further research.
set of A-records, and the subject of the image in it was not the B-
term, the record was a C-record. The image in an A-record had                                                                                                   Application on Multiple Collections
possible implicit connection with the image in C-record.                                                                                                        This experiment focused on a single collection. Applying the method on
4. Evaluation. Manual evaluations were performed to ensure that                                                                                                 multiple collections should reveal more interesting connections. On the
the implicit connections identified were meaningful.                                                              I.W.W.                                        other hand, with the increase in data size and complexity, more noise are
                                                                                                                                                                also expected. How to filter out trivial connections while preserving the
                                                                                                                                                                useful ones will be another important problem to solve.
           Image
A
                               Image
C

                                                                                                                                                                       ACKNOWLEDGEMENT AND REFERENCE
                                                                                                                                                                This poster reflects topics discussed in the Literature Based Discovery
                                                                          …… She was elected to the Seattle
                                                                                                                             …… A hod carrier by trade,         course given at GSLIS in Fall 2008. The author is grateful to Vetle Torvik and
                          Connec&ng
Term
B1
                              School Board but later was recalled
                                                                                                                             he became active in I.W.W.         Qin Wei for their review of this paper and the valuable suggestions.
                                                                          because of her socialist politics and
        Descrip(on
A
                           Descrip(on
C
                                                                circles about 1913. ……
                                                                          her support of the Industrial
                          Connec&ng
Term
B2
                                                                                                                    Ins&tute
of
Museum
and
Library
Services
Digital
Collec&on
and
Content.
Retrieved

                                                                          Workers of the World (I.W.W.). ……
                                                                                                                                                                December
5,
2008,
from
hGp://imlsdcc.grainger.uiuc.edu/.


                                                                                                                                                                King
county
snapshots:
a
photographic
heritage
of
SeaGle
and
surrounding

                          Connec&ng
Term
B3
                                                                                                                    communi&es.
Retrieved
December
5,
2008,
from

                                                                                                                                                                hGp://imlsdcc.grainger.uiuc.edu/collec&ons/FullDisplay.asp?cid=2384.

                                                                                                                                                                Stanford
Named
En&ty
Recognizer.
Retrieved
December
5,
2008
from

                                                                                                                                                                hGp://nlp.stanford.edu/ner/index.shtml.



				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:1
posted:1/23/2012
language:
pages:1