Querying, Navigating and Visualizing a Digital Library Catalog

ABSTRACT

• Using a visualization scheme, the interface provides visual feedback to the user about how the query words influence the ranking of retrieved documents.
• By simple drag-and-drop operations of objects on the screen, the interface facilitates a naive end-user in constructing complex structured queries and in providing relevance feedback.
• To suit the evolving information needs of the user, the interface supports navigational features such as browsing documents by specific authors and browsing the Table of Contents of publications.
• The interface integrates an online thesaurus which provides words related to the query that can be used by the user to expand the original query.

KEYWORDS: Visualization of results, visual query languages, query processing, information retrieval

WALK-THROUGH OF A TYPICAL USER SESSION

Figure 1. Sample querying session. The window titled ``Positive Objects'' is colored green and the window titled ``Negative Objects'' is colored red. All ``incantations'' of an object in the display are colored green/red whenever it is classified as positive/negative.

• The user types in his/her free form textual query in the query window. In the example shown in figure 1, the query is ``ozone depletion and melanoma''
• As every query word is typed in, the system consults an on-line thesaurus and displays words and phrases related to the query word in an adjacent window.
• At any point during the session the user can ``drag-and-drop'' (using the mouse) any of the related words/phrases into the positive and negative windows. Internally the system expands the query by treating the positive words/phrases as synonyms of the corresponding query word. The negative words/phrases are included in the query with a NOT operator. For example, if for a query word ``bank'', the phrase ``financial institution'' is classified as positive and ``river bed'' is classified as negative, the corresponding internal query would be ``#SYNONYM( bank #2( financial institution )) #NOT( 2 river bed))''. The interface facilitates construction of such structured queries by simple ``drag-and-drop'' operations of the mouse. In the example in figure 1, a phrase, namely ``skin cancer'' that is related to the query word ``melanoma'' has been classified as positive. Internally the systems treats the phrase as a synonym of ``melanoma''.
• After the user types in the query, the system evaluates the query and displays the titles of top-ranked documents in the ``Query Results'' window.
• The user examines the query result. Clicking any title with the mouse will bring up the full document.

Figure 2. Visualization of results for the base query.

• Figure 2 is a visualization of the query results for the base query ``ozone depletion and melanoma''. The leftmost column of bars corresponds to the top-ranked document, with the columns progressing to the right representing progressively lesser ranked documents. We can see that almost all of the 150 documents were retrieved because they contained the query words ``ozone'' and ``depletion''. Only 15 of the top 150 documents have anything to do with melanoma. Further, of those 15 documents, only one discusses ozone (the top-ranked document -- leftmost column in Figure 2.) Thus we can clearly see that either there are not many documents dealing with melanoma and ozone or the ozone-layer concept drowns out melanoma during retrieval.
• The user can classify any document as being relevant or non-relevant by ``drag-and-drop''ping the document into positive and negative windows. In the example in figure 1, the user has classified two documents titled ``CFC-free integral skin foams for steering wheels.'' and ``Video comparator system for early detection of cutaneous malignant melanoma'' as positive. The document titled ``Symposium on chemistry of the Atmosphere'' has been classified as negative.
• The user can also highlight a portion of a document and ``drag-and-drop'' that portion into the positive and negative windows. The words in the highlighted document portion are used to expand the query in the next iteration.
• During the next iteration, the reformulated query with the feedback information is processed by the system resulting in an improved ranking of documents.

Figure 3. Visualization of results for query with feedback information.

• Figure 3 is a visualization of the results of the revised query (i.e., the query with relevance feedback information). The figure shows that there are four documents dealing with melanoma and ozone. (Note that the documents which deal with melanoma and it's synonym skin cancer are displayed in the same histogram titled ``melanoma'', since melanoma and skin cancer represent the same query concept). Thus there are three additional documents retrieved due to the effect of classifying the phrase ``skin cancer'' as a synonym of ``melanoma''. But still there are not many documents about melanoma compared to ozone depletion. Our experience with this visualization scheme has shown it to be a useful tool for identifying different facets of the query, as in this case, the facets are melanoma and ozone.
• Using any document as a starting point, the user can browse through the list of other articles in the same journal issue or conference proceedings with a help of a Table-of-Contents which is generated automatically. This is useful in many cases such as when the user comes across a special-issue of a journal devoted to the search topic.
• The user can also browse through the list of articles written by the same author. For example, an author who has written an article about the effects of ozone layer depletion on skin cancer has probably authored more articles along the same lines, and the user might want to see them.

CONCLUSION & FUTURE WORK

The interface also supports a visualization scheme which illustrates how the query results are related to the query words. Visualizing the results of the query keeps the user more informed on how the system computed the ranking of documents. With this information, the user is better equipped to reformulate the query for the next iteration. The interface also has facilities to browse the Table of Contents of publications and to browse the list of articles written by a specific author. It is our opinion that integrating all of the above features in a seamless interface leads to an interplay between different items that is much more beneficial than the sum of the individual items in isolation.

We are in the final stages of implementation, and in future, we intend to test the effectiveness of the interface by conducting studies on how library users, experts looking for detailed information as well as naive users, interact with the interface and how they react to ranked output systems as opposed to existing boolean systems. We plan to include a domain-specific thesaurus for the engineering domain from Compendex and a collection-specific word-association thesaurus if possible.

ACKNOWLEDGEMENTS

REFERENCES

[Miller90a] George A. Miller, Richard Beckwith, Christiane Fellbaum, Derek Gross, and Katherine J. Miller. Introduction to WordNet: An on-line lexical database. Journal of Lexicography, 3(4):235--244, 1990.

[Ousterhout] John K. Ousterhout. Tcl and the Tk Toolkit. Addison-Wesley, 1994.

[VHN95] A. Veerasamy, S. Navathe, and S. Hudson. Visual interface for textual information retrieval systems. To appear in Proceedings of the Third Conference on Visual Database Systems. IFIP 2.6, 1995.