Millions of users of information retrieval systems each day seek useful items, but many are frustrated with the results received. Often, better results can be found when a query is split into parts that cover different but connected aspects of the information need. Sometimes a small set of (2-3) documents is needed as an answer instead of a single document.
This project researches an alternative interpretation of user queries and presentation of the results. It has the potential to improve retrieval results whenever there is a mismatch between a user's understanding of the collection and the actual collection content. A probabilistic retrieval scheme is employed with: a framework based on belief networks which can combine multiple sources of evidence; and user feedback at document, cluster (group), and relationship (e.g., citation or hypertext link) levels.
Query results in this interpretation are networks of document groups representing topics, each group relating and connected to other groups in the network that partially answer the user's information need. New and more effective representations and techniques help users visualize these results. The user is part of the retrieval process, and can manipulate the network of content. The system can provide deeper support of the user's need, in a way that goes well beyond traditional relevance feedback.
This project will involve scores of students (including those in minority and under-represented groups) in information retrieval studies, leading to several world-accessible electronic theses and dissertations, and improved search methods. Because of this research, scientists and the public in general should be able to more easily search for answers to complex questions. Software and collections developed will be available to other researchers. Results will be shared at workshops and conferences and with companies such as AOL, and made available at the project's Web site (http://fox.cs.vt.edu/SSP/).
