The information seeking community seems to be experiencing a renewed interest in session-based approaches to information seeking after years of focusing on single-query interaction. Yesterday at CHI2010, Anne Aula, Rehan M. Khan, and Zhiwei Guan reported on a study of searchers’ behavior. Rather than looking at single-query performance, they analyzed searchers’ query reformulation tactics to characterize difficult vs. easy search tasks. They found a variety of indicators that correlate with users’ difficulties in articulating efficient queries. This work is important as it hints at the possibility that web search engines can diagnose user behavior and alter the interaction with users to facilitate their search process.
In some cases, query performance can be improved automatically by combining queries form a single session to identify potentially useful documents. In a recent article, for example, Kalervo Järvelin reported that fusing the results of several poorly-performing queries produced dramatic improvements in query effectiveness. Queries were combined using systematic strategies (similar to the kind described in Aula et al’s paper) to produce significant improvements in precision.
Finally, a paper by Jones and Klinkner published at CIKM 2008 showed that query logs could be mined to generate session groupings, including situations with overlapping and hierarchical units, with an accuracy of up to 92%.
The combination of these findings implies that it should be possible for search engines to organize users’ queries in topical sessions, to detect situations when search is not going well, and to offer alternative ranking of documents that may improve the quality of results even from otherwise ineffective queries. While this is, as this time, a hypothetical combination of features that still needs to pass a variety of engineering hurdles, there is reason to believe that these automatic processes, when taken in combination, can improve the user experience.
Next steps to look forward to should include connecting search episodes over time to build longer-lasting sessions with evolving information needs. One way to proceed in this direction was suggested by another CHI 2010 paper that modeled a user’s evolving information need in a programming task. It’ll be interesting to see if some of these ideas can be applied to the less-constrained space of document retrieval.