Interactive reformulation of long queries
We present new ways of interacting with a user based on query analysis and reformulation. Our goal is to not only improve retrieval performance but also help the user understand the retrieval process and collection she is searching. We do this by providing users information reflecting the potential impact their decisions will have on the retrieval process. This way, users can make more informed choices from the options presented to them by the retrieval system.
Unlike most previous work in user interaction where a one-procedure-fits-all strategy was pursued, user interaction must be invoked only when there is potential for improvement. This is important as tedious user interaction can have an unfavorable impact on user experience. We present techniques for selective user interaction and show their utility in the context of two interaction techniques we have developed. Our results show that user interaction can be avoided in a vast number of cases without much deterioration in performance.
User interaction can be made more productive by providing users with an optimally-sized set of high quality options. We present efficient techniques to determine such a set. When faced with a decision to interact with a user given a particular query, it is beneficial to determine the best interaction technique suited for that query. We solve this problem by obtaining implicit feedback from the user. By utilizing all the interaction-related techniques described in this thesis, we show through simulations and user studies that users can obtain better performance with less effort.