Modeling Decision Points in User Search Behavior

Understanding and modeling user behavior is critical to designing search systems: it allows us to drive batch evaluations, predict how users would respond to changes in systems or interfaces, and suggest ideas for improvement. In this work we present a comprehensive model of the interactions between a searcher and a search engine, and the decisions users make in these interactions. The model is designed to deal only with observable phenomena. Based on data from a user-study, we are therefore able to make initial estimates of the probabilities associated with various decision points.

More sophisticated estimates of these decision points could include probabilities conditioned on some amount of search activity state. In particular, we suggest that one important part of this state is the amount of utility a user is seeking, and how much of this they have collected so far. We propose an experiment to test this, and to elucidate other factors which influence user actions.