Predicting task difficulty for different task types


  • Copyright is held by the author/owner(s).


This paper reports our investigation of differences in users' behavior between difficult and easy search tasks, as well as how these differences vary with different types of tasks. We also report how behavioral predictors of task difficulty vary across task types. In addition, we explored how whole-task-session level user behaviors and within-task-session level behaviors differ in task difficulty prediction. Data were collected in a controlled lab experiment with 48 participants, each completing 6 search tasks of three types: single-fact finding, multiple-fact finding and multiple-piece information gathering. Results show that task type affects the relationships between task difficulty and user behaviors and that prediction of task difficulty should take account of task type. Results also show that both whole-session level and within-session level user behaviors can serve as task difficulty predictors. Whole-session level variables show higher prediction accuracy, but within-session level factors have the advantage of enabling real-time prediction. These findings can help search systems predict task difficulty and adapt to users.