Rutgers at the TREC 2012 Session Track.

摘要

At Rutgers, we approached the Session Track task as an issue of personalization, based on both the behaviors exhibited by the searcher during the course of an information seeking episode, and a classification of the task that led the person to engage in information-seeking behavior. Our general approach is described in detail at the Web site of our project (http://comminfo. rutgers.edu/imls/poodle) and in the papers available there. In the TREC 2011 Session Track, we tested preliminary results of predictive models of document usefulness using recursive partitioning models learned from user studies of task session information behaviors. In this year s TREC Session Track, we tested predictive models of document usefulness based on user behaviors by using logistic regression. This was combined with predictive models of task type derived from a multinomial logistic regression model learned from the 2012 Session Track data.