Topic-independent modeling of user knowledge in informational search sessions

Abstract Web search is among the most frequent online activities. In this context, widespread informational queries entail user intentions to obtain knowledge with respect a particular topic or domain. To serve learning needs better, recent research in field of interactive information retrieval has advocated importance moving beyond relevance ranking results and considering user’s state within oriented sessions. Prior work investigated use supervised models predict gain from interactions during session. However, characteristics resources that interacts have neither been sufficiently explored, nor exploited task. work, we introduce novel set resource-centric features demonstrate their capacity significantly improve for task predicting users We make important contributions, given reliable training data such tasks sparse costly obtain. various feature selection strategies geared towards selecting limited subset effective generalizable features.