Webis at TREC 2014: Web, Session, and Contextual Suggestion Tracks.

摘要

In this paper we give a brief overview of the Webis group s participation in the TREC 2014 Web, Session and Contextual Suggestion tracks. All our runs for the Web and the Session track are on the full ClueWeb12 and use the online Indri retrieval system hosted at CMU. Our runs for the Contextual Suggestion track are based on the open web. As for the Web track, our runs are aimed at one research question: whether using axioms for re-ranking a baseline result list improve the retrieval performance. Therefore, we implement the axioms available in the axiomatic IR literature and combine them with new axioms aimed at term proximity. Trained on the TREC 2013 Web track data, three promising combinations of axioms are identified in a large-scale experiment and used for our three runs. As for the session track, we tackle three research questions in three different runs. First, similar to the Web track, we examine whether an axiom combination helps to improve session retrieval. Second, we examine the effect of presenting relevant documents from previous years when they seem to be related to the current queries of the 2014 data. Our third question is whether the user interactions can be used to train an activation model to predict relevant documents for new queries.