Laval University and Lakehead University at TREC Dynamic Domain 2015: Combination of Techniques for Subtopics Coverage

This paper describes the joint submission by Laval University and Lakehead University to the TREC 2015 Dynamic Domain track. We submitted four runs for the main track and one run for the judged-only track. In our view, the Dynamic Domain process can be separated into two phases: a traditional static information retrieval phase to generate an initial set of documents, followed by a dynamic information retrieval phase which takes user feedback into account to improve the results. We developed an algorithm that combines keyword search, Latent Dirichlet Allocation, and K-Means clustering to take advantage of each technique to generate the best subset of results for the user. Our systems performed mostly over the median of the group of participants and our system for the “judged-only” task had the best score for a wide range of queries.