Cluster-Based Relevance Feedback: Legal Track 2011

This is our second participation in the TREC Legal Track. The TREC Legal Track 2011 featured only the Learning Task. We participated in Topics 401 and 403. We used Lemur 4.11 for Boolean retrieval and followed it with a clustering technique, where we chose members from each cluster (which we called seeds) for relevance judgement by the TA and assumed all other members of the cluster whose seeds are assessed as relevant to be relevant. Based on the relevance information from seeds and their clusters, we applied Rocchio relevance feedback technique implemented in Terrier 3.0. Then, we used the feedback terms for the expansion of both the text queries and the Boolean queries. Finally, we used Z-fusion[4], a data fusion technique, on two of our runs.