Experiments with the Negotiated Boolean Queries of the TREC 2007 Legal Discovery Track

We analyze the results of several experimental runs submitted for the TREC 2007 Legal Track (also sometimes known as the Legal Discovery Track). We submitted 4 boolean query runs (the initial proposal by the defendant, the rejoinder by the plaintiff, the final negotiated query, and a variation of the final query which had proximity distances doubled). We submitted 2 vector query runs (one based on the keywords of the final negotiated query, and another based on the (natural language) request text). We submitted a blind feedback run based on the final negotiated boolean query. Finally, we submitted a fusion run of the final boolean, request text and final vector runs. We found that none of the runs had a higher mean estimated Recall@B than the original final negotiated boolean query.