Melboune at the TREC 2011 Legal Track

The Melbourne team was a collaboration of the University of Melbourne, RMIT University, and the Victorian Society for Computers and the Law. The approach taken was to train a support vector machine based upon textual features using active learning. Two sources of relevance annotations were used for different runs: the official annotations, provided by the topic authorities; and annotations provided by a member of the Melbourne team with e-discovery experience (though not legal training). We describe the SVM method used in Section 1.1, the run using official annotations in Section 1.2, and the run using the internal annotations in Section 1.3.