The Melbourne Team at the TREC 2010 Legal Track

The Melbourne team was a collaboration between academic and industry groups. The team participated in both the learning and the interactive tasks of this year’s Legal Track. The baseline run for the learning track employed true-relevance feedback, achieving respectable outcomes; the experimental runs added additional features and employed an SVM classifier, with poor results. The techniques developed for the learning task were then deployed in the interactive task. The classifier again achieved poor predictive quality, although final results place our production (non-significantly) first. We describe the learning task efforts in Section 2, and the interactive task in Section 3.