University of Waterloo at the TREC 2013 Temporal Summarization Track.

Gaurav Baruah, Rakesh Guttikonda, Adam Roegiest, Olga Vechtomova

Abstract: The University of Waterloo participated in the Temporal Summarization Track at TREC 2013 and submitted 8 runs for the Sequential Update Summarization Task. Methods like query likelihood ranking, pseudo relevance feedback, BM25 and cosine similarity, as well as, algorithms for passage retrieval and term expansion using distributional similarity to a set of seed words, were used for returning relevant sentences from a stream of time-ordered documents. Higher scores relative to the average score for all submitted runs were achieved on the Latency Comprehensiveness Metric (returning as many nuggets as possible), however, submitted runs performed poorly on the Expected Latency Gain Metric (speediness of updates).