Using paraphrases for improving first story detection in news and Twitter

2012 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 338–346, Montr´eal, Canada, June 3-8, 2012. c⃝2012 Association for Computational Linguistics Using paraphrases for improving first story detection in news and Twitter Saˇsa Petrovi´c School of Informatics University of Edinburgh sasa.petrovic@ed.ac.uk Miles Osborne School of Informatics University of Edinburgh miles@inf.ed.ac.uk Victor Lavrenko School of Informatics University of Edinburgh vlavrenk@inf.ed.ac.uk Abstract First story detection (FSD) involves identify- ing first stories about events from a continuous stream of documents. A major problem in this task is the high degree of lexical variation in documents which makes it very difficult to de- tect stories that talk about the same event but expressed using different words. We suggest using paraphrases to alleviate this problem, making this the first work to use paraphrases for FSD. We show a novel way of integrat- ing paraphrases with locality sensitive hashing (LSH) in order to obtain an efficient FSD sys- tem that can scale to very large datasets. Our system achieves state-of-the-art results on the first story detection task, beating both the best supervised and unsupervised systems. To test our approach on large data, we construct a cor- pus of events for Twitter, consisting of 50 mil- lion documents, and show that paraphrasing is also beneficial in this domain. 1 Introduction First story detection (FSD), sometimes also called new event detection (NED), is the task of detecting the first story about a new event from a stream of documents. It began as one of the tasks in Topic Detection and Tracking (TDT) (Allan, 2002) where the overall goal of the project was to improve tech- nologies related to event-based information organi- zation tasks. Of the five TDT tasks, first story de- tection is considered the most difficult one (Allan et al., 2000a). A good FSD system would be very useful for business or intelligence analysts where timely discovery of events is crucial. With the sig- nificant increase in the amount of information being produced and consumed every day, a crucial require- ment for a modern FSD system to be useful is effi- ciency. This means that the system should be able to work in a streaming setting where documents are constantly coming in at a high rate, while still pro- ducing good results. While previous work has ad- dressed the efficiency (Petrovi´c et al., 2010) aspect, there has been little work on improving FSD perfor- mance in the past few years. A major obstacle is the high degree of lexical variation in documents that cover the same event. Here we address this problem, while keeping in mind the efficiency constraints. The problem of lexical variation plagues many IR and NLP tasks, and one way it has been addressed in the past is through the use of paraphrases. Para- phrases are alternative ways of expressing the same meaning in the same language. For example, the phrase he got married can be paraphrased as he tied the knot. Paraphrases were already shown to …

Content truncated to 500 words