DualSum: a Topic-Model based approach for update summarization

Proceedings of the 13th Conference of the European Chapter of the Association for Computational Linguistics, pages 214–223, Avignon, France, April 23 - 27 2012. c⃝2012 Association for Computational Linguistics DualSum: a Topic-Model based approach for update summarization Jean-Yves Delort Google Research Brandschenkestrasse 110 8002 Zurich, Switzerland jydelort@google.com Enrique Alfonseca Google Research Brandschenkestrasse 110 8002 Zurich, Switzerland ealfonseca@google.com Abstract Update summarization is a new challenge in multi-document summarization focusing on summarizing a set of recent documents relatively to another set of earlier docu- ments. We present an unsupervised proba- bilistic approach to model novelty in a doc- ument collection and apply it to the genera- tion of update summaries. The new model, called DUALSUM, results in the second or third position in terms of the ROUGE met- rics when tuned for previous TAC competi- tions and tested on TAC-2011, being statis- tically indistinguishable from the winning system. A manual evaluation of the gen- erated summaries shows state-of-the art re- sults for DUALSUM with respect to focus, coherence and overall responsiveness. 1 Introduction Update summarization is the problem of extract- ing and synthesizing novel information in a col- lection of documents with respect to a set of doc- uments assumed to be known by the reader. This problem has received much attention in recent years, as can be observed in the number of partic- ipants to the special track on update summariza- tion organized by DUC and TAC since 2007. The problem is usually formalized as follows: Given two collections A and B, where the documents in A chronologically precede the documents in B, generate a summary of B under the assumption that the user of the summary has already read the documents in A. Extractive techniques are the most common approaches in multi-document summarization. Summaries generated by such techniques consist of sentences extracted from the document collec- tion. Extracts can have coherence and cohesion problems, but they generally offer a good trade- off between linguistic quality and informative- ness. While numerous extractive summarization techniques have been proposed for multi- document summarization (Erkan and Radev, 2004; Radev et al., 2004; Shen and Li, 2010; Li et al., 2011), few techniques have been specifically designed for update summarization. Most exist- ing approaches handle it as a redundancy removal problem, with the goal of producing a summary of collection B that is as dissimilar as possible from either collection A or from a summary of collec- tion A. A problem with this approach is that it can easily classify as redundant sentences in which novel information is mixed with existing informa- tion (from collection A). Furthermore, while this approach can identify sentences that contain novel information, it cannot model explicitly what the novel information is. Recently, Bayesian models have successfully been applied to multi-document summarization showing state-of-the-art results in summarization competitions (Haghighi and Vanderwende, 2009; Jin et al., 2010). These approaches offer clear and rigorous probabilistic interpretations that many other techniques lack. Furthermore, they have the advantage of operating in unsupervised settings, which can …

Content truncated to 500 words