Summarization of Historical Articles Using Temporal Event Clustering

2012 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 631–635, Montr´eal, Canada, June 3-8, 2012. c⃝2012 Association for Computational Linguistics Summarization of Historical Articles Using Temporal Event Clustering James Gung Department of Computer Science Miami University Oxford, Ohio 45056 gungjm@muohio.edu Jugal Kalita Department of Computer Science University of Colorado Colorado Springs CO 80920 jkalita@uccs.edu Abstract In this paper, we investigate the use of tempo- ral information for improving extractive sum- marization of historical articles. Our method clusters sentences based on their timestamps and temporal similarity. Each resulting clus- ter is assigned an importance score which can then be used as a weight in traditional sentence ranking techniques. Temporal im- portance weighting offers consistent improve- ments over baseline systems. 1 Introduction Extensive research has gone into determining which features of text documents are useful for cal- culating the importance of sentences for extractive summarization, as well as how to use these features (Gupta and Lehal, 2010). Little work, however, has considered the importance of temporal information towards single document summarization. This is likely because many text documents have very few explicit time features and do not necessarily describe topics in chronological order. Historical articles, such as Wikipedia articles de- scribing wars, battles, or other major events, tend to contain many explicit time features. Historical arti- cles also tend to describe events in chronological or- der. In addition, historical articles tend to focus on a single central event. The importance of other events can then be judged by their temporal distance from this central event. Finally, important events in an ar- ticle will be described in greater detail, employing more sentences than less important events. This paper investigates the value of a temporal- based score towards automatic summarization, specifically focusing on historical articles. We in- vestigate whether or not such a score can be used as a weight in traditional sentence ranking techniques to improve summarization quality. 2 Related Work Event-based summarization is a recent approach to summary generation. (Filatova and Hatzivas- siloglou, 2004) introduced atomic events, which are named entities connected by a relation such as a verb or action noun. Events are selected for summary by applying a maximum coverage algorithm to mini- mize redundancy while maintaining coverage of the major concepts of the document. (Vanderwende et al., 2004) identify events as triples consisting of two nodes and a relation. PageRank is then used to de- termine the relative importance of these triples rep- resented in a graph. Sentence generation techniques are applied towards summarization. Limited work has explored the use of temporal information for summarization. (Lim et al., 2005) use the explicit time information in the context of multi-document summarization for sentence extrac- tion and detection of redundant sentences, ordering input documents by time. They observe that impor- tant sentences tend to occur in in time slots contain- ing more documents and time slots occurring at the end and beginning of the documents set. They se- lect topic sentences for each …

Content truncated to 500 words