Summarizing Emails with Conversational Cohesion and Subjectivity

Proceedings of ACL-08: HLT, pages 353–361, Columbus, Ohio, USA, June 2008. c⃝2008 Association for Computational Linguistics Summarizing Emails with Conversational Cohesion and Subjectivity Giuseppe Carenini, Raymond T. Ng and Xiaodong Zhou Department of Computer Science University of British Columbia Vancouver, BC, Canada {carenini, rng, xdzhou}@cs.ubc.ca Abstract In this paper, we study the problem of sum- marizing email conversations. We first build a sentence quotation graph that captures the conversation structure among emails. We adopt three cohesion measures: clue words, semantic similarity and cosine similarity as the weight of the edges. Second, we use two graph-based summarization approaches, Generalized ClueWordSummarizer and Page- Rank, to extract sentences as summaries. Third, we propose a summarization approach based on subjective opinions and integrate it with the graph-based ones. The empirical evaluation shows that the basic clue words have the highest accuracy among the three co- hesion measures. Moreover, subjective words can significantly improve accuracy. 1 Introduction With the ever increasing popularity of emails, it is very common nowadays that people discuss spe- cific issues, events or tasks among a group of peo- ple by emails(Fisher and Moody, 2002). Those dis- cussions can be viewed as conversations via emails and are valuable for the user as a personal infor- mation repository(Ducheneaut and Bellotti, 2001). In this paper, we study the problem of summariz- ing email conversations. Solutions to this problem can help users access the information embedded in emails more effectively. For instance, 10 minutes before a meeting, a user may want to quickly go through a previous discussion via emails that is go- ing to be discussed soon. In that case, rather than reading each individual email one by one, it would be preferable to read a concise summary of the pre- vious discussion with the major information summa- rized. Email summarization is also helpful for mo- bile email users on a small screen. Summarizing email conversations is challenging due to the characteristics of emails, especially the conversational nature. Most of the existing meth- ods dealing with email conversations use the email thread to represent the email conversation struc- ture, which is not accurate in many cases (Yeh and Harnly, 2006). Meanwhile, most existing email summarization approaches use quantitative features to describe the conversation structure, e.g., number of recipients and responses, and apply some general multi-document summarization methods to extract some sentences as the summary (Rambow et al., 2004) (Wan and McKeown, 2004). Although such methods consider the conversation structure some- how, they simplify the conversation structure into several features and do not fully utilize it into the summarization process. In contrast, in this paper, we propose new summa- rization approaches by sentence extraction, which rely on a fine-grain representation of the conversa- tion structure. We first build a sentence quotation graph by content analysis. This graph not only cap- tures the conversation structure more accurately, es- pecially for selective quotations, but it also repre- sents the conversation structure at the finer granular- ity of sentences. As a second contribution of this pa- per, …

Content truncated to 500 words