Extracting Keywords from Multi-party Live Chats

Copyright 2012 by Su Nam Kim and Timothy Baldwin 26th Pacific Asia Conference on Language,Information and Computation pages 199–208 Extracting Keywords from Multi-party Live Chats Su Nam Kim School of Information Technology Monash University Clayton, VIC, Australia su.kim@monash.edu.au Timothy Baldwin Dept. of Computing and Information Systems The University of Melbourne VIC, Australia tb@ldwin.net Abstract Live chats have become a popular form of communication, connecting people all over the globe. We believe that one of the simplest approaches for providing topic information to users joining a chat is keywords. In this paper, we present a method to automatically extract contextually relevant keywords for multi-party live chats. In our work, we identify keywords that are associated with specific dialogue acts as well as the occurrences of keywords across the entire conversation. In this way, we are able to identify distinguishing features of the chat based on structural information derived from live chats and predicted dialogue acts. In evaluation, we find that using structural infor- mation and predicted dialogue acts performs well, and that conventional methods do not work well over live chats. 1 Introduction Keywords or keyphrases1 are an effective way of representing the core topic of a document, and can effectively summarize and/or help index documents. They are usually found in the form of either sim- plex nouns (e.g. library) or noun phrases (e.g. social issue). They have been studied in the past to pro- vide topic-related information for many applications such as text summarizers, search engines and index- ers. For example, Barzilay and Elhadad (1997) used keywords as semantic meta-information for summa- rizers. D ´Avanzo and Magnini (2005) used them to 1In this work, we use the term keywords for consistency, while noting that it can be used to refer to multiword terms. organize documents for search engines. Dredze et al. (2008) used keywords as summaries of email in order to better manage and prioritize emails. Ham- mouda et al. (2005) used keywords extracted from multiple documents in order to discover the topics of documents for clustering. Gutwin et al. (1999) used automatically extracted keywords to refine the queries to improve precision of search in an online library browser. There has been much research on automatic key- word extraction (Frank et al., 1999; Turney, 1999; Hulth, 2003, inter alia). The majority of work has been done over specific domains such as sci- entific articles and newspapers, including the re- cent SemEval-2010 shared task on keyword extrac- tion (Kim et al., 2010b). A small minority of re- searchers have used different sources of data such as email (Dredze et al., 2008) and HTML docu- ments (Mori et al., 2004), as outlined in Section 2. However, existing approaches tend not to work well when applied to different target sources, and are of- ten susceptible to domain-specific features of the tar- get documents (e.g. structure). In this paper, our aim is to automatically ex- tract keywords for multi-party live chats. Live chats are essentially text-based dialogues, with less dis- fluencies than spoken dialogues …

Content truncated to 500 words