Predicting Thread Discourse Structure over Technical Web Forums

Proceedings of the 2011 Conference on Empirical Methods in Natural Language Processing, pages 13–25, Edinburgh, Scotland, UK, July 27–31, 2011. c⃝2011 Association for Computational Linguistics Predicting Thread Discourse Structure over Technical Web Forums Li Wang,♠♥ Marco Lui,♠♥ Su Nam Kim,♠♥ Joakim Nivre♦ and Timothy Baldwin♠♥ ♠ Dept. of Computer Science and Software Engineering, University of Melbourne ♥ NICTA Victoria Research Laboratory ♦ Dept. of Linguistics and Philology, Uppsala University li.wang.d@gmail.com, saffsd@gmail.com, sunamkim@gmail.com, joakim.nivre@lingfil.uu.se, tb@ldwin.net Abstract Online discussion forums are a valuable means for users to resolve specific information needs, both interactively for the participants and statically for users who search/browse over historical thread data. However, the com- plex structure of forum threads can make it difficult for users to extract relevant informa- tion. The discourse structure of web forum threads, in the form of labelled dependency re- lationships between posts, has the potential to greatly improve information access over web forum archives. In this paper, we present the task of parsing user forum threads to deter- mine the labelled dependencies between posts. Three methods, including a dependency pars- ing approach, are proposed to jointly clas- sify the links (relationships) between posts and the dialogue act (type) of each link. The proposed methods significantly surpass an in- formed baseline. We also experiment with “in situ” classification of evolving threads, and es- tablish that our best methods are able to per- form equivalently well over partial threads as complete threads. 1 Introduction Web user forums (or simply “forums”) are online platforms for people to discuss information and ob- tain information via a text-based threaded discourse, generally in a pre-determined domain (e.g. IT sup- port or DSLR cameras). With the advent of Web 2.0, there has been an explosion of web authorship in this area, and forums are now widely used in various areas such as customer support, community devel- opment, interactive reporting and online eduction. In addition to providing the means to interactively participate in discussions or obtain/provide answers to questions, the vast volumes of data contained in forums make them a valuable resource for “support sharing”, i.e. looking over records of past user inter- actions to potentially find an immediately applica- ble solution to a current problem. On the one hand, more and more answers to questions over a wide range of domains are becoming available on forums; on the other hand, it is becoming harder and harder to extract and access relevant information due to the sheer scale and diversity of the data. This research aims at enhancing information ac- cess and support sharing, by mining the discourse structure of troubleshooting-oriented web user fo- rum threads. Previous research has shown that sim- ple thread structure information (e.g. reply-to struc- ture) can enhance tasks such as forum information retrieval (Seo et al., 2009) and post quality assess- ment (Lui and Baldwin, 2009). We aim to move be- yond simple threading, to predict not only the links between posts, but also show the manner of each link, in the form of the …

Content truncated to 500 words