Making Conversational Structure Explicit: Identification of Initiation-response Pairs within Online Discussions

Human Language Technologies: The 2010 Annual Conference of the North American Chapter of the ACL, pages 673–676, Los Angeles, California, June 2010. c⃝2010 Association for Computational Linguistics Making Conversational Structure Explicit: Identification of Initiation-response Pairs within Online Discussions Yi-Chia Wang Carolyn P. Rosé Language Technologies Institute Language Technologies Institute Carnegie Mellon University Carnegie Mellon University Pittsburgh, PA 15213, USA Pittsburgh, PA 15213, USA yichiaw@cs.cmu.edu cprose@cs.cmu.edu Abstract In this paper we investigate how to identify initiation-response pairs in asynchronous, multi-threaded, multi-party conversations. We formulate the task of identifying initia- tion-response pairs as a pairwise ranking problem. A novel variant of Latent Semantic Analysis (LSA) is proposed to overcome a li- mitation of standard LSA models, namely that uncommon words, which are critical for sig- naling initiation-response links, tend to be deemphasized as it is the more frequent terms that end up closer to the latent factors selected through singular value decomposition. We present experimental results demonstrating significantly better performance of the novel variant of LSA over standard LSA. 1 Introduction In recent years, research in the analysis of social media (e.g., weblogs, discussion boards, and mes- sengers) has grown in popularity. Unlike exposito- ry text, the data produced through use of social media is often conversational, multi-threaded, and more complex because of the involvement of nu- merous participants who are distributed both across time and across space. Recovering the multi- threaded structure is an active area of research. In this paper, we form the foundation for a broader study of this type of data by investigating the basic unit of interaction, referred to as an initi- ation-response pair (Schegloff, 2007). Initiation- response pairs are pairs of utterances that are typi- cally contributed by different participants, and where the first pair part sets up an expectation for the second pair part. Types of common initiation- response pairs include question-answer, assess- ment-agreement, blame-denial, etc. Note that al- though sometimes discussion forum interfaces make the thread structure of the interaction expli- cit, these affordances are not always present. And even in forums that have these affordances, the apparent structure of the discourse as represented through the interface may not capture all of the contingencies between contributions in the unfold- ing conversation. Thus, the goal of this investiga- tion is to investigate approaches for automatically identifying initiation-response pairs in conversa- tions. One of the challenges in identifying initiation- response pairs is that the related messages are not necessarily adjacent to each other in the stream of contributed messages, especially within the asyn- chronous environment of social media. Further- more, individual differences related to writing style or creative expression of self may also complicate the identification of the intended connections be- tween contributions. Identification of initiation- response pairs is an important step towards auto- matic processing of conversational data. One po- tential application of this work is conversation summarization. A summary should include both the initiation and response as a coherent unit or it may fail to capture the intended meaning. We formulate the task of …

Content truncated to 500 words