Making conversational structure explicit: identification of initiation-response pairs within online discussions

In this paper we investigate how to identify initiation-response pairs in asynchronous, multi-threaded, multi-party conversations. We formulate the task of identifying initiation-response pairs as a pairwise ranking problem. A novel variant of Latent Semantic Analysis (LSA) is proposed to overcome a limitation of standard LSA models, namely that uncommon words, which are critical for signaling 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.