Online Conversation Disentanglement with Pointer Networks

Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, pages 6321–6330, November 16–20, 2020. c⃝2020 Association for Computational Linguistics 6321 Online Conversation Disentanglement with Pointer Networks Tao Yu¶ and Shafiq Joty¶† ¶Nanyang Technological University, Singapore †Salesforce Research {tao003, srjoty}@ntu.edu.sg Abstract Huge amounts of textual conversations occur online every day, where multiple conversations take place concurrently. Interleaved conver- sations lead to difficulties in not only follow- ing the ongoing discussions but also extracting relevant information from simultaneous mes- sages. Conversation disentanglement aims to separate intermingled messages into detached conversations. However, existing disentangle- ment methods rely mostly on handcrafted fea- tures that are dataset specific, which hinders generalization and adaptability. In this work, we propose an end-to-end online framework for conversation disentanglement that avoids time-consuming domain-specific feature engi- neering. We design a novel way to embed the whole utterance that comprises timestamp, speaker, and message text, and propose a cus- tom attention mechanism that models disen- tanglement as a pointing problem while effec- tively capturing inter-utterance interactions in an end-to-end fashion. We also introduce a joint-learning objective to better capture con- textual information. Our experiments on the Ubuntu IRC dataset show that our method achieves state-of-the-art performance in both link and conversation prediction tasks. 1 Introduction With the fast growth of Internet and mobile devices, people now commonly communicate in the virtual world to discuss events, issues, tasks, and personal experiences. Among the various methods of com- munication, text-based conversational media, such as Internet Relay Chat (IRC), Facebook Messenger, Whatsapp, and Slack, has been and remains one of the most popular choices. Multiple ongoing con- versations seem to occur naturally in such social and organizational interactions, especially when the conversation involves more than two partici- pants (Elsner and Charniak, 2010). For example, Time Sp Message Text 02:26 system ===zelot joined the channel 02:26 zelot hi, where can i get some help in regards to issues with mount? 02:26 TuxThePenguin After taking it out 02:26 hannasanarion TuxThePenguin, try booting with monitors connected to motherboard 02:26 pnunn TuxThePonguin, sounds like there is on board graphics as well, so try that without the card 02:26 pnunn Yeh, just one monitor though 02:27 hannasanarion pnunn, right 02:27 TuxThePenguin Makes sense to me :) 02:27 pnunn process of elimination. 02:27 TuxThePenguin Along with Occam’s Razor 02:27 Bashing-om zelot: If you are on a supported release of ’buntu, this is a good place to ask. 02:27 TuxThePenguin Any solution is most likely the simplest one 02:28 wllrt I’m a emacs newb and looking to prevent rsi. Figure 1: An excerpt of a conversation from the Ubuntu IRC corpus (best viewed in color). Same color reflects same conversation. Mentions of names are highlighted. consider the excerpt of a multi-party conversation in Figure 1 taken from the Ubuntu IRC corpus (Lowe et al., 2015). Even in this small excerpt, there are 4 concurrent conversations (distinguished by different colors) among 4 participants. Identifying or disentangling individual conver- sations is often considered as a prerequisite for downstream dialog tasks …

Content truncated to 500 words