Neural Conversation Recommendation with Online Interaction Modeling

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 4633–4643, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 4633 Neural Conversation Recommendation with Online Interaction Modeling Xingshan Zeng1,2, Jing Li3∗, Lu Wang4, Kam-Fai Wong1,2 1The Chinese University of Hong Kong, Hong Kong, China 2MoE Key Laboratory of High Confidence Software Technologies, China 3Department of Computing, The Hong Kong Polytechnic University, Hong Kong, China 4Khoury College of Computer Sciences, Northeastern University, Boston, United States 1,2{xszeng,kfwong}@se.cuhk.edu.hk 3jing-amelia.li@polyu.edu.hk, 4luwang@ccs.neu.edu Abstract The prevalent use of social media leads to a vast amount of online conversations being pro- duced on a daily basis. It presents a con- crete challenge for individuals to better dis- cover and engage in social media discussions. In this paper, we present a novel framework to automatically recommend conversations to users based on their prior conversation behav- iors. Built on neural collaborative filtering, our model explores deep semantic features that measure how a user’s preferences match an ongoing conversation’s context. Furthermore, to identify salient characteristics from inter- leaving user interactions, our model incorpo- rates graph-structured networks, where both replying relations and temporal features are encoded as conversation context. Experimen- tal results on two large-scale datasets collected from Twitter and Reddit show that our model yields better performance than previous state- of-the-art models, which only utilize lexical features and ignore past user interactions in the conversations. 1 Introduction Social media has profoundly revolutionized peo- ple’s social interactions, as many individuals now turn to online platforms to voice opinions and ex- change ideas. Meanwhile, the abundance of in- formation brings the problem of information ex- plosion — the huge volume of online discussions produced every day has far outpaced any individ- ual’s capability of digesting them. It is hence diffi- cult for one to discover online discussions that are potentially of interest. To address this issue, we study the problem of online conversation recom- mendation, with the goal of identifying conversa- tions that fit a user’s preferences, hence likely to result in the user’s future engagement. ∗This work was mainly conducted when Jing Li was af- filiated with Tencent AI Lab, Shenzhen, China. T1[U1]: The official 9/11 story is a complete and total lie. Also, it is being used to destroy the U.S. T2[U2]: …… The depressing thing is that it clearly intends to take the rest of the world down with it. T4[U1]: It is not the U.S. it is in fact a shadow global government ...... T3[U3]:Only if the rest of the world allows these lunatics to do whatever they want ...... T7[U5]: Yeah, like these people: <URL> ? T6[U5]: Why? Reddit users tend to be quite intelligent. Why a blind spot about 9/11? T5[U4]:No, there is a 9/11 clause in logic….. Conversation 1 Conversation 2 [U1]: With all of your fears how often do you change your underwear? …… [U0]: Why would you want limited government? …… [U0]: we need homeland security to …

Content truncated to 500 words