You Sound Like Someone Who Watches Drama Movies: Towards Predicting Movie Preferences from Conversational Interactions

Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 3091–3096 June 6–11, 2021. ©2021 Association for Computational Linguistics 3091 You Sound Like Someone Who Watches Drama Movies: Towards Predicting Movie Preferences from Conversational Interactions Sergey Volokhin*, Joyce Ho*, Oleg Rokhlenko+, and Eugene Agichtein*+ *Emory University, GA USA +Amazon, Seattle, WA, USA {svolokh, joyce.c.ho}@emory.edu, {olegro, eugeneag}@amazon.com Abstract The increasing popularity of voice-based per- sonal assistants provides new opportunities for conversational recommendation. One partic- ularly interesting area is movie recommenda- tion, which can benefit from an open-ended in- teraction with the user, through a natural con- versation. We explore one promising direc- tion for conversational recommendation: map- ping a conversational user, for whom there is limited or no data available, to most simi- lar external reviewers, whose preferences are known, by representing the conversation as a user’s interest vector, and adapting collabora- tive filtering techniques to estimate the cur- rent user’s preferences for new movies. We call our proposed method ConvExtr (Conver- sational Collaborative Filtering using External Data), which 1) infers a user’s sentiment to- wards an entity from the conversation context, and 2) transforms the ratings of "similar" ex- ternal reviewers to predict the current user’s preferences. We implement these steps by adapting contextual sentiment prediction tech- niques, and domain adaptation, respectively. To evaluate our method, we develop and make available a finely annotated dataset of movie recommendation conversations, which we call MovieSent . Our results demonstrate that Con- vExtr can improve the accuracy of predicting users’ ratings for new movies by exploiting conversation content and external data. 1 Introduction and Background With the increasing popularity of voice-assistants, there has been a lot of research on making well- established user experiences, like recommenda- tions, conversational (Allan et al., 2012; Culpepper et al., 2018; Radlinski et al., 2019). One such area is movie recommendation. Movie recommendation in general has been an actively researched area in conversational recom- mendation systems (Bennett et al., 2007; Khatri et al., 2018) and has been explored with a variety of approaches, most popular of which include col- laborative filtering (CF) (Katarya and Verma, 2017; He et al., 2019), content-based filtering (Elahi et al., 2017), incorporating user reviews (Zhao et al., 2017; Dubey et al., 2018). Several attempts at both conversational recommendation (Christakopoulou et al., 2016; Sun and Zhang, 2018; Torbati et al., 2021), and specifically conversational movie rec- ommendations (Dalton et al., 2018) have been made. However, establishing the new user’s preferences through a conversation, in order to make an ef- fective recommendation, remains an open ques- tion, and there exists little conversational data for such a task. In our initial exploration, we exam- ine if current well-researched approaches and al- gorithms can be adapted and used to help solve this problem and to establish a baseline of such ap- proaches for future improvement and reference. We explore a new approach to conversational recom- mendation by incorporating preferences of other, external users with established preferences, via …

Content truncated to 500 words