CRFR: Improving Conversational Recommender Systems via Flexible Fragments Reasoning on Knowledge Graphs

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 4324–4334 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 4324 CRFR: Improving Conversational Recommender Systems via Flexible Fragments Reasoning on Knowledge Graphs Jinfeng Zhou1,2, Bo Wang2,1, Ruifang He2,1, Yuexian Hou1 1State Key Laboratory of Communication Content Cognition, People’s Daily Online, Beijing, China 2College of Intelligence and Computing, Tianjin University, Tianjin, China {jfzhou, bo_wang, rfhe, yxhou}@tju.edu.cn Abstract Although paths of user interests shift in knowl- edge graphs (KGs) can benefit conversational recommender systems (CRS), explicit reason- ing on KGs has not been well considered in CRS, due to the complex of high-order and in- complete paths. We propose CRFR, which ef- fectively does explicit multi-hop reasoning on KGs with a conversational context-based rein- forcement learning model. Considering the in- completeness of KGs, instead of learning sin- gle complete reasoning path, CRFR flexibly learns multiple reasoning fragments which are likely contained in the complete paths of inter- ests shift. A fragments-aware unified model is then designed to fuse the fragments informa- tion from item-oriented and concept-oriented KGs to enhance the CRS response with enti- ties and words from the fragments. Extensive experiments demonstrate CRFR’s SOTA per- formance on recommendation, conversation and conversation interpretability. 1 Introduction Different from traditional one-shot recommenda- tion systems (Jannach et al., 2020), conversational recommender systems (CRS) obtain users’ interests through multi-turn conversation, and make recom- mendations with responses. Typical CRS consists of two parts: recommender and response genera- tion. The recommender aims to understand users’ dynamic preference from contextual utterances to find the items matching the preference best. Subse- quently, the response generation aims to generate appropriate sentences asking for more information or exhibiting the recommended items and related explanation. Recommender and response genera- tion are expected to be mutually beneficial. However, contextual utterances are usually insuf- ficient to understand users’ preference (Zhou et al., 2020a). External knowledge, especially knowledge graphs (KGs), helps CRS to alleviate the problem Context: I'm looking for a suspense movie like Shutter Island, and The Prestige is also my favorite. Response: I recommend Inception with Christopher Nolan and Leonardo DiCaprio, it has an unpredictable story. Shutter Island Inception The Prestige Leonardo DiCaprio Christopher Nolan starring director suspense uncertain soap_opera story unpredictable unsure film plot Figure 1: An example of conversational movie recom- mendation. In item-oriented and concept-oriented KGs, multiple high-order reasoning paths contain fragments, from which items and concepts support the recommen- dation and the interpretability of response. (Chen et al., 2019; Zhou et al., 2020a). Existing KG-based CRS still have some issues. The first issue is the lack of explicit reasoning, especially high-order explicit reasoning, in KG to track the deep shift of user interest in conversations. For example, in Figure 1, the user prefers "Shutter Island" and "The Prestige" due to certain attributes, i.e., actor "DiCaprio" and director "Nolan". In this case, both "DiCaprio" and "Nolan" lead to a 2- order explicit reasoning of the user’s interest shift to "Inception" in the KG. Such multiple high-order explicit reasoning …

Content truncated to 500 words