You May Like This Hotel Because ...: Identifying Evidence for Explainable Recommendations

Proceedings of the 1st Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics and the 10th International Joint Conference on Natural Language Processing, pages 890–899 December 4 - 7, 2020. c⃝2020 Association for Computational Linguistics 890 You May Like This Hotel Because ...: Identifying Evidence for Explainable Recommendations Shin Kanouchi1, Masato Neishi2, Yuta Hayashibe1, Hiroki Ouchi3, Naoaki Okazaki4 1Megagon Labs, Tokyo, 2The University of Tokyo, 3RIKEN, 4Tokyo Institute of Technology {shin187nlp,hayashibe}@megagon.ai, neishi@tkl.iis.u-tokyo.ac.jp, hiroki.ouchi@riken.jp, okazaki@c.titech.ac.jp Abstract Explainable recommendation is a good way to improve user satisfaction. However, explain- able recommendation in dialogue is challeng- ing since it has to handle natural language as both input and output. To tackle the challenge, this paper proposes a novel and practical task to explain evidences in recommending hotels given vague requests expressed freely in natu- ral language. We decompose the process into two subtasks on hotel reviews: evidence iden- tification and evidence explanation. The for- mer predicts whether or not a sentence con- tains evidence that expresses why a given re- quest is satisfied. The latter generates a rec- ommendation sentence given a request and an evidence sentence. In order to address these subtasks, we build an Evidence-based Expla- nation dataset, which is the largest dataset for explaining evidences in recommending hotels for vague requests. The experimental results demonstrate that the BERT model can find ev- idence sentences with respect to various vague requests and that the LSTM-based model can generate recommendation sentences. 1 Introduction Recently, dialog systems using Natural Lan- guage Processing technology have been adopted in interactive services such as call centers (Zumstein and Hundertmark, 2017). One challenging issue in a real-world scenario is vague requests1 from users. For example, in a hotel booking service, users often ask operators for “a child-friendly hotel” or “a convenient inn.” To respond to such vague requests, human operators need to explain the reason why the given request 1In this study, a vague request means one that does not specify a specific product, experience or service. is satisfied. An example response would be, “This hotel has a large kids’ space, so I recommend it for families with children like you.” Responding to vague requests with evidences is effective because it not only strengthens the recommendation, but also urges users to make more concrete requests such as “I don’t need a kids’ space but want a baby stroller rental service.” Several studies have addressed explainable recommendations that produce natural language sentences (Zhao et al., 2014; Zhang et al., 2014; Wang et al., 2018; Zhao et al., 2019). One major approach is feature-based explanations. Zhang et al. (2014) generated explanation sen- tences using templates with slots, for example, “You might be interested in [feature], on which this product performs well.” However, by handling only predefined and limited features, this study cannot explain detailed evidences for each hotel such as “a view of Mount Fuji and Lake Kawaguchi.” Furthermore, this study does not ac- cept natural language requests as inputs, which is a major bottleneck for building …

Content truncated to 500 words