E-ConvRec: A Large-Scale Conversational Recommendation Dataset for E-Commerce Customer Service

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 5787–5796 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 5787 E-ConvRec: A Large-Scale Conversational Recommendation Dataset for E-Commerce Customer Service Meihuizi Jia∗,†, Ruixue Liu†, Peiying Wang†, Yang Song†, Zexi Xi†, Haobin Li†, Xin Shen†, Meng Chen(�)†, Jinhui Pang∗, Xiaodong He† ∗School of Computer Science and Technology, Beijing Institute of Technology, Beijing, China †JD AI, Beijing, China {jmhuizi24, pangjinhui}@bit.edu.cn {liuruixue, wangpeiying3, songyang23, xizexi, lihaobin8, shenxin10, chenmeng20, xiaodong.he}@jd.com Abstract There has been a growing interest in developing conversational recommendation system (CRS), which provides valuable recommendations to users through conversations. Compared to the traditional recommendation, it advocates wealthier interac- tions and provides possibilities to obtain users’ exact preferences explicitly. Nevertheless, the corresponding research on this topic is limited due to the lack of broad-coverage dialogue corpus, especially real-world dialogue corpus. To handle this issue and facilitate our exploration, we construct E-ConvRec, an authentic Chinese dialogue dataset consisting of over 25k dialogues and 770k utterances, which contains user profile, product knowledge base (KB), and multiple sequential real conversations between users and recommenders. Next, we explore conversational recommendation in a real scene from multiple facets based on the dataset. Therefore, we particularly design three tasks: user preference recognition, dialogue management, and per- sonalized recommendation. In the light of the three tasks, we establish baseline results on E-ConvRec to facilitate future studies. Keywords: Conversational Recommendation, Dialogue Corpus, User Preference Recognition 1. Introduction Recently, Conversational Recommendation System (CRS), has attracted attention in the dialog community as it collects dynamic and interactive information from users’ requirements and provides useful recommenda- tions (Christakopoulou et al., 2016; Sun and Zhang, 2018; Chen et al., 2019a; Radlinski et al., 2019; Lei et al., 2020a; Lei et al., 2020b; Jannach et al., 2021; Gao et al., 2021; Zhou et al., 2021). Intuitively, a high-qualified dataset is essential to facilitate the de- velopment of CRS. To drive the progress of CRS de- velopment, there are some corpora proposed recently. In general, existing corpora are constructed in roughly two ways. Some studies generate the dialogues in a Wizard-of-Oz setting (Shah et al., 2018) by connecting two crowd-workers to engage in a chat session (Moon et al., 2019; Liu et al., 2020; Xu et al., 2020; Hayati et al., 2020; Liu et al., 2021b; Liao et al., 2021). Some other studies construct datasets from user review cor- pus (Fu et al., 2020) or item rating website (Zhou et al., 2020). The task of conversational recommendation in E- commerce domain is far more complex compared with the above mentioned scenarios. Figure 1 shows a real- world E-commerce conversation. Several characteris- tics can be observed from the conversation. Firstly, users describe the sought products in broader terms and a casual way, with either explicit expressions (e.g. 14 inches, 512 SSD) or implicit words (e.g. cheaper one), resulting in difficulties in eliciting users’ prefer- ences. Therefore, accurately recognizing user prefer- (�) Corresponding Author ence words in casual utterances lays the …

Content truncated to 500 words