HieRec: Hierarchical User Interest Modeling for Personalized News Recommendation

Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, pages 5446–5456 August 1–6, 2021. ©2021 Association for Computational Linguistics 5446 HieRec: Hierarchical User Interest Modeling for Personalized News Recommendation Tao Qi1, Fangzhao Wu2, Chuhan Wu1, Peiru Yang1, Yang Yu2, Xing Xie2 and Yongfeng Huang1 1Department of Electronic Engineering & BNRist, Tsinghua University, Beijing 100084, China 2Microsoft Research Asia, Beijing 100080, China {taoqi.qt, wufangzhao, wuchuhan15, peiruyang17}@gmail.com yfhuang@mail.tsinghua.edu.cn {t-yyu,xing.xie}@microsoft.com Abstract User interest modeling is critical for personal- ized news recommendation. Existing news rec- ommendation methods usually learn a single user embedding for each user from their pre- vious behaviors to represent their overall inter- est. However, user interest is usually diverse and multi-grained, which is difficult to be ac- curately modeled by a single user embedding. In this paper, we propose a news recommen- dation method with hierarchical user interest modeling, named HieRec. Instead of a sin- gle user embedding, in our method each user is represented in a hierarchical interest tree to better capture their diverse and multi-grained interest in news. We use a three-level hierarchy to represent 1) overall user interest; 2) user in- terest in coarse-grained topics like sports; and 3) user interest in fine-grained topics like foot- ball. Moreover, we propose a hierarchical user interest matching framework to match candi- date news with different levels of user interest for more accurate user interest targeting. Ex- tensive experiments on two real-world datasets validate our method can effectively improve the performance of user modeling for person- alized news recommendation. 1 Introduction Recently, massive people are habituated to reading news articles on online news platforms, such as Google News and Microsoft News (Khattar et al., 2018; Das et al., 2007). To help users efficiently obtain their interested news information, person- alized news recommendation technique that aims to recommend news according to user interests, is widely used by these platforms (Wu et al., 2020a; Liu et al., 2010; Lin et al., 2014). User interest modeling is a critical step for per- sonalized news recommendation (Wu et al., 2021; Zheng et al., 2018; Wu et al., 2020c). Existing methods usually learn a single representation vector Figure 1: Click and non-click logs of an example user. to model overall user interests from users’ clicked news (Okura et al., 2017; Wu et al., 2020b; An et al., 2019). For example, Okura et al. (2017) used a GRU network to model user interests from clicked news. They used the latest hidden state of GRU as the user interest representation. Wu et al. (2019e) used multi-head self-attention network to capture user interests, and used an attentive pool- ing network to obtain a unified user representation. However, user interest is usually diverse and multi- grained. For example, as shown in Fig. 1, a user may have interest in movies, sports, finance and health at the same time. In addition, for users who are interested in sports, some of them may have general interest in this …

Content truncated to 500 words