Two Birds with One Stone: Unified Model Learning for Both Recall and Ranking in News Recommendation

Findings of the Association for Computational Linguistics: ACL 2022, pages 3474 - 3480 May 22-27, 2022 c⃝2022 Association for Computational Linguistics Two Birds with One Stone: Unified Model Learning for Both Recall and Ranking in News Recommendation Chuhan Wu† Fangzhao Wu‡∗ Tao Qi† Yongfeng Huang† †Department of Electronic Engineering, Tsinghua University, Beijing 100084, China ‡Microsoft Research Asia, Beijing 100080, China {wuchuhan15, wufangzhao, taoqi.qt}@gmail.com yfhuang@tsinghua.edu.cn Abstract Recall and ranking are two critical steps in per- sonalized news recommendation. Most existing news recommender systems conduct personal- ized news recall and ranking separately with different models. However, maintaining mul- tiple models leads to high computational cost and poses great challenges to meeting the on- line latency requirement of news recommender systems. In order to handle this problem, in this paper we propose UniRec, a unified method for recall and ranking in news recommendation. In our method, we first infer user embedding for ranking from the historical news click behav- iors of a user using a user encoder model. Then we derive the user embedding for recall from the obtained user embedding for ranking by us- ing it as the attention query to select a set of basis user embeddings which encode different general user interests and synthesize them into a user embedding for recall. The extensive ex- periments on benchmark dataset demonstrate that our method can improve both efficiency and effectiveness for recall and ranking in news recommendation. 1 Introduction News recommendation techniques are widely used by many online news websites and Apps to pro- vide personalized news services (Wu et al., 2020b). Recall and ranking are two critical steps in person- alized news recommender systems (Karimi et al., 2018; Wu et al., 2021a). As shown in Fig. 1, when a user visits a news platform, the recommender system first recalls a set of candidate news from a large-scale news pool, and then ranks candidate news for personalized news display (Wu et al., 2020b). Both news recall and ranking have been widely studied (Elkahky et al., 2015; Liu et al., 2019, 2020; Wu et al., 2020a; Wang et al., 2020; Wu et al., 2021c; Qi et al., 2021a,b,c,d). In online news recommender systems, recall and ranking are ∗Corresponding author. … Ranking Model ③ Ranking ④ Display User ① Visit ② Recall News Platform Candidate List News Pool User Behavior Logs Top K News Recall Model … … … Figure 1: A typical pipeline of news recommendation. usually conducted separately with different models, as shown in Fig. 1. However, maintaining separate models for news recall and ranking in large-scale news recommender systems usually leads to heavy computation and memory cost (Tan et al., 2020), and it may be difficult to meet the latency require- ment of online news services. Learning a unified model for personalized news recall and ranking would be greatly beneficial for alleviating the computation load of news recom- mender systems. However, it is a non-trivial task because the goals of recall and ranking are not the same (Covington et al., 2016; …

Content truncated to 500 words