CBML: A Cluster-based Meta-learning Model for Session-based Recommendation Jiayu Song jysongsuda@stu.suda.edu.cn School of Computer Science and Technology, Soochow University Suzhou, China Jiajie Xu ∗ xujj@suda.edu.cn School of Computer Science and Technology, Soochow University Suzhou, China Rui Zhou rzhou@swin.edu.au Swinburne University of Technology Australia Lu Chen luchen@swin.edu.au Swinburne University of Technology Australia Jianxin Li jianxin.li@deakin.edu.au Deakin University Australia Chengfei Liu cliu@swin.edu.au Swinburne University of Technology Australia ABSTRACT Session-based recommendation is to predict an anonymous user’s next action based on the user’s historical actions in the current session. However, the cold-start problem of limited number of ac- tions at the beginning of an anonymous session makes it difcult to model the user’s behavior, i.e., hard to capture the user’s various and dynamic preferences within the session. This severely afects the accuracy of session-based recommendation. Although some existing meta-learning based approaches have alleviated the cold- start problem by borrowing preferences from other users, they are still weak in modeling the behavior of the current user. To tackle the challenge, we propose a novel cluster-based meta-learning model for session-based recommendation. Specially, we adopt a soft- clustering method and design a parameter gate to better transfer shared knowledge across similar sessions and preserve the charac- teristics of the session itself. Besides, we apply two self-attention blocks to capture the transition patterns of sessions in both item and feature aspects. Finally, comprehensive experiments are con- ducted on two real-world datasets and demonstrate the superior performance of CBML over existing approaches. CCS CONCEPTS · Information systems → Recommender systems. KEYWORDS session-based recommendation; meta-learning; soft-clustering; con- tent information ∗ Corresponding author Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for proft or commercial advantage and that copies bear this notice and the full citation on the frst page. Copyrights for components of this work owned by others than ACM must be honored. Abstracting with credit is permitted. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specifc permission and/or a fee. Request permissions from permissions@acm.org. CIKM ’21, November 1ś5, 2021, Virtual Event, QLD, Australia © 2021 Association for Computing Machinery. ACM ISBN 978-1-4503-8446-9/21/11. . . $15.00 https://doi.org/10.1145/3459637.3482239 ACM Reference Format: Jiayu Song, Jiajie Xu, Rui Zhou, Lu Chen, Jianxin Li, and Chengfei Liu. 2021. CBML: A Cluster-based Meta-learning Model for Session-based Rec- ommendation. In Proceedings of the 30th ACM International Conference on Information and Knowledge Management (CIKM ’21), November 1ś5, 2021, Virtual Event, QLD, Australia. ACM, New York, NY, USA, 10 pages. https://doi.org/10.1145/3459637.3482239 1 INTRODUCTION Recommender systems play an important role in providing users required information in a timely and efective manner. Most exist- ing recommendation methods assume that user profles and past activities are constantly recorded. However, in many scenarios, a user is usually anonymous and only a few user historical actions in an ongoing session can be used to predict the user’s next click. This motivates session-based recommendation, which has become an important sub-area of recommender systems. Existing solutions utilize deep neural networks like improved LSTM [13], GRU [2] and self-attention mechanism [31] to capture user preferences within sessions. However, since anonymous sessions tend to contain few interactions, this cold-start problem severely limits the performance of session-based recommendation. As a representative few-shot learning method, meta-learning is proposed in [5] and shows promising results in many cold-start ap- plications, such as few-shot image classifcation [7]. Inspired by this, some recent studies [3, 12] have adopted meta-learning in recom- mendation tasks for addressing cold-start problems. In these models, each user is regarded as a learning task. These models frst learn well-generalized global parameters that can reasonably initialize the parameters of all tasks. When processing each recommendation to a user [3], local updates are conducted on initialized parame- ters using the user’s own data to derive personalized parameters, which represents user-specifc preferences and enables meaningful recommendation. Since meta-learning turns out to be efective in cold-start scenarios [3, 12, 37], it provides great opportunity for session-based recommendation. However, in session-based recom- mendation where each session becomes a learning task, directly applying existing meta-learning based recommendation methods would incur inaccuracy due to two limitations discussed below. First, previous meta-learning based recommendation models [3, 4, 12, 37] for cold-start problems assume that the same global parameters are used to guide parameter initialization for all tasks. Full Paper Track CIKM ’21, November 1–5, 2021, Virtual Event, Australia 1713