MCSRec: Modeling Cognitive Similarity in Sequential Recommendation with Social Networks

Abstract

Combining user social relationships in sequence recommendation helps model users' potential preferences and improves the performance of the recommendation system. However, recommendation with social networks faces two challenging problems, and has not been well-studied in most existing works. The first is cognitive differences. Even users with similar preferences face the same recommended objects, they will make different choices due to cognitive differences. Therefore, modeling the cognitive similarity of users is crucial. The second is the influence strength of her friends might be different. To solve the above problems, this paper proposes a new deep learning model called MCSRec. Specifically, based on users' long short- term personal preferences, design a memory cognitive module to model the cognitive similarity between users and their friends. Then, after obtaining friends' preferences which are similar to users' cognition, model social influence with a graph attention network. The experimental results on three public data sets prove the effectiveness of our proposed MCSRec model on several competitive baselines, including state-of-the-art models.