Sequential Emotion-Aware Recommendation: A Preliminary Study on Data Acquisition and Modeling

Abstract

The rapid development of recommender systems makes researchers pay more and more attention to the utilization of users’ subjective factors. Among various subjective factors, emotion plays an important role that can truly reflect the temporal changes of users’ preferences. However, the acquisition of user emotion is difficult, and is obtained passively in many methods. Moreover, there are few datasets suitable for continuous emotion modeling in existing recommendation research. To address these shortcomings, this work collects and organizes a dataset of users watching movies containing their temporal emotional data, which is appropriate for recommendation tasks based on sequential emotions. In addition, we propose a sequential emotion-aware recommendation method suitable for continuous emotion modeling, which outperforms other advanced sequential modeling approaches on our dataset. The contribution of this work is to conduct preliminary study and exploration on the modeling of users’ sequential emotion, so that this research direction can be further developed in future. The Sequential EMotion-Aware Recommendation (SEMAR) dataset we collected is available at: https://github.com/LinZheng666/Sequential-Emotion-Dataset-for-Recommendation.