2025年03期 v.46 9-16页
卢振业 杜玉晓
（广东工业大学自动化学院，广东 广州 510006）
摘要：针对匿名用户或新用户因缺少历史数据，仅根据当前会话进行音乐推荐方式单一、未考虑情感因素影响用户选择等问题，提出融合情感的异构图神经网络音乐会话推荐算法。该算法根据所有用户的历史数据和当前会话构造基于图神经网络的会话推荐，结合音乐情感因素，为匿名用户或新用户推荐更加准确的音乐。实验结果表明，在Nowplaying数据集上，该算法与次优的基于图神经网络的会话推荐算法相比，P@20提高了2.1%，MRR@20提高了6.8%，有效提升了算法推荐性能。
关键词：会话推荐；异构图神经网络；音乐情感；匿名用户推荐
中图分类号：TP391.3 文献标志码：A 文章编号：1674-2605(2025)03-0002-08
DOI：10.12475/aie.20250302 开放获取
Emotion-enhanced Heterogeneous Graph Neural Network for Music
Session-based Recommendation Algorithm
LU Zhenye DU Yuxiao
(School of Automation, Guangdong University of Technology, Guangzhou 510006, China)
Abstract: To address the limitations of current music session-based recommendation methods for anonymous or new users—such as simplistic recommendations based solely on short-term sessions and neglect of emotional factors influencing user choices—this study proposes an emotion-enhanced heterogeneous graph neural network for music session-based recommendation algorithm. The algorithm constructs a session-based recommendation system using historical data from all users and current sessions via graph neural networks, integrating musical emotional semantics to provide more accurate recommendations for anonymous/new users. Experimental results on the Nowplaying dataset demonstrate that, compared to the suboptimal GNN-based session recommendation method, the proposed algorithm achieves a 2.1% improvement in P@20 and a 6.8% increase in MRR@20, effectively enhancing recommendation performance.
Keywords: session-based recommendation; heterogeneous graph neural network; music emotion; anonymous user recom-mendation
