计算机工程与应用 ›› 2022, Vol. 58 ›› Issue (21): 131-141.DOI: 10.3778/j.issn.1002-8331.2201-0116

• 模式识别与人工智能 • 上一篇    下一篇

融合全局信息的注意力增强会话推荐方法

王永贵,王阳,陶明阳,蔡永旺   

  1. 辽宁工程技术大学 电子与信息工程学院,辽宁 葫芦岛 125105
  • 出版日期:2022-11-01 发布日期:2022-11-01

Global Information Attention Enhancement for Session-Based Recommendation

WANG Yonggui, WANG Yang, TAO Mingyang, CAI Yongwang   

  1. School of Electronics and Information Engineering, Liaoning Technical University, Huludao, Liaoning 125105, China
  • Online:2022-11-01 Published:2022-11-01

摘要: 近年来,基于会话推荐系统(session-based recommender system,SRS)的应用和研究是推荐系统的一个热门方向。如何利用用户会话信息进一步提升用户满意度和推荐精确度,是基于会话推荐系统的主要任务。目前大多数SBR模型仅基于目标会话对用户偏好建模,忽略了来自其他会话的物品转换信息,导致无法全面了解用户偏好。为了解决其局限性,提出融合全局上下文信息注意力增强的图神经网络模型(global context information graph neural networks for session-based recommendation,GCI-GNN)。该模型利用所有会话上的物品转换关系,更准确地获取用户偏好。具体而言,GCI-GNN从目标会话和全局会话学习物品向量表示。使用位置感知注意网络,将反向位置信息纳入物品嵌入中。考虑会话长度信息学习用户表示进而达到更有效的推荐。在Diginetica和Yoochoose数据集上进行实验,实验结果表明,相对最优的基准模型,GCI-GNN模型在Diginetica数据集各项指标上的提高超过2个百分点,在Yoochoose数据集上,GCI-GNN模型在各项指标上的提高超过1个百分点,验证了GCI-GNN模型的有效性。

关键词: 会话的推荐, 图神经网络, 用户兴趣, 注意力网络

Abstract: In recent years, the application and research of session-based recommender system(SRS) is a popular direction of recommender system. How to use user session information to further improve user satisfaction and recommendation accuracy is the main task of session-based recommendation system. At present, most SBR models only model user preferences based on the target session, ignoring the item conversion information from other sessions, thus failing to fully understand user preferences. To address its limitations, this paper proposes a global context information graph neural networks for session-based recommendation(GCI-GNN). This model takes advantage of the item conversion relationship across all sessions to capture user preferences more accurately. Specifically, GCI-GNN first learns object vector representations from both the target session and the global session. Secondly, location awareness attention network is used to incorporate reverse location information into object embedding. Finally, more effective recommendation can be achieved by learning user representation based on session length information. In this paper, experiments are carried out on Diginetica and Yoochoose data sets. The experimental results show that, compared with the optimal benchmark model, the GCI-GNN model improves the indexes of Diginetica data set by more than 2 percentage points. On Yoochoose data set, the GCI-GNN model improves more than 1 percentage point in each index, which verifies the effectiveness of the GCI-GNN model.

Key words: session recommendation, graph neural network, user interest, attention network