计算机工程与应用 ›› 2023, Vol. 59 ›› Issue (13): 247-258.DOI: 10.3778/j.issn.1002-8331.2206-0190

• 大数据与云计算 • 上一篇    下一篇

基于图卷积自注意力机制的神经协同推荐算法

王巍,杜雨晅,郑小丽,张闯   

  1. 1.河北工程大学 信息与电气工程学院,河北 邯郸 056038
    2.河北工程大学 河北省安防信息感知与处理重点实验室,河北 邯郸 056038
    3.江南大学 物联网工程学院,江苏 无锡 214122
  • 出版日期:2023-07-01 发布日期:2023-07-01

Collaborative Filtering Recommendation Algorithm Based on Graph Convolution Attention Neural Network

WANG Wei, DU Yuxuan, ZHENG Xiaoli, ZHANG Chuang   

  1. 1.School of Information & Electrical Engineering, Hebei University of Engineering, Handan, Hebei 056038, China
    2.Hebei Key Laboratory of Security & Protection Information Sensing & Processing, Hebei University of Engineering, Handan, Hebei 056038, China
    3.School of Internet of Things Engineering, Jiangnan University, Wuxi, Jiangsu 214122, China
  • Online:2023-07-01 Published:2023-07-01

摘要: 随着信息技术的快速迭代发展,信息过载问题日益严重,推荐算法在一定程度上可以解决信息过载,但是传统推荐算法无法有效解决数据稀疏性和推荐准确性等相关问题。提出一种基于注意力的图卷积神经协同推荐方法(GCACF)。获取用户和项目的相关交互信息,并将其转换为相应的特征向量;将特征向量使用图卷积神经网络的传播方式聚合本地化信息,同时使用注意力机制重新分配聚合后的权重系数;最后将聚合后的特征向量使用BPR损失函数优化相关参数并得出最终推荐结果。在MovieLens-1M和Amazon-baby两个公开数据集进行对比实验,结果表明,GCACF在准确率、召回率、Mrr、命中率和NDCG五个指标上均优于基线方法。

关键词: 推荐系统, 深度学习, 协同推荐, 注意力, 图卷积神经网络

Abstract: With the rapid iterative development of information technology, the problem of information overload is becoming more and more serious. The recommendation algorithm can solve the information overload to a certain extent, but the traditional recommendation algorithm can not effectively solve the related problems such as data sparsity and recommendation accuracy. This paper proposes a graph convolution attention collaborative filtering(GCACF) recommendation method. Firstly, the model obtains the relevant interactive information of users and projects and transforms into corresponding feature vectors. Secondly, the feature vector aggregates with the propagation of graph convolution neural network and the attention mechanism redistributes the aggregated weight coefficients. Finally, the BPR loss function optimizes aggregated eigenvector and the model obtains the final recommendation result. Through the comparative experiments on Movielens-1M and Amazon-baby on two public datasets, GCACF is superior to the baseline method in precision, recall, Mrr, hit and NDCG.

Key words: recommendation system, deep learning, collaborative recommend, attention mechanism, graph convolution neural network