基于邻域聚合的协同过滤推荐模型

Collaborative Filtering Recommendation Model Based on Neighborhood Aggregation

摘要: 用户和待预测物的嵌入表示是推荐系统的核心，这种嵌入一般通过映射的方式获得，然而上述方法不能有效地利用到用户交互的协作信号，因此生成的嵌入不能很好地发挥协同过滤的效果。为了解决这个问题，本文研究了基于邻域聚合的协同过滤(NACF)模型的方法，该方法将用户交互集成到嵌入中，再利用嵌入传播将协作信号以高阶连通性的形式编码。最后该模型与贝叶斯个性化排序的矩阵分解(BPRMF)和图神经网络协同过滤(NGCF)在MovieLens数据集上的实验结果表明，本文的方法取得的效果更加优越。

Abstract: The embedded representation of users and objects to be predicted is the core of the recommendation system. Such embedding is generally obtained through mapping. However, the above method cannot effectively utilize the cooperative signals of user interaction, so the generated embedding cannot give good play to the effect of collaborative filtering. To solve this problem, this paper studies a collaborative filtering (NACF) model based on neighborhood aggregation, which integrates user interaction into embedding and encodes cooperative signals in the form of high-order connectivity by embedding propagation. Finally, the experimental results of this model, Bayesian personalized ordering matrix decomposition (BPRMF) and graph neural network cooperative filtering (NGCF) on MovieLens data set show that the proposed method achieves better results.