基于会话的图卷积递归神经网络推荐模型

Session-based graph convolutional recurrent neural networks recommendation model

基于会话的推荐是为了解决匿名用户的推荐问题,是推荐系统中的一个重要分支.现有的采用图神经网络的研究方法尽管已经取得了不错的效果,但是它们无法捕获更准确的用户会话间的潜在信息.针对上述问题,论文提出了基于会话的图卷积递归神经网络(GCRNN)推荐模型,通过图卷积网络层捕捉用户会话图的结构信息,利用递归神经网络层来获得会话的时序信息和会话之间的依赖关系,以此捕获更丰富更准确的用户会话间潜在信息,从而提升推荐效果.模型在两个公开数据集上进行广泛的实验,结果表明GCRNN优于现有的研究方法.

The session-based recommendation is a subtask of recommendation system,which addresses the recommendation problem about anonymous users.Although the existing methods with the graph neural network for recommendation have achieved good results,which are insufficient to capture more accurate potential information in user’s sessions.To solve the above problem,a novel recommendation model,session-based Graph Convolutional Recurrent Neural Networks(GCRNN)is proposed in this paper to capture more potential information in user’s sessions and enhance the recommendation effects.In the proposed model,the graph convolutional neural network layer is used to capture structural information in the user graphs,as well the recurrent neural network layer is utilized to obtain the temporal information and the dependency relationship between sessions to acquire more affluent and accurate potential information in sessions.We conducted extensive experiments on two public datasets,and the results show that GCRNN is superior to the state-of-the-art methods in the session-based recommendation.