间接交互信息与因式分解机融合的推荐方法研究

Research on recommendation method of fusion of indirect interactive information and factorization machine

针对现有推荐方法存在交互信息应用不充分和推荐性能不佳的问题,充分利用用户和项目之间的间接交互信息,采用可达矩阵来表达用户和项目之间的间接交互关系,通过可达矩阵与因式分解机有机融合,构建了一个新的推荐方法。在Amazon-Book、Last-FM和Yelp2018数据集上的实验表明,所提方法在推荐效果上既优于传统的基于因式分解机的推荐方法,又好于最新的基于神经网络因式分解机的推荐模型,在推荐的时间效率上比基于知识图谱注意力网络的推荐方法具有明显优势。同时,相对其他推荐方法,该方法还具有更好的可解释性。

Aiming at the problems of insufficient interactive information application and poor recommendation performance in existing recommendation methods,this paper made full use of the indirect interaction information between users and items,and used reachable matrix to express the indirect interaction relationship between users and items.Meanwhile,it constructed a new recommendation method by organically integrating the reachable matrix and the factorization machine.Experiments on the Amazon-Book,Last-FM and Yelp2018 datasets show that the recommended performance of the proposed method is not only better than the traditional recommendation methods based on factorization machine,but also better than the latest factorization machine based on neural network.In terms of the time efficiency of recommendation,this method has obvious advantages over the recommendation methods based on the knowledge graph attention network.At the same time,compared with other recommended methods,this method has better interpretability.