融合显隐式特征的深度协同过滤推荐算法

A deep collaborative filtering recommendation algorithm on fusing explicit and implicit features

针对当前协同过滤算法中的用户和项目耦合关系利用不足、显式反馈与隐式反馈之间的互补关系考虑不充分等问题,提出一种融合显隐式特征的深度协同过滤推荐算法。构建一个双通道卷积神经网络用于学习用户/项目属性之间的显式关联,搭建一个深度潜在特征表示网络,间接利用用户/项目评分学习两者之间的隐式关联。将两个分支的结果联合训练,预测当前用户和项目的交互概率。在公开数据集MovieLens 1M和Tafeng上进行了充分的实验,结果表明本文所提出的方法有效提高了推荐性能。

In the current collaborative filtering algorithm,the rich coupling between users and items was always underutilized.Simultaneously,the existing research has not considered the complementary relations between explicit feedback and implicit feedback adequately.To address this issue,a deep collaborative filtering recommendation algorithm on fusing explicit and implicit features is proposed.Firstly,the double-channel convolutional neural network is established to learn the explicit associations between user/item attributes.Then,the deep latent feature representation network is constructed to learn the implicit associations by using the user/item ratings indirectly.Finally,through the combination of the results above,the predictions of user-item interaction can be obtained by training.Experimental results on MovieLens 1M and Tafeng datasets show that the proposed method can improve the recommendation performance effectively.