融合用户潜在特征的深度跨域推荐方法

Deep Cross-Domain Recommendation Method Fusing Potential Features of Users

冷启动用户推荐是跨域推荐中具有挑战性的问题。目前提出的跨域模型大多使用评分信息获得冷启动用户在目标域中的潜在特征,然而评分信息又相对稀疏。为更好地获得冷启动用户的偏好特征,提出同时为源域与目标域融合评论信息的新型跨域推荐模型DRCDR。该模型使用两个源域的信息为用户提供数据,并且在处理评分信息的同时,考虑到具有时序性的评论中蕴含的上下文信息,进而为源域与目标域同步添加评论信息。DRCDR同时采用新的融合方法融合两个源域信息,并使用多层感知机获得源域与目标域之间的非线性映射函数,最终实现跨域推荐。真实数据集上的实验结果表明,该跨域推荐方法的RMSE和MAE相较于以往跨域方法提高了2%~4%,能够有效实现跨域推荐。

Solving the recommendation of cold start users is a challenging problem in cross-domain recommendation. At present,most of the proposed cross-domain models use scoring information to obtain the potential characteristics of cold start users in the target domain,but the scoring information is relatively sparse. In order to better obtain the preference characteristics of cold-start users,this paper proposes a new cross-domain recommendation model DRCDR,which fuses comment information for the source domain and target domain at the same time.This model uses the information from two source domains to provide data for users. While dealing with the scoring information,it takes into account the context information contained in the temporal comments,and then synchronously adds the comment information to the source domain and the target domain. DRCDR also uses a new fusion method to fuse two source domain information,and uses a multilayer perceptron to obtain a non-linear mapping function between the source domain and the target domain to achieve cross-domain recommendation. The experimental results on real data sets show that the RMSE and MAE of the cross-domain recommendation method proposed are improved by 2%~4%compared with the previous cross-domain methods,and it can effectively realize cross-domain recommendation.