基于知识聚合和迁移的跨领域推荐算法

Cross-Domain Recommendation Algorithm Based on Knowledge Aggregation and Transfer

传统的推荐算法受限于单领域中用户和项目的稀疏关系,也存在用户冷启动等问题.跨领域推荐能够通过学习辅助领域的知识去丰富目标领域的稀疏数据以提高推荐准确率.本文提出了一种知识聚合和迁移相结合的跨领域推荐算法ATCF.与已有算法不同,在对共性知识和个性知识的表示学习中,ATCF均充分融合了辅助域和目标域的知识,通过基于矩阵分解的两级矩阵拼接和两次矩阵填充,得到在群集矩阵及评分矩阵上的共性知识表示;通过知识迁移,构建了重叠用户和非重叠用户的个性知识表示,有效避免了负迁移.在两个跨领域数据集上开展的实验表明,ATCF算法与已有单领域和跨领域推荐算法相比RMSE降低了3%~7%,准确率召回率增加了8%~15%.

Cross-domain recommendation can study the knowledge of the auxiliary domains to enrich the knowledge in the target domain,so as to improve the recommendation precision in the target domain.This paper proposes an aggregation and transfer collaborative filtering algorithm for cross-domain recommendation(ATCF).In order to represent the sharing knowledge in different domains,the knowledge in the auxiliary domain and the target domain are fully aggregated,through two levels of matrix concatenation.Moreover,the personalized knowledge of the target domain is represented by knowledge transferring from auxiliary domain.By fusion the sharing and the personalized knowledge,we can obtain the final rating.Two different cross-domain datasets are used for the experiments.Our efforts show that the ATCF algorithm has better recommendation performance.