Recommendation Algorithm Based on Dual Attention Mechanism and Explicit Feedback

EasyChair Preprint no. 2249

Abstract

The recommendation algorithms are popular in intelligent applications, and the algorithms seamless integration with the knowledge graph has attracted much attention in recent years. However, the existing methods can not make full use of auxiliary information in KG and characterize user-item interaction behavior violence which lead to the recommendation algorithms are still limited by sparse or even cold start issue, and the recommendation results are weakly interpretable. To address the problem, this paper proposes an enhanced CTR recommendation algorithm based on knowledge graph dual attention mechanism and user explicit feedback. Here,(1) The dual attention mechanism is for KG, which can be divided into inter-item attention mechanism and inter-layer attention mechanism. The inter-item attention mechanism calculates the correlation between the item clicked by the user and the entity connected to the item in the KG. Meanwhile, the inter-layer attention mechanism calculates the correlation between different hops in the KG. (2) The user explicit feedback is the user's degree of preference for the item expressed in numerical form and is used to quantify user-item interaction behavior. Finally, in order to evaluate the usability of our proposed method, three datasets with different sparsity, another three datasets which rich in new items, and a new evaluation task-interpretive visualization were designed to conduct multi-view experimental verification.through three real datasets, MovieLens-20M, Book-Crossing and Last.FM. The results show that the proposed model is superior to the state-of-the-art baseline.

Keyphrases: Attention Mechanism, click-through rate, explicit rating, Knowledge Graph, Recommendation

@Booklet{EasyChair:2249,
  author = {Jia-Le Li and Zhi-Juan Du and Jian-Tao Zhou},
  title = {Recommendation Algorithm Based on Dual Attention Mechanism and Explicit Feedback},
  howpublished = {EasyChair Preprint no. 2249},

  year = {EasyChair, 2019}}