Knowledge graph preference migration network for recommendation

doi:10.1016/j.eswa.2023.121256

Knowledge graph preference migration network for recommendation

2024-03-01

1

PRE

AI

DOI：10.1016/j.eswa.2023.121256

原文链接

原文求助

摘要

En 中文

Knowledge graph (KG) is being introduced into recommender systems in more and more scenarios. However, the previous recommendation methods based on KG usually focus on mining the relative signals between a user and an item, while ignoring the exploration of collaborative signals between users. In addition, the traditional recommendation methods based on KG completely overlooks the discrepancies among users in the modeling process, which will lead to the problem of insufficient information in the obtained cold-start users' representations. To tackle these, a novel model named Knowledge Graph Preference Migration Network (KGPM) is proposed by us, aiming to address the lack of information in cold-start users' representations in the knowledge-aware recommendation. To be specific, the method aggregates the representations from a node's neighborhoods of different hops to renew the node representation on a collaborative knowledge graph (CKG) consisting of a KG and a user-item interaction graph (UIIG), and differentiates the importance of different neighbors via employing an attention mechanism. Next, to achieve preference migration by mining more collaborative signals among users, the neighbor users of the cold-start users should be further propagated in our elaborately constructed user logical interaction space. We apply two datasets accompanying specific scenarios to verify our proposed model, and the experimental results show that KGPM achieves significant improvements over the most advanced baselines.

Keyword:

Knowledge graph

Collaborative signals

Cold-start users

User logical interaction space

Preference migration

Knowledge graph preference migration network for recommendation

摘要

期刊

机构

引用论文