融入逻辑规则的知识图谱推荐模型研究

Research on Knowledge Graph Recommendation Models Integrated with Logical Rules

知识图谱嵌入技术已在推荐系统领域引起广泛关注,将结构化知识图谱中的信息融入到推荐模型中,可以提高推荐的个性化程度。然而,因为初始数据的不准确性会导致推荐结果不正确,现存的知识图谱推荐模型中仍存在误差传播问题。针对这个问题,该文提出了RR-KGE模型,由知识图谱嵌入模块和推荐算法模块组成;其中聚焦于知识图谱嵌入框架,将规则嵌入和知识图谱嵌入进行联合学习,通过规则给予模型更多的约束条件,以减少误差传播;并结合此框架将推荐算法ALS(Alternating Least Squares)和RNN(Recurrent Neural Network)相融合来获得更加精确的推荐结果;最后将RR-KGE与不同基准模型进行比较,在两个数据集上多项指标均优于对比模型,证明了推荐方法的有效性。

Knowledge graph embedding technology has attracted widespread attention in the field of recommendation systems.Integrating information from structured knowledge graphs into recommendation models can enhance the personalization of recommendations.However,existing knowledge graph recommendation models still face the issue of error propagation due to the inaccuracy of initial data,which leads to incorrect recommendation results.To address this problem,we propose the RR-KGE model,consisting of a knowledge graph embedding module and a recommendation algorithm module.The focus is on the knowledge graph embedding framework,where rule embedding and knowledge graph embedding are jointly learned.Rules provide the model with additional constraints to reduce error propagation.This framework is combined with the recommendation algorithms ALS(Alternating Least Squares)and RNN(Recurrent Neural Network)to obtain more accurate recommendation results.Finally,RR-KGE is compared with different baseline models,and multiple metrics on two datasets demonstrate its superiority over the comparison models,confirming the effectiveness of the recommendation approach.