基于CM-RippleNet的社区发现知识图谱推荐算法

Recommended Algorithm for Community Detection Knowledge Graphs Based on RippleNet

针对RippleNet未充分考虑实体聚集的关联性,导致推荐实体不够准确的问题,提出基于社区发现的模型CM-RippleNet,该模型通过构建知识图谱复杂网络,使同一个社区内的成员具有较高的相似度;利用Louvain算法对网络中的节点进行社区划分,得到节点-社区映射矩阵,将其加入到实体节点的嵌入表示中,融合用户表示最终计算出推荐结果。在两个公共知识图谱数据集上的实验结果表明:与KGCN、RippleNet、PER、MKR模型相比,CM-RippleNet模型的ACC、AUC、准确度与召回率指标有极大的提升,验证了模型的有效性。

Aiming at the problem that RippleNet failed to sufficiently think over the correlation of entity aggregation and it incurred the inaccuracy of recommended entities,a CM-RippleNet model based on community detection was proposed.This model constructs a knowledge graph complex network to make the members in the same community have a high degree of similarity,including having louvain algorithm used to divide nodes in the network into communities so as to obtain a node-community mapping matrix,and then having it added to the embedded representation of entity nodes and having the user representation fused to finally calculate recommendation result.Experimental results on two public knowledge graph datasets show that,as compared to the KGCN,RippleNet,PER and MKR models,the ACC,AUC,accuracy and recall index values of the CM-RippleNet model are greatly improved,which verifies the effectiveness of the model.