半监督元路径的异构信息网络社区发现算法

Semi-supervised Meta-path-based Algorithm for Community Detection in Heterogeneous Information Networks

基于语义的异构信息网络社区发现算法,大多采用元路径计算目标对象相似性,基于元路径的目标对象相似性度量非常便捷、有效,但是,其语义表达并不完整,不能完全真实地反映目标对象的关联,而目前又缺乏更加准确的表达目标对象相似性的方法.基于语义的异构信息网络社区发现算法往往忽略了异构信息网络复杂的拓扑结构,本文通过谱聚类分析异构信息网络的拓扑结构,半监督校正目标对象的相似性,使用非负矩阵分解法划分异构信息网络的社区,能够有效提高异构信息网络社区发现的准确率.通过对仿真数据和真实数据实验,结果显示本文算法确实有效提高了异构信息网络社区发现的准确率.

Similarity between target objects is mostly calculated based on meta-paths for semantic-based community detection algorithm in heterogeneous information networks.The meta-path-based similarity between target objects is very efficient,but the semantics of meta-path-based similarity is not integrity and can not truly express relevance between target objects.And now there is lack of better expression for similarity semantics between target objects in heterogeneous information networks.The complex topology structure is usually neglected in semantic-based community detection algorithms for heterogeneous information networks.To effectively improve accuracy of community detection algorithm in heterogeneous information networks,a semi-supervised meta-path-based algorithm for community detection is proposed in this paper.First,spectral method is used to analyse the topology structure of heterogeneous information networks.Then the similarity between target objects is adjusted by representatives in every clusters.Last,NMF method is used to detect communities.Through experiments in simulation datasets and real datasets,the experimental results showed the proposed algorithm is effective.