A Meta Graph-Based Top-k Similarity Measure for Heterogeneous Information Networks

Abstract

Studies have demonstrated that real-world data can be modeled as a heterogeneous information network (HIN) composed of multiple types of entities and relationships. Similarity search is a basic operation requiring many problems in HINs. Similarity measures can be used in various applications, including friend recommendation, link prediction, and online advertising. However, most existing similarity measures only consider meta path. Complex semantic meaning cannot be expressed through meta path. In this paper, we study the similarity search problem of complex semantics meaning between two HIN objects. In order to solve the problem, we use meta graphs to express the semantic meaning between objects. The advantage of meta graphs is that it can describe the complex semantic meaning between two HIN objects. And we first define a new meta graph-based relation similarity measure, GraphSim, which is to measure the similarity between objects in HINs, then we propose a similarity search framework based on GraphSim. The experiments with real-world datasets from DBLP demonstrated the effectiveness of our approach.