PROJECT TITLE :

Effective Similarity Search on Heterogeneous Networks A Meta-path Free Approach

ABSTRACT:

In the modeling of information systems that contain multiple object types and relational types, heterogeneous information networks, or HINs, are typically utilized. Homogeneous graphs, on the other hand, are typically referred to as such because they only contain a single variety of nodes and edges. In data mining applications like web search, link prediction, and clustering, one of the most important tasks is to measure the degree to which different objects are similar to one another. At the moment, several different measures of similarity have been defined for HINs. The majority of these measurements are dependent on meta-paths, which are graphs that show the progression of node classes and edge types along the paths that connect two nodes. However, it can be difficult to enumerate and select meta-paths with regard to the quality of similarity scores. This is because meta-paths are typically designed by domain experts. Because of this, employing pre-existing similarity measures in practical contexts is challenging. In order to solve this issue, we have extended SimRank, which is a well-known similarity measure on homogeneous graphs, to HINs by introducing the idea of a decay graph. This will allow us to compare HINs more effectively. HowSim is the name of the recently proposed similarity measure, which has the advantage of not requiring any meta-paths and simultaneously capturing both the structural and semantic similarities between the items being compared. Extensive testing provides evidence that demonstrates both the general applicability and efficacy of HowSim as well as the effectiveness of the algorithms that we have proposed for computing HowSim scores.