Relrank: An Algorithm for Relevance-Based Ranking of Meta-Paths in a Heterogeneous Information Network

摘要

Heterogeneous Information Networks (HIN) are extensively used for mapping heterogeneous data to solve complex real life problems computationally. Meta-path is a key concept used in HIN based research. The efficiency and accuracy of an HIN based task heavily depend on the identification and effective utilisation of meta-paths that are relevant to the problem at hand. Here, we propose an algorithm, Relrank to rank the meta-paths in an HIN in order of their relevance. The algorithm assigns a relevance score to each meta-path under a specific length threshold. The meta-paths are ranked based on the relevance score. The degree of closeness of each meta-path to the actual links present in the HIN acts as the key parameter in determining the rank. We tested the efficiency of the algorithm by applying it on an real world HIN with biological interaction data, to derive novel findings from existing knowledge. The results obtained, justified the correctness of the proposed algorithm. Furthermore, the algorithm design is generic enough to encompass heterogeneous data across various domains.