2021/10/30 by Govind K. Sharma, Sharma, Govind, Paarth Gupta +3
Computer Science · Physics and Astronomy · Biochemistry, Genetics and Molecular Biology · #Advanced Graph Neural Networks #Complex Network Analysis Techniques #Bioinformatics and Genomic Networks
paper · pdf · doi:10.48550/arxiv.2111.00256
The problem of node-similarity in networks has motivated a plethora of such\nmeasures between node-pairs, which make use of the underlying graph structure.\nHowever, higher-order relations cannot be losslessly captured by mere graphs\nand hence, extensions thereof viz. hypergraphs are used instead. Measuring\nproximity between node pairs in such a setting calls for a revision in the\ntopological measures of similarity, lest the hypergraph structure remains\nunder-exploited. We, in this work, propose a multitude of hypergraph-oriented\nsimilarity scores between node-pairs, thereby providing novel solutions to the\nlink prediction problem. As a part of our proposition, we provide theoretical\nformulations to extend graph-topology based scores to hypergraphs. We compare\nour scores with graph-based scores (over clique-expansions of hypergraphs into\ngraphs) from the state-of-the-art. Using a combination of the existing\ngraph-based and the proposed hypergraph-based similarity scores as features for\na classifier predicts links much better than using the former solely.\nExperiments on several real-world datasets and both quantitative as well as\nqualitative analyses on the same exhibit the superiority of the proposed\nsimilarity scores over the existing ones.\n