Hypergraph Embedding Based on Random Walk with Adjusted Transition Probabilities

Kazuya Nagasato, Satoshi Takabe, Kazuyuki Shudo

Abstract: In this paper, we consider embedding hypergraphs using random walks. By executing a random walk on a hypergraph and inputting the resulting node sequence into a skip-gram used in natural language processing, a vector representation that captures the graph structure can be obtained. We propose a random walk method with adjustable transition probabilities for hypergraphs. As a result, we argue that it is possible to embed graph features more appropriately. Experimental results show that by tuning the parameters of the proposed method appropriately, highly accurate results can be obtained even for large hypergraphs for machine learning tasks such as node label classification.