Link Prediction Using Power Law Clique Distribution and Common Edges Distribution
Link Prediction is an interesting problem and is concerned with predicting important edges in a social network based on the current link structure. This prediction is based on the similarity between the two nodes; similarity is captured typically using some function of the degree of the common neighbors of the two nodes. The well-known power law degree distribution is helpful in designing relevant functions used in computing similarity functions. We show that cliques of nodes in the graph also follow a power law distribution in terms of their size. We call this power law clique distribution. It prompts us to consider small size cliques in computing similarity. We specifically use cliques of size three in an appropriately weighted form to compute the similarity. Cliques of size three correspond to common edges. By using the proposed similarity functions, we show experimentally an improvement in performance in terms of classification accuracy over the state-of-the-art local similarity functions using benchmark datasets.
KeywordsCommon Neighbors Resource Allocation Index Cliques of size two and three Power law
- 1.Nowell, L., Kleinberg, J.: The Link Prediction Problem for Social Networks. In: CIKM, pp. 556–559. ACM Press, New York (2003)Google Scholar
- 5.Soundarajan, S., Hopcroft, J.: Using Community information to Improve the Precision of Link Prediction Methods. In: WWW, pp. 607–608. ACM Press, New York (2012)Google Scholar
- 6.Leskovec, J., Kleinberg, J.: Faloutsos: Graph Evolution: Densification and shrinking Diameters. Technical report, arXiv.org (2007)Google Scholar