An Efficient Modified Common Neighbor Approach for Link Prediction in Social Networks

Link prediction in social networks aims at estimating the likelihood of the appearance of a new link between two nodes, based on the existing links and the attributes of the nodes. Many methods for link prediction problem in social networks have been proposed in literature. We especially analyze the shortcomings of common neighbor leading method. Accordingly we generate a new modified common neighbor approach for link prediction in social networks. Our approach efficiently works under the integrated analysis of features along with topological structure of a social network. As a co-authorship network is a true social network, we have considered the co-authorship networks for verifying the effectiveness of the existing leading methods as well as our proposed link prediction method. We have implemented the leading methods as well as our proposed method on two different data domains of co-authorship networks obtained from author lists of papers at five sections of Physics e-Print arXiv, www.arXiv.org. In the first data domain, the papers in the periods (1994 – 1996) and (1997 – 1999) are taken as the training set and testing set respectively. Similarly the papers in the periods (2007 – 2009) and (2010 – 2012) are taken as the training set and testing set for the second data domain. Experimental results show that all the methods are found to perform much better over the random predictor. Again we find that our modified common neighbor approach outperforms over all the existing leading methods considered.