Link Prediction Computational Models: A Comparative Study

Abstract

The continuous creation of digital media mainly video has caused a tremendous growth of digital content. Social networks are the way to represent the relationship between entities. Analysis of a social network involves predicting links between actors or finding relationships, finding important nodes or actors, detecting communities of similar actors, etc. The data in social networks, which involves nodes, relations or contents, are huge and dynamic. There is a need of some data mining techniques to analyse such data and perform analysis. Social network analysis from a data mining perspective is also called link mining or link analysis. Earlier social networks are used only for interacting and sharing information, but it has a concept of research work. Now-a-days there are various social networking sites and it’s the choice of the user to select any one of them. The competition to attract many users or actors is always there. Users want more friends or relations suggested by social networks. And users are switching to adopt the one where he/she finds more relations. To this end link prediction became the core and heart of social network analysis. There is lots of work done in social network analysis towards link prediction, but some issues and challenges are there. In this work these issues and challenges are trying to solve out. Uncertainty of social network data is the main issue, and the fuzzy soft set is applied for it. Interval-valued fuzzy soft is another variation of fuzzy soft set, applied to deal with uncertain data. Markov model is used for user’s behavior prediction in the web, the same concept adopted here to find relations. Genetic algorithm-based approach considers the social graph structure into account to predict relation. Scalability is another issue; all the proposed techniques are scalable techniques of link prediction for social network analysis.

Although the proposed work is for social network data only, but it can be applied for some other applications like molecular biology, telecommunication and criminal investigation where link prediction is necessary. Adopting appropriate link prediction techniques, social networks can enhance their performance and link many people. The network, which links many people will become more popular and win the competition. Experiments are done and proposed work is compared with existing technique. It is observed that the proposed work is better than the previous approaches.