Use of Classification Models for Optimize Link Prediction in the Ego-Social Networks

Social propositional systems are a new generation of systems that use the social network as a user modeling platform to maximize some challenges by using rich interactive data volumes. To make Social networking sites offer new friends to registered users based on local graph features. The main purpose of the link prediction problem on social networks is to suggest a list of users to a particular user that they will probably be communicating in the future. In this research, a prediction method for the link is presented based on the characteristics of classification models. Here, the prediction problem of the link is transformed into a classifying problem with two positive and negative classes, where the positive class represents the relationship and the negative class indicates that the two users are not communicating. Three classical classes DT, NN and NB are used for classification work. To create the dataset, the features of credibility, optimism, number of neighbors, the number of paths of different lengths, the number of shared tweets, the number of internal and external axes are used. Although self-centered grids do not have much overlap in the rings, experiments show that the consideration of self-directed pathways significantly improves predictive performance. The DT classification has recorded the best performance with an average accuracy of 99.85%.