Measuring relationship strength in online social networks based on users' activities and profile information

Abstract

Using online social networks as a means of interchanging information and ideas is developing daily so that internet users spend too much time in social network websites. This fact has turned social networks into very rich sources of information for analyzing users' activities and relationships. In recent years, most research in analysis of social networks has dealt with the friendship network. Since the friendship network is binary, the analysis thereof is of no difference between two users who are very close friends or farther acquaintances. Therefore, the result of the analysis would be not real in many cases. In this paper, we tried to measure relationship strength based on users' activities and profile information. For this purpose, a Facebook application was developed so as to collect users' information. Then based on the users' answers to the questionnaire, we measured their relationship strength at four levels. After outlier removal and normalization of reminder data, some of the features were removed based on the Information Gain criterion. Then considering the collected data, two models of Decision Tree and multilayer perceptron were used so as to measure the users' relationship strength. Having applied 10-fold Cross Validation, the validity of the learned model was evaluated. Based on our experiments, multilayer perceptron could predict the users' relationship strength with 72% accuracy and the accuracy of decision tree with ID3 learning algorithm is a bit lower. Finally In two-class case, we compared MLP with SVM and found that MLP achieved 87% of classification accuracy. Based on accuracy, recall and precision we concluded that MLP can classify relationship strength better.