Understanding and Prediction of User Behavior in Online Social Networks: the Case of GitHub

Abstract

Along with the continuous development of Internet technology, Online Social Networks (OSNs) have gradually become the most popular platforms for content creation, information sharing and communications between users on the Internet. Understanding and prediction of user behavior in OSNs are essentially valuable. In the past few decades, machine learning is widely used and has become incredibly powerful in user behavior prediction. However, few researchers have considered the combination of machine learning and network analysis, especially using the hidden network information as features for prediction problem. In this thesis, we propose a novel method for user activity prediction by using machine learning algorithms as well as network properties from Github. The prediction is based on previous activities of not only the user him/herself but also his/her neighbors on GitHub. The results of prediction and performance evaluation demonstrate that the neighbor activity information from both one-layer and two-dimension networks of GitHub indeed can help to improve the performance of predicting a user’s active level. Additionally, the massive analysis of how our methods can help to improve the prediction accuracy is given from both the network and time series analysis perspective.