Research On User Behavior Mining And Application For Social Network

Compared with general counterparts, Social Networks involving enormous social individ-uals pay more attention on the social relationships between individuals, as well as the interac-tions and influences existing in the individuals.Social Network Analysis not only cares about the ingredients of network forming and network’s topological properties, but also focuses on the characteristics of individuals’rela-tionships and interactions. Among these studies, mining user’s behavior in social networks has become a hot branch. Thanks to its abundant application values and business opportu-nities on user experience enhancement, client relationship management, marketing, intelligent search and etc., mining user’s behavior has attracted more and more attentions in academic and industry societies. Although many research works on user behavior in social networks have emerged, yet the deficiencies existed in these works can not been neglected. On the other hand, with the help of computers, many researchers in computer science often emphasized borrow-ing the previous conclusions during processing the big data of online social networks, howbeit ignored implementation with different scenarios. For example, some scholars have claimed that Social Influence and Homophily are the two important forces driving the individuals in the network to accelerate information diffusion in social networks. However, there are not sufficient quantification of these two forces, especially lacking appropriate transformation in different application settings. Moreover, some scholars on modeling user behaviors paid more attentions to refine the methodologies rather than fitting their models with the real settings, in-cluding further improving the performance of model prediction to justify the application values of the models.Addressing these deficiencies, in this paper we focus the users in several typical social networks as well as their behavior. By extensive mining and analysis, we try to uncover the key ingredients influencing the individuals in the networks. Meanwhile, we take efforts to model different types of user behavior for the objectives of real applications. Besides adopt-ing reasonable and effective methodologies for the concrete problems, we also emphasize the empirical studies and making related conclusions to justify the effectiveness of our findings on different real applications. Specifically, our major work and conclusions include the following points.1. The networks consisting of the users in online forums are the rudiments of social net-works, we propose a method to quantify the Collective Attention of users in such net- works. We find some key factors influencing collective attention in the forum which did not attracted sufficient concentration. Based on our findings, we design a model to char-acterize the evolution of collective attention that is verified to have good performance on predicting the future trend of a newly posted article in the forum. Thus our model can be applied in tracking and predicting public opinion on Web.2. Due to the lack of social relationships between individuals in the forum dataset, the re-lated conclusions neglect social influence. Thus we step further into a more structured network with explicit social ties, Scientific Collaboration Network (SCN). In such net-work, we focus the user behavior of topic following that can be viewed as a process of information diffusion in social networks. We not only quantify social influence and homophily in topic following, but also justify their mixed effects on driving topic fol-lowing by empirical studies. Furthermore, we build a Multiple Logistical Regression model to predict the probability that a given user will follow a certain topic. The eval-uation experiments prove our model’s value on some real applications such as call for paper/participation and special issue proposal of workshop/conference.3. Towards enhancing the prediction performance of our model on user’s topic following, we extend the homogeneous SCN into a Heterogeneous Information Network (HIN) which has more complex structure and informative relationships. Based on the meta path proposed by the predecessors, we add the neighbor’s tendency for a certain topic when predicting the probability of a user’s topic following. Thanks to involve more key features, our new model can achieve higher accuracy on both micro-level prediction and macro-level prediction.4. The emergency of microblogging system has announced a new phase of social media. Studying user behavior in microblogging society deserves investigation and owns broad application values. Facing the most famous Chinese microblogging service, i.e., Sina Weibo, we keep eyes on the behavior of tagging Weibo users that can be categorized into social tagging. Most previous of social tagging works addressed tagging Web objects, e.g., photos and URLs, seldom focused tagging a person. In this paper, we borrow the homophily existing in Weibo users to solve cold start problem in tag recommendation. Furthermore, we try to remove semantic redundancy in tags by a semantic network con-structed from the data of online encyclopedia web-sites. Our experiments justify that our algorithm of recommending tags for Weibo users can be utilized to accurately infer user’s profiles.