Research On Key Problem In Social Computing Based On Networked Data Analysis

With the rapid development of information technology, especially Cloud Computing, Internet of Things, Social Network and advanced information acquisition emerging, the Big Date era is coming. According to the data scale rapidly growth, social computing emerges as social science and interdisciplinary computational science.Social computing devotes to bridge the gap between virtual network and reality social using computer technology and social theory. It reveals trend and rule for social functioning according to network data analysis which can help people cognize and research social science problems. This dissertation researches on the key issues of social computing with the goal of analyze and study network community behaviors, network community detection, network community behavior dynamics and network community opinion evolution.Novelties of this dissertation are stated as follows:1. Compare Instant Message community behavior to micro-blog, the characters analysis of transmission range, privacy, instantaneity, User engagement, and session feature are given. Then, by analyzing real-life dataset, we found that user behavior which follows seasonal and paroxysmal obeys power law distribution at group level in IM system. The analysis reveals that human behavior in IM system is mainly driven by three elements:information contribution, community interactivity and community sluggishness. To further understand the mechanism of dynamic phenomena in IM community, Interest and Community Interaction based Human Dynamic Model (ICHM) is proposed, which adopts the three elements as attributes of dynamics. ICHM can generate the data follows power law distribution, and community interactivity can influence the power exponents of IM community. Simulation result is consistent with the dynamic feature which compared to the data issued from reality. Consequently, ICHM can offer reasonable explanation for the dynamic mechanism of human dynamics in social network.2. According to the features of exponential cutoff in traditional dynamic empirical study, and also with considering interactive features of session centric in instant message, a session based ICHM (SICHM) is proposed. This model takes individual session as the principle, and also uses session transfer probability Ptrans and session exit probability Pcancei to describe individuals’information publish behavior. According to the simulations, it can be concluded that the session driving based interaction follows a power law distribution with exponential cutoff. SICHM can generate the data with distribution of single or double power exponents by adjusting Ptrans, which can influences the position of exponential cutoff and power exponents. Simulation result is consistent with the empirical study which indicates that the feature of session is one of the factors to lead the phenomenon of power law distribution v/ith exponential cutoff in human behavior.3. To improve the efficiency of the community detection algorithm in the large directed networks, a sparse method is proposed. The sparse method compute the similarity between any two neighbors by unifying the co-citation relationship, propagation relationship and the bibliographic coupling relationship of them and use Minwise to improve the efficiency in the computation. Based on this sparse method, the local sparse algorithm LSDN is proposed. It makes the degree of the network obeys the power law after the sparsification. According to the simulations, LSDN can simplify directed network effectively, and it can ensure the accuracy of community detection with keeping the characteristics of whole network unchanged. So it can improve of the efficiency of community detection in the networks. Hence, the problem of considering shortage of expandability and low efficiency in traditional community detection algorithm is solved.’4. In traditional opinion model, the group uniformity pressure and individual driving force are not considered which impact the efficiency of large scale nodes analysis. Therefore, a decision offset based dynamics opinion model (D02M) is proposed. This model can establish state transfer and three kinds of opinion decision for opinion formation analysis with considering multiple factors which are internal expectation traction, external group uniformity pressure and herding effect. Then, improved upon the HK model base on D02M (HK-D02M) is proposed. Simulations results show the two models can simulate unification and polarization of opinion well. D02M and HK-D02M are more correspond to the characteristics of opinion formation and individual interaction with comparing Deffuant model and HK model respectively. Consequently, D02M and HK-D02M can offer reasonable explanation for the community opinion evolution, and also provide a theoretical model for analyzing the opinion formation mechanism in social network.