Link Analysis In Complex Networks And Social Media Prediction

Complexity science is regarded as the science of the twenty-first century. It takes widespread complex systems as research objects. As the topological abstraction of complex systems, networks has become a powerful theoretical tool to characterize and analyze the relation between objects within complex systems. With the arrival of Big Data Era,massive data-driven research turns into a new approach to understand the complex systems. In this paper, we work on both theoretical and practical problems related to network modeling, analysis and forecasting, which includes the following four aspects.First, combining the power-law degree distribution with community structure of the real networks, we propose a link prediction method based on degree-corrected stochastic block model, in which the probability of nodes connection depends not only on the blocks to which they belong, but also on the degree distribution within the group. Experimental results show that the prediction accuracy of this method is higher than both the method based on traditional stochastic block model and classic local similarity method.Second, taking advantage of asymmetry of interactions in directed networks, we extend some representative link prediction methods to directed version including local similarity indices and the method based on stochastic block model. Experimental results show that compared with other adapted methods and Bi-fan predictor, the method based on stochastic block model performs much better in both prediction accuracy and stability.While preferential attachment index has limited prediction ability in undirected networks,it outperforms other local similarity indices in directed networks.Third, we present a theoretical model to analyze the team creativity and show how to measure creative performance of teams through dynamic semantic social network analysis. We establish email communication network for Collaborative Chronic Care Network members, and analyze how the communication pattern and expression affect the team creativity. Experimental results show that the more actors are involved, more e-mails are exchanged, and more outspoken language is used, the more successful of the team. And group density decrease over time, while team members become more focus.Fourth, we discuss the possibility to predict financial market movement with social media. We measure the public emotion and opinion through twitter buzz, and use correlation analysis and Granger causality test to analyze whether they are correlated with stock market trend and asset value changes. Experimental results show that certain emotional tweet percentage significantly negatively correlated with stock market indicators,and the number of tweets containing keywords dollar, oil, gold is positively correlated with the movement of stock market indicators, crude oil price and currency exchange rate respectively. Moreover, the result of Granger causality test show that the above public opinion time series Granger-cause the changes in financial market valuation. This work is reported in Sina Finance(see appendix).