Key Technologies Research On Information Diffusion Analysis For Public Opinion In Online Social Networks

With the development of computer science and information science, the Internet has entered the Web2.0 era. For the openendedness and the freedom of the information diffusion and dissemination process, the activity of the information dissemination in online social networks has reached an unprecedented level. Studying the information diffusion mechanism in the online social networks will greatly help us to understand the network structure and the behaviors of the user group. Besides, it can also help the government to guide the online social opinions. Hence, it has important theoretical and realistic significance.The analysis methods and computing models used in online social networks and traditional social networks are very different, because online social networks have high scalability, dynamic evolution and heterogeneity. In this thesis, with the motivation of public opinion applications and based on the related work, we carry out our work on the topic diversity, dynamic evolution, data noisy and the nonlinear of the variation of the heat in the topics. The contributions in this paper as listed as follows.1. Since the social relationships are rich semantic and varies over time, this thesis conducts research on the social relationships extracting problem and proposes a topic and time sensitive model to compute the link strength in micro-blogging.A great proportion of the information diffusion in the web 2.0 eras is based on the social relationships in the social networks. Messages flow among the users in the e-texture format, and the texture is topic related, so the links in the social networks are rich semantics. Moreover, social network is a dynamic network, and the link strength between individuals is changing over time. Most of the existed methods do not consider the two factors described above, which will affect the accuracy of the information diffusion model. To get the accurate diffusion probability of the users, this thesis takes both the topic and the time of the communication history into consideration, and proposes a topic and time sensitive model for link strength predicting in time-aware online social network. Comprehensive experimental studies on real world micro-blog data set show that our approach outperforms existing ones and well matches the practice, and the performance enhancement is about one third quantitatively.2. Since user-generated contents(UCG) in social networks are usually short and noisy, this thesis studies the interest related information extraction problem of micro-blogging users, and develops a new social characteristics based approach to detect users’ interests in micro-blogging.Micro-blogging texture has the characteristics such as short content, bad normative and many new popular words, which makes it more difficult for texture mining in the online micro-blogging. Based on the micro-blogging texture, we try to find the users’ interests, and develop a new approach to extract the key words representing the users’ interests. The key words of users’ interests are a collection of words that represent the users’ interests. In the method, we first collect all the micro-blogs of the users to form a long document, then we use the characteristics of the micro-blogs, such as the retweet number, the tags and the releasing time, to design the candidate list of the characteristics, next, with the help of the classification algorithms in machine learning, we are able to extract the key words to represent the users’ interest and furthermore, we are able to find the users’ interests. Finally,we use the classification problem of the micro-blogging users to test the performance of our method. Experimental results show that: our method has a high accuracy about 89.79% in finding the users’ interests, which is about 20% higher than traditional methods. And in the user-interest keyword-based classification test, because it is able to filter the insignificant words, its classification accuracy can reach as high as 91.26%, which gets about 18.7% improvements than traditional methods.3. This thesis studies the information diffusion mechanism in the micro-groups,and proposes a new method to predict the behaviors of the information dissemination, which takes both connection strength and users’ interests into consideration.Micro-group is a new social network application. It is similar to the “QQ group” and absorbs the characteristics in BBS and micro-blogs. In micro-groups,the interactions among the users are affected by both the connection strength and the users’ interests. So, in this thesis, we propose a new model, which considers the two factors, to dynamically predict the users’ behaviors. Based on the historical behaviors of users and content similarity, a method mining personal interest has been proposed. Then, we estimate the topic influence network based on the reply number,reply frequency and the content of the posts. Then, we propose an automatic restart random walk model to sort the topic interest measure and to predict the users’ behaviors. The LDA based method used in the content similarity calculation is able to overcome the high-dimension drawbacks in the word based texture representation model. Besides, it takes full advantage of the characteristic that texture in the successive time slices has the same context to update the topic content timely.Experimental results show that the prediction accuracy of our method is about35% higher than traditional methods.4. Considering the transmission capacity difference of individuals, this thesis studies the evaluation mechanism of the public opinion topics spreading on the social networks, and proposes a Neural Network based model to predict the hot trend of the topics.Based on gray correlation factors, a quantitative method to mine key factors which affect hot trend is proposed. Moreover, For lack of neural networks that exist in local optima, we propose an improved model based on genetic algorithm. The results on Tianya Forum dataset show that the proposed model can improve the prediction precision of network hotspot topic change trend compared with other prediction models and can accurately describe the hot topic change trend.In conclusion, this paper studies the characteristics of social network data,delves into key technologies of public opinion information diffusion in the online social networks, the evaluation mechanism of the public opinion in the online social networks, and proposes several novel algorithms with lots of practical experiments.This paper has important theoretical and realistic significance for social network analysis and modeling information diffusion. Meanwhile, it exhibits significant applicable value in lots of applications like personalization recommendation and Social Media Marketing.