Research On User Emotion Analysis Method Based On Social Network

In the resent decade, online social networks achieved rapid development. Facebook was founded in 2004 and has more than one billion registered users. The Twitter, established in 2006 with five hundred million registered users, and the latest data released by the sina show that sina micro-blog has more than 560 million registered users. Network users and the information interaction and the interaction between the users left a various "footprint" on social networking sites, led directly to the arrival of Internet era of big data. People's behavior in the social space directly reflects its activities and emotion in the real world. The emotion research which is based on the subjective psychology experiment presents the changes to the social space emotion computing trend.There are two main directions on the study of micro-blog emotion classification, one is a kind of emotional dictionary method based on rules, another is the method based on machine learning. because of the micro blog this is shorter. There are sparse and the characteristics of high dimension problems on the method based on machine learning because the micro-blog is short. And the classification result is not very ideal. With the emergence of a variety of media forms, providing possibility to solve this problem. This article is based on the study and summary on the basis of existing research, in view of the differences performed by the image characteristics in different categories data set. Selecting the most optimal reflect characteristics of the combination of class differences presents a two-stage classification strategy of emotional polarity classification method. There are problems about the spread of the emotion in social network, which is that the user's emotion can be influenced by friends, the text studied and analyzed the probability factor graph model, and put it into social network users emotional prediction model. This model can put a variety of factors such as the user's attribute dependent, time dependent and friends depends in modeling process, to optimize the probability of makes forecasting category. The research content of this article mainly as follows:(1) Research the image characteristics, find out the optimal features combination which can distinguish polar category attributes. Putting forward a new strategy which is the combination of image characteristics and the text characteristic. Aiming at the shortcomings of the micro-blogging emotion classification method, we can do contrast experiments of different classification methods on the basis of same experiment corpus. Putting forward a kind of strategy which can combine na?ve-bayes and support the new classification which is based on two-step polarity of vector machine.(2) Study probability factor model. The user's attributes, such as text message, the location information, can be moded as a factor function. And the emotion affected by friends also can be moded as the exponential decay function; User's mood changes over time, this paper modeled as a markov-chain. In learning the parameters of the model, this paper adopts the Metropolis- Hastings algorithm for sample training data set.Finally, gain user data on the network as the experimental test data. And do analysis and experience about the two methods in this paper, the results shows that the proposed two methods are useful.