Public Opinion Analysis Based On Sentiment Polarity And Structural Balance

With the development of computers,the number of Chinese netizens is increasing.Netizens are more and more willing to establish their own communication circles on social platforms,such as Sina Micro-blog,Renren,Tencent Micro-blog,Facebook and so on.Because of the openness,randomness and real-time of these social platforms,netizens are willing to publish texts on such platforms,to convey their emotions and express their views and ideas about something.With the popularity of the social platform,the number of netizens has increased,and a large number of data are produced on these platforms.It is unrealistic to analyze these data manually.So,how to obtain useful information from these text data effectively has become a research hotspot.Public opinion analysis,as one of the important research points,is of great significance to understanding the latest hot spot of public opinion and mastering the trend of public opinion.However,most of the traditional public opinion analysis methods only aim at text information and do not take into account structural information.Therefore,this thesis proposes a public opinion analysis method which fusing structure and text information.The main tasks are as follows:(1)Considering that sentiment analysis can extract useful information from a large number of texts effectively,this thesis proposes a hybrid sentiment analysis approach SAFCM(Sentiment Analysis For Chinese Micro-blog)to solve the problem of sentiment analysis in Chinese Micro-blog.This hybrid approach combines the basic technology of Natural Language Processing and machine learning to determine the semantic orientation of Chinese Micro-blog.Firstly,builds a mixed lexicon based on the features of Chinese Micro-blog,and mainly combines basic vocabulary,negative word,degree adverbs and the emoticons.Considering the randomness of Micro-blog text,there will be a lot of unknown words in Micro-blog text.It is not feasible to analyze text based on these words.Therefore,the approach combining How Net and PMI(Pointwise Mutual Information)is used to extend the lexicon.Secondly,the vector representation of each word in Chinese corpus is obtained by using the word2 vec.Finally,the SVM(Support Vector Machine)approach is used to divide the corpus into positive and negative two categories.The approach is verified on multiple data sets,and the experimental results show that the hybrid approach proposed in this thesis is effective.(2)A approach named MDPSO(Micro-blog Data Particle Swarm Optimization)to study the structural balance of the network is proposed in this thesis.Firstly,models the problem of social network structure balance as a mathematical optimization problem.Secondly,designs a new energy function according to the principle of structural balance.Finally,redefines the updating rule of particle velocity and position from discrete point ofview to solve discrete optimization problems in view of the fact that standard particle swarm optimization algorithm can not deal with discrete problems will.Experiments on real data sets prove that MDPSO is effective.It can not only analyze the structural balance of the network,but also find out the unbalanced edge in the network.(3)This thesis proposes a public opinion analysis method POA-SP-SB(Public Opinion Analysis based on Sentiment Polarity and Structural Balance)combining content one and content two.Firstly,obtains the emotional polarity of Micro-blog text and analyzes the social mood of public opinion events based on SAFCM,and the signed network is constructed according to the result of emotional polarity.Then,analyzes the structural balance according to MDPSO,finds out the unbalanced edges in the network,and finds the key nodes in the network using the improved Page Rank algorithm.Relevant government departments or crisis public relations can take corresponding measures to regulate the guidance of public opinion according to the social mood,unbalanced edge and key nodes.Finally,in order to test and verify the effectiveness of the SAFCM,MDPSO and POA-SP-SB,the relevant experiments and analysis are carried out on the real data sets.Experimental results show that the proposed method in this thesis can analyze text emotion well,and find important nodes and edges of public opinion events.Through these nodes operation,related personnel can guide public opinion to develop in a good direction.