Research On Sentiment Analysis And User's Personality Prediction Based On Microblog

In recent years,the domestic web portals continue to explore and deepen the application of micro-blog.Thus micro-blog platform gathers a large number of public opinion information,which contains a large number of emotional information.Text mining on the micro-blog platform can not only get the sentiment polarity of posts,but also build the corresponding user porfiling according to the user's language style and the registered information such as gender,age,region,educational background and other information of the user.In the aspects of user profiling,personality analysis is a very important and difficult problem.To identify sentiment polarity of the micro-blog posts correctly will help the government to collect public opinion and help businesses to adjust marketing strategy.Meanwhile,user profiling plays an important role on improving the user experience and improving economic efficiency in all fields.Furthermore,sentiment analysis and user profiling construction are the key researches among the cross domains of linguistics,text mining and psychology.They also have important scientific and application values in exploring the hidden sentiment information contained in micro-blog and to construct a reasonable user profiling.The main research works of this thesis are as follows:1)This thesis proposes a classification method of short text sentiment polarity based on LSTM(Long-Short Term Memory).Firstly,this thesis selects the method of text feature with good performance,and determines the number of hidden layers and the number of neurons of each layer.Then the classifier is trained by a larger number of micro-blog data,and finally the performance of this model is compared with other machine learning algorithms such as support vector machine and bayesian network.2)This thesis proposes a user personality classification method based on KNN algorithm.Firstly,based on the popular psychological model and personality analysis knowledge base,the personality scoring model is constructed to score the personality of users.Then,the appropriate threshold is selected to classify the user's personality.Finally,the classified user text data is used as the training set of the KNN classifier,and the K value which makes the KNN classifier performance best is found through experiments.3)In the aspect of user profiling,firstly,this thesis proposes a method of user genderprediction based on model fusion,and quantifies sentiment values of short texts according to the predict probabilities of sentiment classifier.Then,the degree of sentiment fluctuation of each user is calculated.Finally,taking Pearson correlation as the representative,the relationship between the sentiment,personality and user attributes are reasonably analyzed.In sentiment analysis section,the average of F-measure of the proposed model can reach 91.1%;In personality predicting section,the average of F-measure of the proposed model can reach 90.0%;In the section of user profiling construction,the relationship among user's sentiment fluctuation,personality,gender,time and other factors can be calculated reasonably.To conclude,the methods mentioned in this study are feasible.The problems that remain and the future study plans are introduced at the end of this thesis.