The Research Of Sentiment Analysis And Prediction On Microblog

With the development of mobile Internet,it become easier to get information.We produces information by sharing information?commenting and other operations.This information contains a large number of personal views.And it could create inestimable commercial and social value.On the basis of microblog,my paper analyzes the users' emmotion,and predicts its influences.In this paper,it analyzes the features of the microblog.First,I propose a sentiment analysis method based on filed dictionary which is builded by training a lot of basic data.After extending the emotional lexicon and building a number of rules about sentiment analysis,the accuracy of classification imporves.Second,this paper has studied the SVM and Naive Bayesian Model.Through using the TF-IDF method of determining the weights and adding parts of the fixed features,the value of the characteristic dimension reduces and the correct classification rate increases.At Last,this paper propose a improved Levenberg-Marquardt algorithm by studying steepest descend method,momentum method and Newton method.The paper points out that the error function inherited nature of the image of a quadratic function and its descent direction toward its corresponding axis of symmetry(maximum or minimum direction)of a quadratic function.After judging the the second order partial derivatives of positive and negtive,the paper increases the rate of change of the damping factor,and improve its convergence speed.The experiments prove it and the algorithm's prediction also has a good effect on microblog.