Using Deep Learning Method For Chinese Microblog Sentiment Analysis Research

At present,microblog is a very popular social networking tool.People are more and more inclined to share what they see and hear,as well as comment on hot events,etc.These comments often contain commentators' rich emotions,which indicate the commenters' opinions.By analyzing the sentimental tendencies of these microblogging reviews,we can obtain rich information,which can be useful for user analysis,public opinion analysis,and other applications.The object of this paper is Chinese microblog.Compared with English text,Chinese text has its unique language features and word construction methods.First,Chinese word segmentation algorithm was studied for Chinese text.There are some characteristics of Chinese microblog,such as high occurrence frequency of new words and semantic diversity of word combinations.By analyzing and comparing the commonly used word segmentation tools,BosonNLP solves the problems of new word recognition and combination ambiguity through point mutual information.This paper uses Boson NLP to deal with Chinese microblog word segmentation to obtain better segmentation effect.After studying the commonly used text representation method,finally,this paper chose to use Google's Word2 vec to train the word vector that containing contextual semantic information.Based on the previous ideas,considering the influence of degree adverbs on emotional expression,we designed a method for adjusting the emotional weight of word vectors.According to the rules of the HowNet sentiment dictionary,different degrees of adverbs are given different weights,highlighting the influence of degree adverbs on emotions.In order to verify the effectiveness of the sentiment word vector adjusted by emotional weights on sentiment classification,the emotional word vectors and the word vectors obtained by Word2 vec were used to learn the features of the LSTM model and used in the microblog sentiment classification.Experimental results showed that the word vectors were adjusting emotional weights improves the accuracy of microblog's sentiment classification.At present,most of the sentiment analysis of Chinese microblog is divided into two categories: positive and negative,but people's emotional expression is diverse.For the problem of the rich emotion expression of microblog,NLPCC2014 has provided microblog data with multiple types of emotional tags,this paper uses LSTM model to learn the characteristics of microblog data.And use Softmax to calculate the probability of multi-classification,and finally achieved the three classes and four classes of Chinese microblog sentiment.