Chinese Microblog Sentiment Analysis Based On Deep Learning

Network sentiment analysis is important in the field of opinion mining and natural language processing.Through the analysis of complex text on the Internet,netizens' thought dynamic or emotional preference can be mined and further used in opinion poll,opinion monitoring,opinion guidance,commodity recommendation,decision support,market positioning,and so on.It will contribute to social stability and facilitate people's lives.At present,a large number of scholars have carried out related researches,having made some achievements.However,most of the existing methods have referred to traditional text analysis methods,resulting in data sparseness and restriction from the choices of rules and features.In addition,they haven't fully used the characteristics of network texts,unable to effectively tap the potential information.Based on the above considerations,this paper treats the sentiment analysis of Chinese microblog as a classification problem and a test bed,uses convolutional neural network in deep learning to analyze network texts.The main work includes the following three aspects:(1)Chinese microblog has free form,nonstandard syntax and short content.In view of this,a context-based convolutional neural network model is proposed in this paper.Firstly,use the interactive context to link current microblog with the original and the next-level ones,so as to expand its content under different topics,and map it into a low-dimensional dense vector.Then use the convolutional neural network model for the extraction and combination of text features.After training the model layer by layer,microblogs are represented as vectors,and sentiment categories are achieved with the classification function.The experimental results show that our method is better than the benchmark in the overall accuracy and FI-score.(2)Chinese microblog usually omits the topic.Existing methods haven't fully utilized users' habits and preferences and the possible similarities among microblogs under the same topic.This paper proposes a convolutional neural network model which combines user and topic information.It constructs three low-dimensional dense vectors to represent user,topic and microblog text respectively,and uses matrix vector multiplication and vector concatenation to integrate them.The processes of feature extraction and sentiment classification are also implemented using the convolutional neural network model.Experimental results show that this method can mine view information more effectively and improve the analysis performance.(3)The distribution of microblogs of different viewpoints under the same topic is unbalanced,leading to the under-learning problem and the drop of analysis results.So a convolutional neural network model based on over-sampling is proposed in this paper.From the perspective of data,the training data scales of positive and negative classes are extended moderately based on the over-sampling technique.Then the model in(2)are used in the sentiment analysis of Chinese microblog.Experimental results show that this method can solve the problem of data unbalance to a certain extent,and further improve the classification accuracy on our dataset.