Research On Chinese Micro-blog Sentiment Analysis With Deep Learning

With the coming of the information age,more and more people are using the internet,data has become an important resource in business activities,massive date has a lot of valuable information.Micro-blog,as a new social media,has been widely accepted by the public.In terms of the content,the topics of micro-blog are varied and high participation.In terms of the user,micro-blog can meet user's personalization requirements,real-time communication and emotional pouring.Its ability to disseminate information is strong,and produces massive data every day.Through the study of the sentiment analysis on Chinese micro-blog,it can play an important role in network opinion Monitoring,marketing management and feedback from policiesAt present,most of the existing research on sentiment analysis is based on traditional machine learning methods and rule-dictionary methods.However,there are many shortcomings,such as high cost of marking and poor transplantability.What's more,the Chinese micro-blog has the characteristics of short content,particular grammar and new vocabulary,the existing methods can not process micro-blog data well.To solve the above problems,this thesis will use deep learning methods,and research from two aspects of opinion target extraction and sentiment classification,the details are as follows:1.A method of opinion target extraction for Chinese micro-blog based on long short term memory network is proposed.First,through the establishment of a bidirectional long short term memory network model to translate the task of extracting opinion target into a task of sequence labeling;Then,the model calculates the probability distribution of attention by using attention mechanism,improves the expressive ability of sequence;Finally,the conditional random fields algorithm is used to plan the optimal tagging path of text sequence,so as to improve the accuracy of opinion target extraction.2.A method of sentiment classification for Chinese micro-blog based on convolutional neural network is proposed.First,through the establishment of a convolutional neural network as a classification model;Then,the model calculates the probability distribution of attention by using attention mechanism,highlights the contribution of the key content,reduce the loss of information;Finally,the highway algorithm is used to control the information flow,so as to solve the difficulties in training,improve the quality of the training process and the accuracy of sentiment classification.3.Based on the above research results,a deep learning environment is constructed and collect the Chinese micro-blog corpus for experiment.The experimental results prove the validity of methods proposed in this thesis,At the same time,summarizes the methods and puts forward the future research plan.