Research On Sentiment Classification System Based On Car Reviews

With the rapid development of Internet and information technology,social media have gradually become an important channel for people to disseminate their ideas,and an open platform to provide rich content of social trends for researchers,such as car topic review community.As the emergence of automobile network reputation,enterprises can find the overall needs of users and identify the implicit risk of the car,and it can grasp the user psychology and build a risk warning system to provide stable protection.But as a new feedback method,collecting the effective information of the massive data quickly and accurately and identifying their emotional tendencies is the main factor.In this paper,we proposed a framework of sentiment classification system based on social media,and use deep neutral network model to carry on the emotional tendency of the comment to improve the utilization rate of companies on car reviews.This paper mainly completed the following work:(1)Proposed a multi-input CNN model.For the car review that the length is less than 140 characters,the model uses pinyin sequences to characterize the feature at the character level and Word Embedding technology to express the car comment.It can enhance the model's feature presentation ability by using convolution operation and kmax down-sampling method.(2)Used LSTM to model the long comment text sequence bidirectionally.For the car review that the length is great than 140 characters,the model can make effective use of the long distance dependency information in the text sequence data,and achieve the sentiment classification of the long car reviews by extracting the keywords and context information in the text.(3)Based on the analysis of the current main media platforms,we proposed a framework of the automobile industry sentiment classification system which is suitable for the multi data sources.The system mainly includes data acquisition,data processing and sentiment classification modules.For the key issues in these modules,we focus on the introduction of car reviews extraction and aggregation plan from the multi data sources and the implementation of training and deployment plan of the deep learning models.