Research On Sentiment Analysis Of Chinese Microblogs Based On Two-layer Convolutional Neural Network And Extended Feature Matrix

With the development of mobile Internet,Weibo platform has gradually become one of the important platforms for daily communication and communication of Chinese netizens.As an important part of the Chinese social network,the social and media attributes of the Weibo platform enable users to understand social hot events in real time and express their opinions and opinions.In terms of public opinion communication,due to the large base of Weibo users,both positive and negative speech can be quickly spread on the Weibo platform.Whether it is for social events or commodity reviews,Weibo's public opinion can often influence or even change the direction of events.Therefore,how to quickly dig out the public opinion tendency of Weibo users under a single Weibo topic,provide decision-making reference for government and enterprises,effectively guide social public opinion,and has strong Realistic and economic significance.The traditional method of sentiment analysis,the main method is to build an emotional dictionary based on linguistics.However,the establishment and maintenance of a language dictionary often takes a lot of time.In order to adapt to the sparse data in Chinese microblog sentiment analysis tasks,neglecting the expressions and word features in microblog texts,in recent years,the research on text classification algorithms based on machine learning methods has become more and more in-depth.This paper first compares different shallow learning models,and then proposes a micro-blog sentiment analysis algorithm Extended-Dual-CNN that combines two-layer convolutional neural networks and extended feature matrices to try to solve the microblog sentiment analysis problem in the field of deep learning.This paper compares the effects of shallow learning networks on the micro-blog sentiment analysis tasks under different generated word vector models.Based on this research,a microblog sentiment analysis algorithm based on double-layer convolutional neural network and extended feature matrix is proposed.Specifically,first,through the One-Hot Encoding and Word2 Vec models,the word vector is generated by the Weibo statement and input as a feature vector into the shallow learning model such as Naive Bayes,Maximum Entropy Model and Support Vector Machine,and the shallow layer is compared.The advantages and disadvantages of the learning models,as well as the influence of the word vector model on the sentiment analysis of the shallow learning model,were obtained as the best model for the microblog emotional analysis task experiment.Then,the Extended-Dual-CNN model is proposed to establish an extended feature matrix for various word features such as Weibo expressions,affirmative or negative words,and punctuation marks expressing emotions commonly used by Weibo users.Then,the features of the splicing of the word vector and the extended feature matrix are respectively input into two layers of the convolutional neural network using static and non-static text,and finally the sentiment classification result is obtained.Through the comparison experiments on COAE2014 task 4,the Extended-Dual-CNN algorithm achieved a classification accuracy of 93.35%.Compared with traditional machine learning algorithms such as single-layer convolutional neural network algorithm and SVM,Extended-Dual-CNN model has obvious advantages.