Sentiment Analysis Based On Dictionary And Neural Network

In recent years,with the development of information science and technology,artificial intelligence has caused the trend of the information times again and again.At present,sentiment analysis is already one of the most active technologies in the field of artificial intelligence.As of June 2019,the number of Internet users in China has reached 854 million.Next,with the gradual maturity of social networks and the development and popularization of technologies such as mobile phones,many users have become accustomed to publishing comments on the Internet.There are commercial and economic values that cannot be ignored in these comments.For example,Weibo comments,Weibo managers can analyze the comment data to monitor public opinion;shopping sites,website managers can summarize historical comment data analysis to formulate the company's business strategy;car forum,consumers can make decisions by taking relevant data analysis Your own purchase intention.This makes sentiment analysis widely used in today's information society.The existing text sentiment analysis methods mainly include: sentiment dictionary-based methods,machine learning-based methods and deep learning-based methods.The dictionarybased sentiment analysis method relies heavily on the detail and coverage of the sentiment dictionary,while the machine learning-based one relies on manual construction to select features and annotate relevant information.In recent years,with the rapid development of deep learning,more and more researchers have devoted themselves to the study of sentiment analysis methods of deep learning to improve the structure of related network models in order to achieve better results.First of all,in the first chapter,the dissertation introduces the current situation of sentiment analysis technology in detail,and gives the main research content and innovations of this dissertation.Secondly,the text preprocessing,feature extraction,sentiment analysis process,etc.were elaborated.Then,the specific research contents of the main two chapters of this dissertation are as follows:(1)Aiming at the phenomenon that many dictionaries in China do not have the sentiment value attribute,which makes it impossible to calculate sentiment values for sentences or documents.First of all,chapter 3 proposes a method of labeling sentiment values for unlabeled sentiment dictionaries using word similarity.That is,by calculating the semantic similarity between the word without sentiment value and the seed word,and then multiplying with the sentiment weight of the seed word,the label value is obtained.Then,the sentiment score of the sentence is calculated according to the marked sentiment dictionary,thereby implementing sentiment analysis based on the dictionary.Finally,in the third chapter,the experiment proves the effectiveness of this method,which has a certain effect on the construction and improvement of specific domain dictionaries.(2)At present,there are several classic neural network models for text processing at home and abroad.Chapter 4 proposes a neural network sentiment classification model.This model combines CNN and two-way LSTM network,uses the output of CNN as the input of two-way LSTM,and then uses the output of two-way LSTM as the input of the self-attention mechanism layer.And compared with the previous various neural network models on Auto and Amazon Musical Instruments review text classification experiment data,the results show the effectiveness of this neural network sentiment analysis model in sentiment analysis.Finally,it summarizes the research work and research content of this dissertation,and gives the thinking direction and research content of the next step in the future.