Sentiment Analysis Of Chinese Reviews Based On Domain Dictionary And Machine Learning Methods

With the rapid development of the Internet,online platforms represented by weibo,BBS and e-commerce are on the rise.People are becoming more and more accustomed to publishing these opinions on the hot topics and the use of products on these platforms.This phenomenon has led to an explosion of comment data.In these user comments,there are lots of valuable emotional information for businessmen,sellers and governments.How to extract and apply these emotional information make sentiment analysis emerge as the times require.Sentiment analysis based on dictionary is the most basic method for emotion analysis.However,there is no suitable dictionary for specfic fields,and the scoring rules need to be optimized.So,this paper constructs an emotional dictionary based on the SO-PMI algorithm and puts forward a set of scoring rules based on Chinese grammatical structure.The preprocessed data is calculated with the basic dictionaries,and the emotional score is obtained according to the scoring rule,so as to judge the sentence's emotional polarity.The experimental result shows that the proposed analysis method based on emotion words has high accuracy.In Chinese text,context is complex,and there exists a phenomenon of polysemy.It makes the sentiment analysis method based on emotion dictionary produce some error.As a high performance method for machine learning,deep learning can be more expressive in the complex Chinese text,and such a performance is based on the premise of constructing excellent Chinese word vectors.Based on the wiki-chinese data set,this paper constructs a set of word vectors applicable to all Chinese using the tool of Word2 Vec.In the comparison experiment,the proposed Chinese word vector construction method has a good performance.After obtaining the word vector with excellent performance,this paper,based on the third party library of Python,builds a sentiment analysis model based on the LSTM recurrent neural network.Using the activation function Sigmoid as the output layer,cross entropy is used as the loss function.When the word vector is trained,it is used as model input to solve the sentiment analysis task of sentence level review text.The accuracy rate is better than the traditional methods of sentiment analysis,and the result is improved.