Sentiment Analysis Based On Women Clothing E-commerce Reviews

With the advent of the web2.0 era and the rapid development of e-commerce industry,a large amount of online commodity reviews on the network contains huge commercial value.This paper uses a variety of analysis and modeling methods to mine and analyze a large number of women's clothing e-commerce review texts,which not only provide reference information for other users,but also provide feedback information for businesses.At the same time,the e-commerce platform can use this in personalized recommendation.The main work carried out in this paper are as follows.First of all,this paper seek out high-frequency words or phrases of these texts so as to observe the characteristics of users' reviews;Second,this paper identify the emotion words and their corresponding negative words and degree adverbs in the text by matching sentiment lexicon named SentiWordNet,and calculate the emotion value of the emotion words by rules,and get the sentiment polarity value of the sentence and the whole comment text by calculation,so as to judge the emotion tendency.Third,after preprocessing the data such as cleaning,normalization,deleting stop words,word stem extraction and word segmentation,this paper consider the different importance of words to sentiment analysis,the tf-idf-based encoding method will map the comment text to the vector space,and the Naive Bayes,Logistic Regression and LightGBM algorithm will be used to train classification models.Fourth,a deep learning classification model is established by using the Recurrent Neural Networks(RNNs)algorithm.Fifth,this paper use existing data set to test the models we established before and make a judgment of the classification effect of each model according to experimental results.The result shows that the model trained by LightGBM algorithm has the best classification effect.