Research On Emotional Classification Method Of E-Commerce Online Comments

With the rapid development of e-commerce,various e-commerce platforms have accumulated a large amount of online consumer comment data,which contains very high commercial value.It is of great significance for businesses and consumers to analyze their content,especially to automatically identify the praise and derogation of the comment content and to classify the comment content emotionally.However,in the face of massive online comment data,it is obvious that manual processing alone can not meet the needs,which makes online comment text content automatic classification technology become very important.In this study,the existing text categorization technology is applied to emotional classification of online reviews in e-commerce,and the advantages and disadvantages of various methods in solving this problem are compared through experiments.In the research,the main work is as follows:Firstly,the web crawler technology is used to collect the real e-commerce website online comment text data and score data,then the word vector model is established by Word2vec tool after the data is pretreated,and the emotion dictionary for e-commerce online comment text is established,and the appropriate feature extraction method is selected for feature extraction,and then the feature extraction is carried out separately.Using dictionary-based classification methods,machine learning-based classification methods such as K-nearest neighbor,decision tree,naive Bayesian,support vector machine,and deep learning classification methods such as convolution neural network and long-term and short-term memory model,the online customer review text data are classified.Finally,the accuracy,recall and F-measure indexes of various classification methods are compared,so as to classify online customer review text data.The advantages and disadvantages of various classification methods are compared and analyzed.The experimental results show that the method based on emotional dictionary is obviously inferior to other classification methods in terms of each index,It can be seen that although the implementation of this method is simple,the classification effect is unsatisfactory,and it has a great dependence on the quality of dictionaries.Secondly,the classification effect of each method based on machine learning is quite different,the performance of decision tree and K-nearest neighbor is not as good as Naive Bayes,and support vector machine is the best in machine learning method.The two methods based on in-depth learning perform well and belong to the best classification effect of the three methods,and the convolutional neural network is the best one among all the methods.In addition,the dimension of word vector and feature selection methods have great influence on the classification effect of machine learning and deep learning methods.This study designed different experiments to compare these two factors.It was found that for machine learning methods,the mean of word vector is the best choice for feature selection,and for deep learning methods,the information gain method is the best choice for feature selection.