Research On Product Views Emotional Analysis Based On Word Vectors And Dictionaries

In recent years,the rapid development of the Internet has made information transmission extremely fast.Due to data sharing,e-commerce platforms have brought many opportunities and challenges.More and more people choose to buy goods on mainstream e-commerce platforms.Among them,the evaluation system of e-commerce platform is extremely important,which mainly guides users to give true evaluation after consumption to make the information of commodities more comprehensive.Through in-depth research on the product reviews,consumers can have a better understanding of the product and help them make purchase decisions.For the seller,it can make them understand the consumers' feelings and attitudes towards the goods and improve the deficiencies of the goods.The manual analysis of a large number of commodity comments is laborious and time-consuming.In this paper,the classification method of machine learning is used for data mining and analysis.In the classification method based on machine learning,different feature selection methods will lead to differences in emotion analysis results.In this paper,the feature representation based on neural network is studied.Firstly,to solve the problem of scarcity of Chinese corpus with subjective description in real situations,this paper adopts web crawler technology to acquire commodity comment data on e-commerce platform,and divides the acquired data set into training set and test set,and conducts emotional polarity labeling and preprocessing.Secondly,using the Skip-gram Word2 vec algorithm model after pretreatment of text to quantify,and adopting the improved TF-IDF add weight information algorithm,then,to use Word2 vec training of the semantic information of Word vector containing only the Word,not in the text of the package contains the emotional information,this paper adopts the SO-PMI algorithm to rich emotional dictionary,feature selection based on the expansion of the emotional dictionary,put forward combined with the feature of Word vector and emotional information representation method of Word-Sen-T.Finally,the corpus of mobile phone comments crawled on JD platform was used as experimental data,use the Word-Sen-T feature extraction method and basic Word vector representation method in machine learning model of NB,KNN and SVM sentiment analysis on the experiment,and the model and other methods of text extraction model in this area in the comment text corpus contrast experiment on the model.Through comparative experiments,the results show that,compared with the traditional Word vector representation model,the proposed feature representation method of Word-Sen-T combining Word vector and emotion information is more obvious in improving the classification effect.There are other excellent models in the classification of affective tendency of comment corpus.The feature extraction model proposed in this paper iscompared with the two models in this field,and the F1 value in the classification results is increased by 6.3% and 4.3% respectively,The validity of the model is verified.