Research On Emotion Classification Features Based On Keyword Weighting

In recent years,with the rapid development of Internet Finance and e-commerce,all kinds of network malls have been in an environment of explosive growth of information.More and more people like to put forward their own reviews on goods and business services when shopping online,which makes the number of commentary information of Internet malls increase in a geometric progression.Customers'comments on commodities represent their attitudes towards merchants'products and services.Therefore,text mining and emotional analysis of the comments on these products are of great help to the study of the public praise among consumers,the recommendation of commodities,the filtering of spam information,the understanding of users' psychology,the mastery of the first dynamic of the market and the improvement of merchants' services.Generally,the main method of traditional text emotional classification is to pre-process the text,construct word vectors,extract features,and finally use classification or clustering method to classify these data emotionally.In the traditional emotional classification method,feature engineering is the most important link,which has the greatest impact on the classification results.However,the traditional TF-IDF and Word2vec methods extract text features with limited content and weak expression ability.It is difficult to give consideration to the relationship between word vectors and context and the word frequency weight of word vectors in short text.The feature words extracted by these methods are not representative of the whole short text and affect the classification effect.To solve this problem,the following work has been done in this paper:First,skip-gram model of Word2vec is used to calculate the probability of deducing the whole sentence from each word.Then each word is sorted in descending order of probability,and these words are modeled twice;Secondly,TF-IDF is improved by adding the concept of the length of words to TF-IDF to calculate the TF-IDF value of keywords extracted in the first step;Thirdly,weigh the Word2vec word vector which has been created in the first step,and create a new T-Word2vec feature representation.Finally,support vector machine is used to classify sample datas.A large number of experiments show that using T-Word2vec as feature to classify the emotions,the indexes of the classification results are significantly improved compared with the traditional TF-IDF and Word2vec word vector representation,and the running time is also lower than that of Word2vec model.