| As a new entertainment mode,live broadcast is more and more popular with users.With the development and popularization of live broadcasting,a large number of live online comments,bullet screen,are produced.These barrages are not only related to the content of live broadcast,but also have the characteristics of online real-time,simple language and Internet,which are easy to produce some new network popular words;At the same time,the real-time interaction between users affects each other's expression and emotion.Therefore,the barrage generally carries the user's various views,and can timely and accurately reflect the user's emotional state when watching the live broadcast.Emotional analysis of barrage text,mining valuable emotional information,can timely and accurately grasp user preferences and behavior,has important reference value for precision marketing and service of live website.Although researchers at home and abroad have done a lot of in-depth research on the emotion analysis of traditional texts,and some mature results have been achieved,there are few researches and Analysis on the bullet screen emotion.Moreover,due to the unique online realtime,simple language and Internet features of barrage,the existing methods are difficult to be directly used for barrage emotion analysis.In order to improve the accuracy of barrage emotion analysis and better analyze the user's preference for live content,this paper makes an in-depth analysis of the related technology of live barrage and emotion analysis.This paper expounds the technology of text sentiment analysis from two aspects of sentiment dictionary based and machine learning based,and summarizes the methods of text representation and text classification.Aiming at the lack of barrage corpus and the characteristics of language brevity and Internet,this paper constructs barrage's exclusive emotion dictionary.By obtaining the background barrage text data of Huya live website,10000 barrage data are selected and preprocessed to build barrage corpus,and through optimizing the corpus to build barrage exclusive emotion dictionary.Then,four kinds of features,such as word vector,emotional word,negative word and punctuation,are selected to extract the corpus features.According to the characteristics of live barrage language,an emotion analysis model based on Improved SVM is proposed.By introducing a classification factor and a gradient descent factor,the generalization error of the predictor is reduced.On this basis,the fusion methods of word vector,emotion word,negative word and punctuation are proposed.The fusion results are mapped to the vector space,and then the emotion is classified by the classifier.By optimizing and adjusting the model parameters,the feature combination suitable for live broadcast Barrage is obtained,which improves the accuracy of classification.Experimental results show that,compared with SVM algorithm,naive Bayes algorithm and maximum entropy algorithm,the proposed method has better performance in accuracy rate,F1 value and recall rate.The result of emotion calculation is used to generate emotion trend graph,and the word cloud is generated according to the barrage heat.The barrage text is analyzed visually from multiple angles. |
PDF Full Text Request