Research On Sentiment Analysis Of Video Bullet Chat Text

The 21st century is the era of information explosion,and people are surround-ed by data every day and received a large amount of information.It has gradually drawn our attention that how to get information we need from a large amount of them,and transform passive acceptance to active acquisition.Video barrage,as an instant text message,gradually enters the public's field of vision.When watching a video,users can comment on the current plot of the video through the barrage,inter-acting with other viewers and enhancing the sense of enjoyment.Barrage comments have the characteristics of openness,diversification,and emotionality.Bullet-screen text analysis plays a supporting role for films and TV series recommendation,pro-gram optimization,and hot event management.Therefore,the sentiment analysis research on video barrage can not only provide a basis in decision-making for users,but also help network platforms realize the supervision of public opinion and purify the network environment.In order to analyze the bullet-screen text,this paper combines Word2vec with the improved SO-PMI algorithm to build a sentiment dictionary suitable for bullet-screen text.Firstly,this paper integrates the existing sentiment dictionary,take out the union of multiple sentiment dictionaries,performs word segmentation and part-of-speech tagging on the text corpus,filters out adjectives and verbs to form candidate word sets,and finds the intersection with the integrated sentiment dictio-nary.Secondly,this paper utilizes Word2vec to train the word vector and transform them into a language that can be recognized by the computer,and filters the seed word set by the TF-IDF value and the similarity between words.Thirdly,the words are classified through the improved SO-PMI algorithm,and the final sentiment dic-tionary is obtained.Finally,this paper compares the effects of Word2vec combined with the improved SO-PMI algorithm,the SO-PMI algorithm and the improved SO-PMI algorithm on word classification.The result of analysis shows that the method based on the combination of Word2vec and the improved SO-PMI algorithm has the highest accuracy of word classification.Perform sentiment analysis based on the constructed sentiment dictionary.The bullet-screen text contains a large number of slangs.Different from traditional word-s,they are more streamlined and colloquial,some of them with strong emotional colors.Therefore,the Internet terms in the bullet-screen text which is sorted out by this article can be added to the sentiment dictionary.Based on this,adverbs of degree and negative words are introduced to calculate sentiment value.In addition,this paper labels sentences manually,and compares the results of sentiment classi-fication based on sentiment dictionary,naive Bayes,and support vector machine.This paper visually analyzes the changes in the word cloud map of the bullet-screen text,the number of barrage,and the sentiment value over time.Moreover,because sentiment value only shows the emotional tendency of audience's comments,and fails to see the audience's attention,via the LDA topic model,this paper mines the potential topics in the bullet-screen text,and then analyze the concentrated attention as well as related feature words in the bullet-screen text data.The results imply that the effect of sentiment classification based on the sentiment dictionary is the best,and the audience's evaluation of the character image,plot and scene production in the play is relatively high.