Sentiment Polarity Analysis And User Classification Based On Danmaku Text

With the continuous development of information technology,danmaku,a mode of video commenting,has gradually emerged on major video sites and are preferred by the general public.Different from traditional comment forms,danmaku enables video users to post comments on the current content while watching,which has the characteristic of immediacy and can more accurately reflect video users' views and sentiment.However,most of the current research on danmaku is more concentrated on the perspectives of communication science and user psychology,and there is less research on the sentiment analysis of danmaku texts.This thesis aims to study the sentiment polarity of danmaku texts by using natural language processing related techniques,and to analyze certain applications based on actual scenarios.The specific research contents are as follows.Firstly,the danmaku data of specified videos in Bilibili website was obtained by using web crawlers,and a series of data pre-processing was carried out based on the characteristics of informality and complexity of danmaku texts,and feature extraction of the text data was completed by using pre-trained word2 vec word vector model.Secondly,danmaku text sentiment polarity classification models were established based on traditional machine learning and deep learning neural network respectively,and after comparative analysis,it was found that the bidirectional LSTM model had the best overall performance compared to other models in terms of various evaluation criteria and had an advantage in this sentiment classification problem.Finally,the danmaku users in the video "The Post Epidemic Era" were selected as the research object,and the time series of the danmaku sentiment distribution of the users were obtained by applying the sentiment analysis model.Based on the study of noisy danmaku,the users were filtered,the optimal time interval of the emotion distribution was selected,and the Kmeans clustering method based on the DTW algorithm was used to establish a classification label for the video user group,and the effectiveness of the method was verified through analysis.With reference to the findings obtained from the analysis of this thesis,it enables video content creators to understand the trend of emotional changes of audiences,video hot content and audience group classification,which provides them with the direction of video content adjustment.For video websites,it provides new guidelines for classifying and labeling danmaku video users,which is a supplementary basis for them to make more accurate video recommendations.For advertising investors,they can learn about the ability of different videos to mobilize the sentiment of danmaku users and provide them with decision support.Overall,it can promote the production and reception of quality video content and create a better video content community.