Research On Sentiment Analysis Of Barrage Comments Based On Deep Learning

Time-sync comments is a popular video commentary in recent years,which can more accurately and concretely reflect the immediate feelings and commendatory and derogatory evaluation of the user when watching video.This paper proposes a deep-learning time-sync video shot sentiment analysis model as a video highlight detection.Provide evidence.Firstly,this paper analyzes the related studies of time-sync comments;secondly,it analyzes the LSTM network model and the AT-LSTM model based on attention model,and then proposes an SIS-LSTM sentiment analysis model that combines video importance scoring and LSTM network model.The final experimental results show that the model proposed in this paper can effectively detect highlight shots contained in video.The research content of this paper has three parts as follows:(1)For text preprocessing problems,based on word-embedding method,and construct a popup commentary emotion dictionaryBecause traditional deep learning methods deal with the problem of time-sync comments commentary sentiment analysis,there are problems such as high dimension and gradient disappearance.This paper uses the word-embedding method to construct emotion dictionary of the time-sync comments.The network model better understands the semantics of time-sync comments.(2)Extracting the deep features of time-sync comments based on LSTM model to highlight the emotional weights of key emotion wordsBecause the time-sync comments contains time series information,this paper extracts the deep features of the time-sync comments based on the LSTM model.The use of LSTM neural network to deal with time-sync comments can effectively utilize the long-distance dependencies of the time-sync comments text and highlight the emotional weights of emotion keywords.(3)SIS-LSTM model based on topic concentration and emotional concentrationThe extraction of video highlights has always been a challenge in the field of video analysis.This paper proposes a video highlight detection method,which uses the topical concentration and emotional concentration of time-sync video to calculate the importance scores of video shots(Shot Importance Score,SIS)extracts the highlight shots of the video.Based on this,combined with the characteristics of time-sync comments,this paper proposes an SIS-LSTM model,using the LSTM model as the coding model,and adding the importance score of the video shots to calculate whether the video shot is a highlights video probability.In the experimental results,this paper uses four groups of comparative experimental models to conduct sentiment analysis on the popup review data.Compared with other models,the experimental results of the SIS-LSTM model proposed in this paper are more superior,which verifies the effectiveness of the combination of time-sync comments topic concentration and emotion intensity for detecting highlight shots of the video.