Research On Fine-grained Sentiment Analysis Of Video Reviews

In recent years,many Internet users have chosen to watch videos online and have left a lot of video reviews to show their personal views.How to extract valuable information in a large number of video reviews has become a problem that needs to be resolved.Fine-grained sentiment analysis can effectively extract the evaluation object in the comment text and realize the corresponding sentiment polarity judgment,which has shown a good application value.Therefore,this article conducts a fine-grained sentiment analysis study of video reviews,and analyzes the sentiment tendencies of video from comments.It can help the user decide whether the video is worth watching and provide users with better services;it can also make video providers more able to Good for user evaluation to make corresponding improvements to video products.As deep learning technology continues to make progress in the field of natural language processing such as speech recognition,machine translation,etc.,the deep learning model has proved to have great advantages in text feature extraction.This paper mainly uses the deep learning model to study the fine-grained sentiment analysis of video reviews.Firstly,it collects online video reviews and preprocesses them,then uses the deep learning model to extract the evaluation objects in the reviews,and then evaluates the emotional tendencies of each evaluation object.Based on the results of fine-grained sentiment analysis,two movie reviews were used as examples for visual analysis.The main innovations of the project are as follows:(1)Target extraction study.In view of the deficiency of using traditional word vector information when extracting evaluation objects from traditional LSTM models,this paper proposes a context-based CNN-BLSTM model.The model uses CNN to map the text context features into a real vector,namely Context Feature,to fully acquire the feature information between words.The Context Feature is used as an input to the BLSTM as an additional feature along with the word vector,so that the model can fully exploit video comment context information.The experimental results show that the model can effectively improve the accuracy of the evaluation object extraction compared to the LSTM model.(2)Sentiment analysis based on the Target.Currently,the TAB-LSTM model is one of the state-of-the-art models based on the evaluation of the emotional disposition of the evaluation object.Compared to the LSTM model,which only considers the above information,the BLSTM model can consider contextual information at the same time and have stronger feature extraction capabilities.Therefore,this paper proposes a TAB-BLSTM model by improving the TAB-LSTM model.The TAB-BLSTM model extracts Target information features and context information features through the BLSTM model.Then the Attention model is trained according to the results of BLSTM model extraction.This model can assign feature weights to BLSTM nodes,highlighting words that are more important for emotional polarity,and improving the effect of affective analysis.The experimental results show that the TAB-BLSTM model can further improve the accuracy of the analysis of emotional tendency compared to the TAB-LSTM model.