Font Size: a A A

Research On Emotional Tendency Classification Based On Online Video Website Reviews

Posted on:2021-01-28Degree:MasterType:Thesis
Country:ChinaCandidate:X Q LiaoFull Text:PDF
GTID:2428330623484371Subject:Information and Communication Engineering
Abstract/Summary:PDF Full Text Request
With the development of modern information technology,people are able to enjoy more leisure activities.The rise of online video websites has made people no longer limited to watching films and television programs on televisions.Now people can watch their favorite videos online anytime,anywhere.These online video websites provide audiences with an interactive video viewing experience where viewers can make comments and discuss while watching the video.There is a lot of valuable emotional information hidden in these comments,and how to effectively mine and analyze these information is a valuable field of research.These information can not only provide reference information for viewers,but also the decision-making assistance for website owners and those working in film and television industries,so as to produce works more in line with public taste.Therefore,focuses on the sentiment tendency classification of online video website reviews,this paper deeply studies its research status in the field of deep learning,and develops innovations in research methods.Major works and contributions of this paper are as follows:1.This paper explores the application of hybrid neural networks to the classification of sentiment tendency of texts.Based on bidirectional gated recurrent unit network and convolutional neural network,this paper explores different combinations of the networks and proposes three different network models--CNN-BIGRU model,BIGRU-CNN model,and BIGRU-P-CNN Model.These models are applied to the video review dataset in this article.The experiment shows that the three hybrid neural network model proposed in this paper is better than that of the single network model in classifying.And the BIGRU-CNN model delivers the best performance with an accuracy and F1-score of 0.8804 and 0.8806.2.In terms of the classification of sentiment tendency in video reviews,this paper optimizes convolutional layers of ordinary convolutional neural networks,and develops a pointwise convolutional neural networks.The simple recurrent unit network is introduced into the problem of sentiment analysis.This paper proposes aBSP-CNN model combining bidirectional simple recurrent unit network and pointwise convolutional neural network.The experiments show that the BSP-CNN model performs better than other models in the two-class and three-class datasets of video reviews.The F1-score of the model on the long text two-class and three-class datasets are 0.9000 and 0.7015 respectively.The model performs better in the long text dataset than in the short text dataset.The result indicates that the BSP-CNN model in this paper can effectively overcome long-term dependency problems and can use the context of the text to improve the classification performance of the model.Comparing the BSP-CNN model with the BILSTM-CNN model,it was found that the model training parameters in this paper are reduced by 37056.In addition,the accuracy in classifying the three-class total dataset of film reviews is improved by 0.56%,and the training time shortened by 13 s.
Keywords/Search Tags:text sentiment analysis, hybrid neural network, gated recurrent unit network, simple recurrent unit network, convolutional neural network
PDF Full Text Request
Related items