Research On Emotional Polarity Analysis Based On Text Mining Of Barrages

With the increasing scale of users in online video industry,massive online video users have generated a large number of reviews.Barrage is a popular online video reviewing method in recent years.It is highly praised by major online video websites,because its content is easy to generate a sense of dialogue and has a strong active atmosphere.The barrages often contain a large number of personal opinions,which can more accurately and concretely reflect users’ instant emotions when watching video than the traditional reviews below the video.This thesis aims to crawl the video barrages,analyze the emotional polarity of the barrages by sentiment analysis technology,and apply the analysis results to personalized video recommendation and online video public opinion analysis.Firstly,this thesis crawls the barrages of the hot variety show Creation 101 in Tencent Video Website as corpus,based on Python 2.7.Select 20,000 of them for pretreatment randomly,and construct training and test sets according to 9:1 ratio.Then,combined with the characteristics of simplification and fragmentation of barrage language,this thesis proposes a barrage emotional polarity analysis model based on TF-IDF(Term Frequency-Inverse Document Frequency)and SVM(Support Vector Machine).The feature words of the barrages are extracted by the TF-IDF algorithm,and the unstructured texts in the training set and the test set are transformed into structured TF-IDF values,which are mapped to vectors and expressed in the form of VSM(Vector Space Model).The SVM algorithm is used to analyze the emotional polarity of the data,by adjusting the parameters,the performance of the model is optimized and the accuracy of the model is improved.In order to verify the validity of the model,this thesis compares Accuracy,Precision,Recall and Macro F1 measure based on the dictionary and the Naive Bayesian emotional polarity analysis method.Finally,this thesis applies the model to personalized video recommendation and online video public opinion analysis.In the aspect of personalized video recommendation,this thesis adopts the idea of content-based recommendation algorithm and proposes a personalized video recommendation method based on the emotional polarity of the barrages.By analyzing the barrages sent by the user,the user’s attention and degree of attention are understood,by analyzing the barrages of the video,a corpus of video content features is obtained.By matching the video features and user concerns,calculating the video recommendation index,and combining the real-time characteristics of the barrages,taking 60 seconds as a unit of time,the video recommendation can be accurately timed,so that users can watch their favorite videos quickly and effectively,thus helping the website to improve the visits and user viscosity.In the aspect of online video public opinion analysis,this thesis takes the variety show as an example,collecting the barrages of actors’ performance time points,calculating the performance score by using the barrage emotional polarity analysis model,and obtaining the popularity and heat of actors,so as to provide data support for commercial decision-making such as post-production of programs on the network video platform,media contracted artists,advertisers’ selected roles and so on.