Sentiment Classification Algorithm for Vtuber Comments Based on BERT-CLS-ATT Model
The webcasting of the video network platform BiLiBiLi is in a booming stage.Among them,Vtuber as a unique classification of webcasters,also shows a trend of rapid development.However,the development of Vtuber is currently facing some limitations,such as lack of intuitive data analysis,lack of efficient methods to grasp hot live content,and lack of methods to avoid controversial content.Therefore,a text classification algorithm is proposed to solve these problems.Aiming at the relevant features of Vtu-ber comments,a BERT-CLS-ATT text sentiment classification model is proposed based on the BERT pre-trained model with whole-word pre-training of incremental domain data and the attention mechanism.The BERT-CLS-ATT model well solves the task of sentiment classification of Vtuber comment texts;Then TF-IDF text keyword extraction algorithm is used to extract keywords for the text of sentiment classification re-sults.Finally,the results of sentiment classification and the extracted keywords are combined to provide scientific guidance for the operation of Vtuber accounts.The experimental results show that the modeling approach achieves more than 83％F1 value and more than 84％accuracy in the task of sentiment classi-fication of Vtuber comment texts.Further experimental results on public datasets show that the BERT-CLS-ATT model structure has a certain universality in the sentiment classification task of texts.
text sentiment classificationdeep learningpre-trained modelattention mechanism
NETL
NSTL
万方数据
