Improvement And Application Of Barrage Recognition Based On BERT-DPCNN

Online video has occupied an important position in the process of watching videos and has become the mainstream of mass video consumption."Barrage" has begun to enter peoples field of vision because of its high interactivity and entertainment.Barrage is a display method that users can directly send comments and display them at the top of the screen when watching videos on video websites.Nowadays,although the domestic application of barrage is dizzying,the overall development of domestic barrage culture and technology is relatively short,and the current research on barrage is still relatively scarce.With the continuous development of deep learning technology,garbage barrage recognition models based on neural network have proven to have unique advantages.Based on the summary and research of traditional classification models and word vector models,this paper discusses the identification of garbage barrage text.Based on the unique characteristics of garbage barrage,a BERT-DPCNN model is constructed to improve the identification method of garbage barrage.This paper uses DPCNN as a garbage barrage recognition model to obtain more levels of information,and then improves on this model.The text vector trained by the BERT model is used as the input to the improved DPCNN model.Then introduce batch normalization into the model and construct a BERT-DPCNN garbage barrage recognition model,which can not only extract more information,but also avoid the problem of the disappearance of deep neural network gradients as much as possible.In this paper,we use the barrage data from the four categories of TV series,movies,variety shows and live broadcasts as data sources for experimental analysis to verify the performance of the proposed model.After obtaining the sample set,use the BERT-DPCNN model to identify the garbage barrage,and compare it with the Text CNN,Bi LSTM,Bi LSTM-Attention,and DPCNN models based on the word2 vec word vector.It can be seen that the improved model proposed in this paper can understand more Multiple semantics are available from the models recall,precision,and F1 indicators.The BERT-DPCNN model has the highest of these three indicators and can more effectively identify the garbage barrage.Moreover,it can be seen from the experimental data that some feature extraction methods involving deep models,such as the BERT model in this paper,can show more obvious advantages than word2vec-based feature extraction methods.The BERT-DPCNN model constructed in this paper can store more semantic environment information,provide more basis for text classification,and can also extract deeper text features.It is the model with the best comprehensive performance and the recognition of garbage barrage.