基于LERT-RCNN的中文弹幕文本情感多分类研究

Multiply classification of Chinese barrage screen based on LERT-RCNN

为提高弹幕文本情感分类准确率,提出基于LERT-RCNN的弹幕文本情感多分类模型。首先使用LERT预训练语言模型获取文本动态特征表示,减少弹幕文本中一词多义对情感分类准确率带来的影响;其次使用BiLSTM和CNN提取更深层次语义特征;最后经全连接层后送入softmax函数得到情感分类结果。实验结果表明,基于LERT-RCNN的弹幕文本情感分类模型的准确率、精确率、召回率及F1值分别为96.17%、94.54%、92.56%及93.51%,与传统文本情感分析模型及单一预训练语言模型LERT相比有明显提升。

In order to improve the classification accuracy of barrage screen sentiment,a multi-classification model of barrage screen sentiment based on LERT-RCNN is proposed.Firstly,the LERT pre-training language model is used to obtain the dynamic feature representation of text to reduce the influence of polysemy on the accuracy of sentiment classification for barrage screen.Secondly,BiLSTM and CNN are used to extract deeper semantic features.Finally,the sentiment classification results are obtained through the full connection layer and fed into the sofamax function.The experimental results show that the accuracy,precision,recall and F1-score of the barrage screen sentiment classification model based on LERT-RCNN are 96.17%,94.54%,92.56%and 93.51%,respectively,which is significantly improved compared with the traditional text sentiment analysis model and single pretraining language model.