Research on Recognition Method of Offensive Speech in Social Media Based on Hybrid Neural Networks
In the task of offensive speech recognition in social media,the existing research fails to give full play to the potential and advantages of different neural networks,resulting in limited recognition ac-curacy.To address this issue,an offensive speech recognition model(BERT-BiLSTM-SA-MCNN)was proposed,which integrated the BERT pre-trained model,Bidirectional Long Short-Term Memory net-works(BiLSTM),self-attention mechanisms(SA),and Multiscale Convolutional Neural Networks(MC-NN).Initially,the BERT pre-trained model was used to encode and convert the input text data.Subse-quently,the BiLSTM network and self-attention mechanism were employed to capture the global semantic features of the text.Thirdly,the Multiscale Convolutional Neural Network was utilized to extract impor-tant local features within the text.Finally,a fully connected layer was applied for the classification and i-dentification of offensive speech.Experimental results demonstrate that the BERT-BiLSTM-SA-MCNN model exhibits superior performance in the task of recognition of offensive speech in social media,achie-ving an accuracy rate,precision,recall and F1 score of 86.67％,84.20％,89.74％,and 86.79％,respectively,with higher accuracy and generalization capabilities.
NETL
NSTL
万方数据
