Emotional Text Classification Based On Deep Semantic Fusion Features

With the rapid development of social media,a large number of netizens express their views on various fields such as news and products on social media,resulting in a large amount of data with emotional categories.Therefore,emotion classification technology is increasing rapidly.As emotional text has insufficient semantic information and insensitive emotional words in the feature expression,it affects the accuracy of emotional text classification.Based on this,the main research content of this article has the following three aspects:(1)Aiming at the emotion text classification technology,the ERNIE pre-training model is proposed to represent the emotion text vector,and combined with the softmax regression classification algorithm to classify the emotion text.The ERNIE pre-training model is a multi-stage knowledge masking strategy based on the BERT pre-training model,which raises word masking to phrase and entity-level masking.The masking strategy is used to enrich the semantic information of emotional text to improve the accuracy of emotional text classification.The experimental results show that the ERNIE pre-training model is 3.56% higher than the BERT pre-training model and 21.44%higher than the Attention-Bi LSTM model in the classification accuracy of sentiment text published by netizens on Weibo during the epidemic.This proves the effectiveness of the proposed model.(2)Aiming at the emotion text classification technology,an ERNIE-DPCNN fusion model emotion text classification algorithm is proposed.Based on the above research,the algorithm inputs the emotional text vectors pre-trained by ERNIE into the DPCNN model for downstream task training,and combines the softmax regression classification algorithm to classify the emotional text.As the DPCNN model continues to increase with the number of network layers,the extracted emotional text features contain deep-level context information.If combined with the ERNIE pre-training model,it can not only obtain rich prior knowledge of emotional text features,but also obtain deeper contextual semantic information according to downstream tasks,thereby improving the accuracy of emotional text classification.The experimental results show that the proposed ERNIE-DPCNN fusion model is3.64% higher than the BERT pre-training model and 21.52% higher than the Attention-Bi LSTM model and 4.55% higher than the BERT-Text CNN fusion model and 0.08% higher than the ERNIE pre-trained model in the classification accuracy of sentiment text published by netizens on Weibo during the epidemic.This proves the effectiveness of the proposed model.(3)Aiming at the emotion text classification technology,based on the above research and introducing the Attention mechanism,an ERNIE-AT-DPCNN model emotion text classification algorithm is proposed.The algorithm inputs the emotion text vector pre-trained by ERNIE into the Attenion mechanism to obtain the attention word vector,and then inputs it into the DPCNN model for downstream task training,and combines the softmax regression classification algorithm to classify the emotion text.Because the Attention mechanism allows the hidden layer of the neural network to assign different probability weights,the key node information similar to emotional words can be effectively paid attention to to improve the accuracy of emotional text classification.The experimental results show that the proposed method of introducing the Attention mechanism achieves the highest accuracy in the classification of emotional texts published by netizens on Weibo during the epidemic compared with the comparison model.This proves the effectiveness of the proposed model.