Research On Emotion Cause Extraction Algorithm Of News Text Based On Deep Learning

With the vigorous development of major social media,netizens have launched their own discussions on various hot topics online.A large number of comments with subjective emotions will be generated in the discussion of hot news events,at the same time,there are comments that deliberately create rumors.It is crucial to dig out valuable information from these texts with rich emotional characteristics.In order to make emotion prediction better applied,decision makers need to know the causes of emotion.However,due to the very complex semantic connection between emotion and cause in the text,and the lack of related emotion cause corpus,there is very little work on deep-level emotion-induced causes mining.The task of emotion cause extraction is to mine the words and sentences behind the root causes from the texts with emotional characteristics.In this paper,aiming at emotion cause extraction,the main research work is as follows:(1)For news texts that have been marked with emotional sentences,in order to extract the potential causes.This paper makes fine-tuning on the pre-trained language model Bidirectional Encoder Representations from Transformers(BERT)and also use the dilation convolution,and proposes the causes extraction model BERT?CEDCNN(Context Emotion Dilation CNN),the model integrates the context of the emotional sentence that the previous researchers have neglected.The final experimental results prove the superiority of our emotion cause extraction model based on BERT+DCNN over a series of baselines.(2)For news texts with emotions that are not marked,in order to extract emotion cause pairs and to solve the ambiguity of the emotion cause pairs,this paper uses a more granular perspective on the Chinese news public sentiment data set.Emotion cause keyword labeling converts the sentencelevel emotion cause pairs extraction into a sequence labeling problem.Firstly,the experiment uses the word fusion model ZC?CNN designed in this paper to extract emotional sentences,and then uses the emotional dilation convolution gated CNN sequence annotation model EDGCNN(Emotion Dilation Gated CNN)proposed in this paper to extract emotion cause keywords,and finally obtains emotion causes pairs.Experiments show that the model proposed in this paper has the best recall rate compared with the recently proposed sentence-level emotion cause pairs extraction model,and the F1 score is also equivalent to the current state-of-the-art model;At the same time,experimental comparison found that the maximum decoding interval length of the final output sequence of the model has a direct impact on the decoding of emotion cause keywords.