Research On Emotion-Cause Pair Extraction Based On Position Information And Sentiment Analysis

Emotion-Cause Pair Extraction is a new task of natural language processing,which aims to mine the semantic relationship between emotion and reason in documents.It has important application value in emotional reasoning,risk prediction and other fields.The current emotion reason extraction model usually uses pipeline structure or end-to-end structure.The pipeline model is limited by error propagation,which affects the performance of the model.The insufficient degree of end-to-end model mining text semantics also leads to low accuracy of the model.To further improve the accuracy of Emotion-Cause Pair through semantic mining,this paper proposes an Emotion-Cause Pair Extraction model based on position information and sentiment analysis.The specific research contents are as follows:（1）In order to solve the problem that semantic information lacks word order and position information,the Emotion-Cause Pair Extraction based on Position Information（PIECPE）model is proposed,which adds position features to the model to improve the extraction accuracy.In order to make full use of the position information,the task transformation rules of the Emotion-Cause Pair Extraction task and the annotation strategy of the benchmark dataset is proposed.Firstly,in order to obtain more abundant semantic information,the pre-training language model BERT（Bidirectional Encoder Representation from Transformers）is used to generate word vectors,and the self-attention mechanism is used to calculate the relative position information of the text.The feature fusion of word granularity and clause granularity is used as text feature.Then,a prediction network is constructed to predict the starting positions of emotional clause and cause clause;finally,the emotional cause clause is assembled into emotional cause relation pairs by assembly algorithm to obtain the extraction results.The experiments show that position information can effectively improve the extraction performance of the model,but the emotional features in the document are not fully excavated.（2）In order to solve the problem that PIECPE does not fully mine emotional features,the Emotion-Cause Pair Extraction model PISAECPE（Emotion-Cause Pair Extraction based on Position Information and Sentiment Analysis）is proposed based on position information and emotion analysis,which integrates emotional features and linear features to improve the extraction accuracy.Firstly,the sentiment clause dataset is constructed,and the text clustering analysis is performed to quantify the emotional features in the document,which providing auxiliary screening indicators for the subsequent extraction process.Then,construct candidate emotional cause relation pairs and calculate probability;finally,three-dimensional convolutional neural network is used for fusion extraction.The experiment found that the PISAECPE model integrated with emotional characteristics significantly improved the extraction accuracy compared with the PIECPE model.The overall results show that the extraction accuracy of PIECPE model is better than that of most mainstream models,and the F₁value reaches 66.85%.The PISAECPE model further integrates emotional features on the basis of PIECPE,and the F₁value is increased by 2.46%,which has achieved the highest accuracy of emotion clause extraction.Table[11]Figure[21]Reference[84]...