| The persuasive essay is a genre designed to convey the author’s perspective,argue a position,and influence the reader’s attitude and behavior.It plays a significant role in academic,business,political,and social contexts.The Toulmin model is a classic dialectical model used to analyze and assess the validity and rationality of arguments.Argument mining of persuasive essays based on the Toulmin model can provide a deeper understanding and analysis of the author’s argumentative process.Currently,the main method of identifying the elements of the Toulmin model in argumentative essays still relies heavily on manual labor,which not only requires high standards for the annotators but also consumes a significant amount of time and effort.At the same time,important information about Toulmin elements is sometimes omitted in the text,and the elements are often mixed in the same paragraph,which poses a challenge to the exploration of Toulmin elements in the article.This thesis employs deep learning-based approach can automatically identify the Toulmin elements in persuasive essays by mining the implicit features in the essays.To provide support for the automatic scoring system in assessing the argumentative ability of essays,a method for argument analysis should be created for the argumentative writing support system.This thesis focuses on argumentative essays written by second language writers(people learning a second language).The data was collected from second-year English majors at a comprehensive university.Two approaches are proposed for argument mining tasks:(1)Developing an argument mining method that incorporates inter-sentence information and document information.Since different Toulmin elements correspond to different levels of information,we use feature mining based on different levels,the sentence level and document level respectively,utilizing Bi LSTM and attention models.We then use a feature weighting method to fuse inter-sentence features and document features to better represent text features and thereby extract Toulmin elements from the essay.(2)Using key indicators to assist in argument mining.The article identifies certain words that can indicate the category of Toulmin elements within paragraphs or sentences.These words are referred to as key indicators.In this thesis,CNN is used to capture shortdistance dependency information,and Bi LSTM-Attention is used to capture long-distance dependency information to mine the key markers in the text.The model was tested on a dataset of 180 persuasive essays,and the automatic annotation accuracy reached 69%.It was also compared with the latest research results on the PE dataset,and the experimental results show that the model in this thesis outperforms existing models in persuasive mining. |
