Argumentation Quality Evaluation of the Chinese Language Based on Semantic and Structural Features

The evaluation of argumentation quality aims to automatically evaluate the overall quality of an argumentative process. This paper investigates the discourse-level argumentation in Chinese student essays, in which there is a thesis sentence and a discourse supporting the claim described as evidence. This paper introduces the first annotated dataset for Chinese discourse-level argumentation quality evaluation, derived from authentic student essays. Furthermore, a model is proposed for evaluating the argumentation quality based on both semantic and structural encoding. This model employs a language model as the text encoder to extract semantic information, while for the structural features, a graph embedding algorithm based on Abstract Meaning Representation is utilized. Next, a fusion function facilitates evaluation of argumentation quality. The experiments demonstrate that the proposed model outperforms existing methods.