(AIMS 2024) Arg-T5: A Multi-Perspective Argument Generation Method Based on Debate Topic

This paper introduces an innovative approach to multi-perspective argument generation in the context of debate topics. Traditional text generation models, such as T5, often fall short in producing diverse arguments, leading to a lack of depth and diversity in debate simulations. To address this, we developed a method integrating perspective features into the dataset and leveraging the pre-trained Mengzi-T5 model to generate compelling arguments from multiple perspectives. Our method which is named Arg-T5, enhances the diversity of generated arguments and maintains relevance and persuasiveness. Through extensive experiments and comparisons with other models, including Mengzi-T5, ChatGLM, and GPT-2, we demonstrate the superior performance of our approach. The outcomes underscore a notable enhancement in argument diversity, addressing a pivotal challenge within computational argumentation. Our work contributes to advancing text generation and computational argumentation, offering a new solution for generating rich and varied debate content.