Empowering Inclusive E-Deliberation by Harnessing Collective Wisdom and Minority Perspectives using AI and LLMs

This project seeks to enhance inclusive e-deliberation by developing AI tools that construct and explore stance and dialectic trees, empowering citizens and policymakers with structured insights into diverse perspectives for better decision-making and stronger social systems.

About

This project aims to pragmatically contribute to inclusive e-deliberation by developing AI tools that facilitate this process, enabling citizens—not just experts or politicians—to be deeply involved in community problem-solving and public decision-making. Towards this end, it proposes the construction of stance trees, which are hierarchical structures representing the various stances on a particular topic. By exploring stance trees, policymakers can gain a comprehensive and well-structured understanding of individual perspectives.

Folders description

• LLM as a Judge: Contains code and data to validate synthesized arguments using LLMs.
• Validation 4 Human Experts vs LLM: Contains validation data for 36 arguments, evaluated by four human experts and LLM models.
• Validation Human Expert vs LLM as Judge: Contains validation data for 81 arguments, evaluated by one human expert and LLMs.
• BERTopic: Colab notebook for generating topics using the BERTopic library.
• Synthesized Arguments: Contains code to synthesize arguments based on opinions using the Gemma, LLaMA, and DeepSeek LLMs, with both 0-shot and 1-shot prompts for deniers and believers.