Facilitating Dynamic Group Discussions with LLM Models: An NLP-Driven Evaluation Strategy
Jomina John, Sanghamitra Menon, Khushi Vijay, Susan Mathew, Sanoy Boby · 2025
This work is an interactive forum built on the LangChain framework, where multiple pre-trained Large Language Models (LLMs) and a human user engage in turnbased discussions. A structured turn-taking system ensures each LLM contributes sequentially, while the human user interacts through a button interface. To maintain engagement, LLMs are periodically reset, introducing fresh perspectives. An NLP-based scoring mechanism evaluates the human user's performance on metrics like speed, clarity, and content quality, offering real-time feedback for improvement. Deepfake technology enhances the experience by giving each LLM a distinct face, adding realism to interactions. By combining LLM intelligence, structured turntaking, NLP-based scoring, and deepfake enhancements, this forum provides an engaging and educational platform.