AI-Powered Question Management in Conferences

Effective management of audience questions are crucial for enhancing speaker-audience interaction in conferences. However, this process is often hindered by irrelevant or repetitive questions and a lack of participation from introverted attendees. This paper presents an AI-Powered Question Management System designed to address these challenges. Leveraging the DistilBERT model, enhanced with a Question-Topic Relevance Prediction Model, the system predicts whether a question is relevant to the discussion topic using deep learning and NLP techniques. This model combines relevance scoring with semantic similarity matching to deliver highly accurate predictions, ensuring only pertinent questions reach the speaker. The backend, implemented using Node.js and MongoDB, ensures robust scalability and seamless performance, while the React.js-based frontend delivers user-friendly interfaces tailored to speakers and listeners. Conference speakers benefit from real-time access to prioritized, topic-relevant questions, enabling more focused and impactful discussions. Listeners, on the other hand, can submit queries anonymously, fostering inclusivity and encouraging participation from all attendees. The Question-Topic Relevance Prediction Model integrates key processes: fine-tuned DistilBERT for binary classification, curriculum learning for optimized training, and similarity-based scoring using topic-specific terms. The model achieves relevance predictions by combining model-generated probabilities and similarity metrics, ensuring a balanced evaluation. Experimental evaluations demonstrate significant improvements in question handling efficiency, including reduced redundancy and enhanced user satisfaction. The proposed solution establishes a scalable, intelligent framework for equitable and effective conference question management, setting a new benchmark in audience engagement technologies.