KNOWLEDGEBOT: ENHANCING CHATBOT INTELLIGENCE VIA FEDERATED LEARNING ON WIKIPEDIA LLM

Abstract

This paper presents KnowledgeBot, a smart chatbot leveraging federated learning for privacy-preserving, accurate medical conversations in both French and English. Combining large language models (LLMs) with a fact-checking system, KnowledgeBot excels at generating informative responses, detecting LLM errors, and mitigating biases through simulated and real user data. Demonstrating strong multilingual capabilities, it achieves factual accuracy rates of 95.2% and 96.1% in human and simulated conversations, outperforming existing baselines. The English model showed higher accuracy than the French counterpart. This research lays the groundwork for secure, reliable, and adaptable chatbots that integrate federated learning with LLMs to advance conversational AI.