KNOWLEDGEBOT - ADVANCING CHATBOT INTELLIGENCE: FEDERATED LEARNING WITH LLM MODEL ON WIKIPEDIA CORPUS

Abstract

This paper discussed KnowledgeBot, a smart chatbot that utilizes the technology of federated learning for a privacy-preserving experience as well as for the converse interactions while offering accurate and essential responses. The KnowledgeBot makes use of large language models (LLMS) and a fact-checking system that enable it to specialize in the medical domain in both French and English languages. Through the combination of simulated and real user utterance generation, alongside biases screening   and mitigation plans, KnowledgeBot proves the good talent in producing informative claims, efficiently looking up information and recognizing wrong LLM outputs. The aforementioned conquest to French shows the flexibility of the brand to appeal across languages. The data collected show that KnowledgeBot achieves an amazing accuracy of factual responses of 95.2% and 96.1% during the human conversations about recent topics and in simulated conversations bypassing other baselines and improving the recent state-of-the-art models. The accuracy of pretrained model computed against English dataset was greater as compared to the accuracy computed against Frenchdataset. This research plans to be a precursor to secure, informative yet adaptable chatbots paving the way of the combination of federational and LLM technologies for advanced conversational experiences.