Experimental evaluation of RAG Components and Their Impact on the Performance of Customer Support Chatbots
Вантажиться...
Дата
2026
Автори
Науковий керівник
Назва журналу
Номер ISSN
Назва тому
Видавець
КПІ ім. Ігоря Сікорського
Анотація
This paper presents an experimental study of the components of RetrievalAugmented Generation (RAG) systems aimed at developing an intelligent support chatbot capable of processing complex textual documents. The research focuses on evaluating various frameworks, vector databases, and text-chunking strategies to identify the optimal configuration that ensures high accuracy, contextual completeness, and computational efficiency. Experimental results demonstrate that the LangChain framework provides superior accuracy and context coverage compared to LlamaIndex, while FAISS outperforms other vector stores in answer relevancy, faithfulness, and processing speed. Additionally, a chunk size of 1000 and chunk overlap of 50 achieve the most balanced performance in terms of precision, recall, and response time. The combination of LangChain, FAISS, and the 1000/50 chunking configuration forms an effective foundation for implementing a high-performance RAG-based support chatbot, capable of delivering accurate, faithful, and contextually relevant responses across various domains.
Опис
Ключові слова
Retrieval-Augmented Generation, RAG, LangChain, LlamaIndex, FAISS, chatbot, vector database, context retrieval, chunking strategy, LLM, доповнена пошуком генерація, RAG, LangChain, FAISS, чат-бот, векторна база даних, пошук контексту, великі мовні моделі
Бібліографічний опис
Oliinyk, V. Experimental evaluation of RAG Components and Their Impact on the Performance of Customer Support Chatbots / V. Oliinyk, P. Ponochovnyy // Адаптивні системи автоматичного управління : міжвідомчий науково-технічний збірник. – 2026. – № 1 (48). – С. 20-29. – Бібліогр.: 9 назв.