Developing a natural language processing-based smart university assistant: Bilgetekin chatbot

This thesis aims to develop a chatbot (intelligent university assistant) system to facilitate and accelerate information access for students and staff at Çankırı Karatekin University. Within the scope of this study, a language prototype was initially created using the RASA NLU framework, a rule-based question-answering bot approach. However, due to limitations encountered with this approach, a shift was made to a hybrid and multi-model-based approach utilizing Large Language Models (LLMs) and the Retrieval Augmented Generation (RAG) technique. In this process, the architecture of both systems, data collection and processing procedures, training details, performance metrics, and user experience tests are comprehensively presented. The developed "BilgeTekin Chatbot" aims to create an effective and useful dialogue system specifically for Çankırı Karatekin University in the Turkish language, highlighting the advantages and disadvantages of different chatbot development approaches. The agglutinative structure, rich morphological features, and contextual complexity of the Turkish language make the direct application of general-purpose NLP models challenging. In light of these difficulties, the necessity of specialized natural language processing models and approaches for the Turkish language is emphasized. In this thesis, a RASA-based model adapted for the Turkish language and an LLM-RAG-based model were developed and comparatively evaluated. The results obtained demonstrate that the LLM-RAG approach provides significant improvements in answer accuracy and user satisfaction compared to the rule-based approach. In the final section, the unique contributions of the study are discussed, and suggestions for future work are provided.