FT-LLM: Development of a Retrieval Augmented Generation based Language Model Framework for Assisting Data Analysis in the Football Industry

In the football industry, data analysis is gradually gaining more recognition as a key tool for developing match and club management strategies and analyzing players. Consequently, large language models (LLMs), with the ability to offer customized responses to user inquiries, could be a valuable data analysis resource. In this paper, we propose FT-LLM, a framework that enhances the language generation capability of language models by incorporating a partially cleansed, football-related documents and a retrieval. Compared to existing RAG methodologies, FT-LLM provides more suitable answer for football related queries by effectively incorporating external documents with the help of Query Refinement and Confidence Check modules. By employing a chain of thought-based prompting strategy to prevent hallucination, language models can convey more relevant and reliable football related insights to users. As a result, FT-LLM has the potential to assist data-driven decision making and operations in the football industry by formulating a match strategy or advising player acquisition.