Leveraging Large Language Models for Project Generation in Category Management

The continuous advancements made on Large Language Models (LLMs) present promising avenues for applications within category management. This thesis aims to investigate the application of LLMs for project generation within category management. To this purpose a project generation system is proposed based on the Purchasing Chessboard framework. Several methods are evaluated for the system, which includes Few-shot learning, Chain-of-Thought prompting, Prompt Chaining, and External Context Integration. Additionally, further investigation is made into the use of LLM-based evaluation methods, and an analysis is made of the correlation with human evaluation. The results show that LLMs can effectively generate relevant projects, with the combination of Few-shot learning and Chain-of-Thought prompting producing the best results according to human evaluation. The LLM-based evaluation shows a low to moderate correlation with human evaluation, suggesting it has limited utility. The results indicate the potential of leveraging LLMs for project generation in category management and highlight the need for further research into the use of more advanced architectures as well as the use of more category-specific data.