Three-Tier Network-to-Prompt Architecture for Fine-Tuning ChatGPT for the Singapore Student Accommodation Platform

In this paper, we present a novel methodology for the supervised fine-tuning of ChatGPT, specifically designed to address domain-specific queries related to accommodations in Singapore for international students who have recently arrived. The domain-specific knowledge used for fine-tuning ChatGPT was gathered through a qualitative survey conducted with property agents actively engaged in the Singaporean real estate market. This specialized knowledge will be transformed into prompts utilizing the proposed Three-Tier Network-to-Prompt Architecture (N2P). The N2P will distill abstract knowledge into structural knowledge through a syntax-to-semantics graphbased encoding, represented by a Network Relational Model (NRM), to construct the query prompts for the supervised fine-tuning process. Thus, this improves ChatGPT’s capability to address questions from international students about accommodation in Singapore. Finally, we will conduct experiments using this proposed architecture through a developed prototype website, AcomoSG, featuring a chatbot that interfaces with finetuned ChatGPT to evaluate the efficacy and effectiveness of the domain-specific knowledge in addressing accommodationrelated inquiries within the context of Singapore, as posed by international students.