Using LLM to Convert Bahasa Indonesia Commands into JSON Structures for GIS System API

Geographic Information Systems (GIS) often require complex Application Programming Interface (API) interactions that are not easily accessible to non-technical users. This research proposes a proof-of-concept system that uses large language models (LLMs) to convert Bahasa Indonesia user commands into JSON structures for GIS system APIs. A dataset was generated by creating pairs of Bahasa Indonesia commands and corresponding JSON structures using ChatGPT based on a predefined template. The dataset was split into 90% for fine-tuning and 10% for testing, using two open pre-trained LLMs: Llama 3.1-8b and Gemma 2-9b. The performance was evaluated across several scenarios, including non-fine-tuning and fine-tuning with different epoch options. Results show that fine-tuning improves accuracy, with Gemma 2-9b achieving the highest accuracy of 94.23% after 20 epochs, slightly outperforming Llama 3.1-8b. This proof-of-concept demonstrates the potential of LLMs to simplify natural language command processing for GIS systems.