<![CDATA[Accurate meteorological information is crucial for the tourism and agriculture sectors in Bali Province. However, global weather Application Programming Interfaces (APIs) often have limitations in recognizing hyper-local areas and lack natural language interaction capabilities. This study aims to develop the Bali Weather Bot, an intelligent Telegram-based assistant using Clean Architecture and Hybrid Natural Language Processing (NLP). The methodology combines the Gemini 1.5 Flash Large Language Model (LLM) for entity extraction with a Context-Aware Dictionary Mapping fallback mechanism to standardize local abbreviations and Indonesian temporal metaphors. Evaluation was conducted using a dataset of 250 test query scenarios evaluated through a confusion matrix and latency benchmarking. System evaluation demonstrates that this hybrid architecture successfully extracts spatial and temporal parameters with an overall accuracy of 82.80%. Quantitative evaluation shows a Precision of 92.00%, Recall of 89.22%, and an F1-score of 90.58%, while maintaining a 0% system crash rate. In conclusion, the hybrid approach effectively mitigates AI hallucination and enhances hyper-local data retrieval reliability in conversational interfaces.]]>
Copyrights © 2026
