<![CDATA[This study discusses the development of a Natural Language Processing (NLP)-based weather chatbot dashboard capable of receiving natural Indonesian language queries and displaying weather information interactively. The system combines Sentence-BERT (SBERT) paraphrase-multilingual-MiniLM-L12-v2 as a text similarity engine, Gemini 2.5-flash Large Language Model (LLM) as a natural language summary generator, and the OpenWeather API as a real-time weather data source. A zero-shot semantic similarity approach is used without fine-tuning, with intent determination based on cosine similarity and a threshold of 0.5 optimized through testing several threshold values. The development method used is the Waterfall model with stages of requirements analysis, architectural design, NLP module implementation and API integration, React dashboard frontend development, as well as functional black-box testing and performance metric evaluation. The test results show that SBERT with a threshold of 0.5 produces an intent classification accuracy of 90% in 20 test scenarios, which increases to 100% after being combined with rule-based auto-adjust and fallback mechanisms. City entity extraction with a 1–3 word sliding window against city.list.json achieved 100% accuracy on 15 city-based queries, while macro metrics yielded a precision of 0.9286, a recall of 0.90, and an F1‑score of 0.8958. The integration of the OpenWeather API and Gemini enables the presentation of natural, informative weather summaries and visualization of real-time weather data in the form of interactive graphs on a React-based dashboard.]]>
Copyrights © 2026
