IRFANUDIN, HANIF LUTHFI (2026) IMPLEMENTASI CHATBOT CUACA BERBASIS SBERT DAN LLM GEMINI MENGGUNAKAN OPENWEATHER API. S1 thesis, Universitas Mercu Buana Jakarta.
|
Text (HAL COVER)
Cover.pdf Download (898kB) | Preview |
|
![[img]](https://repository.mercubuana.ac.id/style/images/fileicons/text.png) |
Text (BAB I)
Bab 1.pdf Restricted to Registered users only Download (582kB) |
|
|
Text (BAB II)
Bab 2.pdf Restricted to Registered users only Download (963kB) |
|
|
Text (BAB III)
Bab 3.pdf Restricted to Registered users only Download (1MB) |
|
|
Text (BAB IV)
Bab 4.pdf Restricted to Registered users only Download (1MB) |
|
|
Text (BAB V)
Bab 5.pdf Restricted to Registered users only Download (546kB) |
|
|
Text (DAFTAR PUSTAKA)
Daftar Pustaka.pdf Restricted to Registered users only Download (631kB) |
|
|
Text (LAMPIRAN)
Lampiran.pdf Restricted to Registered users only Download (2MB) |
Abstract
This study aims to implement an interactive weather chatbot dashboard capable of receiving natural Indonesian-language queries and presenting real-time weather information through text and data visualizations. The system combines SentenceBERT (paraphrase-multilingual-MiniLM-L12-v2) as a semantic similarity engine for intent classification, the Gemini 2.5-flash Large Language Model to generate natural language summaries, and the OpenWeather API as the weather data source. The proposed method applies a hybrid SBERT-based cosine similarity approach with an optimal threshold of 0.5, supported by rule-based auto-adjust and fallback mechanisms. The variables examined include intent classification accuracy, city entity extraction accuracy, and the quality of the weather responses displayed on the dashboard. Testing involved 140 representative user query scenarios that reflect various weather-related intents expressed in natural Indonesian language. The evaluation method employed functional black-box testing, covering the intent classification process, threshold performance, city entity extraction, API invocation, and the final presentation on the dashboard. The results show that SBERT-based intent classification achieved an accuracy of 81,43%, increasing to 100% across 140 representative test scenarios after applying rule-based auto-adjust and fallback mechanisms. City entity extraction also reached 100% accuracy for all queries containing city names. The integration of the Gemini LLM with the React-based dashboard successfully generates natural, informative, and easy-to-understand weather summaries through real-time interactive visualization. Based on these findings, it can be concluded that the hybrid approach combining SBERT, rule-based mechanisms, and the Gemini LLM is effective in developing an accurate, responsive, and user-friendly educational weather chatbot dashboard. Keywords: chatbot weather, Natural Language Processing, SentenceBERT, LLM Gemini, OpenWeather API. Penelitian ini bertujuan mengimplementasikan dashboard chatbot cuaca interaktif yang mampu menerima pertanyaan berbahasa Indonesia secara alami dan menyajikan informasi cuaca secara real-time dalam bentuk teks dan visualisasi data. Sistem dikembangkan dengan mengombinasikan Sentence-BERT (paraphrase-multilingual-MiniLM-L12-v2) sebagai mesin semantic similarity untuk klasifikasi intent, Large Language Model Gemini 2.5-flash untuk menghasilkan ringkasan respons bahasa alami, serta OpenWeather API sebagai sumber data cuaca. Pendekatan yang digunakan adalah hybrid SBERT berbasis cosine similarity dengan threshold optimal sebesar 0,5 yang dikombinasikan dengan mekanisme rule-based auto-adjust dan fallback. Variabel yang diteliti meliputi akurasi klasifikasi intent, akurasi ekstraksi entitas kota, serta kualitas hasil respons cuaca yang ditampilkan pada dashboard. Sampel pengujian berupa 140 skenario query pengguna yang disusun untuk merepresentasikan berbagai intent cuaca dengan menggunakan bahasa Indonesia yang natural dan sehari-hari. Metode evaluasi sistem yang digunakan adalah pengujian fungsional black-box testing terhadap alur klasifikasi intent, skor threshold yang optimal, ekstraksi entitas kota, pemanggilan OpenWeather API, hingga penyajian hasil tampilan pada dashboard. Hasil penelitian menunjukkan akurasi klasifikasi intent berbasis SBERT mencapai 81,43% dan meningkat menjadi 100% pada 140 skenario uji setelah diterapkan mekanisme rule-based auto-adjust dan fallback, sedangkan ekstraksi entitas kota mencapai akurasi 100% pada seluruh query yang mengandung nama kota. Integrasi LLM Gemini dan dashboard React mampu menghasilkan ringkasan cuaca yang natural, informatif, dan mudah dipahami dalam bentuk visualisasi interaktif realtime. Berdasarkan hasil tersebut, disimpulkan bahwa pendekatan hybrid SBERT, rule-based, dan LLM Gemini efektif untuk membangun sistem chatbot cuaca edukatif berbasis dashboard yang akurat, responsif, dan mudah digunakan. Kata kunci: chatbot cuaca, Natural Language Processing, Sentence-BERT, LLM Gemini, OpenWeather API.
Actions (login required)
 |
View Item |
