Rancang Bangun dan Evaluasi Sistem Named Entity Recognition Berbasis Transformer untuk Analisis Berita Berbahasa Indonesia

Muhammad Faiz Khoiri, Ir. Alif Subardono, S.T., M.Eng.

2026 | Tugas Akhir | D4 TEKNOLOGI JARINGAN

Named Entity Recognition (NER) merupakan salah satu tugas dalam pemrosesan bahasa alami yang bertujuan untuk mengidentifikasi dan mengklasifikasikan entitas penting pada suatu teks. Proyek akhir ini bertujuan untuk merancang dan membangun sistem NER berbasis model Transformer untuk analisis berita berbahasa Indonesia, serta mengevaluasi kinerja model dan sistem yang dikembangkan. Model XLM-RoBERTa Base di-fine-tuning menggunakan dataset berita berbahasa Indonesia yang telah dianotasi, kemudian diimplementasikan sebagai layanan REST API menggunakan FastAPI dan di-deploy menggunakan Docker pada cloud VPS. Selain itu, dikembangkan antarmuka pengguna berbasis Streamlit untuk memudahkan proses inferensi. Evaluasi model dilakukan menggunakan metrik precision, recall, F1-score, dan accuracy, sedangkan evaluasi kinerja sistem dilakukan melalui performance testing dan load testing menggunakan Locust dengan mengukur response time, throughput, serta penggunaan sumber daya komputasi. Hasil pengujian menunjukkan bahwa model memperoleh F1-score sebesar 0,9120 dengan precision 0,8928, recall 0,9320, dan accuracy 0,9779. Selain itu, sistem berhasil diimplementasikan dan menunjukkan peningkatan response time serta penggunaan CPU seiring bertambahnya beban kerja, sedangkan throughput mencapai kapasitas maksimum layanan dan penggunaan RAM relatif stabil.

Named Entity Recognition (NER) is a task in natural language processing that aims to identify and classify important entities in a text. This final project aims to design and develop a Transformer-based NER system for Indonesian-language news analysis, as well as to evaluate the performance of both the model and the developed system. The XLM-RoBERTa Base model was fine-tuned using an annotated Indonesian news dataset and implemented as a REST API service using FastAPI. The service was deployed with Docker on a Cloud VPS, while a Streamlit-based web interface was developed to facilitate the inference process. Model evaluation was conducted using precision, recall, F1-score, and accuracy metrics. System performance was evaluated through performance testing and load testing using Locust by measuring response time, throughput, and computational resource utilization. The evaluation results show that the model achieved an F1-score of 0.9120 with a precision of 0.8928, a recall of 0.9320, and an accuracy of 0.9779. In addition, the system was successfully deployed and demonstrated increased response time and CPU utilization as the workload increased, while throughput reached the service's maximum capacity and memory usage remained relatively stable.

Kata Kunci : Named Entity Recognition, Transformer, XLM-RoBERTa Base, FastAPI, Cloud VPS