Sentiment Analysis of Student Comments on Facilities and Infrastructure at Instiki Using Retrieval Augmented Generation

Ni Putu Juliana Dewi
I Kadek Dwi Gandika Supartha
I Putu Yoga Indrawan
Ketut Jaya Atmaja

Indonesian Journal of Data and Science
Vol. 6 No. 3 (2025): Indonesian Journal of Data and Science

This research was conducted to analyze the sentiment of student comments on infrastructure facilities at the Indonesian Institute of Business and Technology (INSTIKI) to overcome the problem of comment analysis that was previously done manually. The data used is in the form of student comments in 2024. The method used in this study is Retrieval Augmented Generation (RAG) with data labeling using Lexicon-Based. The test was carried out on three Large Language Models (LLMs), namely indobenchmark/indobert-base-p1, TinyLlama/TinyLlama-1.1B-Chat-v1.0, and w11wo/indonesian-roberta-base-sentiment-classifier. The test results showed that the indobenchmark/indobert-base-p1 model produced the highest accuracy of 80% in both test sessions compared to other models. The TinyLlama/TinyLlama-1.1B-Chat-v1.0 model produced 60% accuracy in session 1 and 65% in session 2, while the w11wo/indonesian-roberta-base-sentiment-classifier model produced 60% accuracy in both test sessions. The difference in the performance of these three LLMs shows that the model's understanding of Indonesian can affect the results of sentiment predictions.