Tifanny Nabarian, Siti Nurhalizah, Salman El Farisi

Analisis Perbandingan Kinerja IndoBERT dan TF-IDF dalam Mengklasifikasikan Sentimen EDOM Menggunakan Algoritma K-Nearest Neighbor

Introduction

Analisis perbandingan kinerja indobert dan tf-idf dalam mengklasifikasikan sentimen edom menggunakan algoritma k-nearest neighbor. Bandingkan IndoBERT & TF-IDF dalam klasifikasi sentimen komentar EDOM mahasiswa STT NF menggunakan KNN. IndoBERT lebih akurat (0.903) untuk analisis objektif.

Abstract

Dalam konteks pendidikan tinggi, dosen berkontribusi besar terhadap peningkatan kualitas pembelajaran. Di Sekolah Tinggi Teknologi Terpadu Nurul Fikri (STT NF), Evaluasi Dosen oleh Mahasiswa (EDOM) dilaksanakan setiap akhir semester dan menghasilkan data komentar mahasiswa. Namun, analisis masih dilakukan secara manual sehingga kurang efisien dan berpotensi subjektif. Selain itu, belum terdapat kajian komparatif mengenai metode representasi yang sesuai untuk digunakan bersama algoritma K-Nearest Neighbor (KNN) khususnya pada data EDOM. Tujuan dari penelitian ini adalah menganalisis dan membandingkan kinerja IndoBERT dan TF-IDF dalam merepresentasikan teks untuk klasifikasi sentimen komentar EDOM menggunakan KNN. Metode penelitian mengacu pada tahapan CRISP-DM dengan dataset komentar EDOM tahun 2024. Hasil penelitian menunjukkan bahwa IndoBERT+KNN menghasilkan accuracy sebesar 0,903 serta menunjukkan nilai precision, recall, dan F1-score yang lebih seimbang antarkelas dibandingkan TF-IDF+KNN yang memperoleh accuracy sebesar 0,820 dengan performa metrik evaluasi yang cenderung kurang seimbang antarkelas. Hasil ini menunjukkan representasi kontekstual IndoBERT lebih efektif dalam menangani kompleksitas komentar EDOM dan algoritma KNN yang berbasis jarak dibandingkan dengan pendekatan berbasis frekuensi kata pada TF-IDF. Berdasarkan temuan tersebut, penelitian ini memberikan pemilihan metode representasi teks yang lebih optimal untuk pengembangan analisis sentimen secara lebih objektif dan efisien

Review

The study effectively addresses a significant challenge within higher education institutions concerning the manual and potentially subjective analysis of student feedback, specifically Evaluasi Dosen oleh Mahasiswa (EDOM) comments at STT NF. The authors clearly articulate the research gap regarding the comparative performance of text representation methods, particularly IndoBERT and TF-IDF, when used with the K-Nearest Neighbor (KNN) algorithm for sentiment classification on this specific type of data. The objective to objectively analyze and compare these methods is well-defined and highly relevant for improving efficiency and objectivity in processing valuable student feedback. Methodologically, the research adheres to the CRISP-DM framework, leveraging a recent dataset of EDOM comments from 2024. The core of the study lies in its direct comparison of two distinct text representation paradigms: IndoBERT's contextual embeddings and TF-IDF's frequency-based approach, both feeding into a KNN classifier. The results strongly indicate IndoBERT+KNN's superior performance, achieving an accuracy of 0.903 compared to TF-IDF+KNN's 0.820. Furthermore, the paper highlights IndoBERT+KNN's more balanced precision, recall, and F1-score across different sentiment classes, underscoring its robustness. This difference is insightfully attributed to IndoBERT's capacity to capture the contextual complexity of EDOM comments, making it more compatible with a distance-based algorithm like KNN than the simpler frequency-based TF-IDF. This research offers a valuable contribution by providing empirical evidence for the optimal text representation method for sentiment analysis within an Indonesian higher education context. The findings provide a clear recommendation for adopting IndoBERT+KNN, which has practical implications for developing more objective, efficient, and reliable automated sentiment analysis systems for EDOM. By offering a data-driven solution, the study facilitates a more insightful and less biased interpretation of student feedback, ultimately aiding in data-informed decision-making to enhance the quality of teaching and learning in educational institutions.

Full Text

Comments

You need to be logged in to post a comment.