PERBANDINGAN KINERJA HYBRID INSET–SVM DAN SENTISTRENGTH_ID–SVM UNTUK ANALISIS SENTIMEN PLATFORM TELEMEDICINE HALODOC DAN ALODOKTER

Abstract

The quality of telemedicine platforms can be assessed through sentiment analysis of user opinions on social media. This study compared two hybrid methods, InSet-SVM and SentiStrength_id-SVM, for sentiment analysis of the Halodoc and Alodokter platforms. Both lexicons were selected because they apply similar sentiment-weighting mechanisms but differ in vocabulary coverage. Data were collected from X (formerly Twitter) using TWINT with the keywords “Halodoc” and “Alodokter” from January 1 to December 31, 2022. The data were labeled using each sentiment lexicon, while TF-IDF was applied for feature weighting before classification using a Support Vector Machine (SVM). The models were evaluated using three-times-repeated 10-fold cross-validation. The results showed that InSet-SVM outperformed SentiStrength_id-SVM across all evaluation metrics for both datasets. For Halodoc, InSet-SVM achieved 85.92% precision, 86.24% recall, an F1-score of 85.76%, and 86.24% accuracy. For Alodokter, it achieved 83.86% precision, 84.28% recall, an F1-score of 83.59%, and 84.28% accuracy. These findings indicate that using InSet as the initial labeling lexicon produces a more consistent SVM model across both datasets.