Skripsi
ANALISIS SENTIMEN PENGGUNA MEDIA SOSIAL X TERHADAP FILM SORE: ISTRI DARI MASA DEPAN MENGGUNAKAN ALGORITMA SUPPORT VECTOR MACHINE (SVM)
Indonesia’s film industry is rapidly expanding with diverse genres and new titles, creating a need for fast, data-driven mapping of audience sentiment. This study maps social media X users’ sentiment toward the film “Sore: Istri dari Masa Depan,” which carries an unconventional theme, while addressing limited manual labels on short and noise tweets. Support Vector Machine (SVM) is selected for its robustness in high-dimensional feature spaces and effectiveness on short texts. A pseudo-labeling scheme replaces large-scale manual annotation by delegating label assignment to the model through calibrated zero-shot predictions combined with lexicon scores as an agreement filter and an abstain mechanism. The workflow covers deduplication, language filtering, preprocessing, TF-IDF feature extraction on word and character n-grams with meta-features, chi-square feature selection, and SVM training with an 80:20 stratified split and K-Fold Cross-Validation. Evaluation on the test set yields 82.36% accuracy and a macro F1 of 70.06%. The three-class confusion matrix shows dominant positive opinions and more confusion between neutral and negative classes. These findings position pseudo-labeling as a pragmatic trade-off between label quality and annotation efficiency while indicating generally positive public reception.