Analisis Sentimen Opini Masyarakat terhadap Platform Telemedicine menggunakan Metode Kamus Sentimen Gabungan dan Support Vector Machine
Asy Syifaur Roisah Rufaida, Ir. Adhistya Erna Permanasari, S.T., M.T., Ph.D.; Ir. Noor Akhmad Setiawan, S.T., M.T., Ph.D., IPM.
2023 | Tesis | S2 Teknologi Informasi
Public opinion through social media can enhance the quality of telemedicine applications. Sentiment analysis aids in classifying user opinions of these applications. Previous research has shown that adding lexicons and combining lexicon methods with machine learning can improve sentiment analysis performance. However, relying solely on a sentiment lexicon may not accurately determine the sentiment class of words. Therefore, this study combines the InSet (Indonesia Sentiment Lexicon) and Sentiwords in the hybrid lexicon-Support Vector Machine (SVM) method to enhance sentiment analysis performance in telemedicine. The analyzed telemedicine applications are Halodoc and Alodokter.
The labeled data, using sentiment lexicons and TF-IDF, are used as training and testing data for SVM. Evaluation is conducted using the repeated k-fold cross-validation procedure with 10 folds and three repetitions. Classification reports are generated for each fold, providing precision, recall, F1-score, and accuracy values. A comparison between the proposed method and the method using a single sentiment lexicon is performed using a t-test with ?=0.05. Due to three repetitions, Bonferroni correction is applied, resulting in a corrected alpha value of 0.016.
The comparison results indicate that Sentiwords performs the poorest in all evaluation scenarios, while InSet dominates with the best performance. For the Halodoc document, InSet achieves a precision of 85.92%, recall of 86.24%, F1-score of 85.76%, and accuracy of 86.24%. For the Alodokter document, InSet outperforms in recall and accuracy with values of 84.28%, while the hybrid lexicon performs better in precision with 83.90% and an F1-score of 83.70%. The t-test indicates that the performance of the hybrid lexicon method is significantly superior only compared to the Sentiwords, while compared to the InSet, the performance of the hybrid lexicon is lower, although not significantly. This is due to the merging of lexicons that did not consider the weights of synonymous words, resulting in inconsistency between the weight of a word and the weight of its synonymous words.
Kata Kunci : Analisis Sentimen, Lexicon, Support Vector Machine, Indonesia Sentiment Lexicon, SentiStrength_id