Sentiment Analysis Model for VTuber Live Stream Chat using Decision Tree and Support Vector Machine
H. Yuliansyah, H.A. Raihan, Murinto
Journal of Innovation Information Technology and Application (JINITA)
- background: The short, informal, and unstructured nature of Virtual YouTuber (VTuber) live chat makes sentiment analysis challenging, and studies comparing Decision Tree (DT) and Support Vector Machine (SVM) algorithms in this domain remain limited.
- objective: To propose an optimal sentiment analysis model for VTuber live chat by comparing the performance of DT and SVM.
- method: Live chat data underwent preprocessing and was labeled as positive, neutral, or negative using VADER and AFINN lexicons. The models used TF-IDF for feature extraction and were evaluated via K-Fold cross-validation and a confusion matrix.
- results: A 10-fold cross-validation evaluation showed that the DT + AFINN combination with hyperparameter optimization achieved the highest accuracy of 96.26%.
- conclusion: The combination of DT and AFINN is superior in analyzing VTuber live chat sentiment compared to DT+VADER, SVM+AFINN, and SVM+VADER.
