Short Text Classification with Tolerance-Based Soft Computing Method

Text classification aims to assign labels textual units such as documents, sentences and paragraphs. Some applications of text include sentiment news categorization. In this paper, we present a soft computing technique-based algorithm (TSC) classify polarities tweets well categories from text. The TSC is supervised learning method based on tolerance near sets. Near sets theory more recent methodology inspired by rough where instead set approximation operators used induce classes, the classes are directly induced feature vectors using level parameter distance function. proposed takes advantage advances in efficient extraction vector generation pre-trained bidirectional transformer encoders for creating classes. Experiments were performed ten well-researched datasets which both short long Both SBERT TF-IDF experimental analysis. Results transformer-based demonstrate that outperforms five well-known machine algorithms four datasets, it comparable with all other weighted F1, Precision Recall scores. highest AUC-ROC (Area under Receiver Operating Characteristics) score was obtained two six datasets. ROC-PRC Precision–Recall Curve) one dataset Additionally, significant differences observed most comparisons when examining statistical difference between F1-score classifiers Wilcoxon signed-ranks test.