Text classification is a classical machine learning application in Natural Language Processing, which aims to assign labels to textual units such as documents, sentences, paragraphs, and queries. Applications of text classification include sentiment classification and news categorization. Sentiment classification identifies the polarity of text such as positive, negative or neutral based on textual features. In this thesis, we implemented a modified form of a tolerance-based algorithm (TSC) to classify sentiment polarities of tweets as well as news categories from text. The TSC algorithm is a supervised algorithm that was designed to perform short text classification with tolerance near sets (TNS). The proposed TSC algorithm uses pre-trained SBERT algorithm vectors for creating tolerance classes. The effectiveness of the TSC algorithm has been demonstrated by testing it on ten well-researched data sets. One of the datasets (Covid-Sentiment) was hand-crafted with tweets from Twitter of opinions related to COVID. Experiments demonstrate that TSC outperforms five classical ML algorithms with one dataset, and is comparable with all other datasets using a weighted F1-score measure.
Text classification aims to assign labels to textual units such as documents, sentences and paragraphs. Some applications of text classification include sentiment classification and news categorization. In this paper, we present a soft computing technique-based algorithm (TSC) to classify sentiment polarities of tweets as well as news categories from text. The TSC algorithm is a supervised learning method based on tolerance near sets. Near sets theory is a more recent soft computing methodology inspired by rough sets where instead of set approximation operators used by rough sets to induce tolerance classes, the tolerance classes are directly induced from the feature vectors using a tolerance level parameter and a distance function. The proposed TSC algorithm takes advantage of the recent advances in efficient feature extraction and vector generation from pre-trained bidirectional transformer encoders for creating tolerance classes. Experiments were performed on ten well-researched datasets which include both short and long text. Both pre-trained SBERT and TF-IDF vectors were used in the experimental analysis. Results from transformer-based vectors demonstrate that TSC outperforms five well-known machine learning algorithms on four datasets, and it is comparable with all other datasets based on the weighted F1, Precision and Recall scores. The highest AUC-ROC (Area under the Receiver Operating Characteristics) score was obtained in two datasets and comparable in six other datasets. The highest ROC-PRC (Area under the Precision–Recall Curve) score was obtained in one dataset and comparable in four other datasets. Additionally, significant differences were observed in most comparisons when examining the statistical difference between the weighted F1-score of TSC and other classifiers using a Wilcoxon signed-ranks test.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.