Supervised Contrastive Learning for Short Text Classification in Natural Language Processing

• Published in: International eConference on Computer and Knowledge Engineering (Online) pp. 354 - 358
• Main Authors: Esmaeili, Mitra, Vahdat-Nejad, Hamed
• Format: Conference Proceeding
• Language: English
• Published: IEEE 19.11.2024
• Subjects: Accuracy; Contrastive learning; Hurricanes; Machine learning; Natural language processing; Organizations; Robustness; Semantic understanding; Social networking (online); Stability analysis; Supervised contrastive learning; Text categorization; Text classification
• ISSN: 2643-279X
• DOI: 10.1109/ICCKE65377.2024.10874505

In recent years, the swift progress in information retrieval technologies has positioned text classification as a key area of research. Classifying short texts represents a major challenge within natural language processing. Given the growing prevalence of social media during critical events like hurricanes, accurately categorizing these texts is essential for facilitating relief operations. Tweets are concise and have an informal tone, which creates unique challenges for effective classification. Supervised contrastive learning has recently become popular as a strong machine learning method. It offers significant improvements over traditional approaches, especially in the area of natural language processing. This paper introduces a supervised contrastive learning methodology designed to enhance the accuracy of short-text classification while maintaining the model's generalization capability. Our approach consistently surpasses existing state-of-the-art techniques, delivering better accuracy and stability across different ranges of text classification tasks.