EDA: Easy Data Augmentation Techniques for Boosting Performance on Text Classification Tasks
EDA: Простые методы аугментации данных для повышения качества задач классификации текста
2019-01-01
SCID: 54.1/nev96ukp
Discuss with AI
EDAEasy Data Augmentationdata augmentationnatural language processingtext classification
Figures from the paper
Abstract (AI)
Jason Wei, Kai Zou. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). 2019.
Key Findings
1
Across multiple benchmark text classification tasks, EDA yields consistent performance gains over training without augmentation.
2
EDA is particularly effective for low-resource training data, substantially boosting accuracy when labeled examples are scarce.
3
Easy Data Augmentation (EDA) techniques improve text classification performance by applying simple operations like synonym replacement, random insertion, swap, and deletion.
4
The proposed augmentations are computationally cheap and task-agnostic, requiring no external resources beyond a synonym dictionary.
Research Object
Text classification tasks/datasets
Research Subject
Easy Data Augmentation (EDA) techniques for improving model performance on text classification
Publication Details
Publication Date
2019-01-01
Journal
Publisher
ISSN
Cited by
1960
Open access PDF
Access Type
Author Information
Download PDF
Subscribe to digest