.png){kind=icon}
Return
Leveled Learning: An Interpolation-Based Data Augmentation Method on Few-Shot Text Classification
.png){kind=icon}
PRE
AI
DOI：10.1007/978-981-96-7008-6_13.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Few-shot text classification aims to allocate text information into predefined categories using limited labeled samples. Current data augmentation techniques can mitigate the issue of limited labeled data by increasing sample numbers, yet they often overlook variations between samples. In this paper, we introduce leveled learning (LL), a novel technique based on linear interpolation, to alleviate the above problem. Specifically, we evaluate prediction difficulty for all simples, interpolate similardifficulty samples at the hidden layer, and train them sequentially from easiest to hardest to facilitate the model's learning across different training stages. Additionally, to increase the diversity of samples, we utilize a pre-trained language model to generate descriptive information for categories, subsequently using this information to generate pseudo samples. We conduct comprehensive experiments on seven commonly employed datasets, demonstrating that our proposed approach notably outperforms various interpolation-based data augmentation methods.
Keywords:
Linear Interpolation
Data Augmentation
Few-Shot Text Classification
Journal
N
IF:
0
Papers:
24
Citations:
0
Organization
Cited Papers
.png){kind=icon}
ShareSaveShareSaveSelf-Evolution Learning for Mixup: Enhance Data Augmentation on Few-Shot Text Classification Tasks
PREAI
ShareSaveShareSave
View more