A Semi-Supervised Approach for Multi-Domain Classification

Published in ICEET, 2023

Recommended citation: Anirudh Garg, Kartikey Singh, Radhika Mundra. (2023). " A Semi-Supervised Approach for Multi-Domain Classification." IEEE Xplore Digital Library . https://ieeexplore.ieee.org/document/10525791

This research paper presents an innovative approach for addressing the classification of data originating from multiple domains when there is a scarcity of labeled data. While current technologies have achieved impressive accuracy in single-domain classification, the challenge of multi-domain classification persists due to contextual variations. To tackle this issue, we introduce a multi-task based unsupervised data augmentation (UDA) approach that enables learning of domain - specific data contexts. UDA is widely recognized as one of the most effective semi-supervised frameworks, as it requires only a small amount of labeled data for learning purposes. In our study, we leverage a BERT language model and train it using our proposed approach to acquire domain-aware embeddings for data assessment. By doing so, we enhance the ability to classify data from various domains accurately.

Recommended citation: Anirudh Garg, Kartikey Singh, Radhika Mundra. (2009). “A Semi-Supervised Approach for Multi-Domain Classification.” IEEE Xplore Digital Library . 1(1).