Zero-Shot Text Classification via Self-Supervised Tuning
This addresses the problem of zero-shot text classification for NLP practitioners by offering a less template-sensitive and data-efficient method, though it is incremental as it builds on existing self-supervised and tuning paradigms.
The paper tackles zero-shot text classification by introducing self-supervised tuning with unlabeled data, using a first sentence prediction objective, and reports outperforming state-of-the-art baselines on 7 out of 10 tasks while being less sensitive to prompt design.
Existing solutions to zero-shot text classification either conduct prompting with pre-trained language models, which is sensitive to the choices of templates, or rely on large-scale annotated data of relevant tasks for meta-tuning. In this work, we propose a new paradigm based on self-supervised learning to solve zero-shot text classification tasks by tuning the language models with unlabeled data, called self-supervised tuning. By exploring the inherent structure of free texts, we propose a new learning objective called first sentence prediction to bridge the gap between unlabeled data and text classification tasks. After tuning the model to learn to predict the first sentence in a paragraph based on the rest, the model is able to conduct zero-shot inference on unseen tasks such as topic classification and sentiment analysis. Experimental results show that our model outperforms the state-of-the-art baselines on 7 out of 10 tasks. Moreover, the analysis reveals that our model is less sensitive to the prompt design. Our code and pre-trained models are publicly available at https://github.com/DAMO-NLP-SG/SSTuning .