Zero-Shot Learning For Text Classification: Extending Classifiability Beyond Conventional Techniques

Abstract

Text classification plays a crucial role in organizing and understanding huge amounts of text data. However, traditional text classification methods often face challenges when dealing with unseen or novel classes. Zero-shot learning (ZSL) offers a promising solution to this problem by enabling the classification of text instances into classes that have not been encountered during training. There is a plethora of potential benefits of ZSL in several applications, emphasizing its ability to handle new classes and adapt to evolving domains. In this paper, we have used the AG news dataset which is a commonly used benchmark dataset for text classification tasks. It consists of news articles from the AG's corpus, collected from four different categories: World, Sports, Business, and Science/Technology. Each article is assigned a label corresponding to one of these categories. We applied state-of-the-art deep learning algorithms such as Convolutional Neural Networks and Recurrent Neural Networks to compare the performance with Zero Shot Learning (ZSL). ZSL proved to be robust and performed better compared to the other algorithms in terms of accuracy and F1 Score.