Deep Neural Networks for Text Classification

Abstract

Text classification is a NLP technique that groups open-ended text into a set of predefined categories. Within a certain amount of text, it is a task we can easily handle using our knowledge and common sense. But as the amount of text waiting to be classified becomes larger, manually labeling the text would cost too many resources. People employ algorithms to automate the classification process to account for the massive input text. In terms of algorithms, there are rule-based algorithms and data-driven algorithms. Because of the limitation inherent in the rule-based procedures, it is only useful in very limited circumstances. On the other hand, data-driven algorithms take advantage of the rich available data we have on the internet and make predictions based on the previously observed data. In general, the latter procedure has a wider application and oftentimes produces more accurate results. In the early years, data-driven procedures in text classification were dominated by machine learning algorithms. But with the development of deep learning, DL methods start to take the place of ML methods in text classification. In recent years, graph neural networks(GNN) gained much traction as researchers explore new DL models. Researchers have examined the use of GNNs in many domains, and a few explored the use of GNNs in TC. This paper serves as an overview of the recent development of TC with a focus on GNNs.