Text Classification using Deep Learning

Text Classification is a classical problem in Natural Language Processing (NLP) where certain sentences, paragraphs or documents need to be assigned to one or more predefined categories. Deep learning models based on recurrent structures have been able to surpass issues faced by conventional machine learning models and achieve satisfactory results in classifying text data by utilizing semantic information. While Deep Learning (DL) models have achieved state-of-the-art on many NLP tasks, these models are trained from scratch, requiring large datasets, and days to converge. These major drawbacks of DL models have been addressed through Inductive Transfer learning. Transfer Learning (TL) has thus changed the face of DL in NLP in the recent years by allowing us to take pretrained state of the art models and fine tuning them to suit the task at hand , thus obviating the need for training language models from scratch. This study, apart from exploring some DL models, would be focusing on two of the most popular TL models namely, Universal Language Model Fine Tuning (ULMFiT) and Bidirectional Encoder Representations from Transformers (BERT) that employ transfer learning to classify news articles into predefined categories.