Multi-label classification of text document using deep learning

Recently, studies in the field of Natural Language Processing and some of its related important problem and Applications in the machine learning field continue to mount up. Machine Learning is prove to be predominantly data-driven in the sense that generic model buildings are used and then tailored to a specific application data. Needless to say, this has proven to be a very effective approach to modeling the complicated data dependencies we frequently experience in practice, making very few assumptions and allowing the information to talk for themselves. Examples can be found in chemical process engineering, climate science, systems, healthcare, and linguistic processing of natural language, to name a few. Moreover, text classification is one of the important aspect of Natural Language Processing. Text classification is the act of categorizing text or text documents into a given set of labels. While on the other hand, multi-label text classification deals with classifying text or documents into one more labels at the same time. Over the years, some methods for classifying text and documents have been proposed, including popularly known Bag of Words (BoW) method, Supervised Machine Learning, tree induction and label-vector embedding, to mention a few. These kind of tools can be used in many digital applications, such as document filtering, search engines, document management systems, etc. Lately, Deep Learning based methods is getting more attention, especially in an Extreme Multi-Label text classification. Deep learning is one of the major solutions to many machine learning applications that involve high-dimensional and unstructured data, such as pictures and text documents. However, it is of paramount importance in many of these applications to be able to reason accurately about the uncertainties associated with the predictions of these models. Therefore in this studies, we explore multi-label classification of text documents using deep learning methods such as CNN, RNN, LSTM, and even GRU. We investigate two scenarios in the studies. Firstly, multi-label classification models with plane embedding layer, and secondly with a Glove, Word2vec, and FastText as pre-trained embedding corpus for our models. We evaluate and compare these different neural network models performances in terms of multi-label evaluation metrics with respect to the two approaches.