Deep Neural Network in Text Classification

Inhalt:
Text classification is the task to assign labels to the related piece of text. Nowadays, texts are believed to be among the most informative types of unstructured data, waiting to be explored. Many researchers have introduced deep learning models to do better on text classifications. Compared with traditional methods like SVM or KNN, deep learning models have their unique advantages. Focusing on text preprocessing and word embedding algorithms, many of them had achieved remarkable benchmark performances. Multiple feature extraction techniques and novel classifier models have all been tailored to fit textual data rather well. With such a great step forward in this field of research, a detailed summary is in need. This review covered the most recent milestones of applying deep learning models to text classification tasks. We briefly overview some common approaches for deep learning models to deal with texts, mostly Graph Convolution Networks (GCN) and Graph Attention Network (GAT). Some of their outstanding variants are also discussed in this review. We have also proposed some unsolved problems to be potential research interests, pushing the application of this valuable work in real-world cases.