Persian text classification via character-level convolutional neural networks

Text classification is an important task in the field of text mining. Due to the ever-increasing volume of text data on the Internet, text classification is needed more than ever before. In the last decades, researchers have been trying to offer accurate models to distinguish data of each class from the other class. Many researches aimed to find a set of features that would increase the accuracy of classification. Recently, deep learning has been considered as the ideal approach in big data issues, and has proved the impact of using convolutional networks on improving the results. These models mainly represent text based on words. Due to the complexity of natural languages and the importance of having extra knowledge, researchers are still trying to achieve a more efficient method. In this research, we used a deep convolutional neural network and applied it on the Persian corpus. A remarkable feature of this model is the use of character-based convolutional layers to extract features from text, which has a higher accuracy (0.49) than other character-based representations.