Bengali Documents Categorization Using Deep Learning

Abstract

In today's world, the amount of information and data is increasing day by day. After all, as the internet is readily available, a large amount of information and data is being stored in online cloud servers. Nowadays due to the availability of internet no one waits for newspapers anymore, everyone is more inclined towards online news portals. Online news portals are social media, Facebook, twitter, WhatsApp, telegram, Instagram, messenger, LinkedIn, blog etc. The amount of news on these online news portals is constantly increasing at a huge rate. As a result, the number of online readers is also increasing day by day. The need for data classification is increasing for all these digital data. There is a lot of data in the world that is not classified, such data is known as unusable data. Because we can use usable data but unusable data cannot be used for any purpose. The resulting data needs to be classified to become usable data. This research paper of ours is basically categorization of Bangla documents or data through deep learning. Our dataset had a total of 19137 data from which 18999 data were obtained by cleaning the data. Out of 18999 data, 13679 data have been taken for training, 1900 data for testing and 3420 data for validation check. The documents in our data set are divided into 12 categories, such as Politics, Education, Sports, Entertainment, Crime, Opinion, Accident, International, Environment, Economics, Science_Tech, Art. There are many types of deep learning models like CNN, LSTM, ANN, SBM etc. Among all the deep learning algorithms, the CNN model of our research paper is used. Using CNN model, we got 78.95% accuracy.