Hybrid Model for Efficient Assamese Text Classification using CNN-LSTM

Abstract

In modern times, there has been an exorbitant rise in unstructured digital text, owing to the ever-increasing use of internet. Therefore, to be able to extract knowledge out of it, a perceived need was felt to organize the enormous amount of digital text into different categories. This is why Text Classification is considered a critical task in NLP (Natural Language Processing). This research suggests a hybrid model(C-LSTM-ATC) that combines the benefits of two deep learning models, namely, the Convolutional Neural Network (CNN) and Long and Short Term Memory (LSTM), to categorize Assamese text, a topic that largely remains unexplored till now. Another hybrid model was also tried by combining LSTM with the Support Vector Machine (LSTM-SVM). The C-LSTM-ATC model performs splendidly with an accuracy of 97.2% while the LSTM-SVM model outputs an accuracy of 92.2% when tested on the dataset prepared. The model was trained using as many as 768 Assamese text documents and the test results showed that the proposed C-LSTM-ATC model produces more accurate classification and higher F1 scores, than the LSTM-SVM and also the CNN and LSTM models when used separately.