Tirukkural Couplet Classification with Long Short-Term Memory Neural Network

Abstract

In the modern machine learning world, the text classification is playing major role in organizing the knowledge in a structured way so as to apply the deep analysis over the letters, words, sentences and the structured contextual semantical meanings of the text corpus. In this paper, we have considered the three classes of  Tirukkural corpus labelled as ARAM, PORUL and INBAM with 7 snippet key terms in each couplet. Tirukkural comprises of 1330 couplets of 133 chapters. As determined by the great author of Tirukkuṟaḷ Tiruvalluvar, these documents were categorized into three classes as mentioned earlier. The meaning for the ARAM, PORUL and INBAM is Morality, Materialism and Love respectively. The collections of classification models were built using various algorithms with the help of the training set of the Tirukkural corpus and then these models were compared with the testing set couplets of Tirukkural. The classification performance accuracy and the errors were visualized with plots and numerical values.  A comprehensive experiment is also conducted using bisectional Long Short-Term Memory (LSTM) deep learning networks and the performance is visualized in this study. The proposed work is the first work in Tamil Language with respect to Tirukkuṟaḷ using deep learning technique like LSTM – RNN. In proposed work, the accuracy of the various classifiers like Linear Regression, Random forest, Naïve Bayes Multinomial, and LSTM – RNN are compared. The accuracy of linear regression classifier and LSTM – RNN are high when compared to other classifiers like Random forest and Naïve Bayes Multinomial. The accuracy value of LSTM – RNN classifier is approximately 65%.