Convolutional neural networks for Thai poem classification

Abstract

In this work, we propose a Convolutional Neural Networks (CNNs) that able to be unsupervised feature learning to classify Thai poem (Klon-8) categories and Thai poem sentiment analysis. Thai poem has prosody, syllable rhyme and rhythm, there are challenges and different from prose text classification. The input of model representation by the vector (word2vec) generated from Thai-Text corpus 5.9 Million words. We perform the experiments by comparing with Support Vector Machine (SVM) and Naïve Bayes. CNNs showed the performance of poem categories 83% and performance of sentiment analysis 61%. CNNs showed a good performance, although unused knowledge about the composition of the poem for feature extraction.