ThaiFBDeep: A Sentimental Analysis Using Deep Learning Combined with Bag-of-Words Features on Thai Facebook Data

Abstract

Thailand has a huge number of Facebook user. Most company has their own public page to communicate with their customers. Thus, it's desirable to perform sentimental analysis on Facebook post messages to understand customer's reaction of specific promotion, event or news. This work aims to propose a novel method to perform sentimental analysis on Thai Facebook data by combining information generated from a classical Bag-Of-Words features and advance deep learning approaches called ThaiFBDeep. Remarkably, according to Thai people usually conduct new words every year, the proposed data preprocessing techniques should be able to handle this kind of words. The experiment results show that ThaiFBDeep achieved a 91.75% of train accuracy and an 83.36% of independent test accuracy which is better than other well-known methods i.e. Naïve Bayes, Support Vector Machine, Multi-Layers Perceptron, Long Short-Term Memory and Convolution Neural Networks. These results also show that the including of Bag-Of-Words features can improve efficiency of Deep Learning based approach for sentimental analysis.