A Classification Model for Thai Statement Sentiments by Deep Learning Techniques

Abstract

At present, many organizations realized the importance of sentiment analysis for consumer reviews. The positive and negative comments can help to evaluate the user satisfaction of products and services to control and improve their qualities. In addition, the deep learning techniques are very interesting methods for current researches in the data mining field. Therefore, this research studied on the deep learning techniques to analyzed user reviews and comments in Thai Language from the TripAdvisor website. To begin with, user comments in four categories: hotels, restaurants, tourist attractions, and airlines were collected and tested on the combination of two basic deep learning technique that are convolutional neural network and long-short term memory. All user comments were divided into individual statements to classify into three groups: positive feelings, negative feelings, non-expressed feelings or neutrality. The research results found that the best classification model is the combination of three convolutional neural networks with 32, 64, and 128 filters, respectively, and the kernel size of 2 equal to the three components. Moreover, the performance of the proposed classification model was evaluated by accuracy, precision, and recall values which were higher than 80% in positive and negative groups, including F1 score about 0.8.