การคัดเลือกคุณลักษณะที่มีประสิทธิภาพในการจำแนกความคิดเห็นสำหรับการปรับปรุงหลักสูตร THE EFFECTIVE FEATURES SELECTION THROUGH OPINION CLASSIFICATION FOR CURRICULUM ADJUSTMENT

Abstract

           This research presents the opinion classification based on machine learning techniques for creating opinion classification model. We compared for finding the effective feature selection in the opinion classification for curriculum improvement. Based on the questionnaire of 1,575 sets, Feature Selection based on Term Frequency-Inverse Document Frequency method, Chi-Square method and Information Gain method. The testing of effective on opinion classification with Naïve Bayes Algorithm, K-Nearest Neighbors, Support Vector Machine.

            It was found that when the value of Threshold was greater than or equal to 1 for opinion classification, TF-IDF method indicated that the Support Vector Machine algorithm was the most accurate, with an accuracy of 90%. In conclusion, the result for comparison of features selection results from the TF-IDF method provided more accurate and effective for curriculum adjustment than other methods. This research result could be the guideline for effectiveness of the development on the recommendation system for adjust curriculum with text mining.