Abstract:
The increasing number of Thai research articles makes it challenging to classify them into sub-categories. This task requires specialists and a lot of time to classify the different types of articles. Therefore, this research presents methods and techniques for multi-label classification of computer science articles in Thai journals. We present a comparison of different methods for multi-label classification, including Binary Relevance (BR), Classifier Chains (CC), and Label Power-set (LP) with a word segmentation method that uses a Support Vector Machine (SVM) classifier. We found that the CC-SVM-RBF kernel method combined with pythainlp word segmentation and TF-IDF produces the best results for both example-based and label-based metrics, with ML-accuracy is 0.578, Subset accuracy is 0.300, Recall is 0.670 and Micro-average recall is 0.670 On the other hand, BR-SVM-RBF combined with pythainlp word segmentation and TF-IDF produces the best results for both example-based and label-based metrics with Hamming loss is 0.106, Precision is 0.735, F-measure is 0.655, Micro-average precision is 0.586 and Micro-average F-Measure is 0.715. In Future work, Subset accuracy should be improved for the multi-label classification model in the Thai language.