Developing Multi-Label Classification Model for Improving Text Categorizing Problems a Case of Traffy* Fondue Platform

Abstract

This paper develops an integrated deep-learning-based model for a multi-label text classification problem in order to enhance the efficiency of Traffy* Fondue which is a comprehensive platform for diverse issues of citizen’s complaints over the Bangkok metropolitan area. The dataset of Traffy* Fondue has been found to have inaccuracy problems of label categorization, especially in the 'Others' category, which results in delaying of coordination and problem-solving to address the complaint on time. To overcome this problem, this study develops an integrated deep-learning-based model for multi-label text classification problem. Five main methods have been applied to model text classification, namely, Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), Bi-LSTM, fastText and WangchanBERTa. The performance of developed models has been tested against traditional algorithms. The result of modeling in the 'Others' dataset shows that the Bi-LSTM-, CNN+Bi-LSTM- and WangchanBERTa-based approaches are the best three models that outperform others with higher precision, recall, and F1 score. Our approach offers a promising solution to expedite issue resolution and improve coordination within Bangkok's civic management framework.