Multi-label Aspect Categorization with Convolutional Neural Networks and Extreme Gradient Boosting

Abstract

Aspect categorization is a subtask in aspect-based sentiment analysis. This task is a multi-label classification task since each document may belong to more than one aspect categories. In this paper, we divided the aspect categorization task into two subtasks, feature extraction and multi-label classification. We use Convolutional Neural Network (CNN) as feature extractor and Extreme Gradient Boosting (XGBoost) as the top-level classifier. Dataset used in this paper consists of 9450 hotel reviews. Our model achieved higher F1-score, 0.9316, and lower Hamming Loss, 0.02667, compared to our baselines which are CNN-LSTM (F1 0.9198 & loss 0.03217) and CNN-SVM (F1 0.9295 & loss 0.02719).