Attention-Based Deep Learning Model for Aspect Classification on Vietnamese E-commerce Data

Ngoc-Tu Nguyen, Trong-Dat Nguyen, Duy-Cat Can, Mai-Vu Tran, Ha Luu Manh, Hoang-Quynh Le

Abstract: This article introduces methods for applying Deep Learning in identifying aspects from written commentaries on Shopee e-commerce sites. The used datasets are two sets of Vietnamese consumers' comments about purchased products in two domains. Words and sentences will be performed as vectors, or characteristic matrices through language models such as one-hot, fastText, PhoBERT. We then used Convolutional Neural Network (CNN) and the Fully Connected Neural Network (Multilayer perceptron - MLP) to learn the aspects which are mentioned in the comments. Experimental results showed that our team's methods achieved much better results than traditional learning algorithm using other word-level vectors such as SVM, Naïve Bayes, etc.