An Effective Contextual Language Ensemble Model for Vietnamese Aspect-based Sentiment Analysis

Abstract

Aspect-based sentiment analysis (ABSA) allows finer-grained inferences to provide specific sentiment for each aspect of the same sentence. In this paper, we present an ensemble model combined with multi-task learning based on different pre-trained contextual language models on a compound task as Category-Sentiment Classification (CSC) for the Vietnamese language. Furthermore, we provide the performance of fine-tuning state-of-the-art pre-trained language BERTology models, which are available for the Vietnamese language. Experimental results demonstrate that our ensemble approach consistently achieves the best results in two out of three datasets benchmark datasets compared to previous results and individual models.