Enhancing Aspect-based Sentiment Analysis through Supervised Contrastive Learning with Induced Tree Structures

A facet of sentiment analysis is Aspect-based Sentiment Analysis (ABSA), which entails identifying and assessing sentiments towards particular aspects or attributes within text data. In ABSA, an induced tree can illustrate the connections between various aspects of a text and the corresponding sentiments. By employing supervised contrastive learning, models for ABSA can be trained using both labeled and unlabeled data to boost performance. In this research, we fused induced tree structures from the pre-trained BERT model with supervised contrastive learning to empower the model to effectively distinguish between different aspects and sentiments. This method assists the model in grasping subtle sentiment variations towards various aspects of products or services, leading to more accurate sentiment analysis outcomes. The experimental results showcase the effectiveness of our integrated framework in delivering precise sentiment forecasts and profound insights into sentiment-attribute relationships, as evidenced by the enhanced classification accuracy on SemEval2014 benchmarks.