This study investigates a semi-supervised approach to ABSA, addressing the challenges posed by limited labeled data through data augmentation and pseudo-labeling. Utilizing Sentence BERT with contextual embedding ‘RoBERTa-large’, our framework combines labeled data, augmented data, and pseudo-labeled data to enhance model performance. The data augmentation technique employs nlp textual augmentation for synonym replacement with words, which expands the training set while preserving sentiment labels. The pseudo-labeling process uses a threshold to assign sentiment labels to unlabeled data, allowing for more extensive model training without manual annotation. Our results demonstrate that the semi-supervised approach yields significant improvements in accuracy, with the model achieving 77.9% accuracy.