Prompt-based data labeling method for aspect based sentiment analysis

Kun Bu, Yuanchao Liu

Abstract: ABSA aims to extract aspect terms and corresponding sentiment from unstructured texts. Supervised approaches are widely used in existing ABSA models because of their model maturity, and most of them usually need large-scale training data to deal with over-fitting. However, in real scenarios, the labeled data is difficult to obtain, thus the performance is adversely influenced. To address these issues, this paper proposes a prompt-based data augmentation method, enabling it to overcome small data problems by expanding the sample size in the training corpus. Our approach computes the relationship between the prompt templates and unlabeled data and then assigns labels to expand the training data. To achieve this, we formulate it as a data filtering problem and implement it with Natural Language Inference models. The experimental results on four well-studied datasets demonstrate that our model not only achieves results on par with existing state-of-the-art data augmentation methods on a few occasions but also significantly improves the effectiveness of existing ABSA models on most occasions, indicating its strong robustness in various base ABSA models. Further discussion shows that prompt learning can help the model mark data from unlabeled datasets, which explains its effectiveness in data augmentation.