STAR: Stepwise Task Augmentation with Relation Learning for Aspect Sentiment Quad Prediction

Aspect-based sentiment analysis (ABSA) aims to identify four sentiment elements, including aspect term, aspect category, opinion term, and sentiment polarity. These elements construct a complete picture of sentiments. The most challenging task, aspect sentiment quad prediction (ASQP), requires predicting all four elements simultaneously and is hindered by the difficulty of accurately modeling dependencies among sentiment elements. A key challenge lies in the scarcity of annotated data, which limits the model ability to understand and reason about the relational dependencies required for effective quad prediction. To address this challenge, we propose a stepwise task augmentation framework with relation learning that decomposes ASQP into a sequence of auxiliary subtasks with increasing relational granularity. Specifically, STAR incrementally constructs auxiliary data by augmenting the training data with pairwise and overall relation tasks, enabling the model to capture and compose sentiment dependencies in a stepwise manner. This stepwise formulation provides effective relational learning signals that enhance quad prediction performance, particularly in low-resource scenarios. Extensive experiments across four benchmark datasets demonstrate that STAR consistently outperforms existing methods, achieving average F1 improvements of over $2%$ under low-resource conditions.

💡 Research Summary

The paper tackles the most demanding sub‑task of aspect‑based sentiment analysis, Aspect Sentiment Quad Prediction (ASQP), which requires simultaneous prediction of four elements: aspect term, aspect category, opinion term, and sentiment polarity. Existing approaches either treat the problem as a single end‑to‑end generation task or rely on data‑augmentation techniques that permute element orders. While these methods improve performance, they do not explicitly model the relational dependencies among the four elements, and they remain heavily dependent on large annotated corpora.

To address these issues, the authors propose STAR (Stepwise Task Augmentation with Relation learning), a framework that decomposes ASQP into three hierarchical subtasks of increasing relational granularity: (1) Quad Prediction, (2) Pairwise Relation, and (3) Overall Relation. The first subtask follows the Multi‑View Prompting (MVP) paradigm, assigning distinct markers (