Generate Then Correct: Single Shot Global Correction for Aspect Sentiment Quad Prediction

TL;DR

This paper introduces Generate-then-Correct (G2C), a novel approach for aspect sentiment quad prediction that drafts and globally corrects predictions in a single pass, addressing order sensitivity and exposure bias issues.

Contribution

The paper proposes G2C, a sequence-level global correction method that improves aspect sentiment quad prediction by reducing error propagation and order sensitivity.

Findings

G2C outperforms strong baselines on Rest15 and Rest16 datasets.

Single-shot correction effectively mitigates error propagation in quad prediction.

The method leverages LLM-synthesized drafts for training the corrector.

Abstract

Aspect-based sentiment analysis (ABSA) extracts aspect-level sentiment signals from user-generated text, supports product analytics, experience monitoring, and public-opinion tracking, and is central to fine-grained opinion mining. A key challenge in ABSA is aspect sentiment quad prediction (ASQP), which requires identifying four elements: the aspect term, the aspect category, the opinion term, and the sentiment polarity. However, existing studies usually linearize the unordered quad set into a fixed-order template and decode it left-to-right. With teacher forcing training, the resulting training-inference mismatch (exposure bias) lets early prefix errors propagate to later elements. The linearization order determines which elements appear earlier in the prefix, so this propagation becomes order-sensitive and is hard to repair in a single pass. To address this, we propose a method, Generate-then-Correct (G2C): a generator drafts quads and a corrector performs a single-shot, sequence-level global correction trained on LLM-synthesized drafts with common error patterns. On the Rest15 and Rest16 datasets, G2C outperforms strong baseline models.