STAR-OPD: Structured Aspect-Cascade-Aware On-Policy Reward Distillation for ABSA Quadruple Extraction
On the map: Knowledge distillation → Multimodal distillation
Abstract (arXiv)
Aspect-based sentiment analysis (ABSA) quadruple extraction requires jointly predicting target, aspect, opinion, and sentiment over reviews that often contain multiple fine-grained sentiment tuples. While large chain-of-thought (CoT) models perform well on this task, distilling them into smaller deployable models remains difficult. We identify a task-specific failure mode in distilled ABSA extraction: student errors at the target-aspect interface create structurally invalid states, such as broken target-aspect bindings and hallucinated targets, which then corrupt downstream predictions. Conventional off-policy distillation is poorly suited to this setting because it trains only on teacher-generated trajectories and provides little supervision on the student-induced structural states that dominate inference. To address this mismatch, we propose STAR-OPD (STructured Aspect-cascade-aware On-Policy Reward Distillation), which builds on generic on-policy distillation and instantiates it for ABSA quadruple extraction with cascade-aware, set-structured rewards. STAR-OPD trains on student rollouts and applies set-structured rewards that directly target binding consistency, target grounding, and fine-grained aspect disambiguation. Experiments on E-ABSA20K and SemEval-2014 show that STAR-OPD consistently outperforms off-policy and general on-policy baselines, reduces target hallucination, and substantially improves performance on structurally hard cases. With Qwen3-4B, STAR-OPD substantially narrows the student-teacher gap while improving inference efficiency, highlighting the importance of on-policy structural correction for distilled ABSA extraction.
Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in Multimodal distillation, one of 12 directions in Knowledge distillation. Builds on 12 of its 20 references that are in the map. First posted on arXiv 21 Aug 2026.
Builds on
12 of 20 references.
References · 12
- 2023Efficient Memory Management for Large Language Model Serving with PagedAttentioncited by 5,944
- 2023On-Policy Distillation of Language Models: Learning from Self-Generated Mistakescited by 939
- 2015Scheduled Sampling for Sequence Prediction with Recurrent Neural Networkscited by 803
- 2016Sequence-Level Knowledge Distillationcited by 801
- 2022Large Language Models Are Reasoning Teacherscited by 329
- 2022Teaching Small Language Models to Reasoncited by 284
- 2026Learning beyond Teacher: Generalized On-Policy Distillation with Reward Extrapolationcited by 184
- 2026Entropy-Aware On-Policy Distillation of Language Modelscited by 175
- 2019Knowing What, How and Why: A Near Complete Solution for Aspect-based Sentiment Analysiscited by 123
- 2022Improving Aspect Sentiment Quad Prediction via Template-Order Data Augmentationcited by 30
- 2023InstructABSA: Instruction Learning for Aspect Based Sentiment Analysiscited by 23
- 2025RLKD: Distilling LLMs' Reasoning via Reinforcement Learningcited by 11
Built on by
Not cited yet.
Similar, not cited
Close in wording, no citation either way
- 2026Dependency Reasoning for Aspect Sentiment Quadruple Prediction
- 2021Unified Instance and Knowledge Alignment Pretraining for Aspect-based Sentiment Analysis
- 2026Large-Scale Aspect-Based Sentiment Analysis with Reasoning-Infused LLMs
- 2023On the Robustness of Aspect-based Sentiment Analysis: Rethinking Model, Data, and Training
- 2025AF-MAT: Aspect-aware Flip-and-Fuse xLSTM for Aspect-based Sentiment Analysis
- 2026DanceHA: A Multi-Agent Framework for Document-Level Aspect-Based Sentiment Analysis
- 2026MASTE: A Multi-Agent Pipeline for Zero-Shot Aspect Sentiment Triplet Extraction
- 2024DS2-ABSA: Dual-Stream Data Synthesis with Label Refinement for Few-Shot Aspect-Based Sentiment Analysis