Research map / Graph learning / Knowledge and temporal graphs
You Only Read Once: Constituency-Oriented Relational Graph Convolutional Network for Multi-Aspect Multi-Sentiment Classification
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes YORO, a one-to-many ABSA framework that predicts sentiments of all aspects in a sentence at once using a BERT encoder and a constituency-oriented relational GCN (CorrGCN) with auxiliary opinion classification and supervised contrastive loss.
Problem
Existing aspect-based sentiment analysis models predict one aspect at a time, which is inefficient and under-models relations among multiple aspects in multi-aspect, multi-sentiment sentences.
Approach
Builds a graph from constituency parse trees with four relation types (aspect-specific, inter-aspect, globally-shared, opinion word), uses a relational GCN with syntactic-distance-weighted edges, and jointly predicts all aspects from one sentence input, treating ABSA as token-level classification.
Key results
- YORO reaches 86.08% accuracy and 85.51 F1 on MAMS, beating the best baseline CHGMAN (85.05 / 84.29).
- On Rest14, YORO gets 87.14% accuracy and 81.83 F1, slightly above CHGMAN (86.88 / 81.62).
- On Lap14, YORO gets 81.82% accuracy (best) but 78.32 F1, slightly below SDGCN's 78.34 F1 and close to APSCL.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use YORO instead of the strongest listed baseline | MAMS · MAMS | accuracy: 85.05 -> 86.08 | Helped |
| Use YORO instead of the strongest listed baseline | MAMS · MAMS | F1: 84.44 -> 85.51 | Helped |
| Use YORO instead of the strongest listed baseline | Rest14 · Rest14 | accuracy: 86.88 -> 87.14 | Helped |
| Use YORO instead of the strongest listed baseline | Rest14 · Rest14 | F1: 81.62 -> 81.83 | Helped |
| Use YORO instead of the strongest listed baseline | Lap14 · Lap14 | accuracy: 81.52 -> 81.82 | Helped |
| Use YORO instead of the strongest listed baseline | Lap14 · Lap14 | F1: 78.47 -> 78.32 | Hurt |
| Add CorrGCN (ablation) | YORO on MAMS · MAMS | accuracy: 84.06 -> 86.08 | Helped |
| Distinguish edge relationships (ablation) | YORO on MAMS · MAMS | accuracy: 82.49 -> 86.08 | Helped |
| Add opinion sentiment classification and supervised contrastive learning (ablation) | YORO on MAMS · MAMS | accuracy: 85.18 -> 86.08 | Helped |
| Add opinion sentiment classification (ablation) | YORO on MAMS · MAMS | accuracy: 85.10 -> 86.08 | Helped |
| Add supervised contrastive learning (ablation) | YORO on MAMS · MAMS | accuracy: 84.43 -> 86.08 | Helped |
| Add opinion-word relation to CorrGCN (ablation) | YORO on MAMS · MAMS | accuracy: 85.63 -> 86.08 | Helped |
Surprising findings
- Replacing the relational GCN with a vanilla GCN degrades performance most sharply, and the parser quality affects results (Berkeley parser beats benepar2).
Limitations
- Lap14 F1 is slightly below SDGCN and APSCL; pseudo-aspect strategy is weaker for single-aspect sentences.
- Efficiency gains are asserted but not measured; single-aspect sentences lack inter-aspect structure.
Experimental setup
Datasets: MAMS SemEval 2014 Restaurants SemEval 2014 Laptops Bing Liu opinion lexicon
Benchmarks: MAMS Rest14 Lap14
Metrics: accuracy macro-average F1
Baselines: CHGMAN APSCL dotGCN RGAT T-GCN R-GAT CapsNet SDGCN
Builds on: R-GCN BERT GCN aspect-oriented tree SupCon
Closest papers by meaning
- TextGT: A Double-View Graph Transformer on Text for Aspect-Based Sentiment Analysis (AAAI 2024 · similarity 0.88)
- A Novel Energy Based Model Mechanism for Multi-Modal Aspect-Based Sentiment Analysis (AAAI 2024 · similarity 0.86)
- Modeling Adaptive Inter-Task Feature Interactions via Sentiment-Aware Contrastive Learning for Joint Aspect-Sentiment Prediction (AAAI 2024 · similarity 0.86)
- Aspect-Based Sentiment Analysis with Explicit Sentiment Augmentations (AAAI 2024 · similarity 0.85)
- Exploring Model Editing for LLM-based Aspect-Based Sentiment Classification (AAAI 2025 · similarity 0.84)
- DEQA: Descriptions Enhanced Question-Answering Framework for Multimodal Aspect-Based Sentiment Analysis (AAAI 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.