Papers › Aspect-Category-Opinion-Sentiment Extraction Using Generative Transformer Model
Aspect-Category-Opinion-Sentiment Extraction Using Generative Transformer Model
Cao Duy Hoang, Quang Vinh Dinh, Ngoc Hong Tran
Sentiment analysis is one of Natural Language Processing's applications that aims to process and extract sentiment information quickly and effectively. To expand upon the previous triplet extraction, that being aspect-opinion-sentiment triplets, Aspect-Category-Opinion-Sentiment (ACOS) quadruple extraction was created. There are several methods to extract quadruples, albeit with several limitations, such as their effectiveness towards implicit information and their overall low-performance score. This paper proposes a method of using BART-Aspect-Based-Sentiment-Analysis (BARTABSA), a sentiment analysis model that aims to unify the previous Aspect Based Sentiment Analysis subtask - namely Aspect-Opinion pair extraction and Aspect-Opinion-Sentiment triplet extraction, and solve them without changing the core algorithm or adding other models to it - to solve the ACOS subtask. After some modification to the data and the model's outer layer, the result shows significant and promising improvements over previous results.
Code
Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.
Code Syntology ran Syntology
Not run by Syntology. Nothing on this page verifies that the listed code works.
Tasks
1 archive task tag without a task page not shown.
Results from the paper archive 2025-07-28
| Task | Dataset | Model | Metric | Value | Rank at snapshot | Leaderboard | Report |
|---|---|---|---|---|---|---|---|
| Aspect-Category-Opinion-Sentiment Quadruple Extraction | Laptop-ACOS | BART-ABSA | F1 | 39.41 | #2 of 2 | Archive leaderboard | report |
| Aspect-Category-Opinion-Sentiment Quadruple Extraction | Restaurant-ACOS | BART-ABSA | F1 | 53.45 | #2 of 2 | Archive leaderboard | report |
Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.
Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections