Large language model augmented syntax-aware domain adaptation method for aspect-based sentiment analysis
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
Cross-domain aspect-based sentiment analysis aims to leverage knowledge from the source domain to identify the sentiment polarity towards a given aspect attribute in the text content from the target domain. Existing domain adaptation approaches either focus on acquiring domain-independent shared feature representations or adjusting the obtained feature distribution to the target domain, which fails to address critical domain-specific attributes, leading to misaligned feature representations. We propose a large language model augmented syntax-aware domain adaptation method that integrates advanced large language models with structured syntactic knowledge to recognize semantic attributes and address the lack of syntax sensitivity in large language models. A domain topic predictor based on adversarial training is developed to enhance the robustness and generalization of the framework across different domains. Additionally, automatic soft prompt learning is conducted based on analysed domain topics and task-relevant feature representations for domain-specific fine-tuning, aiding the architecture in conveying domain-specific semantic information in the cross-domain environment. The feature aggregation approach dynamically fuses six categories of analysed feature representations for fine-grained sentiment classification. To the best of our knowledge, this study represents the pioneering effort to systematically leverage syntactic and cross-domain characteristics to enhance pre-trained large language models in addressing cross-domain aspect-based sentiment analysis tasks. Experimental results on publicly available benchmark datasets validate the effectiveness of the proposed architecture.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.001 | 0.002 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it