{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,9,24]],"date-time":"2026-09-24T06:03:13Z","timestamp":1790229793121,"version":"4.1.0"},"reference-count":41,"publisher":"Frontiers Media SA","license":[{"start":{"date-parts":[[2026,9,24]],"date-time":"2026-09-24T00:00:00Z","timestamp":1790208000000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"content-domain":{"domain":["frontiersin.org"],"crossmark-restriction":true},"short-container-title":["Front. Artif. Intell."],"abstract":"<jats:p>\n                    In natural language processing (NLP), Aspect-Based Sentiment Analysis (ABSA) remains a challenge, particularly when handling ambiguous, domain-specific \u201cneutral\u201d sentiment classifications in cases of severe class imbalance. While there has been a pivot toward deploying generative Small Language Models (SLMs) on resource-limited edge devices, their performance on unevenly distributed data remains under-explored. This paper presents a comprehensive, cross-paradigm evaluation of the following ABSA architectures: a sequence-based BiGRU, a state-of-the-art DeBERTa-v3 encoder, and Alibaba\u2019s Qwen2.5-1.5B generative SLM. To evaluate the feasibility of deploying the SLM on edge devices, we modified the Qwen2.5 model to work on a 6GB VRAM device using BFloat16 quantized low-rank adaptation (QLoRA) and an abstract syntax tree (AST) parser. A closer look at the results shows a fundamental Generative Shortfall. Despite optimizing the SLM for edge devices, its performance was suboptimal (accuracy: 79.67%, macro-F1: 0.71), indicating that the model\u2019s conservative token generation behavior limited its ability to identify sparse and subtle semantic targets within the text. While the DeBERTa-v3 encoder, when balanced with multi-sample dropout and layer-wise learning rate decay (LLRD), achieved a macro-F1 of 0.87, successfully identifying 81.95% of the minority neutral class. Supported by bootstrap resampling and McNemar\u2019s statistical significance test (\n                    <jats:inline-formula>\n                      <mml:math xmlns:mml=\"http:\/\/www.w3.org\/1998\/Math\/MathML\" id=\"M1\">\n                        <mml:mi>p<\/mml:mi>\n                        <mml:mo>&lt;<\/mml:mo>\n                        <mml:mn>0.001<\/mml:mn>\n                      <\/mml:math>\n                    <\/jats:inline-formula>\n                    ), the results show that although generative SLMs can be efficiently deployed on resource-constrained edge devices, structurally regularized extractive encoders provide significantly better performance for precise opinion mining on highly imbalanced data.\n                  <\/jats:p>","DOI":"10.3389\/frai.2026.1930455","type":"journal-article","created":{"date-parts":[[2026,9,24]],"date-time":"2026-09-24T05:44:50Z","timestamp":1790228690000},"update-policy":"https:\/\/doi.org\/10.3389\/crossmark-policy","source":"Crossref","is-referenced-by-count":0,"title":["The Generative Shortfall: a comparative analysis of autoregressive vs. extractive architectures in ABSA"],"prefix":"10.3389","volume":"9","author":[{"given":"Chathurvedi V. Rama","family":"Chandra","sequence":"first","affiliation":[{"name":"School of Computer Science and Engineering, Vellore Institute of Technology","place":["Vellore, Tamil Nadu, India"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"T.","family":"Padmavathy","sequence":"additional","affiliation":[{"name":"School of Computer Science and Engineering, Vellore Institute of Technology","place":["Vellore, Tamil Nadu, India"]}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"1965","published-online":{"date-parts":[[2026,9,24]]},"reference":[{"key":"ref1","doi-asserted-by":"publisher","first-page":"113987","DOI":"10.1016\/j.knosys.2025.113987","article-title":"ATESA-B\u00c6RT: a heterogeneous ensemble learning model for aspect-based sentiment analysis","volume":"326","author":"Apostol","year":"2025","journal-title":"Knowl. Based Syst."},{"key":"ref2","doi-asserted-by":"publisher","first-page":"127322","DOI":"10.1109\/ACCESS.2025.3588301","article-title":"Semantic parsing for aspect-based sentiment analysis","volume":"13","author":"Aqeel","year":"2025","journal-title":"IEEE Access"},{"key":"ref3","doi-asserted-by":"publisher","first-page":"2501","DOI":"10.1007\/s10579-025-09820-9","article-title":"MEMD-ABSA: a multi-element multi-domain dataset for aspect-based sentiment analysis","volume":"59","author":"Cai","year":"2025","journal-title":"Lang. Resour. Eval."},{"key":"ref4","doi-asserted-by":"publisher","first-page":"1001","DOI":"10.1007\/s13198-025-02713-8","article-title":"Combining transfer and ensemble learning models for image and text aspect-based sentiment analysis","volume":"16","author":"Chauhan","year":"2025","journal-title":"Int. J. Syst. Assur. Eng. Manag."},{"key":"ref5","doi-asserted-by":"crossref","first-page":"10088","DOI":"10.52202\/075280-0441","article-title":"QLORA: efficient finetuning of quantized LLMs","volume":"36","author":"Dettmers","year":"2023","journal-title":"Adv. Neural Inf. Proces. Syst."},{"key":"ref6","first-page":"4171","author":"Devlin","year":"2019"},{"key":"ref7","first-page":"4367","article-title":"Boosting large language models with continual learning for aspect-based sentiment analysis","volume":"2024","author":"Ding","year":"2024","journal-title":"Find. Assoc. Comput. Linguist.: EMNLP"},{"key":"ref8","doi-asserted-by":"publisher","first-page":"125933","DOI":"10.1016\/j.eswa.2024.125933","article-title":"PGSO: prompt-based generative sequence optimization network for aspect-based sentiment analysis","volume":"265","author":"Dong","year":"2025","journal-title":"Expert Syst. Appl."},{"key":"ref9","doi-asserted-by":"publisher","first-page":"69255","DOI":"10.1109\/access.2026.3690176","article-title":"Unsupervised aspect-based sentiment analysis through LLM: a case study of an unlabeled Portuguese beer database","volume":"14","author":"Eiras","year":"2026","journal-title":"IEEE Access"},{"key":"ref10","doi-asserted-by":"publisher","first-page":"846","DOI":"10.33395\/sinkron.v10i2.15888","article-title":"From methodologies to metrics: a review of aspect-based sentiment analysis approaches","volume":"10","author":"Esmaeel","year":"2026","journal-title":"Sinkron"},{"key":"ref11","doi-asserted-by":"crossref","first-page":"102556","DOI":"10.1016\/j.is.2025.102556","article-title":"Hierarchical aspect-based sentiment analysis using semantic capsuled multi-granular networks","volume":"132","author":"Gracewell","year":"2025","journal-title":"Inf. Syst."},{"key":"ref12","doi-asserted-by":"crossref","first-page":"384","DOI":"10.1007\/s00530-025-01963-7","article-title":"A prompt-based dual-layer cross-modal distillation learning method for aspect-based sentiment analysis","volume":"31","author":"Guo","year":"2025","journal-title":"Multim. Syst."},{"key":"ref13","author":"He","year":"2023"},{"key":"ref14","doi-asserted-by":"crossref","first-page":"112782","DOI":"10.1016\/j.knosys.2024.112782","article-title":"CABiLSTM-BERT: aspect-based sentiment analysis model based on deep implicit feature extraction","volume":"309","author":"He","year":"2025","journal-title":"Knowl. Based Syst."},{"key":"ref15","first-page":"2106.09685","article-title":"LoRA: low-rank adaptation of large language models","author":"Hu","year":"2021","journal-title":"arXiv preprint arXiv"},{"key":"ref16","doi-asserted-by":"publisher","first-page":"127353","DOI":"10.1016\/j.eswa.2025.127353","article-title":"Multi-axis fusion with optimal transport learning for multimodal aspect-based sentiment analysis","volume":"280","author":"Jia","year":"2025","journal-title":"Expert Syst. Appl."},{"key":"ref17","doi-asserted-by":"publisher","first-page":"7907","DOI":"10.1007\/s40747-024-01570-5","article-title":"DCASAM: advancing aspect-based sentiment analysis through a deep context-aware sentiment analysis model","volume":"10","author":"Jiang","year":"2024","journal-title":"Complex Intell. Syst."},{"key":"ref18","doi-asserted-by":"crossref","first-page":"111654","DOI":"10.1016\/j.engappai.2025.111654","article-title":"Aspect-based sentiment analysis with semantic and syntactic enhanced multi-layer fusion model","volume":"159","author":"Jin","year":"2025","journal-title":"Eng. Appl. Artif. Intell."},{"key":"ref19","doi-asserted-by":"publisher","first-page":"126867","DOI":"10.1016\/j.neucom.2023.126867","article-title":"FABSA: an aspect-based sentiment analysis dataset of user reviews","volume":"562","author":"Kontonatsios","year":"2023","journal-title":"Neurocomputing"},{"key":"ref20","doi-asserted-by":"publisher","first-page":"131037","DOI":"10.1016\/j.neucom.2025.131037","article-title":"Enhancing syntactic and semantic features via TextGINConv and Kolmogorov-Arnold networks for aspect-based sentiment analysis","volume":"651","author":"Li","year":"2025","journal-title":"Neurocomputing"},{"key":"ref21","doi-asserted-by":"publisher","first-page":"111297","DOI":"10.1016\/j.engappai.2025.111297","article-title":"Integrating multiple syntactic structures for enhanced aspect-based sentiment analysis","volume":"158","author":"Liu","year":"2025","journal-title":"Eng. Appl. Artif. Intell."},{"key":"ref22","doi-asserted-by":"publisher","first-page":"60943","DOI":"10.1109\/access.2024.3386969","article-title":"Comparative analysis of deep natural networks and large language models for aspect-based sentiment analysis","volume":"12","author":"Mughal","year":"2024","journal-title":"IEEE Access"},{"key":"ref23","doi-asserted-by":"publisher","first-page":"88342","DOI":"10.1109\/access.2025.3568484","article-title":"ElHBiAt: Electra pre-training network hybrid of BiLSTM, the attention layer to aspect-based sentiment analysis","volume":"13","author":"Nezhad","year":"2025","journal-title":"IEEE Access"},{"key":"ref24","article-title":"Qwen2.5 technical report","year":"2025","journal-title":"arXiv preprint arXiv:2412.15115"},{"key":"ref25","doi-asserted-by":"publisher","first-page":"295","DOI":"10.3233\/web-230096","article-title":"Development of optimized cascaded LSTM with Seq2seqNet and transformer net for aspect-based sentiment analysis framework","volume":"23","author":"Ramasamy","year":"2024","journal-title":"Web Intell."},{"key":"ref26","author":"Setiabudi","year":"2025"},{"key":"ref27","article-title":"Large language models for aspect-based sentiment analysis","author":"Simmering","year":"2024","journal-title":"arXiv preprint arXiv:2310.18025."},{"key":"ref28","doi-asserted-by":"publisher","first-page":"47459","DOI":"10.1109\/access.2025.3549303","article-title":"OptiASAR: optimized aspect-based sentiment analysis of reviews with BiLSTM-GRU and NER-BERT in healthcare decision-making","volume":"13","author":"Singh","year":"2025","journal-title":"IEEE Access"},{"key":"ref29","doi-asserted-by":"publisher","first-page":"111975","DOI":"10.1016\/j.knosys.2024.111975","article-title":"Harnessing domain insights: a prompt knowledge tuning method for aspect-based sentiment analysis","volume":"298","author":"Sun","year":"2024","journal-title":"Knowl. Based Syst."},{"key":"ref30","doi-asserted-by":"publisher","first-page":"10","DOI":"10.1007\/s00530-024-01601-8","article-title":"Multilayer interactive attention bottleneck transformer for aspect-based multimodal sentiment analysis","volume":"31","author":"Sun","year":"2025","journal-title":"Multim. Syst."},{"key":"ref31","doi-asserted-by":"publisher","first-page":"22775","DOI":"10.1007\/s00521-025-11508-8","article-title":"Dependency relationship-enhanced graph convolutional network for aspect-based sentiment analysis","volume":"37","author":"Tian","year":"2025","journal-title":"Neural Comput. & Applic."},{"key":"ref32","doi-asserted-by":"publisher","first-page":"181","DOI":"10.1007\/s10489-024-06011-x","article-title":"WRGAT-PTBERT: weighted relational graph attention network over post-trained BERT for aspect-based sentiment analysis","volume":"55","author":"Verma","year":"2025","journal-title":"Appl. Intell."},{"key":"ref33","doi-asserted-by":"publisher","first-page":"33","DOI":"10.1007\/s13748-024-00352-x","article-title":"Polarity-aware deep attention network for aspect-based sentiment analysis","volume":"14","author":"Wadawadagi","year":"2025","journal-title":"Prog. Artif. Intell."},{"key":"ref34","doi-asserted-by":"publisher","first-page":"112763","DOI":"10.1016\/j.asoc.2025.112763","article-title":"Aspect-based sentiment analysis by knowledge and attention integrated graph convolutional network","volume":"171","author":"Wan","year":"2025","journal-title":"Appl. Soft Comput."},{"key":"ref35","doi-asserted-by":"publisher","first-page":"1421","DOI":"10.1109\/TPAMI.2026.3657354","article-title":"Parameter-efficient fine-tuning methods for pretrained language models: a critical review and assessment","volume":"48","author":"Xu","year":"2024","journal-title":"IEEE Trans. Pattern Anal. Mach. Intell."},{"key":"ref36","doi-asserted-by":"crossref","first-page":"404","DOI":"10.1007\/s40747-025-02034-0","article-title":"Enhancing aspect-based sentiment analysis with multiple-knowledge promotion and multi-perspective noise filtering","volume":"11","author":"Yang","year":"2025","journal-title":"Complex Intell. Syst."},{"key":"ref37","doi-asserted-by":"publisher","first-page":"341","DOI":"10.1007\/s11227-024-06783-6","article-title":"Dynamic position weighting aspect-focused graph convolutional network for aspect-based sentiment analysis","volume":"81","author":"Yu","year":"2025","journal-title":"J. Supercomput."},{"key":"ref38","doi-asserted-by":"publisher","first-page":"64","DOI":"10.1007\/s11227-024-06472-4","article-title":"Aspect-aware semantic feature enhanced networks for multimodal aspect-based sentiment analysis","volume":"81","author":"Zeng","year":"2025","journal-title":"J. Supercomput."},{"key":"ref39","doi-asserted-by":"publisher","first-page":"111934","DOI":"10.1016\/j.engappai.2025.111934","article-title":"Simplified syntax-guided domain-shared representation learning for cross-domain aspect-based sentiment analysis","volume":"160","author":"Zhang","year":"2025","journal-title":"Eng. Appl. Artif. Intell."},{"key":"ref40","doi-asserted-by":"crossref","first-page":"43","DOI":"10.1007\/s12293-025-00475-1","article-title":"Hierarchical prompt fusion and image denoising for multimodal aspect-based sentiment analysis","volume":"17","author":"Zhu","year":"2025","journal-title":"Mem. Comput."},{"key":"ref41","doi-asserted-by":"publisher","first-page":"129472","DOI":"10.1016\/j.neucom.2025.129472","article-title":"Large language model augmented syntax-aware domain adaptation method for aspect-based sentiment analysis","volume":"625","author":"Zou","year":"2025","journal-title":"Neurocomputing"}],"container-title":["Frontiers in Artificial Intelligence"],"original-title":[],"link":[{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/frai.2026.1930455\/full","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,9,24]],"date-time":"2026-09-24T05:44:51Z","timestamp":1790228691000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/frai.2026.1930455\/full"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,9,24]]},"references-count":41,"alternative-id":["10.3389\/frai.2026.1930455"],"URL":"https:\/\/doi.org\/10.3389\/frai.2026.1930455","relation":{},"ISSN":["2624-8212"],"issn-type":[{"value":"2624-8212","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,9,24]]},"article-number":"1930455"}}