{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,9,18]],"date-time":"2026-09-18T06:08:55Z","timestamp":1789711735632,"version":"4.0.1"},"reference-count":94,"publisher":"Frontiers Media SA","license":[{"start":{"date-parts":[[2026,9,18]],"date-time":"2026-09-18T00:00:00Z","timestamp":1789689600000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"funder":[{"DOI":"10.13039\/501100004329","name":"The Slovenian Research and Innovation Agency","doi-asserted-by":"publisher","id":[{"id":"10.13039\/501100004329","id-type":"DOI","asserted-by":"publisher"}]}],"content-domain":{"domain":["frontiersin.org"],"crossmark-restriction":true},"short-container-title":["Front. Artif. Intell."],"abstract":"<jats:p>In the contemporary digital media landscape, the ability to automatically distill public opinion from a vast and continuous stream of information is highly important. Aspect-Based Sentiment Analysis (ABSA) offers this granular capability. In this work, we address a specific, industrially relevant formulation of this task, more formally known as document-level entity-targeted sentiment analysis (TSA) or entity-level sentiment analysis (ELSA), where the \u201caspects\u201d are named entities such as companies and brands. However, its application to morphologically complex, less-resourced languages like those in the South Slavic family, particularly within the long-form news domain, remains a significant challenge. In this work, we introduce AspectBench, a new benchmark dataset for document-level ABSA, comprising real-world online news articles in Slovenian and Serbo-Croatian, designed to test generalization to unseen aspects, with the Serbo-Croatian portion being made publicly available. Using this benchmark, we conduct a comprehensive empirical study evaluating five distinct modeling paradigms, including lightweight document-embedding-based classifiers, fine-tuned multilingual and language-specific pretrained language models (PLMs), hierarchical attention networks (HANs), local large language models (LLMs) in zero and few-shot settings, and learning-to-defer (L2D) policies with multi-expert PLMs, complete LLM deferral, and confidence-gated selective LLM deferral using DSPy-calibrated prompts. Our experiments show that supervised expert models remain the most reliable foundation for this task, with Longformer, mDeBERTa-v3, mT5, language-specific encoders, and hierarchical models providing strong performance depending on language and class balance. Aspect masking is generally useful for supervised expert models, especially the PLM and HAN variants, but is not uniformly beneficial across all model families and evaluation views. Standalone local LLMs are not competitive, and complete LLM deferral is unstable when applied to every instance. However, selective LLM deferral is more promising because routing low-confidence expert predictions to the LLM can improve or stabilize performance while limiting unnecessary LLM calls. We additionally use minimum viable set analysis to show how performance, calibration, and uncertainty scale with annotation volume, providing practical guidance for future media-monitoring annotation campaigns on real-world datasets. Our work therefore fills a crucial gap for less-resourced ABSA and offers practical insights into the trade-offs between model complexity, data characteristics, and the development of robust, learning-to-defer systems for real-world media monitoring.<\/jats:p>","DOI":"10.3389\/frai.2026.1844418","type":"journal-article","created":{"date-parts":[[2026,9,18]],"date-time":"2026-09-18T05:36:25Z","timestamp":1789709785000},"update-policy":"https:\/\/doi.org\/10.3389\/crossmark-policy","source":"Crossref","is-referenced-by-count":0,"title":["Evaluating fine-tuned, embedding-based, and zero-shot models for aspect-based sentiment analysis in South Slavic news"],"prefix":"10.3389","volume":"9","author":[{"given":"Nishan","family":"Chatterjee","sequence":"first","affiliation":[{"name":"Jo\u017eef Stefan Institute","place":["Ljubljana, Slovenia"]},{"name":"Jo\u017eef Stefan International Postgraduate School","place":["Ljubljana, Slovenia"]},{"name":"University of La Rochelle","place":["La Rochelle, France"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Boshko","family":"Koloski","sequence":"additional","affiliation":[{"name":"Jo\u017eef Stefan Institute","place":["Ljubljana, Slovenia"]},{"name":"Jo\u017eef Stefan International Postgraduate School","place":["Ljubljana, Slovenia"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Antoine","family":"Doucet","sequence":"additional","affiliation":[{"name":"University of La Rochelle","place":["La Rochelle, France"]},{"name":"University of Ljubljana","place":["Ljubljana, Slovenia"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Senja","family":"Pollak","sequence":"additional","affiliation":[{"name":"Jo\u017eef Stefan Institute","place":["Ljubljana, Slovenia"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Matthew","family":"Purver","sequence":"additional","affiliation":[{"name":"Jo\u017eef Stefan Institute","place":["Ljubljana, Slovenia"]},{"name":"Queen Mary University of London","place":["London, United Kingdom"]}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"1965","published-online":{"date-parts":[[2026,9,18]]},"reference":[{"key":"B1","doi-asserted-by":"publisher","first-page":"140958","DOI":"10.1109\/ACCESS.2024.3467008","article-title":"Annotation quality versus quantity for object detection and instance segmentation","volume":"12","author":"Agnew","year":"2024","journal-title":"IEEE Access"},{"key":"B2","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2409.20054","article-title":"Evaluating and explaining training strategies for zero-shot cross-lingual news sentiment analysis","author":"Andren\u0161ek","year":"2024","journal-title":"arXiv preprint"},{"key":"B3","unstructured":"Claude 4 Sonnet"},{"key":"B4","unstructured":"Claude opus 4.5 System Card"},{"key":"B5","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2504.12185","article-title":"Salad: Improving robustness and generalization through contrastive learning with structure-aware and llm-driven augmented data","author":"Bae","year":"2025","journal-title":"arXiv Preprint"},{"key":"B6","doi-asserted-by":"crossref","DOI":"10.3115\/1073012.1073017","article-title":"\u201cScaling to very very large corpora for natural language disambiguation,\u201d","volume-title":"Proceedings of the 39th annual meeting on association for computational linguistics","author":"Banko","year":"2001"},{"key":"B7","doi-asserted-by":"crossref","DOI":"10.18653\/v1\/2023.bsnlp-1.10","article-title":"\u201cTarget two birds with one SToNe: entity-level sentiment and tone analysis in Croatian news headlines,\u201d","volume-title":"Proceedings of the 9th workshop on slavic natural language processing 2023 (SlavicNLP 2023)","author":"Bari\u0107","year":"2023"},{"key":"B8","author":"Beltagy","year":"2020","journal-title":"Longformer: The Long-Document Transformer"},{"key":"B9","author":"Bommarito","year":"2025","journal-title":"Binary bpe: A Family of Cross-Platform Tokenizers for Binary Analysis"},{"key":"B10","doi-asserted-by":"publisher","first-page":"5","DOI":"10.1023\/A:1010933404324","article-title":"Random forests","volume":"45","author":"Breiman","year":"2001","journal-title":"Mach. Learn"},{"key":"B11","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2307.00374","author":"Chang","year":"2023","journal-title":"Revisiting Sample Size Determination in Natural Language Understanding"},{"key":"B12","doi-asserted-by":"publisher","first-page":"611","DOI":"10.18653\/v1\/2023.ijcnlp-main.41","article-title":"\u201cA review of datasets for aspect-based sentiment analysis,\u201d","volume":"2023","author":"Chebolu","year":"2023","journal-title":"Proc. IJCNLP-AACL"},{"key":"B13","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2402.03216","article-title":"Bge m3-embedding: multi-lingual, multi-functionality, multi-granularity text embeddings through self-knowledge distilllation","author":"Chen","year":"2024","journal-title":"arXiv Preprint"},{"key":"B14","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2020.acl-main.747","article-title":"\u201cUnsupervised cross-lingual representation learning at scale,\u201d","author":"Conneau","year":"2020","journal-title":"Proc. 58th annual meeting ACL"},{"key":"B15","first-page":"4367","article-title":"\u201cBoosting large language models with continual learning for aspect-based sentiment analysis,\u201d","volume-title":"Findings ACL: EMNLP 2024","author":"Ding","year":"2024"},{"key":"B16","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2505.22137","article-title":"Limited generalizability in argument mining: State-of-the-art models learn datasets, not arguments","author":"Feger","year":"2025","journal-title":"arXiv Preprint"},{"key":"B17","doi-asserted-by":"publisher","first-page":"8","DOI":"10.1186\/1472-6947-12-8","article-title":"Predicting sample size required for classification performance","volume":"12","author":"Figueroa","year":"2012","journal-title":"BMC Med. Inform. Decis. Mak"},{"key":"B18","first-page":"4885","article-title":"\u201cSelective classification for deep neural networks,\u201d","volume-title":"Proc. NeurIPS 2017","author":"Geifman","year":"2017"},{"key":"B19","year":"2025","journal-title":"Gemma 3 Technical Report"},{"key":"B20","first-page":"963","article-title":"\u201cBenchmarking long-tail generalization with likelihood splits,\u201d","volume-title":"Findings ACL: EACL 2023","author":"Godbole","year":"2023"},{"key":"B21","volume-title":"Rusentne-2023: Evaluating Entity-Oriented Sentiment Analysis on Russian News Texts","author":"Golubev","year":"2023"},{"key":"B22","year":"2025","journal-title":"Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities"},{"key":"B23","doi-asserted-by":"publisher","first-page":"1055","DOI":"10.1016\/0895-4356(93)90173-X","article-title":"The analysis of ordinal agreement data: beyond weighted kappa","volume":"46","author":"Graham","year":"1993","journal-title":"J. Clin. Epidemiol"},{"key":"B24","doi-asserted-by":"crossref","DOI":"10.18653\/v1\/2021.eacl-main.142","article-title":"\u201cNewsMTSC: a dataset for (multi-)target-dependent sentiment classification in political news articles,\u201d","volume-title":"Proceedings of the 16th conference of the european chapter of the association for computational linguistics: main volume","author":"Hamborg","year":"2021"},{"key":"B25","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2205.03109","article-title":"\u201cControlled dropout for uncertainty estimation,\u201d","author":"Hasan","year":"2022","journal-title":"IEEE Int. Conf. SMC"},{"key":"B26","volume-title":"The Elements of Statistical Learning. Springer Series in Statistics","author":"Hastie","year":"2009"},{"key":"B27","volume-title":"Debertav3: Improving Deberta Using Electra-Style Pre-training with Gradient-Disentangled Embedding Sharing","author":"He","year":"2023"},{"key":"B28","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2410.08499","article-title":"Studying and benchmarking large language models for log level suggestion","author":"Heng","year":"2024","journal-title":"arXiv Preprint"},{"key":"B29","doi-asserted-by":"publisher","first-page":"296","DOI":"10.1007\/s10462-024-10906-z","article-title":"A systematic review of aspect-based sentiment analysis: domains, methods, and trends","volume":"57","author":"Hua","year":"2024","journal-title":"Artif. Intell. Rev"},{"key":"B30","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2501.15968","article-title":"Multi-view attention syntactic enhanced graph convolutional network for aspect-based sentiment analysis","author":"Huang","year":"2025","journal-title":"arXiv Preprint"},{"key":"B31","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2109.04847","article-title":"Active learning for reducing labeling effort in text classification tasks","author":"Jacobs","year":"2021","journal-title":"arXiv Preprint 2109.04847"},{"key":"B32","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2407.02834","article-title":"Aspect-based sentiment analysis techniques: a comparative study","author":"Jayakody","year":"2024","journal-title":"arXiv Preprint"},{"key":"B33","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.1612.03651","article-title":"Fasttext.zip: compressing text classification models","author":"Joulin","year":"2016","journal-title":"arXiv Preprint"},{"key":"B34","first-page":"427","article-title":"\u201cBag of tricks for efficient text classification,\u201d","volume-title":"Proc. 15th Eur. ACL Short Papers","author":"Joulin","year":"2017"},{"key":"B35","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2310.03714","article-title":"Dspy: compiling declarative language model calls into self-improving pipelines","author":"Khattab","year":"2023","journal-title":"arXiv Preprint"},{"key":"B36","article-title":"\u201cAdam: a method for stochastic optimization,\u201d","volume-title":"Proc. ICLR 2015","author":"Kingma","year":"2015"},{"key":"B37","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2604.13285","article-title":"L2d-clinical: learning to defer for adaptive model selection in clinical text classification","author":"Kondadadi","year":"2026","journal-title":"arXiv Preprint"},{"key":"B38","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.1808.06226","article-title":"Sentencepiece: a simple and language independent subword tokenizer and detokenizer for neural text processing","author":"Kudo","year":"2018","journal-title":"arXiv Preprint"},{"key":"B39","author":"Ljube\u0161i\u0107","year":"2021"},{"key":"B40","first-page":"6150","article-title":"\u201cPredict responsibly: improving fairness and accuracy by learning to defer,\u201d","volume-title":"Proc. NeurIPS 2018, Madras, D., Pitassi, T., and Zemel., R. (2018). \u201cPredict responsibly: improving fairness and accuracy by learning to defer,\u201d in Proc. NeurIPS 2018","author":"Madras","year":"2018"},{"key":"B41","first-page":"394","article-title":"\u201cUzABSA: aspect-based sentiment analysis for the Uzbek language,\u201d","author":"Matlatipov","year":"2024","journal-title":"Proc. SIGUL"},{"key":"B42","doi-asserted-by":"publisher","first-page":"1093","DOI":"10.1016\/j.asej.2014.04.011","article-title":"Sentiment analysis algorithms and applications: a survey","volume":"5","author":"Medhat","year":"2014","journal-title":"Ain Shams Eng. J"},{"key":"B43","volume-title":"Explosion\/Spacy: v3.7.2: Fixes for Apis and Requirements","author":"Montani","year":"2023"},{"key":"B44","article-title":"\u201cLearning-to-defer with expert-conditional advice,\u201d","author":"Montreuil","year":"2026"},{"key":"B45","author":"Mozannar","year":"2020","journal-title":"Consistent Estimators for Learning to Defer to an Expert"},{"key":"B46","first-page":"6804","article-title":"\u201cAspect-based sentiment analysis as fine-grained opinion mining,\u201d","volume-title":"Proc. LREC","author":"Ocampo Diaz","year":"2020"},{"key":"B47","doi-asserted-by":"publisher","DOI":"10.2139\/ssrn.4906218","article-title":"Social media sentiment analysis for brand reputation management","author":"Olukemi","year":"2024","journal-title":"SSRN Electron. J"},{"key":"B48","article-title":"Gpt-4o system card","year":"2024","journal-title":"arXiv preprint"},{"key":"B49","unstructured":"Gpt-5.5 system card\n          \n          2026"},{"key":"B50","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2406.11695","article-title":"Optimizing instructions and demonstrations for multi-stage language model programs","author":"Opsahl-Ong","year":"2024","journal-title":"arXiv Preprint"},{"key":"B51","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2303.09014","article-title":"Art: automatic multi-step reasoning and tool-use for large language models","author":"Paranjape","year":"2023","journal-title":"arXiv Preprint"},{"key":"B52","article-title":"\u201cAutomatic differentiation in pytorch,\u201d","volume-title":"NIPS Autodiff Workshop","author":"Paszke","year":"2017"},{"key":"B53","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.1905.03423","author":"Pei","year":"2019","journal-title":"Targeted Sentiment Analysis: A Data-Driven Categorization"},{"key":"B54","unstructured":"Qwen Team\n              Alibaba Group.\n            \n          \n          Qwen2.5: A Party of Foundation Models"},{"key":"B55","volume-title":"Qwen2 Technical Report","author":"Qwen Team","year":""},{"key":"B56","first-page":"6773","article-title":"\u201cEntity-level sentiment analysis (ELSA): an exploratory task survey,\u201d","volume-title":"Proc. COLING","author":"R\u00f8nningstad","year":"2022"},{"key":"B57","first-page":"999","article-title":"\u201cA hierarchical model of reviews for aspect-based sentiment analysis,\u201d","volume-title":"Proc. EMNLP 2016","author":"Ruder","year":"2016"},{"key":"B58","first-page":"113","article-title":"\u201cBENCHi\u0107-lang: a benchmark for discriminating between Bosnian, Croatian, Montenegrin and Serbian,\u201d","volume-title":"Proc. VarDial 2023","author":"Rupnik","year":"2023"},{"key":"B59","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2008.07267","author":"Schr\u00f6der","year":"2020","journal-title":"A Survey of Active Learning for Text Classification using Deep Neural Networks"},{"key":"B60","volume-title":"Active Learning Literature Survey","author":"Settles","year":"2010"},{"key":"B61","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2402.03300","article-title":"Deepseekmath: Pushing the limits of mathematical reasoning in open language models","author":"Shao","year":"2024","journal-title":"arXiv Preprint"},{"key":"B62","author":"Shen","year":"2022","journal-title":"LexicalRichness: A Small Module to Compute Textual Lexical Richness"},{"key":"B63","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2202.01924","article-title":"Zero-shot aspect-based sentiment analysis","author":"Shu","year":"2022","journal-title":"arXiv Preprint"},{"key":"B64","unstructured":"Iso 639-3 code: hbs\n          \n          2026"},{"key":"B65","first-page":"7687","article-title":"\u201cA sequence-to-structure approach to document-level targeted sentiment analysis,\u201d","volume-title":"Findings ACL: EMNLP 2023","author":"Song","year":"2023"},{"key":"B66","doi-asserted-by":"publisher","author":"Soylu","year":"2024","DOI":"10.18653\/v1\/2024.emnlp-main.597"},{"key":"B67","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":"B68","first-page":"1","article-title":"\u201cAutogluon-multimodal (automm): supercharging multimodal automl with foundation models,\u201d","author":"Tang","year":"2024","journal-title":"Proc. 3rd Int. conf. automated machine learning"},{"key":"B69","doi-asserted-by":"publisher","first-page":"330","DOI":"10.1016\/j.ejor.2023.09.009","article-title":"Explainability through uncertainty: Trustworthy decision-making with neural networks","volume":"317","author":"Thuy","year":"2024","journal-title":"Eur. J. Oper. Res"},{"key":"B70","author":"Tiwari","year":"2026","journal-title":"Learning, Fast and Slow: Towards LLMS That Adapt Continually"},{"key":"B71","doi-asserted-by":"crossref","DOI":"10.63317\/4smeu6r7yan5","article-title":"\u201cApproximating learning curves for active-learning-driven annotation,\u201d","volume-title":"LREC'08","author":"Tomanek","year":"2008"},{"key":"B72","article-title":"\u201cSloberta: Slovene monolingual large pretrained masked language model,\u201d","volume-title":"24th International multiconference information society 2021","author":"Ul\u010dar","year":"2021"},{"key":"B73","article-title":"\u201cGenerative model for less-resourced language with 1 billion parameters,\u201d","volume-title":"Language technologies and digital humanities conference","author":"Vre\u0161","year":"2024"},{"key":"B74","doi-asserted-by":"publisher","first-page":"8975","DOI":"10.63317\/4hb5dcxvoz5q","article-title":"\u201cIn-context example retrieval from multi-perspectives for few-shot aspect-based sentiment analysis,\u201d","author":"Wang","year":"2024","journal-title":"LREC-COLING"},{"key":"B75","first-page":"90","article-title":"\u201cBaselines and bigrams: simple, good sentiment and topic classification,\u201d","volume-title":"Proc. 50th annual meeting ACL short papers","author":"Wang","year":"2012"},{"key":"B76","doi-asserted-by":"crossref","DOI":"10.18653\/v1\/2025.acl-long.127","article-title":"\u201cSmarter, better, faster, longer: a modern bidirectional encoder for fast, memory efficient, and long context finetuning and inference,\u201d","volume-title":"ACL 25","author":"Warner","year":"2025"},{"key":"B77","doi-asserted-by":"publisher","DOI":"10.1109\/JCDL52503.2021.00038","author":"Weeber","year":"2021","journal-title":"Assisted Text Annotation Using Active Learning to Achieve High Quality With Little Effort"},{"key":"B78","doi-asserted-by":"publisher","DOI":"10.52202\/068431-1800","article-title":"\u201cChain-of-thought prompting elicits reasoning in large language models,\u201d","author":"Wei","year":"2022","journal-title":"Proc. NeurIPS 2022"},{"key":"B79","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2020.emnlp-demos.6","article-title":"\u201cTransformers: state-of-the-art natural language processing,\u201d","author":"Wolf","year":"2020","journal-title":"Proc. EMNLP Demo 2020"},{"key":"B80","doi-asserted-by":"publisher","first-page":"241","DOI":"10.1016\/S0893-6080(05)80023-1","article-title":"Stacked generalization","volume":"5","author":"Wolpert","year":"1992","journal-title":"Neural Netw"},{"key":"B81","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2502.11824","article-title":"M-absa: a multilingual dataset for aspect-based sentiment analysis","author":"Wu","year":"2025","journal-title":"arXiv Preprint"},{"key":"B82","doi-asserted-by":"publisher","first-page":"8079","DOI":"10.1007\/s13042-025-02711-z","article-title":"Evaluating zero-shot multilingual aspect-based sentiment analysis with large language models","volume":"16","author":"Wu","year":"2025","journal-title":"Int. J. Mach. Learn. Cybern"},{"key":"B83","year":"2025","journal-title":"Grok 3"},{"key":"B84","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2020.emnlp-main.292","article-title":"\u201cTasty burgers, soggy fries: probing aspect robustness in aspect-based sentiment analysis,\u201d","author":"Xing","year":"2020","journal-title":"Proc. EMNLP 2020"},{"key":"B85","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2021.naacl-main.41","author":"Xue","year":"2021","journal-title":"mt5: A Massively Multilingual Pre-trained Text-to-Text"},{"key":"B86","first-page":"23","article-title":"\u201cModeling complex interactions in long documents for aspect-based sentiment analysis,\u201d","volume-title":"Proc. 14th WASSA workshop","author":"Yan","year":"2024"},{"key":"B87","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/N16-1174","article-title":"\u201cHierarchical attention networks for document classification,\u201d","author":"Yang","year":"2016","journal-title":"Proc. NAACL-HLT"},{"key":"B88","first-page":"3914","article-title":"\u201cBenchmarking zero-shot text classification: Datasets, evaluation and entailment approach,\u201d","volume-title":"Proc. EMNLP-IJCNLP 2019","author":"Yin","year":"2019"},{"key":"B89","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2502.17419","article-title":"From system 1 to system 2: A survey of reasoning large language models","author":"Zhang","year":"2025","journal-title":"arXiv Preprint"},{"key":"B90","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2509.08827","article-title":"A survey of reinforcement learning for large reasoning models","author":"Zhang","year":"2025","journal-title":"arXiv Preprint"},{"key":"B91","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2203.01054","article-title":"A survey on aspect-based sentiment analysis: tasks, methods, and challenges","author":"Zhang","year":"2022","journal-title":"arXiv Preprint"},{"key":"B92","first-page":"649","article-title":"\u201cCharacter-level convolutional networks for text classification,\u201d","volume-title":"Proc. NIPS 2015","author":"Zhang","year":"2015"},{"key":"B93","first-page":"12697","article-title":"\u201cCalibrate before use: Improving few-shot performance of language models,\u201d","volume-title":"Proc. ICML 2021","author":"Zhao","year":"2021"},{"key":"B94","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2503.15133","article-title":"Emograce: aspect-based emotion analysis for social media data","author":"Zorenb\u00f6hmer","year":"2025","journal-title":"arXiv preprint 2503.15133"}],"container-title":["Frontiers in Artificial Intelligence"],"original-title":[],"link":[{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/frai.2026.1844418\/full","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,9,18]],"date-time":"2026-09-18T05:36:39Z","timestamp":1789709799000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/frai.2026.1844418\/full"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,9,18]]},"references-count":94,"alternative-id":["10.3389\/frai.2026.1844418"],"URL":"https:\/\/doi.org\/10.3389\/frai.2026.1844418","relation":{},"ISSN":["2624-8212"],"issn-type":[{"value":"2624-8212","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,9,18]]},"article-number":"1844418"}}