Do We Still Need Fine Tuning? Turkish Sentiment Analysis in the Era of Large Language Model

Fine-tuned BERTurk models outperform prompted large language models in Turkish sentiment analysis, according to a new study that questions whether supervised fine-tuning remains necessary in the era of large language models [1]. The research, posted to arXiv on 28 June 2026, compares classical machine learning methods, fine-tuned pretrained language models, and prompted large language models on a Turkish e-commerce review dataset labeled negative, neutral, and positive [1]. Fine-tuned BERTurk models achieved the best overall results and outperformed all prompted large language models in the full three-class task [1]. The neutral class proved to be the main difficulty: while several large language models were much more competitive in binary positive–negative classification, they degraded substantially in the three-class setting by collapsing neutral reviews into polarized categories [1]. The findings suggest that, in realistic Turkish sentiment classification, prompted large language models do not yet match supervised fine-tuning in the zero-shot setting, and that including the neutral class is crucial for robust evaluation [1]. The study arrives as generative artificial intelligence applications have become widespread, with tools such as ChatGPT and DALL-E enabling users to instantly generate sophisticated texts, images, and media [7]. The arXiv platform itself, where the paper appears, is a repository that provides rapid dissemination of research findings at no cost to readers and submitters [10]. arXiv has also launched a formalized framework called arXivLabs to enable community collaborations that add value for readers and authors, with tools such as the Bibliographic Explorer and CORE Recommender now featured on article record pages [11].