Akademik Veri Yönetim Sistemi
Engineering Applications of Artificial Intelligence, cilt.164, 2026 (SCI-Expanded, Scopus)
Social media is a forum for unrestricted communication, but it also facilitates the spread of hate speech. The number of hateful and disturbing posts on these platforms has increased due to differences in gender, sexual orientation, religion, and other factors. Identifying hate speech is essential to prevent the formation of offensive discourse targeting Lesbian, Gay, Bisexual, and Transgender individuals and to avoid discrimination and marginalization. This study aims to detect hate speech against Lesbian, Gay, Bisexual, and Transgender individuals in Turkish tweets posted on the X/Twitter platform. To achieve this, seven large language models were fine-tuned. By applying both soft voting and hard voting methods to these models, a new approach called “Chosen Deep” is proposed. All analyses were conducted with a newly created dataset developed in this study and a comparable earlier dataset. The proposed method is evaluated against the seven fine-tuned large language models, as well as Multinomial Naïve Bayes, Support Vector Machine, Decision Tree, and the Generative Pre-trained Transformer-4. The findings demonstrate that the proposed method outperforms the other models and algorithms. To the best of our knowledge, this is the first study on hate speech detection in which voting techniques have been applied to Turkish fine-tuned large language models. By integrating these models with voting techniques, the study increases the accuracy of hate speech detection, strengthens the effectiveness of natural language processing methods in artificial intelligence, and contributes to the development of more effective approaches to reduce discrimination in engineering sciences.