5 December 2025 in BRAIN. BROAD RESEARCH IN ARTIFICIAL INTELLIGENCE AND NEUROSCIENCE
Social media platforms enable users to interact and share content through text, images, videos, and links. Sentiment analysis, a key method in Natural Language Processing (NLP), evaluates the emotional tone of social media content. It helps to identify hate speech, which includes harmful, offensive language directed at individuals or groups. In this study, we present the performance comparison of two transformer-based models, RoBERTa-Base and XLM-RoBERTa-Base, for hate speech identification in formal (English) and informal (Roman Urdu and mixed English-Roman Urdu) text. We fine-tune both models on multilingual datasets and evaluate the model's performance. XLM-RoBERTa-Base achieved (94.46%) accuracy for English, (93.77%) for Roman Urdu, and (88.32%) for mixed English-Roman Urdu text, while RoBERTa-Base performed better in English (97.24%) and Roman Urdu (94.28%), but dropped slightly to (88%) for mixed English-Roman Urdu text. These results show RoBERTa-Base excels in English, but struggles with Roman Urdu and code-switched mixed languages, while XLM-RoBERTa-Base performs consistently across all languages.
16 September 2026 in BRAIN. BROAD RESEARCH IN ARTIFICIAL INTELLIGENCE AND NEUROSCIENCE
Automated Urdu hate-speech detection remains challenging because annotated resources are limited, orthography varies, and harmful meaning depends on both lexical and contextual cues. This study evaluates a leakage-safe lexical-semantic framework combining 5,000-dimensional word/bigram TF-IDF features with 768-dimensional multilingual sentence embeddings, followed by binary particle swarm optimisation (BPSO) and classical ensemble learning. Experiments used the Urdu text and binary labels from MMHS11K: 8,800 balanced training records and an untouched balanced test set of 2,200 records. BPSO selected 2,849 of 5,768 hybrid dimensions, reducing dimensionality by 50.61%. On the official test set, Stacking with the complete hybrid representation achieved Macro-F1 = 0.8468 and ROC-AUC = 0.9265; the PSO-selected representation achieved Macro-F1 = 0.8391 and ROC- AUC = 0.9196. After mask selection, classifier fitting time decreased by 52.24%, excluding sentence-embedding extraction and BPSO search. Leakage-safe nested five-fold validation showed a similar trade-off: fitting time decreased by 52.64%, with Macro-F1 = 0.8444 ± 0.0066 after PSO versus 0.8492 ± 0.0110 without PSO. Soft Voting and Stacking were statistically comparable on selected features. Three-seed transformer aggregates performed better: Urdu-RoBERTa achieved the highest Macro-F1 (0.8836), and XLM-R achieved the highest ROC-AUC (0.9529). Holm-corrected exact McNemar tests confirmed that transformer aggregates significantly outperformed PSO Stacking. Overall, BPSO approximately halves representation size and downstream classical-model fitting time, with a small predictive decrease that was not statistically significant relative to complete-hybrid Stacking after correction.