• Dr. T. Murali Mohan
• T V Satya Sheela

Author profiling on social media, particularly gender identification, has become an important task for applications in security, marketing, and forensic analysis. Twitter provides a unique benchmark for this problem, where each user can be represented by both textual content and shared images. Earlier studies found that simple n-gram features with linear models often outperformed deep learning methods, while visual features contributed little when used in isolation. In this work, we propose a transformer-based multimodal approach that leverages BERT embeddings for textual data and Vision Transformer (ViT/CLIP) embeddings for image data. A cross-modal attention module is employed to capture interactions between the two modalities, and the fused representations are classified using a multi-layer perceptron (MLP) head. Experimental results on an English Twitter dataset show that our model achieves an accuracy of 86.4%, outperforming traditional n-gram + SVM baselines 82% and previous multimodal CNN approaches nearly 80%. These findings demonstrate the effectiveness of transformer-based text–image fusion with a lightweight MLP classifier for author profiling on social media.

Gender Identification, Author Profiling, Twitter, PAN 2018 dataset, BERT, Vision Transformer (ViT), Cross-Modal Fusion, MLP.

"Enhancing Gender Identification on Twitter with Transformer-Based Text–Image Fusion for Author Profiling", International Journal of Emerging Technologies and Innovative Research (www.jetir.org), ISSN:2349-5162, Vol.12, Issue 10, page no.c599-c604, October-2025, Available :http://www.jetir.org/papers/JETIR2510277.pdf

"Enhancing Gender Identification on Twitter with Transformer-Based Text–Image Fusion for Author Profiling", International Journal of Emerging Technologies and Innovative Research (www.jetir.org | UGC and issn Approved), ISSN:2349-5162, Vol.12, Issue 10, page no. ppc599-c604, October-2025, Available at : http://www.jetir.org/papers/JETIR2510277.pdf