MULTIMODAL HATEFUL MEME DETECTION USING RESNET AND BERT

The proposed Multimodal Hateful Meme Detection System successfully combines ResNet50 and BERT to analyse both visual and textual components of memes. The system overcomes the limitations of traditional unimodal approaches by capturing contextual relationships between image and text. Experimental results show that the model achieves an accuracy of approximately 85%–90%, demonstrating its effectiveness in identifying both explicit and implicit hateful content. The probability-based predictions further improve interpretability and usability of the system. In addition to high accuracy, the system provides a complete platform with user and admin modules for content moderation, user interaction, and data management. Overall, the proposed system offers a reliable, scalable, and efficient solution for automated hateful meme detection in real-world social media environments.

[1]. Hateful Meme Prediction Model Using Multimodal Deep Learning; Md Rekib Ahmed, Neeraj Bhadani,          Ishita Chakraborty;    2021 International Conference on Computing, Communication and Green Engineering (CCGE);DOI:10.1109/CCGE50943.2021. 9776440; ESSN: null [2]. Improving Multimodal Hateful Meme Detection Exploiting LMM-Generated Knowledge; Maria Tzelepi, Vasileios Mezaris; arXiv preprint arXiv:2504.09914; DOI: 10.48550/arXiv.2504.09914; [3]. Improving Multimodal Hateful Meme Detection Exploiting LMM-Generated Knowledge; Maria Tzelepi, Vasileios Mezaris; 2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW 2025); DOI: 10.1109/CVPRW67362.2025.00025. [4]. Multimodal Hateful Meme Classification Based on Transfer Learning and a Cross-Mask Mechanism; Fan Wu, Guolian Chen, Junkuo Cao, Yuhan Yan, Zhongneng Li; Electronics, Volume 13, Issue           14        (2024); DOI: 10.3390/electronics13142780; ESSN: 2079-9292. [5].Hateful Memes Detection Based on Mu lti-Task Learning; Zhiyu Ma, Shaowen Yao , Liwen Wu, Song Gao, Yunqi Zhang; Math ematics, Volume 10, Issue 23; DOI: 10.339 0/math10234525; ESSN: 2227-7390. ([mdpi.com][1]