Year: 2024
Type: conference-paper
Abstract: With the popularity of social media and internet applications, the distribution of sensitive information via social media is a growing phenomenon.However, existing deep learning methods rarely focus on the relationship between text and image content in an image simultaneously, and tend to ignore the real meaning hidden in text or image content.In this paper, we propose a multimodal deep learning method that combines ResNet and BERT so that the model can focus on the relationship between image content and text information simultaneously.First, a fully connected layer is added to both ResNet and BERT to reduce the high-dimensional features extracted by the model to low-dimensional features and reduce the computational complexity.Second, the text features are pre-processed by self attention mechanism and residual connection to reduce the risk of gradient vanishing and improve the fusion effect.Finally, feature fusion is performed by using Multi-Head Attention mechanism to dynamically assign weights.The results show that the proposed model in this paper has higher accuracy of sensitive content detection compared to other classification models.
Authors: Zheng Linhui, Liu Wei, Liu Qiong
Institution: Beijing Information Science & Technology University
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4411021133