Social Network Images Vulnerability Detection Using Graph Neural Network (GNN)

The empirical study involves detecting vulnerable images in social media. Use a vulnerable image detection dataset with a 78% data accuracy rate using a graph neural network. The study investigates that sharing vulnerable pictures on social networks might cause cyberbullying or stalking victims leads to many stressful events. The inexcusable behavior by many people is encompassing in the digital world. This phenomenon leads to threats of violence, slander, or even leakage of personal images may be included. The study associates vulnerability detection with the pictures posted on social media. In this way, the people of the society will be able to identify the vulnerable images and help to build social awareness. The ability of GNN to model the dependence between nodes in a network allows for a breakthrough in graph analysis research. A graph autoencoder framework can help GNNs solve the network embedding challenge. The approach associate's construction of a custom dataset with two classes, involve Vulnerable and Nonvulnerable images. We declare image size and batch size. After creating the GNN model data is trained using 10 epochs and then plot the accuracy and loss curve. The significance of the accuracy rate reached a satisfactory level. For the statistical approach, a confusion matrix is constructed to compare the actual and prediction data this helps to understand the performance of the proposed model. Graph neural network is very effective to classify images and other things. The findings constructed a Graph Neural Networks (GNNs) is an effective framework for representation learning of graphs.