Abstract
Social media applications are rich sources of information and provide a platform to spread and gather any news. The universal nature of online media has led to the generation of abundant data over networks. This unconditional way to share over social networks leads to the spread of false information. With this rising popularity and the growth of communication over social media, the problem of spreading rumors is increasing. Rumor detection is a trending subject in the social networking world, and it is a difficult problem to solve. Approaches to discovering rumors have evolved highly over the past decade. These approaches significantly impact enterprises that expect to detect rumors, over massive data in social networks, at a large scale. Most efforts use state-of-art machine learning approaches: hand-crafted textual features from the dataset, rumor classification, and rumor prediction. However, with the availability of abundant information and unreliability in the use of hand-crafted features, models are expected to utilize all possible features over data. Deep learning models detect rumors based on the propagation path of rumors but overlook structures of dynamic diffusion over a network. This project demonstrates a model implemented to work on the dynamic features of rumors. The project's primary goal is to capture the influential features of rumors and learn over dynamic structures built over propagation path for early rumor detection. We use Graph Convolutional Networks (GCN) to represent the rumor propagation tree with source and response posts as graphs to achieve the goal. GCN also updates node representations in the graphs based on responses for rumors generated over time. We will use a pattern matching algorithm to detect similar user stance sub-graph patterns generated across the graph for efficient structure reconstructions. Therefore, through such recursive learning over the updated graph structures, we can accurately predict rumors. Through this project we hope to provide an efficient working model that predicts rumors earlier than the existing approaches.