MULTIMODAL RUMOR DETECTION FOR DOMAIN OF HEALTH

In order to solve the problem of rumor detection in the health field in social media, considering textual and visual features, this paper proposes a multimodal rumor detection method. This method integrated the image features obtained by edge detection into the image feature extraction to enhance the feature extraction effect of the image. In addition, a feature recommendation mechanism was proposed to strengthen adversarial neural networks for feature capture of rumors in the health field under multimodal data. The experimental results show that the method has good performance in the accuracy rate, recall rate and FI score indicators, which is higher than the traditional general method, and can effectively identify the health rumor information on social networks.