Multimodal Misinformation Detection Using Early Fusion of Linguistic, Visual, and Social Features