Using the Image-Text Relationship to Improve Multimodal Disaster Tweet Classification