When a disaster occurs, there is an influx of information on Twitter. Humanitarian organizations use Twitter to get in touch and provide aid to those affected. Due to the massive information overload, this task becomes increasingly difficult. Many problems can be visualized as a joint learning exercise between two contexts, for example, learning from text and a social graph. Learning from multiple views has proven to significantly improve prediction accuracy for several classification problems. In this paper, we build a system that helps crisis management authorities in providing timely aid to those in need, in times of a natural disaster. By jointly learning from the graph and textual data, we also look into the interpretability of this joint-learning scheme. Interpretability provides us with a solid justification for the results we obtain from machine learning models.