"What is Relevant in a Text Document?": An Interpretable Machine Learning Approach