Justification Narratives for Individual Classifications

Machine learning models are now used extensively for decision making in diverse applications, but for non-experts they are essentially black boxes. While there has been some work on the explanation of classifications, these are targeted at the expert user. For the non-expert, a better model is one of justification not detailing how the model made its decision, but justifying it to the human user on his or her terms. In this paper we introduce the idea of a justification narrative: a simple model-agnostic mapping of the essential values underlying a classification to a semantic space. We present a package that automatically produces these narratives and realizes them visually or textually.