CrystalCandle: A User-Facing Model Explainer for Narrative Explanations
This addresses the issue of model adoption for end users at platforms like LinkedIn, though it appears incremental as it builds on existing interpretation methods with a focus on user-facing customization.
The paper tackles the problem of low trust in predictive machine learning models due to lack of interpretability by proposing CrystalCandle, a user-facing explainer that creates narrative explanations, resulting in lifts in adoption rates and increases in downstream metrics like revenue compared to previous approaches.
Predictive machine learning models often lack interpretability, resulting in low trust from model end users despite having high predictive performance. While many model interpretation approaches return top important features to help interpret model predictions, these top features may not be well-organized or intuitive to end users, which limits model adoption rates. In this paper, we propose CrystalCandle, a user-facing model explainer that creates user-digestible interpretations and insights reflecting the rationale behind model predictions. CrystalCandle builds an end-to-end pipeline from machine learning platforms to end user platforms, and provides users with an interface for implementing model interpretation approaches and for customizing narrative insights. CrystalCandle is a platform consisting of four components: Model Importer, Model Interpreter, Narrative Generator, and Narrative Exporter. We describe these components, and then demonstrate the effectiveness of CrystalCandle through use cases at LinkedIn. Quantitative performance analyses indicate that CrystalCandle's narrative insights lead to lifts in adoption rates of predictive model recommendations, as well as to increases in downstream key metrics such as revenue when compared to previous approaches, while qualitative analyses indicate positive feedback from end users.