At the Data Storytelling Virtual Summit, Larry Birnbaum, CoFounder of Narrative Science and Professor of Computer Science at Northwestern, shared his expertise on “How to Model Questions?” within a compelling data story.
Key Insights from this talk:
Storytellers need models of what the listener needs to know.
- These models are based on questions that the listener/user has asked and the questions that will be asked down the line. It is typical for more questions to arise based on answers given, so the models need to predict those next questions.
Characteristics are not metrics
- “Highest” and “Best” are often the same, whether it is sales this quarter, new opportunities or deals closed. They are not always the same however, this distinction can be problematic for machine learning. Highest is an objective fact. Best, on the other hand, must be related to a performance goal and the computer must know the goal and understand the relationship.
A list is not a story.
- A list generated by a machine does not answer the question of “why.” It conveys information and we can understand it. However, it lacks the reasoning and depth of a story. It only gives us surface level information.