To serve locally, run python3 -m http.server 9000 and navigate to localhost:9000.
The visualization is also deployed online at vis.csail.mit.edu/pubs/beyond-expertise-roles/framework-connections.
This is an interactive figure that is part of the following paper:
Suresh, Harini; Gomez, Steven R.; Nam, Kevin K.; Satyanarayan, Arvind. 2021. Beyond Expertise and Roles: A Framework for Characterizing Stakeholders of Interpretable Machine Learning and their Needs. In Proceedings of CHI '21: ACM CHI Conference on Human Factors In Computing Systems (CHI '21).
Here, we combine the two halves of the framework (knowledge-contexts and goals-objectives-tasks),and show how our more granular and composable vocabulary is able to describe a set of 58 papers from the literature on ML Interpretability (see section 4.1 for more details on this set of papers).
Each paper was coded by at least two people, who used the framework to identify instances of stakeholder knowledge types and contexts, as well as goals, objectives, and tasks. Where possible, we collected snippets of the papers corresponding to a description or discussion of stakeholder.
In the figure, light grey links represent the set of all 58 papers. Links connect codes that appear in the same paper. The width of the link corresponds to the number of papers that contain both those codes. In some cases, we did not denote a code for a certain level because we did not notice it being explicitly specified. In the figure, these cases contain links to/from the "code undetermined" node.
When hovering over a specific code, a colored overlay of links appear representing the set of papers that contain that code. Below, the papers and any corresponding snippets of that code appear. To scroll through and list of papers, click on the node and the paper list will remain up.