ds001 - Structured Review

Date:
April 2024
Collaborators:
Danny Hillis and SJ Klein
Code:
https://github.com/knowledgefutures/ds001
Demo:
https://ds001.knowledgefutures.org

Overview

Within the narrative content of every scientific publication is a richly connected structure of concepts, observations, and claims. Citations and references to related concepts, observations, and claims made by other publications extends these connections outward, making a rich graph that connects all the work in a field — and at the limit, all human knowledge.

Peer review traditionally provides a qualitative judgment of the narrative content by an expert who has their own mental model of this structured graph — the connections to contradicting claims, supporting claims, best practices, and ramifications. We consider an additional form of peer review, perhaps earlier in the process, that does not seek to pass qualitative judgment, but instead extracts this connected structure of concepts, observations, and claims into a publishable graph. The set of all the connected graphs across all the works in a field would create a knowledge graph capturing the detail, nuance, complexity and perhaps opportunity within a field. Such a graph could serve as the foundation for new types of applications, analyses, and discoveries.

Akin to a patent attorney pulling out and structuring the claims from the description of a new invention - we propose a form of review that pulls out and structures the claims and concepts in scientific publications, albeit in a more technical, data-focused, and interoperable manner than patent claims.

Key Ideas

  • A useful design will have multiple types of structured annotations
  • Structured annotations serve as connections that produce a knowledge graph
  • There will be multiple sources of annotations which can be overlaid on one another
  • There are a multiplicity of possible knowledge graphs based on how one builds their models of trust, relevance, and access.

Important Details

  • Structured annotations could be used for personal or professional use, created manually or computationally.
  • Annotations types could be tuned per content format. Annotating a video, dataset, codebase, etc would be different than annotating an article.
  • Comprehensiveness is in the eye of the use case. What is useful for the purpose?
  • Annotations are sometimes connections to existing content, and sometimes the generation of new content (aka nanopubs).