Skip to content

Use Cases

DRAFT: Work in progress

This material is a work in progress, at "rough draft" stage.

Data Context

"data context"

In a world where organizations must be resilient in the face of abrupt changes, we must adapt more resilient means for reconciling data from a wide variety of sources: vendors, customers, partners, government agencies, standards bodies, and so on.

KGs provide means for a kind of abstraction layer to make the data cohere.

However, it's not clear when Gartner will acknowledge the breadth of industry adoption for KG approaches in enterprise data management.

To paraphrase Natasha Noy, a research scientist at Google Research and highly-regarded practitioner in this field:

An "enterprise knowledge graph" provides ground truth through which we can reconcile our queries and other usage of many disparate data stores.

For example, having persistent identifiers with other metadata attached is a great start.

For several KG projects, we kept reusing a similar working set of libraries:

Each of these libraries provides a useful piece of the puzzle when you need to leverage knowledge representation, graph algorithms, entity linking, interactive visualization, metadata queries, axioms, etc. However, some of them are relatively low-level (e.g., rdflib) or perhaps not maintained as much (e.g., skosify) and there are challenges integrating them. Challenges we kept having to reinvent work-arounds to resolve.

There are general operations that one must perform on knowledge graphs:

  • building triples
  • managing a mix of namespaces
  • serialization to/from multiple formats
  • queries
  • interactive visualization
  • transitivity and other forms of enriching a graph
  • graph algorithms
  • inference (e.g., PSL, Bayesian Networks, Causal, MLN, etc.)
  • quality assurance (e.g., axioms)
  • parallel processing across a cluster
  • embedding (deep learning integration)
  • other ML integrations

Last update: 2020-12-31