Knowledge Graph Development for More Effective AI Systems: Systematic Approaches

• Mike's presentation https://bit.ly/gphv1e5
• Connect with Mike https://www.linkedin.com/in/mikedillinger/
• Interesting reads:
• https://www.linkedin.com/newsletters/knowledge-architecture-7149627777497841664/
• https://www.linkedin.com/pulse/new-years-resolutions-your-knowledge-graphs-mike-dillinger-phd-g7dsc/
• https://www.linkedin.com/pulse/knowledge-graphs-artificial-intelligence-mike-dillinger-phd-zxaac/
• Graph of Relations analysis https://arxiv.org/abs/2305.19987v3
• Taxonomies: Connecting Data with Duct Tape https://www.youtube.com/watch?v=gvoU75vHPQI

About this Episode

Building a knowledge graph isn’t just about gathering and connecting data, it’s about making sense of the entities and relationships represented within it. On the one hand, poorly resolved entities and vague connections will weaken a graph’s ability to deliver meaningful insights. On the other hand, let’s say you have a well-constructed graph and intend to use it for powering AI systems downstream … How do the choices you make about representing your entities and relations have an impact on AI applications?

Mike Dillinger is an expert in knowledge graphs and computational linguistics, with key leadership roles at LinkedIn developing their graph assets. In a recent series of articles, Mike advocates a methodical approach to knowledge graph development to support more effective AI systems through what might be called a “shift-left” strategy: provide structured knowledge input – artificial knowledge – to enable machine intelligence. Build on entity resolution as a required first step, then apply systematic approaches for relation resolution as well. We need to measure the diversity, depth, and density of relations so they can be categorized and clearly defined, and then structure the predicates we use – just as we structure our entities. The result is a best-of-breed mix of knowledge graphs and deep learning, aka neuro-symbolic systems, more effective and scalable.

In this session, Paco Nathan and Mike Dillinger will dive into the critical role of both entity resolution and relation resolution in knowledge graph construction. We’ll explore the challenges of defining clear, accurate relationships and discuss techniques for improving how connections are identified, structured, and maintained over time. Whether you’re building a new graph or enhancing an existing one, you’ll gain practical strategies to ensure your graph remains scalable, flexible, and aligned with evolving data.