I went into Microsoft Ignite 2020 looking for information on Project Cortex. With a background in Library & Information Science and years of experience building knowledge bases and hierarchical taxonomies, I was fascinated to learn how Project Cortex is using Artificial Intelligence (AI) to connect people, ideas, and topics. In this blog post, I’m sharing links to some of my favorite Ignite session videos and initial thoughts on Project Cortex.
#MS Ignite sessions
- Project Cortex: Knowledge discovery and content intelligence in M365
- How to build a Document Understanding Model using Project Cortex
- Expertise & Knowledge Networks: Microsoft’s Vision For a Successful Knowledge System In Your Organization
- Connect people with knowledge and expertise in Microsoft 365
- Rolling out Project Cortex at DXC Technology, Mott MacDonald, and Protiviti
- How to Successfully Activate and Adopt Project Cortex
Knowledge in your organization is like stars in the sky. I LOVED Naomi Moneypenny’s analogy that people, ideas, and resources in our organizations connect together like stars in constellations. Project Cortex helps draw lines and connect all the people and disparate knowledge in our organizations, exposing ideas, data, and content in new and interesting ways.
Microsoft SharePoint Syntex is the first product from Project Cortex. Syntex uses a SharePoint Content Center site to create, manage, and deploy understanding models. These models teach Microsoft’s AI how to review your content and make connections to build knowledge. You can teach the model to understand data the way you do. As different subject matter experts across your organization teach the model what they know, the model is able to look at data from multiple perspectives and deliver the right content to the right user.
Building a model is easier (and faster) than you’d think. I was impressed to learn you need a relatively small set of content to build and train your model. You can build a model with as few as 5 sample documents. And it’s brilliant that they require you to provide the model with both “good examples” and at least one “bad example.” If you’re trying to teach a model how to review organizational purchase orders, for example, you should upload one document that is clearly not a purchase order so it learns how to recognize anomalies.
Training your model is a straightforward 4-step process:
- Add example files (minimum of 5)
- Classify files & run training (label your positive and negative sample files & train the model on keywords and phrases that are important to you in each file)
- Create and train extractors
- Apply the new model to document library(s)
Information architecture is vital. Knowledge and information is meaningless without context. And a solid information architecture is a foundational part of having strong AI experiences. As Naomi Moneypenny shared in her session, “Any investment that you make in information architecture will pay dividends in AI, helping to give it structure, helping to give it seeding, and actually promoting a much better experience.”