<< back Page 2 of 2

Harnessing the Power of AI for the Enterprise: Q&A with Seth Earley

BDQ: What do ontologies offer?

SE: If we start from a conceptual perspective, we need to look at the business imperatives. What are the things that are important to the business? And now, how do we convert those concepts into the tools and data structures that will streamline data flows and workflow processes? Starting conceptually rather than techni­cally allows us to understand how value is created and conveyed without getting hung up with technology.

An ontology is a set of organizing principles and the relationships between them. The idea is that you are cataloging your data struc­tures and sources as they relate to concepts about the business, but in a way that’s very flexible and that captures knowledge relation­ships. In this way, as the ontology evolves, it becomes something of increasing value. The ontology can power many different appli­cations. It becomes an overlay on the various data and content processes throughout the organization.

BDQ: What do ontologies enable for AI?

SE: Any machine learning algorithm will benefit from reference data, because otherwise it doesn’t know what you name things; it doesn’t have a sense of what products are for, what services are, what cus­tomers you have, or what the customer types are. There is a process by which labels can be generated by the AI but those are not always human-friendly. An analyst has to interpret them and get them aligned with approved terminology for them to be practical.

BDQ: How is information in ontologies represented and stored?

SE: When you develop an ontology, you are basically building var­ious vocabularies and hierarchies, and then building the rela­tionships between them. That information can live in a standard database, a specialized database, or in an ontology management system. Some people describe them as being stored as triples, which is a subject-verb-object structure, and that relationship itself is the data that is stored.

In addition, you can have things that are conceptually related. You might have a product and then there might be services for that, but there may be a product and an application for that prod­uct. Maybe you want to look for applications for an industry and then tell me what products you have to enable those applications. You also have the ability to manage, in one place, vocabularies that may have different levels of granularity and specificity.

BDQ: How do graph databases or other database technologies fit in?

SE: People often think of knowledge graphs as ontologies. You can also incorporate operational data and many other pieces. For example, you can build chatbots and have dialogue snippets stored in an ontology to make bots more reusable.

A knowledge graph contains entities and the relationships between them. It becomes a knowledge network with multi-dimensional sets of relationships. A good example is Facebook, where you can find people from the same high school and then find they have an interest in common with you or that they previously worked for the same company. And so it’s all about these connections, like playing the game “Six Degrees of Kevin Bacon” where the goal is to find the shortest path from a random actor to Bacon through films that various actors have in common. These are examples of traversing a knowledge graph. Building a corporate knowledge graph will help to find human expertise and answers more quickly by mapping vari­ous data sources and knowledge bases. Ontologies have to live some­place and ultimately, they’re going to be in a database of some sort.

BDQ: Why are ontologies underappreciated?

SE: People think of them as academic and abstract, and in some ways they are. But just as machine learning algorithms were academic and abstract, they then became more mainstream when they proved their value. It has been difficult to apply ontologies on a practical level. There are ontology management tools which can be challenging to deploy. Frequently, ontologies are behind the scenes. For example, ontologies are behind more powerful search. They can also be embedded in product information management systems. But those are tool-specific applications. A better architecture might be to serve them up to multiple systems using web services.

BDQ: What else is misunderstood about the use of ontologies?

SE: Another misunderstanding is that once you build an ontology you are done. That is never the case. They always evolve, and there is almost endless complexity. It’s like asking when you are done with a taxonomy, or when you’re done with sales and manufacturing and marketing, or business development and customer service. You are never finished because they change and evolve with the business.

Ontologies are very practical, but you have to build them in a practical way, incrementally. Start small; even though there is a wide range of concepts in the ontology, focus on the business problem and do not try to boil the ocean. Use basic informa­tion hygiene, which is part of the world of ontology. Then, the ontology allows you to capture more human expertise in the structures of your data source as time goes on.

<< back Page 2 of 2


Subscribe to Big Data Quarterly E-Edition