Alexis
Welcome to another episode of Talk Tech with Data Dave. And today we’re doing an expert episode.
Before we get started, I’d love to remind our listeners you can always reach out to us with a question for Data Dave by emailing us at talktech@d3clarity.com. You can also send us a question right on the D3Clarity website. Or, you can connect with the Data Dave or I on LinkedIn and send us a question right there on LinkedIn. We love to answer listener questions on the podcast. So please send us some questions. We will be happy to answer them on the podcast, give you a shout out, and maybe even have you send in your questions so you can be featured on the podcast.
But like I said at the beginning, we’re doing an expert episode. Dave, today we have with us Erik Lee, which probably a lot of you remember from a Data Dave Dives Deeper that we did earlier this year.
Dave, Erik, how are you guys today?
Data Dave
Excellent. Very well.
Erik Lee
I’m doing great, thank you.
Alexis
Hopefully you guys remember Erik from his Data Dave Dives Deeper episode earlier this year where we talked a lot about taxonomy. I think we ended somewhere around talking about shoes and that’s how I kind of understood taxonomy. But the place we landed was with Dave. You said, “Hey, I wanted to talk about ontology, and I wanted to talk about knowledge graphs, but we didn’t get to talk about it today.” And I was like, “Well, that’s gonna have to be a question for another podcast.”
So we brought back Erik to join us for another podcast, where hopefully today, we can answer the question, what is ontology? We said it before, and I will say it again. That’s ontology with a T, not oncology with a C, as in not oncology as in cancer studying, but ontology as in whatever the gentlemen are about to explain to me, that probably has something to do with taxonomy.
So. if you didn’t get a chance to listen to that first episode, go back and listen to it. It’s episode 42. Otherwise, I’m gonna hand it over to you guys. Please help me. What is ontology?
Data Dave
So, I’m gonna hand it over to Erik first, because ontology… people often use it somewhat synonymously with taxonomy, and it actually has two meanings according to the dictionary. “A branch of philosophy that deals with the nature of existence” is the first definition, which we’re not necessarily going to talk about because that gets quite involved and so on. The second definition is “a list of concepts and categories in a subject area, shows the relationship between them”, which I think is more relevant to what we’re going to talk about. We’re not going to go into Aristotle and Descartes and various other deep philosophical discussions on the, you know, “I think therefore I am”, which is the other ontological study.
And I’ll hand that to Erik.
So, Erik as a taxonomist, not a taxidermist, but a taxonomist, what is an ontology? What do you consider an ontology? And see whether we can separate some of this confusion that I know exists in the world.
Erik Lee
Cool, thank you very much.
So, it’s interesting that that definition that you pulled up said it was a list. Because really, it’s not so much a list as it is a whole structure. It’s a semantic structure. It’s based on the meaning. And yes, it’s the kind of high-level definition is, it’s a concept model of a domain of knowledge. So when you think about it, it’s the things that are in a domain of knowledge, let’s say, like medicine. So, it’s the things that are in that domain and the relationships between them made explicit by this modeling. So, in medicine, you have things like patients and treatments and conditions and diseases, and they’ll have relations like patients can exhibit symptoms, patients can have a condition, symptoms are demonstrated in this way, conditions have a particular set of symptoms.
So, it’s a way to kind of organize and just model that whole domain in a way that makes it easier to kind of understand and see what those connections are and how the different kind of concepts relate to each other in a way that can also be made machine-readable.
Data Dave
A list of concepts and categories in a subject area that shows the relationship between them. The key part of that is the relationship between them is what I got from what you were saying.
So, we’ve got a list of things like symptoms, like diseases, like physicians, like body parts, anatomy, all these things can be categorized in the domain of medicine. But there isn’t as strict a relationship as there is in a taxonomy. Correct? It’s more a disease can have any number of symptoms. Also many symptoms can be related with many diseases. So, we’ve got many to many type relationships where in taxonomical terms it’s usually more rigidly a hierarchy.
Erik Lee
Right.
Alexis
You had mentioned it was more like a parent-child in taxonomy.
Data Dave
That’s right.
Alexis
Sounds more like, [spider web hand movements].
Erik Lee
And yeah, so say like, I don’t know… a taxonomy usually ends up looking like a tree. And you can have some like, related terms or preferred use. But yeah, really the strength of the ontology is you can really develop, like, what these relationships are. So as opposed to a tree, like I think you were thinking in ontology, if you have it in a graphical format, it looks like a web, basically.
Data Dave
Yeah.
Erik Lee
The things are like, usually in circles or ovals or whatever, and connected by the lines, and the lines are those relationships. And it comes out of a field of mathematics called graph theory, where basically it has these things called vertices connected by what’s called edges. And it’s related to like topology.
And it has several different uses in mathematics where you can kind of calculate these relationships.
Alexis
I liked your medical example there, but I need you to dumb it down even further for me.
Data Dave
Think of cities and roads.
Alexis
Okay, okay.
Data Dave
Cities and roads on a map. And where your thing is a city and your edge, your relationship between those cities is the road. So, the city of Chicago is connected to Detroit by an interstate.
Erik Lee
Yeah, right.
Data Dave
So, yeah, Chicago is a thing, Detroit is a thing. And they’re connected. Related by the road.
Erik Lee
Yeah, by like, uses Highway 111 or whatever.
Data Dave
Right, exactly.
Erik Lee
Or even within a city, like going from your house to like City Hall. There’s different roads. It has these different connections. So, it’s a many-to-many. My house connects to City Hall via Main Street, via First Avenue.
Data Dave
And if you think about it that way, if we relate that, you could start to say that a large thing is a city and a house is a small thing. The house is within the bounds of the city. But your house is also related to your parents’ house because you’re related to your parents. You can build a map that is multifaceted, so to speak.
Erik Lee
Yeah, this is actually a great example because it’s one of the real strengths of an ontological structure is you can make new connections based off of that.
So, like in that medicine one, let’s say you have a patient who has a condition, and the condition has a part of the body it affects, and the condition also has a treatment that treats that condition. You’ve also created this thing where the treatment affects that part of the body that the condition affects because you have those relationships between the condition, the treatment, and that part of the body. So, you can make kind of new connections based off of existing ones. So, through those connections, you can say, “Oh, these things are connected in this path. So if I query along this path, I can get this new connection made. I can get information put together in a new way based on these connections.”
Alexis
You guys kind of just explained the difference between taxonomy and ontology to me a little bit with the tree versus web. But talk to me about it in applicable terms.
When would you use ontology over taxonomy? Or when would you apply them separately or together? Help me understand that from like a business use case.
Erik Lee
An enterprise or an organization, and we kind of touched on it a bit, can have like these multiple systems that are trying to communicate with each other across the entire enterprise. So, having an ontology can connect all these things in a way where it’s a shared language across the entire enterprise. So, it really helps get past data silos and information silos, and also provides this kind of shared understanding.
So, I know we gave the example of the customer talking about taxonomies. So, in an ontology, you can say like, “A customer is a person. The customer has a name. Customer has a user id. Customer has an address. Customer has a marketing profile…” and all these ways that can connect. So, if you’re doing, for instance, analytics, like I said, you can do this, what’s called inference, pull these connections and use it in a bunch of different ways. For instance, you could query, “I want to know every customer that has hometown Chicago,” and pull that whole list just based off that connection and then connect it to each other. It’s like, okay, these people are all in a particular area depending on what’s important and how you decide to model that information.
Data Dave
You talked about the taxonomy of shoes, right? And we got the taxonomy of shoes, where shoes are in a taxonomy of classification. Women’s shoes, men’s shoes, green shoes, black shoes, various taxonomical discussion of shoes. Shirts are not in the taxonomy of shoes.
Right. Makes sense. I can have another taxonomy of properties for shirts. We’re all wearing blue shirts.
Alexis
I noticed that earlier.
Data Dave
That’s a taxonomy of shirts.
Alexis
Right.
Data Dave
Now, the fact that Alexis bought this shirt and this pair of shoes does not cause the shirts to be in the taxonomy of shoes, but it does mean there is a relationship that can infer a relationship between that shirt and that pair of shoes because they are both owned by Alexis.
Alexis
Okay.
Data Dave
So now I’ve got a linkage between that blue shirt and this red pair of shoes because Alexis owns them both.
Erik Lee
Yep. And now if I’m a retailer and I have this thing, I can say, “Alexis has these shirts. Alexis has these shoes.” And I could use this to power recommendations, like, “If she likes those shoes, these shoes have similar qualities, so she might enjoy these shoes. Or we have, like, some sort of a matching engine that will say, like, hey, if you like these shoes, here are some matching, you know, dresses to go with it,” for instance.
Data Dave
So, we’ve inferred a connection that is not part of the taxonomy.
Alexis
And back to Erik’s example… I’m learning here. Tell me if I’m wrong…. If we are using some sort of engine that says, “She bought these shoes, she’s probably going to like these similar shoes.” Those similar shoes are probably being picked because they have similar metadata attached to them, correct?
Erik Lee
Yes.
Data Dave
Yes, exactly.
Erik Lee
You got it.
Data Dave
Exactly. And then you can say things like, “Because Alexis picked those shoes, somebody like Alexis might want this shirt that Alexis liked because Alexis liked these shoes.”
Erik Lee
And how do we know what Alexis is like? Because we’ve got this metadata that comes from a taxonomy that says, “Oh, she’s outdoorsy, or she is business casual, or these ways of describing you as a customer that can also describe other customers”. So, an engine can say, this has the same metadata, these people. So, we can start putting that together for the recommendations.
Data Dave
It’s not strict, but you can pull it. You could say from the evidence presented here is that people on Talk Tech with Data Dave, wear blue shirts. Therefore, future guests on Talk Tech with Data Dave should also wear blue shirts. Therefore, we should sell blue shirts to people who are going to be on Talk Tech with Data Dave. That’s assuming that Talk Tech with Data Dave has any power in retail, but which I doubt.
Alexis
We don’t. We have one shirt design, and it’s actually black, but that’s a different story altogether.
Data Dave
But you see how the connection is made, and then you can start to infer knowledge from the connections.
Alexis
That is the perfect segue to the last question I want to ask.
Okay, you just said we’re inferring knowledge from it. And then, Erik, earlier, you said that this idea was taken from a mathematical concept of data graphs I think you said.
Erik Lee
Graph theory.
Data Dave
Graph theory.
Alexis
Graph theory. Right. And then, Dave, back on the other episode, you had mentioned that you wanted to talk about knowledge graphs. And I was like, what the heck is a knowledge graph? Because, like, to me, I think about It’s Always Sunny in Philadelphia where Charlie Day is trying to connect all of the little things in the mailroom. If you’ve never seen it, it’s very funny. There’s lots of memes about it. That’s my knowledge graph. I move on. Please help me understand what that means.
Erik Lee
Really what a knowledge graph is… at its most basic, it’s applied ontology.
It is an organization or company has taken an ontological structure and put their data and their information and what they want to model into it. And that way, it connects these concepts to concrete items or systems or units or things within the organization. So, we were talking about like, oh, an ontology of the customer. So, a knowledge graph would have like, all right, the individuals within the class of customers. So class is a grouping of individuals that all has common attributes, basically, but each individual has their own value for user ID and stuff. So there’s a list of individuals with the user IDs and it properties.
So it’s actually used like, okay, we have an ontological structure now we put like our data in it so we can use it for the things we want to use it for. And you probably hear it a lot too, used in relation to like large language models or AI, because a lot of these models, they work on statistics, like, “How tightly related are these words that are being used in the query?”
For instance, what a knowledge graph does is it makes these relationships explicit. It says these two are connected in this way. So, it adds a lot of statistical weighting to the stuff that’s modeled in the knowledge graph. Which means the response that you’re going to get is a lot more likely to be reliable and based on your information because you’re specifically giving it relationships to wait to work on.
Data Dave
One example would be think of a knowledge graph in a diagnostic framework. I’ve got a problem, and I’m calling customer support about my car. There’s a whole bunch of knowledge that is factual that can be related to each other. If I’ve got a flat tire: Part is tire. Issue is flat. Symptom is it makes a loud noise and wobbles a bit when I go around a corner. There’s symptoms, there’s issue, and there’s a part. Those three pieces are knowledge. That’s explicit connections that an expert would have. That said, “If it does this, it’s highly likely your tire is flat, and the part that needs to be replaced is the tire.” That is knowledge, that is things that an expert would know.
You can also plot that on a graph. So, you can say, “My symptom is this, my failure is this, and my part is this. Now expand that to say, well, my part is connected to my wheel, connected to my axle, connected to my engine.” You can build up this graph, this structure that starts to describe a whole load of symptoms, failure, and parts on my car. What your customer support rep can now do is walk that knowledge graph to say, “What symptom did you experience? If you experience that symptom, then it’s probably this failure linked to this part. Maybe you should do this.” That can get very, very complex.
And so that’s the concept of a knowledge graph. A graph that describes in that scenario all the failure states, all the symptoms, and all the parts of my car so that now as a user, I can walk that graph to assist in diagnosing. “Okay, I’ve replaced the tire. It still wobbles around the corner. Okay, maybe it’s the wheel bearing. Okay, maybe it’s to this, maybe it’s to that, whatever it might be.” And you can continue to walk this graph and that sort of thing and often in AI and think of an AI chatbot, right? You can automate with this knowledge that was extracted from an expert. You can now automatically walk this graph to say, “Okay, what symptom did you have? Have you checked your tire pressure?”
That’s before I need to get to and go to a mechanic. That’s just mechanical walking my knowledge graph. But I’ve got that AI type, smart, that is now mimicking the expert in the conversation.
Erik Lee
And if you’re an enterprise, like, okay, yeah, you sell car parts or something. If you have a connection between, like, your products and this into the graph, you can say, “Oh, the problem, like, oh, you have a flat tire. Would you be interested in purchasing a tire? And there’s also a connection between, like, this model of car uses this size of tire. Like, these ones will fit your car. Here you go. Are you interested?” And do stuff like that too.
Alexis
So these three things, ontology, taxonomy, and knowledge graphs, it sounds like they are very much used together very, very regularly. And just like when we were talking about taxonomy, it sounds like these are things that are being used in our everyday lives, although we maybe don’t realize they’re being used in our everyday lives.
Data Dave
Absolutely, absolutely.
Erik Lee
Yeah.
Data Dave
These are mathematical terms and structured terms for things that people use all the time. Categorization.
Erik Lee
Yeah.
Data Dave
Relationships, related information. All this is- what we’re talking about- is these are the precise terms for naming some of these areas of study.
Erik Lee
Even, like your morning routine. Like, okay, it is morning. These are activities that happen in the morning. So, I need to go to the bathroom because that is where I brush my teeth, and I brush my teeth before eating, which is an activity that happens in the dining room, et cetera. So, it’s making that kind of just implicit knowledge like, “Okay, I do this,” explicit and able to be calculated, acted upon, machine-readable, and mathematically modeled.
Alexis
Yes. Okay, I like that. I think those are wonderful definitions of those two items. I now have three more items that I’m actually understanding a little bit more, which I really appreciate.
Erik, thank you so much for joining us again on Talk Tech with Data Dave. This has been wonderful. I’m so happy that I’m actually understanding things a little bit more now. When I have breakthroughs like that on these episodes, I’m like, “Yes, that’s what this is all about.”
Erik Lee
Yes.
Data Dave
Excellent.
Alexis
Dave, as always, it’s been wonderful. Thank you so much for being here today. Listeners out there, as always, I will say it one more time. Please send us questions. You can reach out to us via email talktech@d3clarity.com. You can hop on the D3Clarity website, or you can reach out to Data Dave or myself on the LinkedIn and send us questions right there.
Erik, again, thank you so much for being with us today. Dave, thanks again for chatting.
Data Dave
Thank you.
Erik Lee
A real pleasure.
Data Dave
Thank you, Alexis, always a pleasure. Thank you everybody.
