Home » Blogs » AI Recruiting Technology » All-Yo Questions: The Use Of ...

All-Yo Questions: The Use Of Ontology In Recruitment Automation

Ontology Q&A

with Ankit Somani, Cofounder, and Michael Baxter, Senior Information Engineer 

What is ontology?

Ankit:
Hi, this is Ankit your host for all your questions. It’s another Friday afternoon in Mountain View and I’m sitting here today with Michael who is a senior information engineer at AllyO, Michael do you want to say something about who you are and where you come from?

Michael:
I enjoy hard problems and it’s taken time to learn how to solve them but I always think outside the box.

Ankit:
It’s awesome, that’s what we count on for with Michael. So today, Michael is gonna talk about a very important topic and it’s ontology. Michael, do you wanna share what ontology really is? It seems like people keep hearing about it, it’s the word that’s thrown around but there isn’t a good definition in the HR world that I’ve found.

Michael:
Right, so, first of all, we do not mean the philosophical sense of the word ontology, this is much simpler. This word is more in a computer science or systems sense and we worked on it a while and cut it down to a simple definition. Ontology is modeling meaning and relatedness using data. It’s super simple meaning is about things we know or don’t, relatedness is how they connect to each other and data is the representation.

Ankit:
Sounds like it’s some taxonomy on steroids.

Michael:
It’s kind of like that yes. That’s a good way to think about it.

Why is ontology important?

Ankit:
Awesome, so why is ontology important? Why should everyone know about it?

Michael:
I think it’s important because there are certain domains, HR Tech is one, which are not large in a sense of big data. I think there’s a small number of vocabularies around hiring people and things like that, so having a representation that’s able to deal with paucity or lack of information rather than being oversubscribed with tons and tons of data is much more important

How is machine learning different from ontology?

Ankit:
Makes sense, and you know when people think about artificial intelligence and ontology is useful for that, people hear a lot around machine learning. In fact, it’s the pop star of last 5 years. How do you think machine learning differs from ontology and where do you think ontology is more relevant than machine learning?

Michael:
Good questions, I would describe machine learning as a pattern matching discipline. it’s a discipline in which some really good strong people have done the math and cleaned up a lot of the computational problems for pattern matching from something that is unknown into the ability to classify with a label that has taken lots and lots of training data to obtain a super optimum result where the precision of recall is human equal or better. What’s different about ontology is we’re doing a knowledge representation this is actually modeling meaning rather than pattern recognition. So when we model meaning we have the ability to make an inference with meaning. Then we can start to comprehend or understand what text streams actually refer to and that’s different than pattern matching in the sense that it can also identify things that are not known or gaps in the information. Pattern recognition is more about what’s going on, in the sense of does this pattern match a label or not in its classification. This is a little more subtle and it uses logical connectives that are strictly binary so there’s no ambiguity. So the way I like to think of it is, an ontology represents what I know or what do I know, what I call WDIK. It’s kind of interesting in that when something isn’t inferable, it’s immediately in the other class called I don’t know or IDK. This IDK is useful because it’s like a feedback signal, you can actually sense the gap of something that’s missing in terms of representation immediately. And it’s not necessarily a bad thing, it’s how things work.

Ankit:
So I think what I got from that was that with ontology you actually understand how things are related to each other, make an inference from it, and then you can comprehend things.

Michael:
Yes, the system can comprehend things.

Ankit:
So you can even ask the system what you comprehended while you comprehend it, ask those questions. It’s not like it’s looking through a bunch of cat pictures and the system understood that this thing looks like a cat, which is what the current incarnation is doing.

Michael:
That’s correct. So one of the key differences is that there’s no probabilistic inferencing going on in the use of an ontology. An ontology is almost like a linear algebra equation. You can keep cascading on meaning as long as you like. So I talked about kind the density or grid nests of an ontology, how tight of a knowledge representation is it in terms of a particular domain of interest and the idea of that is. You can make it as tight as you like, I can keep adding to it all I want until I get to inferences that are very very finely defined and that’s extremely useful. It’s a lot different than using a probabilistic system, you know Bayesian rules or other various complicated hyperplanes to do pattern matching. The other thing that’s different about it is the logical rules and relatedness matter. I can go backward if I found something I inferred I can ask the question why do ‘I know that?’ and that’s the key to enabling a lot of cool applications, being able to infer the chain of reasoning that led to an inference and then exploiting that further.

What are some use cases for ontology?

Ankit:
Yeah, that sounds fascinating. As you’re eluding towards interesting use cases I’d love to hear what you have in mind for the kind of use cases one can have

Michael:
Some of the things I think about our cross modalities like looking at one piece of information you might not have looked at in juxtaposition to another. So one of them might be looking at an applicant’s CV or at his or her resume but then combining that with other pieces of information at different stages in the hiring and employment process. So one of the things that’s changing is kind of the range of effects, the ability to look at HR track as something is not just time of hire but perhaps much wider in the span of time.

Ankit:
Yeah that sounds exciting. In fact, you know just to build on that one of the things that was interesting for me when you mentioned understanding is that you don’t need to understand a certain keyword you can understand the whole giant concept. So screening can be much better like if I am looking for a sales associate whose great relationship-based sales I could see an answer from them and understand whether that is close enough to the concept of where relationship-based sales versus – far from it or they are more a hunter or a farmer or whatever it may be

Michael:
That’s right. Another example is related to that kind of thing, it’s even is less subtle. One of them might be for example mobile equipment. Mobile equipment can mean one thing if you are an RF engineer but it could mean something entirely different if you work in the field of construction where mobile equipment means like a ditch-digger, something I’d love to have my house forcleaning up the garden, but this kind of thing does distinguish this ability to distinguish those nuances. It’s very interesting and different than machine learning because it’s not like you can have one label that looks like a ditch-digger and also a cat or just digger and a cell phone and connote those two things ,you need a piece of information you need a knowledge: what do I know? What do I know is the ditch digger runs on diesel and the cell phone runs on battery power you know there’s a lot of different ways of looking at it makes sense

Ankit:
All right Michael, I will quickly summarize what I heard. Give me a thumbs up or down whether it makes sense. So ontology is a really important concept, it was almost forgotten in the world of AI, the big data world took over, which is mostly around pattern recognition and call it on steroids because we just applied compute power to it to get something out of it. But this is a really understanding how things are related and how close they are to each other and using that knowledge and its representation in terms of making inferences and using those inferences to create new experiences. As you mentioned it’s even more important because the data is not always available it’s not always accessible given the level of security that needs to happen around it and so it can really create new experiences

Michael:
That’s right so thumbs up, let me add a little bit further. What I like about it as well is that ability to infer is actually extremely important for mobile in different ways. I think it’s a key to what makes the artificial intelligence a more humane experience. In other words for the candidate for the employers for everybody involved this is a way by having knowledge representation to unique fie and individualize an experience in that sense it’s not trying to find a needle in a haystack because this looked like a needle and these are all hay but rather it’s unique to the individual and unique to the context in which it all happens

Ankit:
Awesome thank you Michael for your time if you have any questions please do leave it in the comments we’ll be happy to take it up in our next session thank you very much.

 

 

Wondering how AllyO can transform your recruiting process and delight candidates? Schedule a demo to find out. 

SUBSCRIBE TO ALLYO BLOG

Stay up-to-date with the latest insights and trends from AI recruiting brought to you by AllyO Blog!

Schedule Demo Widget

Ready to transform your HR communications?

Learn More

How AllyO helps you adapt now and prepare for the future of hiringLearn More
+