OWL/SWRL and Machine Learning

classic Classic list List threaded Threaded
4 messages Options
Reply | Threaded
Open this post in threaded view
|

OWL/SWRL and Machine Learning

Michael DeBellis-2
A while back I went offline with someone who was posting questions about how to design a system using SWRL. The more we look at his problem the more it looks to me like a classic Machine Learning problem. I've been advising him that if he uses ML he is better off to start with an environment designed for ML rather than with OWL and SWRL. E.g., Octave, Matlab, Python, etc. I just wanted to double check to make sure I'm not giving him bad info. 

Is there any robust reusable work that implements things like matrices and matrix operations in OWL and SWRL? Or is there any alternative way using SWRL to do machine learning types of things such as take a bunch of data -- e.g., matrices (data property values) from a Trusted Account and matrices for a Fake Account -- and automatically infer the rules to tell one from another?

I'm pretty sure the answer is no and that using an environment designed for ML would be the way to go but just wanted to check if anyone disagrees or has other suggestions. Thanks,

Michael


_______________________________________________
protege-user mailing list
[hidden email]
https://mailman.stanford.edu/mailman/listinfo/protege-user
Reply | Threaded
Open this post in threaded view
|

Re: OWL/SWRL and Machine Learning

Jim McCusker
There's this sort of work, but I don't know how robust it is:

Gajderowicz, Bart & Sadeghian, Alireza & Soutchanski, Mikhail. (2013). Ontology Enhancement Through Inductive Decision Trees. 10.1007/978-3-642-35975-0_14.

The popularity of ontologies for representing the semantics behind many real-world domains has created a growing pool of ontologies on various topics. While different ontologists, experts, and organizations create the vast majority of ontologies, often for narrow application do-mains, they frequently overlap with other ontologies in broader domains, specifically as they pertain to the Semantic Web. These overlapping on-tologies sometimes model similar or matching theories, that may be in-consistent. To assist in the reuse of these ontologies, this paper describes a technique for enriching manually created ontologies by supplementing them with inductively derived rules, and reducing the number of incon-sistencies. The derived rules are translated from decision trees created by executing a tree based data mining algorithm with probability mea-sures over the data being modelled. These rules can be used to revise the ontology adding a higher level of granularity, in order to identify possible similarities missed by the original ontologists. We then discuss how this may be applied to ontology matching. We demonstrate the application of our technique by presenting an example, and discuss how various data types may be treated to generalize the semantics of an ontology for a broader application domain.



On Tue, Jun 11, 2019 at 8:55 AM Michael DeBellis <[hidden email]> wrote:
A while back I went offline with someone who was posting questions about how to design a system using SWRL. The more we look at his problem the more it looks to me like a classic Machine Learning problem. I've been advising him that if he uses ML he is better off to start with an environment designed for ML rather than with OWL and SWRL. E.g., Octave, Matlab, Python, etc. I just wanted to double check to make sure I'm not giving him bad info. 

Is there any robust reusable work that implements things like matrices and matrix operations in OWL and SWRL? Or is there any alternative way using SWRL to do machine learning types of things such as take a bunch of data -- e.g., matrices (data property values) from a Trusted Account and matrices for a Fake Account -- and automatically infer the rules to tell one from another?

I'm pretty sure the answer is no and that using an environment designed for ML would be the way to go but just wanted to check if anyone disagrees or has other suggestions. Thanks,

Michael

_______________________________________________
protege-user mailing list
[hidden email]
https://mailman.stanford.edu/mailman/listinfo/protege-user


--
Jim McCusker

Director, Data Operations
Tetherless World Constellation
Rensselaer Polytechnic Institute
[hidden email]
http://tw.rpi.edu

_______________________________________________
protege-user mailing list
[hidden email]
https://mailman.stanford.edu/mailman/listinfo/protege-user
Reply | Threaded
Open this post in threaded view
|

Re: OWL/SWRL and Machine Learning

Michael DeBellis-2
Thanks very much Jim that is exactly the kind of thing I was looking for. It's also interesting to me for my own work. I've been reading a lot of cognitive science literature on how people organize and use concepts and it was kind of surprising all the ways that people routinely ignore basic set theory (for people who know Cog Sci I'm referring to work by people like Rosch and Lakoff) so I've been wondering if there were ways to represent human ontologies in a formal way that can still follow the kind of logic seen in experiments by people like Rosch. Fuzzy OWL is one thing I've looked at and this paper also looks relevant. 

Michael


On Tue, Jun 11, 2019 at 6:16 AM Jim McCusker <[hidden email]> wrote:
There's this sort of work, but I don't know how robust it is:

Gajderowicz, Bart & Sadeghian, Alireza & Soutchanski, Mikhail. (2013). Ontology Enhancement Through Inductive Decision Trees. 10.1007/978-3-642-35975-0_14.

The popularity of ontologies for representing the semantics behind many real-world domains has created a growing pool of ontologies on various topics. While different ontologists, experts, and organizations create the vast majority of ontologies, often for narrow application do-mains, they frequently overlap with other ontologies in broader domains, specifically as they pertain to the Semantic Web. These overlapping on-tologies sometimes model similar or matching theories, that may be in-consistent. To assist in the reuse of these ontologies, this paper describes a technique for enriching manually created ontologies by supplementing them with inductively derived rules, and reducing the number of incon-sistencies. The derived rules are translated from decision trees created by executing a tree based data mining algorithm with probability mea-sures over the data being modelled. These rules can be used to revise the ontology adding a higher level of granularity, in order to identify possible similarities missed by the original ontologists. We then discuss how this may be applied to ontology matching. We demonstrate the application of our technique by presenting an example, and discuss how various data types may be treated to generalize the semantics of an ontology for a broader application domain.



On Tue, Jun 11, 2019 at 8:55 AM Michael DeBellis <[hidden email]> wrote:
A while back I went offline with someone who was posting questions about how to design a system using SWRL. The more we look at his problem the more it looks to me like a classic Machine Learning problem. I've been advising him that if he uses ML he is better off to start with an environment designed for ML rather than with OWL and SWRL. E.g., Octave, Matlab, Python, etc. I just wanted to double check to make sure I'm not giving him bad info. 

Is there any robust reusable work that implements things like matrices and matrix operations in OWL and SWRL? Or is there any alternative way using SWRL to do machine learning types of things such as take a bunch of data -- e.g., matrices (data property values) from a Trusted Account and matrices for a Fake Account -- and automatically infer the rules to tell one from another?

I'm pretty sure the answer is no and that using an environment designed for ML would be the way to go but just wanted to check if anyone disagrees or has other suggestions. Thanks,

Michael

_______________________________________________
protege-user mailing list
[hidden email]
https://mailman.stanford.edu/mailman/listinfo/protege-user


--
Jim McCusker

Director, Data Operations
Tetherless World Constellation
Rensselaer Polytechnic Institute
[hidden email]
http://tw.rpi.edu
_______________________________________________
protege-user mailing list
[hidden email]
https://mailman.stanford.edu/mailman/listinfo/protege-user

_______________________________________________
protege-user mailing list
[hidden email]
https://mailman.stanford.edu/mailman/listinfo/protege-user
Reply | Threaded
Open this post in threaded view
|

Re: OWL/SWRL and Machine Learning

Jim McCusker
Yes, we are starting to use set theory and pos-hoc classification of instances to solve interesting problems here too.

Jim

On Tue, Jun 11, 2019 at 9:58 AM Michael DeBellis <[hidden email]> wrote:
Thanks very much Jim that is exactly the kind of thing I was looking for. It's also interesting to me for my own work. I've been reading a lot of cognitive science literature on how people organize and use concepts and it was kind of surprising all the ways that people routinely ignore basic set theory (for people who know Cog Sci I'm referring to work by people like Rosch and Lakoff) so I've been wondering if there were ways to represent human ontologies in a formal way that can still follow the kind of logic seen in experiments by people like Rosch. Fuzzy OWL is one thing I've looked at and this paper also looks relevant. 

Michael


On Tue, Jun 11, 2019 at 6:16 AM Jim McCusker <[hidden email]> wrote:
There's this sort of work, but I don't know how robust it is:

Gajderowicz, Bart & Sadeghian, Alireza & Soutchanski, Mikhail. (2013). Ontology Enhancement Through Inductive Decision Trees. 10.1007/978-3-642-35975-0_14.

The popularity of ontologies for representing the semantics behind many real-world domains has created a growing pool of ontologies on various topics. While different ontologists, experts, and organizations create the vast majority of ontologies, often for narrow application do-mains, they frequently overlap with other ontologies in broader domains, specifically as they pertain to the Semantic Web. These overlapping on-tologies sometimes model similar or matching theories, that may be in-consistent. To assist in the reuse of these ontologies, this paper describes a technique for enriching manually created ontologies by supplementing them with inductively derived rules, and reducing the number of incon-sistencies. The derived rules are translated from decision trees created by executing a tree based data mining algorithm with probability mea-sures over the data being modelled. These rules can be used to revise the ontology adding a higher level of granularity, in order to identify possible similarities missed by the original ontologists. We then discuss how this may be applied to ontology matching. We demonstrate the application of our technique by presenting an example, and discuss how various data types may be treated to generalize the semantics of an ontology for a broader application domain.



On Tue, Jun 11, 2019 at 8:55 AM Michael DeBellis <[hidden email]> wrote:
A while back I went offline with someone who was posting questions about how to design a system using SWRL. The more we look at his problem the more it looks to me like a classic Machine Learning problem. I've been advising him that if he uses ML he is better off to start with an environment designed for ML rather than with OWL and SWRL. E.g., Octave, Matlab, Python, etc. I just wanted to double check to make sure I'm not giving him bad info. 

Is there any robust reusable work that implements things like matrices and matrix operations in OWL and SWRL? Or is there any alternative way using SWRL to do machine learning types of things such as take a bunch of data -- e.g., matrices (data property values) from a Trusted Account and matrices for a Fake Account -- and automatically infer the rules to tell one from another?

I'm pretty sure the answer is no and that using an environment designed for ML would be the way to go but just wanted to check if anyone disagrees or has other suggestions. Thanks,

Michael

_______________________________________________
protege-user mailing list
[hidden email]
https://mailman.stanford.edu/mailman/listinfo/protege-user


--
Jim McCusker

Director, Data Operations
Tetherless World Constellation
Rensselaer Polytechnic Institute
[hidden email]
http://tw.rpi.edu
_______________________________________________
protege-user mailing list
[hidden email]
https://mailman.stanford.edu/mailman/listinfo/protege-user
_______________________________________________
protege-user mailing list
[hidden email]
https://mailman.stanford.edu/mailman/listinfo/protege-user


--
Jim McCusker

Director, Data Operations
Tetherless World Constellation
Rensselaer Polytechnic Institute
[hidden email]
http://tw.rpi.edu

_______________________________________________
protege-user mailing list
[hidden email]
https://mailman.stanford.edu/mailman/listinfo/protege-user