- From: Phillip Rhodes <motley.crue.fan@gmail.com>
- Date: Fri, 3 Aug 2018 19:15:33 -0400
- To: Michael F Uschold <uschold@gmail.com>
- Cc: John Leonard <john.leonard@incisivemedia.com>, "semantic-web@w3.org" <semantic-web@w3.org>
There is, indeed, an entire field - and body of research - on "ontology learning". See, for ref: https://en.wikipedia.org/wiki/Ontology_learning https://pdfs.semanticscholar.org/8198/9605fba59a415a28603ce709566cc6ebc45c.pdf https://arxiv.org/pdf/1311.1764.pdf etc... Phil This message optimized for indexing by NSA PRISM On Fri, Aug 3, 2018 at 6:57 PM, Michael F Uschold <uschold@gmail.com> wrote: > This is a very good question. I agree with other responses that ML and the > Semantic Web technologies are not in competition, they do very different > things. The question is whether and how they can work together in concert. > ML is used in a lot of NLP which is, in turn, used to extract triples from > text. The state of the art here is improving all the time, but it is still > not that great. A common approach is for a NLP engine to take as input both > an ontology and a text document and to extract triples using the classes and > properties in the ontology. This works best if the ontology has a lot of > text descriptions of the classes and properties. I'm not sure how much these > triple-extractors make use of the ontology axioms, probably they use the > class hierarchy, and property domains and ranges. > > Another interesting possibility, which I have not seen much written is using > ontology vocabulary to express the features that are learned. I'm not an > ML person, but in ML they talk about models, which to some extent describe > the kinds of things in the subject area and their attributes/features. > Probably someone has looked into this. > > Probably the most common use of ML is to put things into pre-determined > categories. But its just statistics. There is no human-grokkable way to > explain why something goes into one category vs. another. It would be nice > if there was a way to do that, in terms of an ontology classes and > properties. Don't know if that is being done. > > Another possible link up is where ML is used to do automatic creation of > categories. Humans can look at the categories and give them meaningful > names, but to the computer they are just the result of statistics, they have > no meaning. It could be that meaning of these categories could be inferred > and matched against an ontology. General topics would probably start with > the DBpedia ontology. > > Way back in 2005 there was a Dagstuhl workshop on ML and the Semantic Web; > but there is not a lot of documentation on that event. > > Michael > > > > > > > On Fri, Aug 3, 2018 at 4:30 AM, John Leonard > <john.leonard@incisivemedia.com> wrote: >> >> Can someone please fill me in on the latest state of play between the >> symantec web and using machine learning techniques with massive unstructured >> datasets to derive probablistic links between data items. Are the two >> techniques in competition? Are they compatible? Or is it more a case of >> horses for courses? I refer to this 2009 paper >> https://static.googleusercontent.com/media/research.google.com/en//pubs/archive/35179.pdf >> The Unreasonable Effectiveness of Data by Norvig et al. >> Thanks for any pointers >> >> > > > > -- > > Michael Uschold > Senior Ontology Consultant, Semantic Arts > http://www.semanticarts.com > LinkedIn: www.linkedin.com/in/michaeluschold > Skype, Twitter: UscholdM > > >
Received on Friday, 3 August 2018 23:15:59 UTC