3 ms·
> Particularly people saw vocabularies [...] and they felt the W3C was trying to put them in a straightjacket. [...] [T]he goal of the semantic web is just the
by echelon 5y ago
> Particularly people saw vocabularies [...] and they felt the W3C was trying to put them in a straightjacket. [...] [T]he goal of the semantic web is just the opposite. Hypothetically you should be able to take data from multiple sources and various vocabularies, throw them all into one big graph
Absolutely right.
> throw them all into one big graph [...] and then do OWL and RDF inference to infer new facts in a desired vocabulary. [...] What I learned from 10 years of work around the semweb is how to build a very sloppy system and apply 'data lake' techniques to full scan it and refine a process of creating data extracts that completely fit the requirements for things I want to do with data
> You can't count on it being clean at the beginning you need a repeatable and scalable process to do it.
Just to paint the picture for others, this is the "hard" vision for Semantic Web, where all information arises from disparate sources and goes into a giant queryable knowledge graph. This vision has a ton of value to certain users (it's very enterprisey), but 90% of us would benefit tremendously from only the most simple tools and ontologies.
If we focus on ontologies around blogging, threaded conversations, identity, and chat, we'd see immediate use and potential. Grow from simple beginnings.
- PaulHoule 5y agoActually "enterprise" might not be big enough. Google and Facebook might be more like it. I don't think most people understood it at the time, but schema.org, as it is deployed on the web, is adding some predicate and type structure to text that isn't useful at all unless it is paired with an entity matching system which was beyond state of the art at this time. Now neural networks for text are nowhere near as successful at they are for images and audio but it's clear to me today that such a system can be built and that you could even use schema.org metadata to bootstrap the training set for information extraction systems that don't need the metadata. As it was I saw schema.org as much more an effort for "big web" to take as opposed to give.
- zozbot234 5y ago> that isn't useful at all unless it is paired with an entity matching system You can obviously provide entity disambiguation as part of schema.org-formatted data - it's very much part of the spec, for any "entity" type. In most cases, just linking to a well-known description of the entity (such as a wikidata.org/entity/Qnnnn URI if it's generally notable) is enough. The same works to disambiguate named entities as part of running text.