[JimmyRuska]

Semantic web tech solves a common problem. You have a database where you want to have some shared schema among many groups, and you want a way to infer facts based on first order logic. You want to be able to query multiple sources and reason about facts when taking into account multiple sources.

Whether you use semantic web tech or not that's still a common problem that doesn't always have a good plug and play solution. There's still a lot of places using jsonld format for metadata and cataloging information. You can google cooking recipes and get ratings, cook time; search for movies and see how high rated the movie is and who made it with a synopsis of the plot, all of these are product metadata powered by rdfs or jsonld metadata, a relic of the semantic web. It would be incorrect to say semantic web is dead. Any AI that can effectively use wikidata as a fact table would be jeopardy grade. There's still new tools coming out like RDFox that apply first order logic at multicore speed across huge datasets for reasoning. There is work being done to make it horizontally scalable. I think people will just go on an endless loop of getting the same pain points and creating new tools using the trending tech of the day, but even in this day and age, sometimes something like prolog or picat is what you need.

[zozbot234]

> you want a way to infer facts based on first order logic

Isn't that computationally infeasible? Semantic web standards are based on description logics, i.e. multi-modal logics chosen specifically for computational expediency.

Also, I wouldn't describe JSON-LD as a "relic" of anything. It's a fairly recent standard in the grand scheme of things, and many interesting projects these days implicitly rely on it.

[aggerdom]

So not an expert in this area, would love if someone corrects me. My understanding is generally FOL is infeasible. Propositional logic even can be computationaly difficult [1]. My understanding is that most of the semantic web stuff is done using a description logic of some flavor. These will be named based on the properties of the logic. The important thing is that they are generally decidable, and you can use something like MALET or some other solver to infer things from your database or ontology.(You give up some expressivity for decidability) Not sure how much stuff is going on with that these days. Played with a petrology ontology in protegé some back in college, but haven't followed the space. I remember OWL being important, but can't remember why at the moment.

[1] For example if you try to figure out if a formula is satisfiable. You can for sure do this using truth tables. The catch is that you're looking at 2^n complexity where n is the number of propositions in your formula.