Re: What is a Knowledge Graph? CORRECTION

On 17.06.2019 11:39, Patrick J Hayes wrote:
> Hi Dieter
>
> Good to hear from you.
Yes, nice to see that both of us are still alive!
>> On Jun 16, 2019, at 1:16 AM, Dieter Fensel <dieter.fensel@sti2.at> wrote:
>>
>> Hi Pat,
>>
>> thanks for these extremely interesting post. I would like to jump onto your summary:
>>
>>>>> 5. The semantic nets of the 1970s were, almost univerally, /much/ more expressive than knowledge graphs or RDF, or any of the other ‘graph’-like modern notations. They typically had ways of encoding quantifier scopes, disjunction, negation and sometimes such things as modal operators. The grandfather of them all, C.S.Peirce’s ‘existential graphs’  had the full expressivity of first-order logic in 1885 (implemented as ‘conceptual graphs’ by John Sowa about 90 years later http://www.jfsowa.com/cg/cgonto.htm). It has been downhill from there.
>>>>>
>> Indeed, old work on KR was mostly around very expressive Tboxes (and Aboxes were just an uninteresting appendix where you put 2-3 facts to illustrate your approach). Given the expressivism of that Tboxes it was mostly work for the sand box of academic seminars. Turning this work from the" Kopf auf die Füsse" is actually the real contribution of the recent work on KGs. They come with bizarre large grown Aboxes and with extremely simple Tboxes.
> Well, some of the OWL ontologies are moderately complicated… but perhaps you would place all that in the ivory tower?
+1
> If so, how about the fairly complicated Abox reasoning (mostly, I believe, done in Prolog) which Fair Isaac uses to rapidly guess at potentially fraudulent uses of credit cards? That seems to be both real world and non-trivial (though I confess I do not know it in detail, as it is highly proprietary, of course.) I am sure there are other examples.
Yes, here and there an example but not a mainstream. It is still niche 
application stuff.
>> There may be two major reasons for this:
>>
>> 1) World view: It is very hard to think on useful rules that hold on a global scale. I saw trials in recent papers such as "parents and child hold same nationality" (does already no hold in my simple case), "wife and husband stay in the same city" (...). At University I learnt the rule married(a,b) and male(a) implies female(b), and even married(a,b) and married(a,c) implies equal(b,c) does not hold on an universal scale. Maybe in future married(a,b) implies human(b) may even be invalid. Who knows?
> I agree, simple theories of the complicated real world break very quickly. But there are still some useful generalizations to be had. For example, almost all nations have an administrative capital, about which a lot can be said. And you know there are ivory-tower (or at least AI-lab) devices for handling generalities with exceptions, that have been around for decades. And highly complicated ontologies, using full FO logic and even more, are in serious use in (for example) medical informatics and process control industries, in some cases now handled by ISO committees.
>
>> 2) System view: A system with million of rules and trillions of facts will never scale in the foreseeable future of the 21st century. Not even during the building and knowledge curation phase.
> Maybe. I have been impressed by how effective sometimes quite simple reasoners can be even on very large corpora. I think the sheer size is less important than the connection density, in fact.
+1
>> Therefore, we watch a new systems architecture evolving where extremely big (semantic) data lakes are accessed and consolidated by means that define a domain and task specific view on them. Kind of heterogeneous and distributed micro Tboxes (to make a reference to the idea of micro theories in CYC (*). This new view is arising over the last ten years based on excellent work of LOD academics and people in industry on in-house KGs and it is just in its beginning of releasing its full potential.
> Well, I will agree with you here. The size of the Kbases does change the rules of the game, indeed. It is both a problem and an opportunity.
>
> But I will claim that the basic issues of expressivity of the notation, and consistency in the use of terminology (enforcing which is one major use of a powerful Abox) are still relevant in this brave new world, perhaps even more than before.
+2
>> You know it always takes a bit downhill when you leave the Ivory tower.
> :-) Yes, indeed. I have been outside the tower myself for about two decades now, and no doubt my snarkiness in the previous message was fuelled partly by nostalgia :-)
>
> Pat
I know. And you knew I know :-)
>> Btw, I think you need at least multi-sets to even cover simple graphs.
>>
>> (*) with the difference that the KG is not structured by them.
> It could be, though, right? There is an opportunity here, IMO.

I asked my final PhD student to work on it. Lets see how far he will go.

One disagreement we still may have: Obviously you can reduce any maths 
(at least the little piece of it  I know) to set theory. But many people 
in the database community still argue that the relational model that is 
a nearly direct application of set theory  is outdated and should be 
replaced by graphs. See f.e.:R. Angeles and C. Gutierrez: Survey of 
Graph Database Models, /ACM Computing Surveys/, 40(1), 2008. When you 
represent a graph as a set you also obviously destroy a lot of potential 
valid information. Redundancy, locality, etc. Or you make the set 
representation very complex and it starts to become encoding.