Fwd: [TinkerPop] Algebraic Property Graphs

[Ryan Wisnesky]

Hi All,

The message below may be of interest to this list.  Algebraic property graphs are particular kinds of algebraic databases in the sense of FQL/AQL/CQL etc.

Ryan

Begin forwarded message:

From: Ryan Wisnesky <ry...@conexus.ai>

Subject: Fwd: [TinkerPop] Algebraic Property Graphs

Date: September 11, 2019 at 4:31:44 PM EDT

To: categor...@googlegroups.com

Hi All,

The message below may be of interest to this list.  Algebraic property graphs are particular kinds of algebraic databases in the sense of FQL/AQL/CQL etc.

Ryan

Begin forwarded message:

From: "'Joshua Shinavier' via Gremlin-users" <gremli...@googlegroups.com>

Subject: [TinkerPop] Algebraic Property Graphs

Date: September 11, 2019 at 10:35:37 AM EDT

To: gremli...@googlegroups.com

Reply-To: gremli...@googlegroups.com

Users of Gremlin,

I am happy to give you a first look at a paper that has been several months in the making: Algebraic Property Graphs. This has been a great collaboration between Ryan Wisnesky (of Conexus AI) and me, with early contributions from Jan Hidders (active in the Property Graph Schema Working Group).

The paper addresses the long-standing need for a formally specified property graph data model (of which RDF* is another, contrasting example). It is somewhat like a bigger, better Graph.Features, and is originally based on my work on integrating graph and non-graph schemas at Uber, where we are very heavy on loose systems of algebraic datatypes including Avro, Thrift, and Protocol Buffers. Algebraic property graphs allow the majority of our data to be treated as graph data, without the need for manual mappings.

In my Global Graph Summit talk last January, I motivated formalizing the data model in terms of category theory, but it took Ryan's deep expertise to pull this off. See my earlier messages to gremlin-users (esp. here and here) for context. The result is a framework which is conceptually simple, as property graphs should be, but mathematically rigorous and with straightforward connections to past work on database theory and data integration.

Since Ryan and I started writing the paper, mm-ADT has also burst into glorious existence (see Marko's talk at ApacheCon today), and delves deep into process in addition to structure. We will be working to bring the best of both worlds to TinkerPop 4.

Josh

Note: this paper has just gotten through Uber's lengthy internal review process. A more permanent arXiv preprint of the paper should become available later today; we will tweet the link and/or post it here.

--
You received this message because you are subscribed to the Google Groups "Gremlin-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to gremlin-user...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/gremlin-users/CAPc0Out02AGJtJN10zc5gaMA_P97KC_CxcgC8T8cQKdLMc_37g%40mail.gmail.com.

[Paul-Olivier Dehaye]

It is interesting indeed!

I feel I should say hi.

In a previous career I was a research mathematician reading your work
and David's to better understand how to have interoperability between
math data, knowledge and software. See for instance
https://arxiv.org/abs/1603.06424 which uses completely different
language but has benefited from these readings.

In the next iteration of my career I am now a data protection activist
working with individuals to reappropriate their data. This would
include for instance American citizens who want to get back their
Cambridge Analytica data (as featured in the Netflix doc The Great
Hack), but also Uber drivers who, in a Marxist sense, kinda really
want to reappropriate their tool of production (the data). See for
instance:
https://www.citylab.com/transportation/2019/08/uber-drivers-lawsuit-personal-data-ride-hailing-gig-economy/594232/
As you can guess Uber must not like these efforts so much.
In any case, part of that work is leading me to build property graphs
to better understand the type of data collected by Uber (which quickly
gets into algebraic types aspects, given the complexity). Once they
get the data back, I build data processing tooling that supports their
goals, generated from those property graphs (which requires to
annotate the properties, of course).

I suspect that in the future iteration of my career I will be working
with Uber and others to build better systems, providing more autonomy
to the individuals whose data is processed.

Paul

> You received this message because you are subscribed to the Google Groups "Categorical Data" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to categoricalda...@googlegroups.com.
> To view this discussion on the web visit https://groups.google.com/d/msgid/categoricaldata/F636FC00-D5D1-41F8-890C-D02AFD40D89B%40gmail.com.