Personal Knowledge Graphs

Why, what, and where to?

Ivo Velitchkov | July 2nd 2021

• Open Knowledge Graphs

• Enterprise Knowledge Graphs

• Personal Knowledge Graphs

See Linked Data uptake.

Everybody is already using PKG

Nobody is using PKG yet

Personal information is trapped in silos

• “Re-finding”
• 70% save frequently accessed website as bookmarks
• Bookmarks are used for retrieving only
  • 16% of bookmarked sites
    • Possible reason: ^^fail to preserve the context^^ of the original retrieval

    ---

    • Source: Out of sight and out of mind: Bookmarks are created but not used (2021)
    • Olaf Bergman, Steve Whittaker, Joel Schooler
• Also: what is bookmarked is a whole page but we are interested only in certain parts
  • Enter ^^web annotation^^ tools: Diigo, Hypothes.is
• Bookmarking items on social media
  • often with the meaning of “read later”

• Usually don’t have persistent IDs
• Their parts* never do

= not reusable not linkable

• Application and data are coupled

* paragraphs, images, bullets

Everybody is already using PKG

2019-2020: Eruption of “tools for thought”
(networked note-taking)

• Some history
• Interacting with a PKG
• My experience and hacks
• Architecture and technology patterns
• Some other observations
• The limitations of PKG apps

Luhmann’s Zettelkasten

Niklas Luhmann (1927–1998)

• 70 books and 400 articles
• Zettelkasten
  • • 90 000 linked cards
    • 30 years

Every note is only an element that receives its quality only from the network of links and back-links within the system.

As a result of extensive work with this technique, a kind of secondary memory will arise, an alter ego with who we can constantly communicate. It proves to be similar to our own memory in that it does not have a thoroughly constructed order of its entirety, not hierarchy, and most certainly no linear structure like a book. Just because of this, it gets its own life, independent of its author.

My PKG, 14 months old

• 3117 nodes (pages)

But my actual graph is 20 times bigger:

• Daily tasks
• Events (integration with calendar)
  • preparation
  • notes
  • tasks
  • follow-up
• Research
  • Projects
  • Collecting knowledge
    • Bookmarks with highlights from
      • Kindle books
      • Web pages, articles
      • Embebed PDFs
  • Linking ideas
• Writing (drafting and directly publishing where possible)
  • Journaling
  • Articles
  • Books
  • Emails
  • Posts
    • WordPress
    • Twitter
    • Slack
    • LinkedIn
  • Reviews
  • Newsletters
  • Other documents
• Projects
  • Events (integration with calendar)
  • Presentations
  • Planning
  • Issues
  • Risks
  • Test
  • Deliveries
  • Documentation
• Presentations (both making and presenting)
• Relationship management
• Learning
• Home
• Tracking service subscriptions
• Training
• Code

• Reduced or completely stopped using
  • MS Word and Google Doc (now only for reviews)
  • PowerPoint and Google Slides
  • Evernote
  • Trello
  • Plectica
  • Simple Mind
  • Pastebin
  • Google Tasks
  • Zenpen
  • Diigo
  • …

• Most current apps stop at references

• But what kind of reference?

• And what’s the direction?

All graphy tools4thought hide the most valuable feature of directed graphs: the edges’ direction.
They show P1–P2 both when P1 is referred to in P2 and when P2 is referred to in P1.
Should be:
{a}–mention–>P1–mention–>{b}@roamhacker @houshuang @logseq @AthensResearch

— Ivo Velitchkov (@kvistgaard) April 16, 2021

• 
• ROCO = ROam Core Ontology
  • Classes such as Class and Property
  • Properties such as subClassOf and same as
• RIO = Roam Internal Ontology
  • Currently:
  • 111 classes
  • 28 properties
• Demo: https://youtu.be/F1HYaGfqzLM

• General Knowledge
  • Wikidata: 12.5 billion statements
  • DBpedia: 860 million
  • YAGO: 0.478 billion
  • Google Knowledge Graph: 86 billion
• Scientific publications
  • Microsoft Academic Knowledge Graph: 10 billion
  • Open Citations: 8 billion
• Domain knowlegde
  • Life sciences
    • Uniprot: 81 billion
    • IDSM: 14 billion
  • Government
    • EU publications: 0.893 billion
    • UK Parliament
    • Italian Parliament
    • Swiss Federal GeoPortal

• Processing at client side
• Markdown (sign of the third wave)
• Clojure

• Technology stack
  • Backend – Postgres, Clojure
  • API – GraphQL
  • Frontend – React, Redux

• Very high importance of
  • extensions (plug-ins)
  • community
• Growing importance of having full control over personal data
  • Best achieved so far by MD-based apps

• Scope
  • Most of your personal data is always “outside”, not connected
    • Example: When I search for “PKG”
      • I should get results from
        • Text documents (Word, Google doc)
        • PDFs
        • eBooks
        • Emails
          • Containing in the subject
          • Containing PKG in the body
        • LinkedIn
          • My posts containg PKG
          • My bookmars containg PKG
          • The posts containing PKG that I reacted to
        • Twitter
          • My tweets with PKG
          • My RTs with PKG
          • My likes with PKG
          • Tweets with PKG I commented to
        • Discord
          • Messages with PKG
        • Slack
          • Messages with PKG
        • Telegram
          • Messages with PKG
        • Notebooks
        • GitHub
• Own your own data
• Interoperability
  • My workflow on X needs a feature/plug-in only available for Y

• Just like EKG,

• PKG should be ^^data-centric^^

• Principles

  • Data is self-describing and does not rely on an application for interpretation and meaning.
  • Data is expressed in open, non-proprietary formats.
  • Applications are allowed to visit the data, perform their magic and express the results of their process back into the data layer
  • Access to and security of the data is a responsibility of the data layer, and not managed by applications.

Personal Knowledge Graphs (PKG)

Nobody is using PKG yet

• Solid
• Decentralized technologies
  • IPFS
  • dWeb