# Knowledge Graphs

> Knowledge captured as connected notes, entities, and claims.

- HTML version: https://robbiepalmer.me/ideas/knowledge-graphs
- Source: https://web.stanford.edu/class/cs520/2020/notes/What_is_a_Knowledge_Graph.html

A knowledge graph stores knowledge as addressable items connected by relations. An item might be a
note, page, entity, or claim. A relation might be an ordinary hyperlink or a named predicate such
as *cites* or *supersedes*.

Linked notes and machine-readable claim stores both use the term. More explicit relations enable
more precise queries.

## What makes it a knowledge graph

The label describes the graph's purpose. Its items record claims, sources, interpretations, and
decisions. A graph database can store transport networks or software dependencies too. Nodes and
edges alone do not make the contents knowledge.

For personal use, the job is capture. A note keeps its source and its connections to later notes or
decisions. Backlinks recover those connections without forcing the note into one folder or a
finished taxonomy.

## Three common forms

| Form                       | Relation                                         | What it supports                              |
| -------------------------- | ------------------------------------------------ | --------------------------------------------- |
| Linked notes or wiki pages | Hyperlink, with meaning supplied by prose        | Navigation, backlinks, and associative recall |
| Typed content graph        | Named links between kinds of content             | Relation-specific queries and checks          |
| Semantic knowledge graph   | Predicates between identified entities or claims | Precise queries, validation, and inference    |

[Roam Research](https://roamresearch.com/) and
[Obsidian](https://obsidian.md/help/links) fit the first form. Their links and backlinks build a
graph from notes without requiring a formal schema.

Wikipedia is also a linked body of knowledge written for people. Its structured companion,
[Wikidata](https://www.wikidata.org/wiki/Wikidata:Introduction), represents knowledge as identified
items and property-value statements. Wikidata can attach qualifiers and references to a statement,
which makes the statement usable and auditable by software.

The [Stanford CS520 introduction](https://web.stanford.edu/class/cs520/2020/notes/What_is_a_Knowledge_Graph.html)
concentrates on this formal end. It describes a directed, labelled graph with well-defined
meanings for its labels. A personal knowledge graph can still be useful without RDF, global
identifiers, or inference.

## Add structure when it answers a question

Untyped links are cheap to capture and leave interpretation to the reader. Typed links cost more:
the author must choose and maintain a vocabulary. That cost pays off when software needs the
distinction. A *cites* relation can find unsupported notes. *Supersedes* can identify the current
decision. *Contradicts* can collect unresolved disagreement.

[Ontology engineering](/ideas/ontology-engineering) defines the concepts and relations used by a
formal graph. [Data provenance](/ideas/data-provenance) records the source and history of a claim.
[Named entity recognition](/ideas/ner) can propose entities found in prose. Formal graphs often use
these techniques. Linked notes can omit them.

Documents, relational tables, RDF stores, and graph databases can all hold a knowledge graph. The
storage choice affects querying and performance. It does not decide whether the content represents
knowledge.

## Questions it prompts

* What should remain connected after the immediate task ends?
* Which repeated queries or checks justify more structure?
* How will the graph represent sources, uncertainty, and disagreement?

## Sources

* Obsidian. ["Internal links."](https://obsidian.md/help/links)
* Roam Research. ["A note taking tool for networked thought."](https://roamresearch.com/)
* Wikidata. ["Introduction."](https://www.wikidata.org/wiki/Wikidata:Introduction)
* Stanford University CS520. ["What is a Knowledge Graph?"](https://web.stanford.edu/class/cs520/2020/notes/What_is_a_Knowledge_Graph.html).
  2020\.
* Hogan, Aidan, et al. ["Knowledge Graphs"](https://doi.org/10.1145/3447772).
  *ACM Computing Surveys* 54, no. 4, 2021.

## Related ideas

- [Data Provenance](https://robbiepalmer.me/ideas/data-provenance.md): A record of where data came from, who or what changed it, and how a result was produced.
- [Digital Twins](https://robbiepalmer.me/ideas/digital-twins.md): Digital models kept connected to real counterparts so observations, simulations, and decisions can inform one another.
- [Named-Entity Recognition](https://robbiepalmer.me/ideas/ner.md): NER finds spans of text that name entities and assigns them types such as people, organisations, products, or identifiers.
- [Ontology Engineering](https://robbiepalmer.me/ideas/ontology-engineering.md): The practice of defining the concepts, relationships, constraints, and shared vocabulary needed to represent a domain.

## Where it appears

- Project: [Commercial Knowledge Graph](https://robbiepalmer.me/projects/commercial-knowledge-graph.md)
- ADR: [ADR 033: In-Memory Content Graph](https://robbiepalmer.me/projects/personal-knowledge-graph/adrs/033-content-graph-indexes.md)

---

Markdown index of this site: https://robbiepalmer.me/llms.txt