Knowledge graph

Chive uses a community-curated knowledge graph to classify and connect scholarly works. This Wikipedia-style approach allows the research community to build and maintain a structured taxonomy of academic fields.

Overview

The knowledge graph serves three purposes:

Discovery: Find related preprints through field classifications
Context: Understand how a work fits into broader research areas
Navigation: Browse preprints by field, subfield, or topic

                    ┌──────────────────┐
                    │   Mathematics    │
                    └────────┬─────────┘
           ┌─────────────────┼─────────────────┐
           ▼                 ▼                 ▼
    ┌──────────────┐  ┌──────────────┐  ┌──────────────┐
    │   Algebra    │  │   Analysis   │  │   Geometry   │
    └──────┬───────┘  └──────────────┘  └──────────────┘
           │
    ┌──────┴──────┬──────────────┐
    ▼             ▼              ▼
┌────────┐  ┌──────────┐  ┌───────────┐
│ Group  │  │  Ring    │  │  Linear   │
│ Theory │  │  Theory  │  │  Algebra  │
└────────┘  └──────────┘  └───────────┘

Field nodes

A field node represents an academic discipline, subdiscipline, or topic. Each field has:

Property	Description
`name`	Human-readable name (e.g., "Algebraic Geometry")
`description`	Brief explanation of the field's scope
`aliases`	Alternative names (e.g., "Algebraic Geometry" = "AG")
`parentFields`	Broader categories this field belongs to
`childFields`	Narrower specializations within this field
`relatedFields`	Fields with conceptual overlap
`externalIds`	Links to Wikidata, Library of Congress, etc.

Field relationships

Fields connect through three relationship types:

Broader/Narrower (hierarchical)
    Mathematics
        └── Algebra
              └── Group Theory

Related (conceptual overlap)
    Algebraic Topology ──related──► Algebraic Geometry

Cross-disciplinary
    Computational Linguistics
        ├── parent: Linguistics
        └── parent: Computer Science

PMEST classification

Beyond hierarchical fields, Chive uses PMEST (Personality, Matter, Energy, Space, Time) faceted classification. This system allows filtering across orthogonal dimensions:

Facet	Meaning	Example
Personality	Core subject	"Quantum mechanics"
Matter	Material or substance	"Carbon nanotubes"
Energy	Process or action	"Oxidation"
Space	Geographic scope	"Arctic regions"
Time	Temporal scope	"Holocene"

A single preprint can be classified across multiple facets:

Preprint: "Climate-Driven Carbon Nanotube Degradation in Arctic Soils"

Facets:
  Personality: Environmental Chemistry, Materials Science
  Matter: Carbon nanotubes, Soil
  Energy: Degradation, Climate change
  Space: Arctic
  Time: Contemporary (2020-present)

Users can combine facets to narrow searches:

GET /xrpc/pub.chive.graph.browseFaceted?
  personality=materials-science&
  matter=carbon-nanotubes&
  space=arctic

Authority records

Authority records ensure consistency across the knowledge graph. They're like library catalog entries for concepts:

// Example authority record for "Quantum Computing"
{
  "name": "Quantum Computing",
  "aliases": [
    "Quantum Computation",
    "QC"
  ],
  "description": "Computational paradigm using quantum-mechanical phenomena",
  "externalLinks": {
    "wikidata": "Q339",
    "lcsh": "sh2008010405",
    "viaf": "168470861"
  },
  "broaderTerms": ["Computer Science", "Quantum Mechanics"],
  "narrowerTerms": ["Quantum Error Correction", "Quantum Algorithms"],
  "relatedTerms": ["Quantum Information Theory"]
}

Authority records link to external controlled vocabularies:

Vocabulary	Purpose
Wikidata	Multilingual structured knowledge
LCSH	Library of Congress Subject Headings
VIAF	Virtual International Authority File
FAST	Faceted Application of Subject Terminology

Reconciliation

When users tag preprints, Chive reconciles tags against authority records:

User enters: "quantum computing"
         ↓
Chive matches: Authority record "Quantum Computing" (Q339)
         ↓
Preprint linked to canonical concept

This prevents fragmentation ("quantum computing" vs "Quantum Computation" vs "QC" all map to the same concept).

Community governance

The knowledge graph uses Wikipedia-style moderation. Users can:

Propose new fields or changes
Discuss proposals in threaded comments
Vote on whether to accept proposals

Proposal types

Type	What it does	Approval threshold
Create field	Add a new field to the taxonomy	67% with 5+ votes
Update field	Modify name, description, or relationships	60% with 3+ votes
Merge fields	Combine redundant fields	67% with 5+ votes
Deprecate field	Mark a field as obsolete	75% with 7+ votes
Authority change	Update authority records	75% with 7+ votes

Voter tiers

Not all votes carry equal weight. Expertise in the relevant field increases vote weight:

Tier	Vote weight	Criteria
Community member	1.0x	Any authenticated user
Active contributor	1.5x	10+ preprints or reviews
Domain expert	2.5x	Publications in the field
Trusted editor	3.5x	Appointed by governance
Authority editor	4.5x	Library science expertise

Proposal workflow

┌──────────┐     ┌─────────────┐     ┌──────────┐     ┌───────────┐
│  Draft   │────►│  Discussion │────►│  Voting  │────►│  Outcome  │
│          │     │  (7 days)   │     │ (5 days) │     │           │
└──────────┘     └─────────────┘     └──────────┘     └───────────┘
                       │                   │
                       │                   │
                       ▼                   ▼
              Revisions allowed     Threshold met?
                                    ├── Yes → Approved
                                    └── No  → Rejected

User tags vs. authority terms

Chive distinguishes between user-generated tags and authority-controlled terms:

User tags	Authority terms
Free-form text	Controlled vocabulary
Personal organization	Community consensus
No voting required	Proposal + voting
May be reconciled	Canonical concepts

Users can tag preprints freely. Popular tags may be promoted to authority terms through a two-stage process:

Automatic nomination: Tag used on 10+ preprints by 3+ users
Community vote: Standard proposal process

Graph algorithms

The knowledge graph enables advanced discovery features:

Citation analysis

Find papers that:
  - Cite foundational works in the field
  - Bridge multiple subfields
  - Introduce new connections

Semantic similarity

Given a preprint about "quantum error correction":
  - Find semantically similar preprints
  - Suggest related fields to explore
  - Identify key authors in adjacent areas

Field evolution

Track how fields change over time:
  - New subfields emerging
  - Fields merging or splitting
  - Terminology shifts

Integration with search

The knowledge graph enhances search in several ways:

Expansion

A search for "machine learning" automatically includes:

Narrow terms: "deep learning", "neural networks"
Related terms: "artificial intelligence", "statistical learning"

Disambiguation

A search for "network" prompts:

Computer networks?
Neural networks?
Social networks?
Network science?

Faceted browsing

Filter results by any PMEST dimension while staying within a field:

Field: Machine Learning
  Filter by Matter: Medical imaging
  Filter by Time: Last 5 years
  Filter by Space: [any]

Wikidata integration

Chive synchronizes with Wikidata to:

Import established classifications
Link local concepts to global identifiers
Contribute new academic concepts back

# Example SPARQL query to find related concepts
SELECT ?item ?itemLabel WHERE {
  wd:Q339 wdt:P279* ?item .  # Q339 = Quantum computing
  ?item wdt:P31 wd:Q11862829 .  # Instance of academic discipline
  SERVICE wikibase:label { bd:serviceParam wikibase:language "en" }
}

API endpoints

Endpoint	Purpose
`pub.chive.graph.getField`	Get field details
`pub.chive.graph.listFields`	List fields (paginated)
`pub.chive.graph.searchAuthorities`	Search authority records
`pub.chive.graph.getAuthority`	Get authority record details
`pub.chive.graph.browseFaceted`	Faceted search
`pub.chive.graph.getFieldPreprints`	Preprints in a field

Next steps

Data sovereignty: How your data stays yours
Peer review: Review and endorse preprints
Governance: How decisions are made

Overview​

Field nodes​

Field relationships​

PMEST classification​

Facet search​

Authority records​

Reconciliation​

Community governance​

Proposal types​

Voter tiers​

Proposal workflow​

User tags vs. authority terms​

Graph algorithms​

Citation analysis​

Semantic similarity​

Field evolution​

Integration with search​

Expansion​

Disambiguation​

Faceted browsing​

Wikidata integration​

API endpoints​

Next steps​