Gradual Semantic Alignment for SKOS (GSAS)

GSAS - A numeric and perception-aware approach to SKOS mapping relations

This repository provides a reproducible, ontology-driven framework for modelling graded semantic alignment in SKOS using numeric degrees of connection.

It combines three complementary layers:

1. Minimal SKOS mapping degrees (exactMatch, closeMatch, relatedMatch)
2. 4-level SKOS mapping refinement (high, medium, low, dubious)
3. 7-star SKOS mapping refinement (methodological layer)
4. Empirical perceptions of probability (language-based, empirical layer)

All layers are published as property-only RDF (TBox), are fully mergeable into a single ontology, and are accompanied by CSV tables and plots for transparency and reproducibility.

This repository is intended to be cited together with a working paper that explains the modelling decisions, mathematical background, and empirical basis of the approach.

---

Repository structure

fuzzy-wobbly-semantic-alignment/
│
├─ ontology/
│  ├─ create_ontology.py
│  └─ merged_ontology.ttl
│
├─ skos/
│  ├─ skos.py
│  ├─ skos_minimal_degrees.ttl
│  ├─ skos_minimal_degrees.csv
│  └─ skos_minimal_degree_plot.jpg
│
├─ skos_4level/
│  ├─ skos_4level.py
│  ├─ skos_4level_mapping.ttl
│  ├─ skos_4level_degrees.csv
│  └─ skos_4level_degree_plot.jpg
│
├─ skos_7star/
│  ├─ skos_7star.py
│  ├─ probly.csv
│  ├─ skos_7star_mapping.ttl
│  ├─ skos_7star_degrees.csv
│  └─ skos_7star_degree_plot.jpg
│
├─ skos_perceptions/
│  ├─ skos_perceptions.py
│  ├─ probly.csv
│  ├─ skos_perceptions_mapping.ttl
│  ├─ skos_perceptions_stats.csv
│  └─ skos_perceptions_degree_plot.jpg
│
├─ .gitignore
└─ README.md

---

Design principles

Across all scripts and outputs, the following principles apply:

• SKOS-first modelling
  skos:relatedMatch, skos:closeMatch, and skos:exactMatch form the stable semantic backbone.

• Property-only RDF (TBox)
  No individuals are created. All modelling is done using owl:ObjectProperty and owl:AnnotationProperty.

• Single custom namespace
  All extensions use: https://w3id.org/skos-plus/ ensuring stable, consistent URIs across all layers.

• Numeric justification
  Each mapping relation is annotated with a numeric skosplus:degreeOfConnection in the range [0,1].

• Reproducibility & transparency
  Every RDF file is accompanied by:

• a CSV table with numeric values

• a plot visualising the mapping logic

---

skos/skos.py — Minimal SKOS degree model

Purpose

This script defines a baseline numeric interpretation of SKOS mapping relations without any 7-star or perception-based logic.

It answers the question:

How can skos:exactMatch, skos:closeMatch, and skos:relatedMatch be grounded in numeric degrees of connection?

What the script does

• Assigns a numeric degree of connection to:
• skos:exactMatch
• skos:closeMatch
• skos:relatedMatch
• Uses a simple, monotonic mapping in the range [0,1]
• Produces RDF, CSV, and a plot

Outputs

• skos_minimal_degrees.ttl
• skos_minimal_degrees.csv
• skos_minimal_degree_plot.jpg

This layer acts as the semantic anchor for all further extensions.

---

skos_4level/skos_4level.py — 4-level SKOS mapping refinement

Purpose

This script provides a coarse-grained refinement layer for SKOS mappings using four intuitive levels:

• high
• medium
• low
• dubious

It answers the question:

How can SKOS mappings be expressed with a small, human-friendly set of confidence levels while remaining compatible with the 7-star model?

What the script does

• Introduces four sub-properties of skos:relatedMatch
• Derives numeric degrees from the same 7-star exponential function
• Aggregates star levels into bins:
  • dubious → stars 1–2
  • low → stars 3–4
  • medium → star 5
  • high → stars 6–7
• Annotates each property with:
  • skosplus:degreeOfConnection
  • an internal explanatory star bin (annotation only)
• Produces RDF, CSV, and a plot

Modelling role

This layer acts as a bridge between:

• the minimal SKOS baseline (3 relations)
• the fine-grained 7-star refinement

It is intended for use cases where interpretability and simplicity are preferred over maximal granularity.

Outputs

• skos_4level_mapping.ttl
• skos_4level_degrees.csv
• skos_4level_degree_plot.jpg

---

skos_7star/skos_7star.py — 7-star SKOS mapping refinement

Purpose

This script introduces a 7-star refinement of SKOS mappings that allows skos:relatedMatch to be expressed at multiple granular levels.

It answers the question:

How can SKOS mappings be refined without breaking SKOS semantics?

What the script does

• Introduces five sub-properties of skos:relatedMatch:
  • very weak
  • weak
  • moderate
  • strong
  • very strong
• Maps:
  • star level 6 → skos:closeMatch
  • star level 7 → skos:exactMatch
• Computes numeric degrees using a normalised, saturating exponential function
• Annotates each relation with:
  • star level
  • degree of connection
  • an internal explanatory perception band

Important modelling decision

The perception band is modelled as an annotation property, not as part of the object-property hierarchy.
This preserves a clean SKOS-first structure while keeping the internal logic documented.

Outputs

• skos_7star_mapping.ttl
• skos_7star_degrees.csv
• skos_7star_degree_plot.jpg

---

skos_perceptions/skos_perceptions.py — Empirical perceptions mapping

Purpose

This script models natural language probability phrases (e.g. Highly Likely, Almost Certainly) as SKOS-compatible mapping relations based on empirical data.

It answers the question:

How do people interpret probability phrases, and how can this be integrated into SKOS mappings?

Data source

• probly.csv
  Derived from the study published at: https://github.com/zonination/perceptions

What the script does

• Treats each perception phrase as a mapping property
• Computes:
  • median
  • quartiles
  • interquartile range (IQR)
  • Tukey whiskers
• Uses the median (normalised to 0–1) as the degree of connection
• Assigns each phrase as a sub-property of:
  • skos:exactMatch
  • skos:closeMatch
  • skos:relatedMatch based on numeric thresholds
• Produces a plot showing:
  • the 17 empirical medians
  • a smooth approximation curve for generalisation

Outputs

• skos_perceptions_mapping.ttl
• skos_perceptions_stats.csv
• skos_perceptions_degree_plot.jpg

---

ontology/create_ontology.py — Merging all layers

Purpose

This script merges all RDF layers into a single coherent ontology.

It answers the question:

Can minimal SKOS mappings, 7-star refinement, and empirical perceptions coexist without semantic inconsistency?

What the script does

• Loads:
• skos_minimal_degrees.ttl
• skos_4level_mapping.ttl
• skos_7star_mapping.ttl
• skos_perceptions_mapping.ttl
• Merges them into one RDF graph
• Relies on shared URIs, namespaces, and the SKOS backbone

Output

• merged_ontology.ttl

This file can be:

• opened directly in Protégé
• published as an ontology artefact
• referenced from academic publications

---

License

MIT License