Project Overview

The IPCC Figure-Term Knowledge Graph makes figures from the IPCC Sixth Assessment Report AR6 easier to find and explore by linking them to scientific terms mentioned in their captions

The project collects figures and captions from all three IPCC Working Groups, identifies terms from third party export of IPCC AR6 Glossary in their captions, and builds a bipartite knowledge graph linking figures to terms

It provides

• Reusable figure corpus
• Scientific terms vocabulary
• Interactive graph visualization

to help researchers discover and explore related figures. All connections are based on caption text not image analysis

Data Sources

Type	Data	Source	Description
External source	IPCC AR6 figures	WG1, WG2, WG3	Official IPCC figure pages used to collect captions and figure details
External source	IPCC AR6 glossary	TIB Hannover glossary export	Third party HTML export of the official IPCC glossary used to build the vocabulary
Input	Figure captions	wg1.json, wg2.json, wg3.json	Collected figure metadata, captions, image URLs and source links
Input	Glossary vocabulary	ipcc_ar6_vocabulary.json	Extracted scientific terms and aliases used for caption matching
Output	Figure corpus	corpus.db	Cleaned and structured figure records in SQLite
Output	Term matches	concepts.db	Identified caption mentions and accepted figure–term links in SQLite
Output	Knowledge graph	ipcc_visual_knowledge_graph.ttl	Bipartite figure–term knowledge graph in RDF / Turtle format

Build Process

1. Figure Collection

Extract figure metadata, captions and image URLs from IPCC AR6 WG1, WG2 and WG3 pages using Requests and Beautiful Soup, and organize the collected data into separate json files

2. Corpus Preparation

Clean, deduplicate and validate collected figure records using Pydantic then organize figures and captions in a structured SQLite database

3. Term Matching

Identify IPCC glossary terms and aliases in figure captions using normalized, trie based lexical matching with longest match selection and store the matches in SQLite database

4. Knowledge Graph Construction

Use RDFLib to build a bipartite knowledge graph linking IPCC figures to glossary terms found in their captions and save it in RDF/Turtle (.ttl) format

5. Search and Visualization

Search IPCC figures by glossary terms or caption text and explore their connections through interactive HTML knowledge graph

Knowledge Graph

Nodes: Figure and ScientificConcept (glossary terms) Links: mentionsConcept connects figures to terms found in their captions

• Interactive Interface

• Figure Node selection

• Scientific term node selection

• Caption box on Figure node hover

Installation and Usage

Run these three cells in Google Colab to install the project, load the knowledge graph and explore figure–term connections

1. Install

!git clone https://github.com/semanticClimate/semanticlens.git
%cd semanticlens
!pip install -e .

2. Load the knowledge graph

from ipcc_kg import IPCCGraph

kg = IPCCGraph.from_repo()

3. Search and explore

query = "heatwave"

graph = kg.explore(query, limit=10)

graph.show()

4. Export graph

html_path = graph.save("graph.html")
rdf_path = graph.save("graph.ttl")

print(f"HTML visualization: {html_path}")
print(f"RDF Turtle: {rdf_path}")

Results

Result	Count
IPCC AR6 figures collected (WG1, WG2, WG3)	1310
Figure captions collected	1,297
Glossary terms in the vocabulary	920
Unique figure–term links identified	6,178
Figures linked to at least one term	1,215