e5a532a8bc
- Introduced `test_nlp_extractor.py` with unit tests for the InstitutionExtractor, covering various extraction patterns (ISIL, Wikidata, VIAF, city names) and ensuring proper classification of institutions (museum, library, archive). - Added tests for extracted entities and result handling to validate the extraction process. - Created `test_partnership_rdf_integration.py` to validate the end-to-end process of extracting partnerships from a conversation and exporting them to RDF format. - Implemented tests for temporal properties in partnerships and ensured compliance with W3C Organization Ontology patterns. - Verified that extracted partnerships are correctly linked with PROV-O provenance metadata.
472 lines
17 KiB
Python
472 lines
17 KiB
Python
#!/usr/bin/env python3
|
|
"""
|
|
Enrich low-coverage countries using fuzzy name matching with Wikidata.
|
|
|
|
Target countries with <30% Wikidata coverage:
|
|
- Brazil (BR): 1.0%
|
|
- Belgium (BE): 0.0%
|
|
- Italy (IT): 0.0%
|
|
- Thailand (TH): <30%
|
|
- Norway (NO): <30%
|
|
- Vietnam (VN): <30%
|
|
|
|
Strategy:
|
|
1. Query Wikidata for museums/libraries/archives in target country
|
|
2. Fuzzy match institution names (normalized, threshold 0.85)
|
|
3. Verify type compatibility (don't match museum → archive)
|
|
4. Enrich with Wikidata IDs, VIAF, founding dates, websites
|
|
"""
|
|
|
|
import sys
|
|
from pathlib import Path
|
|
from typing import Any, Optional
|
|
from datetime import datetime, timezone
|
|
import time
|
|
import yaml
|
|
from difflib import SequenceMatcher
|
|
import re
|
|
|
|
sys.path.insert(0, str(Path(__file__).parent.parent / "src"))
|
|
|
|
from SPARQLWrapper import SPARQLWrapper, JSON as SPARQL_JSON # type: ignore
|
|
|
|
|
|
def normalize_name(name: str) -> str:
|
|
"""Normalize institution name for fuzzy matching."""
|
|
# Lowercase
|
|
name = name.lower()
|
|
|
|
# Remove common prefixes/suffixes in multiple languages
|
|
name = re.sub(r'^(stichting|gemeentearchief|regionaal archief|museum|museu|museo|biblioteca|bibliotheek|library|archive|archief|archivo)[\s\-]+', '', name)
|
|
name = re.sub(r'[\s\-]+(archief|museum|museo|museu|bibliotheek|biblioteca|library|archive|archivo)$', '', name)
|
|
|
|
# Remove organizational forms
|
|
name = re.sub(r'\b(s\.a\.|sa|nv|bv|vzw|asbl|inc|ltd|gmbh)\b', '', name)
|
|
|
|
# Remove punctuation
|
|
name = re.sub(r'[^\w\s]', ' ', name)
|
|
|
|
# Normalize whitespace
|
|
name = ' '.join(name.split())
|
|
|
|
return name
|
|
|
|
|
|
def similarity_score(name1: str, name2: str) -> float:
|
|
"""Calculate similarity between two names (0-1)."""
|
|
norm1 = normalize_name(name1)
|
|
norm2 = normalize_name(name2)
|
|
return SequenceMatcher(None, norm1, norm2).ratio()
|
|
|
|
|
|
def institution_type_compatible(inst_type: str, inst_name: str, wd_name: str, wd_desc: str) -> bool:
|
|
"""
|
|
Check if institution types are compatible.
|
|
|
|
Prevents mismatches like museum → archive or library → museum.
|
|
"""
|
|
# Define type keywords by language
|
|
museum_kw = ['museum', 'museo', 'museu', 'muzeum']
|
|
archive_kw = ['archief', 'archive', 'archivo', 'archivio']
|
|
library_kw = ['bibliotheek', 'biblioteca', 'library', 'bibliothèque', 'bibliothek']
|
|
|
|
inst_lower = (inst_name + ' ' + inst_type).lower()
|
|
wd_lower = (wd_name + ' ' + wd_desc).lower()
|
|
|
|
# Check institutional type
|
|
inst_is_museum = any(kw in inst_lower for kw in museum_kw)
|
|
inst_is_archive = any(kw in inst_lower for kw in archive_kw)
|
|
inst_is_library = any(kw in inst_lower for kw in library_kw)
|
|
|
|
wd_is_museum = any(kw in wd_lower for kw in museum_kw)
|
|
wd_is_archive = any(kw in wd_lower for kw in archive_kw)
|
|
wd_is_library = any(kw in wd_lower for kw in library_kw)
|
|
|
|
# If both have explicit types, they must match
|
|
if inst_is_museum and not wd_is_museum:
|
|
return False
|
|
if inst_is_archive and not wd_is_archive:
|
|
return False
|
|
if inst_is_library and not wd_is_library:
|
|
return False
|
|
|
|
return True
|
|
|
|
|
|
def query_country_institutions(sparql: SPARQLWrapper, country_code: str) -> dict[str, dict[str, Any]]:
|
|
"""
|
|
Query Wikidata for GLAM institutions in a specific country.
|
|
|
|
Returns: dict keyed by QID
|
|
"""
|
|
# Map ISO 3166-1 alpha-2 to Wikidata QIDs
|
|
country_qids = {
|
|
"BR": "Q155", # Brazil
|
|
"BE": "Q31", # Belgium
|
|
"IT": "Q38", # Italy
|
|
"NO": "Q20", # Norway
|
|
"TH": "Q869", # Thailand
|
|
"VN": "Q881", # Vietnam
|
|
"MX": "Q96", # Mexico
|
|
"CL": "Q298", # Chile
|
|
}
|
|
|
|
qid = country_qids.get(country_code)
|
|
if not qid:
|
|
print(f" ⚠️ No Wikidata QID mapping for country code: {country_code}")
|
|
return {}
|
|
|
|
query = f"""
|
|
SELECT DISTINCT ?item ?itemLabel ?itemDescription ?typeLabel ?isil ?viaf ?coords ?website ?inception
|
|
WHERE {{
|
|
# Institution is in target country
|
|
?item wdt:P17 wd:{qid} .
|
|
|
|
# Institution is a GLAM type
|
|
VALUES ?type {{
|
|
wd:Q7075 # library
|
|
wd:Q166118 # archive
|
|
wd:Q33506 # museum
|
|
wd:Q1007870 # art gallery
|
|
wd:Q28564 # public library
|
|
wd:Q11396180 # academic library
|
|
wd:Q207694 # art museum
|
|
wd:Q2772772 # history museum
|
|
}}
|
|
?item wdt:P31 ?type .
|
|
|
|
# Optional enrichment data
|
|
OPTIONAL {{ ?item wdt:P791 ?isil . }} # ISIL code
|
|
OPTIONAL {{ ?item wdt:P214 ?viaf . }} # VIAF ID
|
|
OPTIONAL {{ ?item wdt:P625 ?coords . }} # Coordinates
|
|
OPTIONAL {{ ?item wdt:P856 ?website . }} # Official website
|
|
OPTIONAL {{ ?item wdt:P571 ?inception . }} # Founding date
|
|
|
|
# Get labels (adjust languages by region)
|
|
SERVICE wikibase:label {{
|
|
bd:serviceParam wikibase:language "en,pt,es,nl,fr,it,no,th,vi" .
|
|
}}
|
|
}}
|
|
LIMIT 2000
|
|
"""
|
|
|
|
sparql.setQuery(query)
|
|
|
|
try:
|
|
raw_results = sparql.query().convert()
|
|
bindings = raw_results.get("results", {}).get("bindings", []) if isinstance(raw_results, dict) else []
|
|
|
|
# Parse results into dict keyed by QID
|
|
results = {}
|
|
for binding in bindings:
|
|
item_uri = binding.get("item", {}).get("value", "")
|
|
qid = item_uri.split("/")[-1] if item_uri else None
|
|
|
|
if not qid or not qid.startswith("Q"):
|
|
continue
|
|
|
|
result = {
|
|
"qid": qid,
|
|
"name": binding.get("itemLabel", {}).get("value", ""),
|
|
"description": binding.get("itemDescription", {}).get("value", ""),
|
|
"type": binding.get("typeLabel", {}).get("value", ""),
|
|
"identifiers": {}
|
|
}
|
|
|
|
if "isil" in binding:
|
|
result["identifiers"]["ISIL"] = binding["isil"]["value"]
|
|
|
|
if "viaf" in binding:
|
|
result["identifiers"]["VIAF"] = binding["viaf"]["value"]
|
|
|
|
if "website" in binding:
|
|
result["identifiers"]["Website"] = binding["website"]["value"]
|
|
|
|
if "inception" in binding:
|
|
result["founding_date"] = binding["inception"]["value"].split("T")[0]
|
|
|
|
if "coords" in binding:
|
|
coords_str = binding["coords"]["value"]
|
|
if coords_str.startswith("Point("):
|
|
lon, lat = coords_str[6:-1].split()
|
|
result["latitude"] = float(lat)
|
|
result["longitude"] = float(lon)
|
|
|
|
results[qid] = result
|
|
|
|
return results
|
|
|
|
except Exception as e:
|
|
print(f"\n❌ Error querying Wikidata: {e}")
|
|
import traceback
|
|
traceback.print_exc()
|
|
return {}
|
|
|
|
|
|
def fuzzy_match_institutions(
|
|
institutions: list[dict[str, Any]],
|
|
wikidata_results: dict[str, dict[str, Any]],
|
|
threshold: float = 0.85
|
|
) -> list[tuple[int, str, float, dict[str, Any]]]:
|
|
"""
|
|
Fuzzy match institutions with Wikidata results.
|
|
|
|
Returns: List of (institution_idx, qid, confidence_score, wd_data)
|
|
"""
|
|
matches = []
|
|
|
|
for idx, inst in enumerate(institutions):
|
|
inst_name = inst.get("name", "")
|
|
inst_type = inst.get("institution_type", "")
|
|
if not inst_name:
|
|
continue
|
|
|
|
# Skip if already has real Wikidata ID
|
|
has_wikidata = any(
|
|
id_obj.get("identifier_scheme") == "Wikidata" and
|
|
id_obj.get("identifier_value", "").startswith("Q") and
|
|
int(id_obj.get("identifier_value", "Q999999999")[1:]) < 100000000
|
|
for id_obj in inst.get("identifiers", [])
|
|
)
|
|
if has_wikidata:
|
|
continue
|
|
|
|
# Find best match
|
|
best_score = 0.0
|
|
best_qid = None
|
|
best_data = None
|
|
|
|
for qid, wd_data in wikidata_results.items():
|
|
wd_name = wd_data.get("name", "")
|
|
wd_desc = wd_data.get("description", "")
|
|
if not wd_name:
|
|
continue
|
|
|
|
# Check type compatibility
|
|
if not institution_type_compatible(inst_type, inst_name, wd_name, wd_desc):
|
|
continue
|
|
|
|
score = similarity_score(inst_name, wd_name)
|
|
if score > best_score:
|
|
best_score = score
|
|
best_qid = qid
|
|
best_data = wd_data
|
|
|
|
# Only include matches above threshold
|
|
if best_score >= threshold and best_qid and best_data:
|
|
matches.append((idx, best_qid, best_score, best_data))
|
|
|
|
return matches
|
|
|
|
|
|
def enrich_institution(inst: dict[str, Any], wd_data: dict[str, Any]) -> bool:
|
|
"""Enrich an institution with Wikidata data. Returns True if enriched."""
|
|
enriched = False
|
|
|
|
if "identifiers" not in inst or not inst["identifiers"]:
|
|
inst["identifiers"] = []
|
|
|
|
identifiers_list = inst["identifiers"]
|
|
existing_schemes = {i.get("identifier_scheme", "") for i in identifiers_list if isinstance(i, dict)}
|
|
|
|
# Add Wikidata ID
|
|
if "Wikidata" not in existing_schemes:
|
|
identifiers_list.append({
|
|
"identifier_scheme": "Wikidata",
|
|
"identifier_value": wd_data["qid"],
|
|
"identifier_url": f"https://www.wikidata.org/wiki/{wd_data['qid']}"
|
|
})
|
|
enriched = True
|
|
|
|
# Add other identifiers
|
|
wd_identifiers = wd_data.get("identifiers", {})
|
|
for scheme, value in wd_identifiers.items():
|
|
if scheme not in existing_schemes:
|
|
id_obj = {
|
|
"identifier_scheme": scheme,
|
|
"identifier_value": value
|
|
}
|
|
|
|
if scheme == "VIAF":
|
|
id_obj["identifier_url"] = f"https://viaf.org/viaf/{value}"
|
|
elif scheme == "Website":
|
|
id_obj["identifier_url"] = value
|
|
|
|
identifiers_list.append(id_obj)
|
|
enriched = True
|
|
|
|
# Add founding date
|
|
if "founding_date" in wd_data and not inst.get("founding_date"):
|
|
inst["founding_date"] = wd_data["founding_date"]
|
|
enriched = True
|
|
|
|
# Add coordinates if missing
|
|
if "latitude" in wd_data and "longitude" in wd_data:
|
|
locations = inst.get("locations", [])
|
|
if isinstance(locations, list) and len(locations) > 0:
|
|
first_loc = locations[0]
|
|
if isinstance(first_loc, dict) and first_loc.get("latitude") is None:
|
|
first_loc["latitude"] = wd_data["latitude"]
|
|
first_loc["longitude"] = wd_data["longitude"]
|
|
enriched = True
|
|
|
|
# Update provenance
|
|
if enriched:
|
|
prov = inst.get("provenance", {})
|
|
if isinstance(prov, dict):
|
|
existing_method = prov.get("extraction_method", "")
|
|
if existing_method:
|
|
prov["extraction_method"] = f"{existing_method} + Wikidata fuzzy enrichment"
|
|
else:
|
|
prov["extraction_method"] = "Wikidata fuzzy enrichment"
|
|
|
|
return enriched
|
|
|
|
|
|
def main():
|
|
base_dir = Path(__file__).parent.parent
|
|
input_file = base_dir / "data" / "instances" / "global" / "global_heritage_institutions_wikidata_enriched.yaml"
|
|
output_file = base_dir / "data" / "instances" / "global" / "global_heritage_institutions_wikidata_enriched.yaml"
|
|
|
|
# Target countries with low coverage
|
|
target_countries = ["BR", "BE", "IT", "NO", "TH", "VN"]
|
|
|
|
print("="*80)
|
|
print("🌍 LOW-COVERAGE COUNTRIES FUZZY MATCHING")
|
|
print("="*80)
|
|
print(f"\n🎯 Target countries: {', '.join(target_countries)}\n")
|
|
print(f"📖 Loading dataset...\n")
|
|
|
|
start_time = time.time()
|
|
|
|
with open(input_file, 'r', encoding='utf-8') as f:
|
|
institutions = yaml.safe_load(f)
|
|
|
|
load_time = time.time() - start_time
|
|
print(f"✅ Loaded {len(institutions):,} institutions in {load_time:.1f}s\n")
|
|
|
|
# Setup SPARQL
|
|
sparql = SPARQLWrapper("https://query.wikidata.org/sparql")
|
|
sparql.setReturnFormat(SPARQL_JSON)
|
|
sparql.setMethod('POST')
|
|
sparql.addCustomHttpHeader("User-Agent", "GLAM-Extractor/0.2")
|
|
|
|
total_enriched = 0
|
|
|
|
# Process each country
|
|
for country_code in target_countries:
|
|
print("="*80)
|
|
print(f"🌍 Processing {country_code}")
|
|
print("="*80)
|
|
|
|
# Filter institutions for this country
|
|
country_institutions_idx = [
|
|
idx for idx, inst in enumerate(institutions)
|
|
if any(loc.get('country') == country_code for loc in inst.get('locations', []))
|
|
]
|
|
|
|
if not country_institutions_idx:
|
|
print(f" ⚠️ No institutions found for {country_code}\n")
|
|
continue
|
|
|
|
print(f" Found {len(country_institutions_idx):,} institutions")
|
|
|
|
# Count those without Wikidata
|
|
country_without_wikidata = [
|
|
idx for idx in country_institutions_idx
|
|
if not any(
|
|
id_obj.get("identifier_scheme") == "Wikidata" and
|
|
id_obj.get("identifier_value", "").startswith("Q") and
|
|
int(id_obj.get("identifier_value", "Q999999999")[1:]) < 100000000
|
|
for id_obj in institutions[idx].get("identifiers", [])
|
|
)
|
|
]
|
|
|
|
current_coverage = (len(country_institutions_idx) - len(country_without_wikidata)) / len(country_institutions_idx) * 100
|
|
print(f" Current Wikidata coverage: {current_coverage:.1f}%")
|
|
print(f" Institutions needing enrichment: {len(country_without_wikidata):,}\n")
|
|
|
|
# Query Wikidata
|
|
print(f"🔍 Querying Wikidata for {country_code} institutions...")
|
|
print(" (This may take 30-60 seconds)\n")
|
|
|
|
wikidata_results = query_country_institutions(sparql, country_code)
|
|
print(f"✅ Found {len(wikidata_results):,} {country_code} institutions in Wikidata\n")
|
|
|
|
if not wikidata_results:
|
|
print(f" ⚠️ No Wikidata results for {country_code}, skipping\n")
|
|
continue
|
|
|
|
# Fuzzy match
|
|
print("🔗 Fuzzy matching names (threshold: 0.85)...\n")
|
|
|
|
country_insts = [institutions[idx] for idx in country_without_wikidata]
|
|
matches = fuzzy_match_institutions(country_insts, wikidata_results, threshold=0.85)
|
|
|
|
print(f"✨ Found {len(matches):,} high-confidence matches\n")
|
|
|
|
# Show sample matches
|
|
if matches:
|
|
print(f"📋 Sample matches (showing first 5):")
|
|
for i, (local_idx, qid, score, wd_data) in enumerate(matches[:5]):
|
|
inst = country_insts[local_idx]
|
|
print(f"\n{i+1}. Confidence: {score:.3f}")
|
|
print(f" Local: {inst.get('name')}")
|
|
print(f" Wikidata: {wd_data.get('name')} ({wd_data.get('qid')})")
|
|
print(f" Type: {wd_data.get('type', 'Unknown')}")
|
|
|
|
print(f"\n✅ Applying {len(matches)} matches for {country_code}...\n")
|
|
|
|
country_enriched = 0
|
|
for local_idx, qid, score, wd_data in matches:
|
|
global_idx = country_without_wikidata[local_idx]
|
|
if enrich_institution(institutions[global_idx], wd_data):
|
|
country_enriched += 1
|
|
|
|
print(f"✨ Enriched {country_enriched:,} institutions")
|
|
|
|
# Calculate new coverage
|
|
new_coverage = (len(country_institutions_idx) - len(country_without_wikidata) + country_enriched) / len(country_institutions_idx) * 100
|
|
print(f" Coverage: {current_coverage:.1f}% → {new_coverage:.1f}% (+{new_coverage - current_coverage:.1f}%)\n")
|
|
|
|
total_enriched += country_enriched
|
|
else:
|
|
print(f" ❌ No matches found for {country_code}\n")
|
|
|
|
# Rate limiting between countries
|
|
time.sleep(2.0)
|
|
|
|
# Write output
|
|
if total_enriched > 0:
|
|
print("="*80)
|
|
print("💾 Writing enriched dataset...")
|
|
print("="*80 + "\n")
|
|
|
|
with open(output_file, 'w', encoding='utf-8') as f:
|
|
header = f"""---
|
|
# Global Heritage Institutions - Low-Coverage Countries Fuzzy Enriched
|
|
# Generated: {datetime.now(timezone.utc).isoformat()}
|
|
#
|
|
# Total institutions: {len(institutions):,}
|
|
# Countries processed: {', '.join(target_countries)}
|
|
# Total new matches: {total_enriched:,}
|
|
|
|
"""
|
|
f.write(header)
|
|
yaml.dump(institutions, f, allow_unicode=True, default_flow_style=False, sort_keys=False, width=120)
|
|
|
|
print(f"✅ Complete! Output: {output_file}\n")
|
|
|
|
# Final report
|
|
print("="*80)
|
|
print("📊 ENRICHMENT REPORT")
|
|
print("="*80)
|
|
print(f"\n✨ Total institutions enriched: {total_enriched:,}")
|
|
print(f"⏱️ Processing time: {(time.time()-start_time)/60:.1f} minutes")
|
|
print("="*80 + "\n")
|
|
else:
|
|
print("❌ No institutions enriched\n")
|
|
|
|
|
|
if __name__ == "__main__":
|
|
main()
|