test(recognition): reproduce Dutch address overcapture contexts

tests/test_dutch_address_span_precision.py

@@ -0,0 +1,89 @@
+from presidio_analyzer import RecognizerResult
+
+from dutch_address_span_precision import tighten_dutch_address_results
+from dutch_recognizers import get_dutch_entity_names, get_dutch_recognizers
+
+
+LIVE_CONTEXTS = [
+    "Beschrijving Polderweg 8",
+    "De inspectie bezoekt Polderweg 8",
+    "Nu Polderweg 8 de",
+    "Op Polderweg 8",
+    "Polderweg 8 een",
+    "Polderweg 8 en",
+    "Polderweg 8 in",
+    "Polderweg 8 is",
+    "Polderweg 8 na",
+    "Polderweg 8 op",
+]
+
+
+def _raw_address_results(text: str):
+    entities = get_dutch_entity_names(include_legal=True)
+    results = []
+    for recognizer in get_dutch_recognizers(supported_language="en"):
+        if "NL_ADDRESS" not in getattr(recognizer, "supported_entities", []):
+            continue
+        if hasattr(recognizer, "load"):
+            recognizer.load()
+        results.extend(recognizer.analyze(text, entities=entities, nlp_artifacts=None))
+    return [result for result in results if result.entity_type == "NL_ADDRESS"]
+
+
+def _values(text: str, results):
+    return [text[result.start : result.end] for result in results]
+
+
+def test_live_polderweg_contexts_resolve_to_exact_address_only():
+    for text in LIVE_CONTEXTS:
+        raw = _raw_address_results(text)
+        assert raw, f"baseline recognizer unexpectedly missed address in: {text!r}"
+        resolved = tighten_dutch_address_results(text, raw)
+        values = _values(text, resolved)
+        assert "Polderweg 8" in values, (text, values)
+        assert all(value == "Polderweg 8" for value in values), (text, values)
+
+
+def test_house_number_does_not_absorb_short_adjacent_words():
+    for suffix in ("een", "en", "in", "is", "na", "op", "de"):
+        text = f"Polderweg 8 {suffix}"
+        broad = RecognizerResult(entity_type="NL_ADDRESS", start=0, end=len(text), score=0.66)
+        [resolved] = tighten_dutch_address_results(text, [broad])
+        assert text[resolved.start : resolved.end] == "Polderweg 8"
+
+
+def test_legitimate_prefix_address_with_postcode_and_city_is_preserved():
+    text = "Verweerder woont aan Laan van Meerdervoort 55, 2517 AM Den Haag."
+    start = text.index("Laan")
+    end = text.index(".")
+    broad = RecognizerResult(entity_type="NL_ADDRESS", start=start, end=end, score=0.70)
+    [resolved] = tighten_dutch_address_results(text, [broad])
+    assert text[resolved.start : resolved.end] == "Laan van Meerdervoort 55, 2517 AM Den Haag"
+
+
+def test_common_suffix_street_forms_remain_detectable():
+    examples = (
+        "Polderweg 8",
+        "Kerkstraat 12A",
+        "Nieuwe Kerkstraat 14",
+        "Westersingel 101-2",
+    )
+    for address in examples:
+        broad = RecognizerResult(entity_type="NL_ADDRESS", start=0, end=len(address), score=0.66)
+        [resolved] = tighten_dutch_address_results(address, [broad])
+        assert address[resolved.start : resolved.end] == address
+
+
+def test_unknown_address_shape_is_preserved_fail_safe_instead_of_dropped():
+    text = "Onbekende adresvorm 42"
+    original = RecognizerResult(entity_type="NL_ADDRESS", start=0, end=len(text), score=0.66)
+    [resolved] = tighten_dutch_address_results(text, [original])
+    assert resolved.start == original.start
+    assert resolved.end == original.end
+
+
+def test_non_address_results_are_untouched():
+    text = "Polderweg 8"
+    original = RecognizerResult(entity_type="LOCATION", start=0, end=len(text), score=0.7)
+    [resolved] = tighten_dutch_address_results(text, [original])
+    assert resolved is original