fix: [Fatal Bug]: Incorrect deduplication of entities with same tit...

packages/graphrag/graphrag/index/operations/finalize_entities.py

@@ -17,7 +17,7 @@ async def finalize_entities(
 ) -> list[dict[str, Any]]:
     """Read entity rows, enrich with degree, and write back.
 
-    Streams through the entities table, deduplicates by title,
+    Streams through the entities table, deduplicates by title and type,
     assigns degree from the pre-computed degree map, and writes
     each finalized row back to the same table (safe when using
     truncate=True, which reads from the original and writes to
@@ -36,14 +36,16 @@ async def finalize_entities(
             Sample of up to 5 entity rows for logging.
     """
     sample_rows: list[dict[str, Any]] = []
-    seen_titles: set[str] = set()
+    seen_entities: set[tuple[str, str]] = set()
     human_readable_id = 0
 
     async for row in entities_table:
         title = row.get("title")
-        if not title or title in seen_titles:
+        entity_type = row.get("type", "")
+        entity_key = (title, entity_type)
+        if not title or entity_key in seen_entities:
             continue
-        seen_titles.add(title)
+        seen_entities.add(entity_key)
         row["degree"] = degree_map.get(title, 0)
         row["human_readable_id"] = human_readable_id
         row["id"] = str(uuid4())

tests/unit/indexing/test_finalize_graph.py

@@ -191,12 +191,12 @@ async def test_missing_degree_defaults_to_zero(self):
         assert len(table.written) == 1
         assert table.written[0]["degree"] == 0
 
-    async def test_deduplicates_by_title(self):
-        """Duplicate titles should be skipped."""
+    async def test_deduplicates_by_title_and_type(self):
+        """Duplicate title and type combinations should be skipped."""
         table = FakeTable([
-            _make_entity_row("A"),
-            _make_entity_row("A"),
-            _make_entity_row("B"),
+            _make_entity_row("A", "PERSON"),
+            _make_entity_row("A", "PERSON"),
+            _make_entity_row("B", "ORGANIZATION"),
         ])
         degree_map = {"A": 1, "B": 2}
         await finalize_entities(table, degree_map)
@@ -205,6 +205,22 @@ async def test_deduplicates_by_title(self):
         titles = [r["title"] for r in table.written]
         assert titles == ["A", "B"]
 
+    async def test_preserves_entities_with_same_title_different_type(self):
+        """Entities with same title but different types should not be deduplicated."""
+        table = FakeTable([
+            _make_entity_row("PARENT AREA", "PERSON"),
+            _make_entity_row("PARENT AREA", "ORGANIZATION"),
+            _make_entity_row("B", "ENTITY"),
+        ])
+        degree_map = {"PARENT AREA": 1, "B": 2}
+        await finalize_entities(table, degree_map)
+
+        assert len(table.written) == 3
+        titles = [r["title"] for r in table.written]
+        types = [r["type"] for r in table.written]
+        assert titles == ["PARENT AREA", "PARENT AREA", "B"]
+        assert types == ["PERSON", "ORGANIZATION", "ENTITY"]
+
     async def test_skips_empty_title(self):
         """Rows with empty or missing title should be skipped."""
         table = FakeTable([