fix: deduplicate entities by (title, type) to preserve same-name different-type entities

.semversioner/next-release/patch-20260428025412857613.json

@@ -0,0 +1,4 @@
+{
+  "type": "patch",
+  "description": "Fix entity deduplication to use (title, type) compound key instead of title only, preserving entities with the same name but different types"
+}

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, 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 +36,15 @@ 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", "")
+        if not title or (title, entity_type) in seen_entities:
             continue
-        seen_titles.add(title)
+        seen_entities.add((title, entity_type))
         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

@@ -192,7 +192,7 @@ async def test_missing_degree_defaults_to_zero(self):
         assert table.written[0]["degree"] == 0
 
     async def test_deduplicates_by_title(self):
-        """Duplicate titles should be skipped."""
+        """Duplicate (title, type) pairs should be skipped."""
         table = FakeTable([
             _make_entity_row("A"),
             _make_entity_row("A"),
@@ -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_same_title_different_type(self):
+        """Entities with the same title but different types must not be merged."""
+        table = FakeTable([
+            _make_entity_row("Python", entity_type="LANGUAGE"),
+            _make_entity_row("Python", entity_type="CONCEPT"),
+            _make_entity_row("Java", entity_type="LANGUAGE"),
+        ])
+        degree_map = {"Python": 2, "Java": 1}
+        await finalize_entities(table, degree_map)
+
+        assert len(table.written) == 3
+        pairs = [(r["title"], r["type"]) for r in table.written]
+        assert ("Python", "LANGUAGE") in pairs
+        assert ("Python", "CONCEPT") in pairs
+        assert ("Java", "LANGUAGE") in pairs
+
     async def test_skips_empty_title(self):
         """Rows with empty or missing title should be skipped."""
         table = FakeTable([