Fix: do not return embeddings in HybridRetriever (#187)

* Fix: do not return embedding property in HybridRetriever

* Update changelog

* Fix changelog

Author: stellasia

CHANGELOG.md

@@ -8,6 +8,8 @@
 
 ### Changed
 - Vector and Hybrid retrievers used with `return_properties` now also return the node labels (`nodeLabels`) and the node's element ID (`id`).
+- `HybridRetriever` now filters out the embedding property index in `self.vector_index_name` from the retriever result by default.
+
 
 ## 1.1.0
 

src/neo4j_graphrag/retrievers/base.py

@@ -131,8 +131,12 @@ def _verify_version(self) -> None:
         if version_tuple < target_version:
             raise Neo4jVersionError()
 
-    def _fetch_index_infos(self) -> None:
-        """Fetch the node label and embedding property from the index definition"""
+    def _fetch_index_infos(self, vector_index_name: str) -> None:
+        """Fetch the node label and embedding property from the index definition
+
+        Args:
+            vector_index_name (str): Name of the vector index
+        """
         query = (
             "SHOW VECTOR INDEXES "
             "YIELD name, labelsOrTypes, properties, options "
@@ -141,7 +145,7 @@ def _fetch_index_infos(self) -> None:
             "options.indexConfig.`vector.dimensions` as dimensions"
         )
         query_result = self.driver.execute_query(
-            query, {"index_name": self.index_name}, database_=self.neo4j_database
+            query, {"index_name": vector_index_name}, database_=self.neo4j_database
         )
         try:
             result = query_result.records[0]

src/neo4j_graphrag/retrievers/hybrid.py

@@ -117,6 +117,9 @@ def __init__(
             else None
         )
         self.result_formatter = validated_data.result_formatter
+        self._embedding_node_property = None
+        self._embedding_dimension = None
+        self._fetch_index_infos(self.vector_index_name)
 
     def default_record_formatter(self, record: neo4j.Record) -> RetrieverResultItem:
         """
@@ -187,6 +190,7 @@ def get_search_results(
         search_query, _ = get_search_query(
             SearchType.HYBRID,
             self.return_properties,
+            embedding_node_property=self._embedding_node_property,
             neo4j_version_is_5_23_or_above=self.neo4j_version_is_5_23_or_above,
         )
 

src/neo4j_graphrag/retrievers/vector.py

@@ -123,7 +123,7 @@ def __init__(
         self._node_label = None
         self._embedding_node_property = None
         self._embedding_dimension = None
-        self._fetch_index_infos()
+        self._fetch_index_infos(self.index_name)
 
     def default_record_formatter(self, record: neo4j.Record) -> RetrieverResultItem:
         """
@@ -287,7 +287,7 @@ def __init__(
         self._node_label = None
         self._node_embedding_property = None
         self._embedding_dimension = None
-        self._fetch_index_infos()
+        self._fetch_index_infos(self.index_name)
 
     def get_search_results(
         self,

tests/e2e/test_hybrid_e2e.py

@@ -39,6 +39,7 @@ def test_hybrid_retriever_search_text(
     assert len(results.items) == 5
     for result in results.items:
         assert isinstance(result, RetrieverResultItem)
+        assert "'vectorProperty': None," in result.content
 
 
 @pytest.mark.usefixtures("setup_neo4j_for_retrieval")

tests/unit/retrievers/test_hybrid.py

@@ -158,9 +158,11 @@ def test_hybrid_cypher_retriever_invalid_database_name(
     assert "Input should be a valid string" in str(exc_info.value)
 
 
+@patch("neo4j_graphrag.retrievers.HybridRetriever._fetch_index_infos")
 @patch("neo4j_graphrag.retrievers.HybridRetriever._verify_version")
 def test_hybrid_search_text_happy_path(
     _verify_version_mock: MagicMock,
+    _fetch_index_infos_mock: MagicMock,
     driver: MagicMock,
     embedder: MagicMock,
     neo4j_record: MagicMock,
@@ -176,13 +178,17 @@ def test_hybrid_search_text_happy_path(
         driver, vector_index_name, fulltext_index_name, embedder
     )
     retriever.neo4j_version_is_5_23_or_above = True
+    retriever._embedding_node_property = (
+        "embedding"  # variable normally filled by fetch_index_infos
+    )
     retriever.driver.execute_query.return_value = [  # type: ignore
         [neo4j_record],
         None,
         None,
     ]
     search_query, _ = get_search_query(
         SearchType.HYBRID,
+        embedding_node_property="embedding",
         neo4j_version_is_5_23_or_above=retriever.neo4j_version_is_5_23_or_above,
     )
 
@@ -208,9 +214,11 @@ def test_hybrid_search_text_happy_path(
     )
 
 
+@patch("neo4j_graphrag.retrievers.HybridRetriever._fetch_index_infos")
 @patch("neo4j_graphrag.retrievers.HybridRetriever._verify_version")
 def test_hybrid_search_favors_query_vector_over_embedding_vector(
     _verify_version_mock: MagicMock,
+    _fetch_index_infos_mock: MagicMock,
     driver: MagicMock,
     embedder: MagicMock,
     neo4j_record: MagicMock,
@@ -288,9 +296,11 @@ def test_hybrid_search_retriever_search_missing_embedder_for_text(
         )
 
 
+@patch("neo4j_graphrag.retrievers.HybridRetriever._fetch_index_infos")
 @patch("neo4j_graphrag.retrievers.HybridRetriever._verify_version")
 def test_hybrid_retriever_return_properties(
     _verify_version_mock: MagicMock,
+    _fetch_index_infos_mock: MagicMock,
     driver: MagicMock,
     embedder: MagicMock,
     neo4j_record: MagicMock,