Fix "NOT IN" operator (#327)

* Fix "NOT IN" operator

* Update CHANGELOG

* Ruff

Author: stellasia

CHANGELOG.md

@@ -13,6 +13,11 @@
 
 - Improved log output readability in Retrievers and GraphRAG and added embedded vector to retriever result metadata for debugging.
 
+### Fixed
+
+- Fixed a bug where the `$nin` operator for metadata pre-filtering in retrievers would create an invalid Cypher query.
+
+
 ## 1.6.1
 
 ### Added

examples/customize/retrievers/use_pre_filters.py

@@ -4,26 +4,29 @@
 from neo4j_graphrag.embeddings import OpenAIEmbeddings
 from neo4j_graphrag.retrievers import VectorRetriever
 
-URI = "neo4j://localhost:7687"
-AUTH = ("neo4j", "password")
-
-INDEX_NAME = "embedding-name"
+URI = "neo4j+s://demo.neo4jlabs.com"
+AUTH = ("recommendations", "recommendations")
+DATABASE = "recommendations"
+INDEX_NAME = "moviePlotsEmbedding"
 DIMENSION = 1536
 
-# Connect to Neo4j database
-driver = neo4j.GraphDatabase.driver(URI, auth=AUTH)
-
 
-# Initialize the retriever
-retriever = VectorRetriever(driver, INDEX_NAME, embedder=OpenAIEmbeddings())
+# Connect to Neo4j database
+with neo4j.GraphDatabase.driver(URI, auth=AUTH) as driver:
+    # Initialize the retriever
+    retriever = VectorRetriever(driver, INDEX_NAME, embedder=OpenAIEmbeddings())
 
-# Perform the search
-query_text = "Find me a book about Fremen"
-pre_filters = {"int_property": {"$gt": 100}}
-print(
-    retriever.search(
+    # Perform the search
+    query_text = "Find me a movie about love"
+    pre_filters = {"int_property": {"$gt": 100}}
+    # pre_filters = {
+    #     "year": {
+    #         "$nin": ["1999", "2000"]
+    #     }
+    # }
+    retriever_result = retriever.search(
         query_text=query_text,
         top_k=1,
         filters=pre_filters,
     )
-)
+    print(retriever_result)

src/neo4j_graphrag/filters.py

@@ -100,7 +100,9 @@ def cleaned_value(self, value: list[Union[str, int, float]]) -> Any:
 
 
 class NinOperator(InOperator):
-    CYPHER_OPERATOR = "NOT IN"
+    def lhs(self, field: str) -> str:
+        lhs = super().lhs(field)
+        return f"NOT {lhs}"
 
 
 class LikeOperator(Operator):
@@ -114,8 +116,8 @@ def cleaned_value(self, value: str) -> str:
 
 class ILikeOperator(LikeOperator):
     def lhs(self, field: str) -> str:
-        safe_field_cypher = self.safe_field_cypher(field)
-        return f"toLower({self.node_alias}.{safe_field_cypher})"
+        lhs = super().lhs(field)
+        return f"toLower({lhs})"
 
     def cleaned_value(self, value: str) -> str:
         value = super().cleaned_value(value)
@@ -368,6 +370,5 @@ def get_metadata_filter(
         contains the query parameters
     """
     param_store = ParameterStore()
-    return _construct_metadata_filter(
-        filter, param_store, node_alias=node_alias
-    ), param_store.params
+    query = _construct_metadata_filter(filter, param_store, node_alias=node_alias)
+    return query, param_store.params

tests/unit/test_filters.py

@@ -165,7 +165,7 @@ def test_single_condition_cypher_nin(param_store_empty: ParameterStore) -> None:
     generated = _single_condition_cypher(
         "field", NinOperator, ["a", "b", "c"], param_store=param_store_empty
     )
-    assert generated == "node.field NOT IN $param_0"
+    assert generated == "NOT node.field IN $param_0"
     assert param_store_empty.params == {"param_0": ["a", "b", "c"]}
 
 
@@ -552,7 +552,7 @@ def test_get_metadata_filter_in_operator() -> None:
 def test_get_metadata_filter_not_in_operator() -> None:
     filters = {"field": {"$nin": ["a", "b"]}}
     query, params = get_metadata_filter(filters)
-    assert query == "node.field NOT IN $param_0"
+    assert query == "NOT node.field IN $param_0"
     assert params == {"param_0": ["a", "b"]}