Support Metadata filtering with Neo4J #114

vga91 · 2025-03-25T13:34:14Z

Issue

Change

Migration from langchain4j core repo PR: langchain4j/langchain4j#2577

Added Filter implementation to Neo4j embedding.
- If filter is null we can find the relevant documents using the existing vector index,
- otherwise, we start preparing the base Cypher statement that will be used for the pre-filtered metadata approach and then a vector.similarity.cosine will be executed.
  As explained. here for the langchain python implementation.
Upgraded neo4j container to latest 5.26.x version
Added Neo4jFilterMapperTest.java

Similar to https://github.com/langchain-ai/langchain/blob/master/libs/community/langchain_community/vectorstores/neo4j_vector.py#L1064.

General checklist

There are no breaking changes
I have added unit and integration tests for my change
I have manually run all the unit tests in all modules, and they are all green
I have manually run all integration tests in the module I have added/changed, and they are all green

I have added/updated the documentation
I have added an example in the examples repo (only for "big" features)
I have added/updated Spring Boot starter(s) (if applicable)

Checklist for adding new maven module

I have added my new module in the root pom.xml and langchain4j-community-bom/pom.xml

Checklist for adding new embedding store integration

I have added a {NameOfIntegration}EmbeddingStoreIT that extends from either EmbeddingStoreIT or EmbeddingStoreWithFilteringIT
I have added a {NameOfIntegration}EmbeddingStoreRemovalIT that extends from EmbeddingStoreWithRemovalIT

Checklist for changing existing embedding store integration

I have manually verified that the {NameOfIntegration}EmbeddingStore works correctly with the data persisted using the latest released version of LangChain4j

jexp · 2025-03-28T10:15:46Z

...neo4j/src/main/java/dev/langchain4j/community/store/embedding/neo4j/Neo4jEmbeddingStore.java

+
+        try (var session = session()) {
+            String statement = String.format(
+                    "CALL { MATCH (n:%1$s) WHERE n.%2$s IS NOT NULL AND size(n.%2$s) = toInteger(%3$s) AND %4$s DETACH DELETE n } IN TRANSACTIONS ",


doesn't seem to apply the filter?

The filter is applied via filterEntry.getKey() in the String.format and filterEntry.getValue() in the params.
For example, if the Filter is IsEqualTo(key=type, comparisonValue=a) , the filterEntry.getKey() is "n.type = $param_1" and the filterEntry.getValue() is Map.of("param_1", "a") .

Therefore the result is:

session.run( "CALL { MATCH .... AND n.type = $param_1 DETACH DELETE n } IN TRANSACTIONS ", Map.of("param_1", "a") )

so that we can handle any neo4j data type

jexp · 2025-03-28T10:16:10Z

...neo4j/src/main/java/dev/langchain4j/community/store/embedding/neo4j/Neo4jEmbeddingStore.java

-            String statement =
-                    String.format("CALL { MATCH (n:%1$s) DETACH DELETE n } IN TRANSACTIONS", this.sanitizedLabel);
+            String statement = String.format(
+                    "CALL { MATCH (n:%1$s) WHERE n.%2$s IS NOT NULL AND size(n.%2$s) = toInteger(%3$s) DETACH DELETE n } IN TRANSACTIONS",


why do we need to check for embedding and embedding size match? this will make it super expensive?

jexp · 2025-03-28T10:19:27Z

...neo4j/src/main/java/dev/langchain4j/community/store/embedding/neo4j/Neo4jEmbeddingStore.java

+        final AbstractMap.SimpleEntry<String, Map<?, ?>> entry = new Neo4jFilterMapper().map(filter);
+        final String query =
+                """
+                CYPHER runtime = parallel parallelRuntimeSupport=all


you forgot to apply the filter in the query?

oh, it's this? AND %4$s entry.getKey(),

we should probably switch to cypher-dsl soon (separate PR), so all this text-formatting doesn't get out of hand.

Yeah it makes sense, we'll switch to cypher-dsl 👍

jexp · 2025-03-28T11:48:30Z

...neo4j/src/main/java/dev/langchain4j/community/store/embedding/neo4j/Neo4jEmbeddingStore.java

        }
    }

    /*
    Private methods
    */
+    private EmbeddingSearchResult getSearchResUsingVectorSimilarity(
+            EmbeddingSearchRequest request, Filter filter, Value embeddingValue, Session session) {
+        final AbstractMap.SimpleEntry<String, Map<?, ?>> entry = new Neo4jFilterMapper().map(filter);


A map.entry as value is quite odd? shouldn't it be a list of record QueryFilter or something?
Or the appropriate thing from cypher-dsl in the future

jexp · 2025-03-28T12:01:55Z

...y-neo4j/src/main/java/dev/langchain4j/community/store/embedding/neo4j/Neo4jFilterMapper.java

+        }
+    }
+
+    private String getOperation(String key, String operator, Object value) {


usually this is an enum of operators which is then also able to format it
and a record(property, Operator, value) and then a list of that

but I think Cypher DSL has this stuff out of the box.

jexp · 2025-03-28T12:06:02Z

...y-neo4j/src/main/java/dev/langchain4j/community/store/embedding/neo4j/Neo4jFilterMapper.java

+
+    public static final String UNSUPPORTED_FILTER_TYPE_ERROR = "Unsupported filter type: ";
+
+    public static class IncrementalKeyMap {


a list might be better

jexp · 2025-03-28T12:09:05Z

...o4j/src/test/java/dev/langchain4j/community/store/embedding/neo4j/Neo4jEmbeddingStoreIT.java

+
+        final And filter = new And(
+                new And(new IsEqualTo("key1", "value1"), new IsEqualTo("key2", "10")),
+                new Not(new Or(new IsIn("key3", asList("1", "2")), new IsNotEqualTo("key4", "value4"))));


do we test teh formatting somewhere explicitely including the parentheses and parameters?

Yes, we added some complex condition, like in the should_map_or_not_and test

jexp

LGTM see my small comments

Martin7-1 · 2025-03-30T13:34:31Z

@vga91 There are some conflicts, could you please resolve it?

vga91 · 2025-03-30T14:18:23Z

@Martin7-1 just resolved it, thanks!

Martin7-1

@vga91 Thank you!

Martin7-1 · 2025-03-31T13:14:59Z

...neo4j/src/main/java/dev/langchain4j/community/store/embedding/neo4j/Neo4jEmbeddingStore.java

+                ORDER BY score DESC
+                LIMIT $maxResults
+                """
+                        .formatted(


Same .formatted problem like previous PR.

Changed all of them

Martin7-1 · 2025-03-31T13:15:37Z

...neo4j/src/main/java/dev/langchain4j/community/store/embedding/neo4j/Neo4jEmbeddingStore.java

+                                this.dimension,
+                                entry.getKey(),
+                                embeddingValue);
+        final Map params = entry.getValue();


nit: Could you please specify the generic type? like Map<String, Object>?

Martin7-1 · 2025-03-31T13:17:23Z

...y-neo4j/src/main/java/dev/langchain4j/community/store/embedding/neo4j/Neo4jFilterMapper.java

+            return new RuntimeException(invalidSanitizeValue);
+        });
+
+        return "n.%s %s $%s".formatted(sanitizedKey, operator, param);


Same .formatted problem like previous PR.

Martin7-1 · 2025-03-31T13:17:26Z

...y-neo4j/src/main/java/dev/langchain4j/community/store/embedding/neo4j/Neo4jFilterMapper.java

+
+    public String mapNotIn(IsNotIn filter) {
+        final String inOperation = getOperation(filter.key(), "IN", filter.comparisonValues());
+        return "NOT (%s)".formatted(inOperation);


Same .formatted problem like previous PR.

Martin7-1 · 2025-03-31T13:17:32Z

...y-neo4j/src/main/java/dev/langchain4j/community/store/embedding/neo4j/Neo4jFilterMapper.java

+    }
+
+    private String mapAnd(And filter) {
+        return "(%s) AND (%s)".formatted(getStringMapping(filter.left()), getStringMapping(filter.right()));


Same .formatted problem like previous PR.

Martin7-1 · 2025-03-31T13:17:35Z

...y-neo4j/src/main/java/dev/langchain4j/community/store/embedding/neo4j/Neo4jFilterMapper.java

+    }
+
+    private String mapOr(Or filter) {
+        return "(%s) OR (%s)".formatted(getStringMapping(filter.left()), getStringMapping(filter.right()));


Same .formatted problem like previous PR.

Martin7-1 · 2025-03-31T13:17:41Z

...y-neo4j/src/main/java/dev/langchain4j/community/store/embedding/neo4j/Neo4jFilterMapper.java

+    }
+
+    private String mapNot(Not filter) {
+        return "NOT (%s)".formatted(getStringMapping(filter.expression()));


Same .formatted problem like previous PR.

Martin7-1 · 2025-03-31T13:21:02Z

...y-neo4j/src/main/java/dev/langchain4j/community/store/embedding/neo4j/Neo4jFilterMapper.java

+
+        private int counter = 1;
+
+        public String put(Object value) {


Looks like it should use AtomicInteger to keep thread-safe. WDYT?

changed to private final AtomicInteger integer = new AtomicInteger(); 👍

Martin7-1 · 2025-03-31T13:22:20Z

...neo4j/src/main/java/dev/langchain4j/community/store/embedding/neo4j/Neo4jEmbeddingStore.java

+            String statement = String.format(
+                    "CALL { MATCH (n:%1$s) WHERE n.%2$s IS NOT NULL AND size(n.%2$s) = toInteger(%3$s) AND %4$s DETACH DELETE n } IN TRANSACTIONS ",
+                    this.sanitizedLabel, this.embeddingProperty, this.dimension, filterEntry.getKey());
+            final Map params = filterEntry.getValue();


nit: Could you please specify the generic type? like Map<String, Object>?

Martin7-1 · 2025-04-03T16:36:07Z

@vga91 Looks like AtomicInteger does not initialize correctly, and thus cause theNeo4jFilterMapperTest failed. Could you please check?

Martin7-1

@vga91 Thank you!

vga91 mentioned this pull request Mar 25, 2025

Support Metadata filtering with Neo4J langchain4j/langchain4j#2577

Closed

12 tasks

Martin7-1 added enhancement New feature or request P3 Medium priority theme: embedding store Issues/PRs related to embedding store labels Mar 25, 2025

jexp reviewed Mar 28, 2025

View reviewed changes

jexp approved these changes Mar 28, 2025

View reviewed changes

vga91 mentioned this pull request Mar 30, 2025

Neo4jGraph features #119

Open

11 tasks

vga91 added 2 commits March 30, 2025 15:32

Support Metadata filtering with Neo4J

6722ae5

optimized removes methods like search one

46cdbb4

changes after rebase

3cc1a24

vga91 force-pushed the issue-core-repo-1252 branch from 87782f5 to 3cc1a24 Compare March 30, 2025 13:44

Martin7-1 reviewed Mar 31, 2025

View reviewed changes

changes 2nd review

979dfc0

vga91 and others added 3 commits April 4, 2025 14:47

fix param enumeration

1084341

Merge branch 'main' into issue-core-repo-1252

6e5dff9

format

d6a1aa8

Martin7-1 approved these changes Apr 4, 2025

View reviewed changes

Martin7-1 merged commit c314460 into langchain4j:main Apr 4, 2025
4 checks passed

vga91 mentioned this pull request Apr 11, 2025

[DRAFT] Issue 111 cypher dsl #129

Draft

11 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Support Metadata filtering with Neo4J #114

Support Metadata filtering with Neo4J #114

vga91 commented Mar 25, 2025 •

edited

Loading

jexp Mar 28, 2025

vga91 Mar 28, 2025

jexp Mar 28, 2025

jexp Mar 28, 2025 •

edited

Loading

jexp Mar 28, 2025

vga91 Mar 28, 2025

jexp Mar 28, 2025

jexp Mar 28, 2025

jexp Mar 28, 2025

jexp Mar 28, 2025

vga91 Mar 30, 2025

jexp left a comment

Martin7-1 commented Mar 30, 2025

vga91 commented Mar 30, 2025

Martin7-1 left a comment

Martin7-1 Mar 31, 2025

vga91 Apr 3, 2025

Martin7-1 Mar 31, 2025

Martin7-1 Mar 31, 2025

Martin7-1 Mar 31, 2025

Martin7-1 Mar 31, 2025

Martin7-1 Mar 31, 2025

Martin7-1 Mar 31, 2025

Martin7-1 Mar 31, 2025

vga91 Apr 3, 2025

Martin7-1 Mar 31, 2025

Martin7-1 commented Apr 3, 2025

Martin7-1 left a comment


		public static final String UNSUPPORTED_FILTER_TYPE_ERROR = "Unsupported filter type: ";

		public static class IncrementalKeyMap {

Support Metadata filtering with Neo4J #114

Support Metadata filtering with Neo4J #114

Conversation

vga91 commented Mar 25, 2025 • edited Loading

Issue

Change

General checklist

Checklist for adding new maven module

Checklist for adding new embedding store integration

Checklist for changing existing embedding store integration

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jexp Mar 28, 2025 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jexp left a comment

Choose a reason for hiding this comment

Martin7-1 commented Mar 30, 2025

vga91 commented Mar 30, 2025

Martin7-1 left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Martin7-1 commented Apr 3, 2025

Martin7-1 left a comment

Choose a reason for hiding this comment

vga91 commented Mar 25, 2025 •

edited

Loading

jexp Mar 28, 2025 •

edited

Loading