community[patch]: Fix RRF normalization and lucene characters for neo4j vector

libs/langchain-community/src/vectorstores/neo4j_vector.ts

@@ -614,7 +614,7 @@ export class Neo4jVectorStore extends VectorStore {
       k: Number(k),
       embedding: vector,
       keyword_index: this.keywordIndexName,
-      query,
+      query: removeLuceneChars(query),
     };
     const results = await this.query(readQuery, parameters);
 
@@ -717,7 +717,10 @@ function getSearchIndexQuery(searchType: SearchType): string {
     hybrid: `
           CALL {
               CALL db.index.vector.queryNodes($index, $k, $embedding) YIELD node, score
-              RETURN node, score UNION
+              WITH collect({node:node, score:score}) AS nodes, max(score) AS max 
+              UNWIND nodes AS n 
+              // We use 0 as min
+              RETURN n.node AS node, (n.score / max) AS score UNION 
               CALL db.index.fulltext.queryNodes($keyword_index, $query, {limit: $k}) YIELD node, score
               WITH collect({node: node, score: score}) AS nodes, max(score) AS max
               UNWIND nodes AS n
@@ -729,3 +732,32 @@ function getSearchIndexQuery(searchType: SearchType): string {
 
   return typeToQueryMap[searchType];
 }
+
+function removeLuceneChars(text: string): string {
+  // Remove Lucene special characters
+  const specialChars = [
+    "+",
+    "-",
+    "&",
+    "|",
+    "!",
+    "(",
+    ")",
+    "{",
+    "}",
+    "[",
+    "]",
+    "^",
+    '"',
+    "~",
+    "*",
+    "?",
+    ":",
+    "\\",
+  ];
+  let modifiedText = text;
+  for (const char of specialChars) {
+    modifiedText = modifiedText.split(char).join(" ");
+  }
+  return modifiedText.trim();
+}

libs/langchain-community/src/vectorstores/tests/neo4j_vector.int.test.ts

@@ -5,7 +5,7 @@ import { Neo4jVectorStore } from "../neo4j_vector.js";
 
 const OS_TOKEN_COUNT = 1536;
 
-const texts = ["foo", "bar", "baz"];
+const texts = ["foo", "bar", "baz", "This is the end of the world!"];
 
 class FakeEmbeddingsWithOsDimension extends FakeEmbeddings {
   async embedDocuments(documents: string[]): Promise<number[][]> {
@@ -469,3 +469,50 @@ test.skip("Test fromExistingGraph multiple properties hybrid", async () => {
   await neo4jVectorStore.close();
   await existingGraph.close();
 });
+
+test.skip("Test escape lucene characters", async () => {
+  const url = process.env.NEO4J_URI as string;
+  const username = process.env.NEO4J_USERNAME as string;
+  const password = process.env.NEO4J_PASSWORD as string;
+
+  expect(url).toBeDefined();
+  expect(username).toBeDefined();
+  expect(password).toBeDefined();
+
+  const embeddings = new FakeEmbeddingsWithOsDimension();
+  // eslint-disable-next-line @typescript-eslint/no-explicit-any
+  const metadatas: any[] = [];
+
+  const neo4jVectorStore = await Neo4jVectorStore.fromTexts(
+    texts,
+    metadatas,
+    embeddings,
+    {
+      url,
+      username,
+      password,
+      preDeleteCollection: true,
+      searchType: "hybrid",
+    }
+  );
+
+  const output = await neo4jVectorStore.similaritySearch(
+    "This is the end of the world!",
+    2
+  );
+  console.log(output);
+  const expectedResult = [
+    new Document({
+      pageContent: "This is the end of the world!",
+      metadata: {},
+    }),
+    new Document({
+      pageContent: "baz",
+      metadata: {},
+    }),
+  ];
+
+  expect(output).toStrictEqual(expectedResult);
+  await dropVectorIndexes(neo4jVectorStore);
+  await neo4jVectorStore.close();
+});