BM25Retriever: escape regex metacharacters in getTermFrequency to prevent crashes (#8749)

Author: hntrl

.changeset/major-lamps-walk.md

@@ -0,0 +1,5 @@
+---
+"@langchain/community": patch
+---
+
+BM25Retriever: escape regex metacharacters in getTermFrequency to prevent crashes

libs/langchain-community/src/retrievers/tests/bm25.test.ts

@@ -1,6 +1,7 @@
 import { expect, test } from "@jest/globals";
 import { Document } from "@langchain/core/documents";
 import { BM25Retriever } from "../bm25.js";
+import { getTermFrequency } from "../../utils/@furkantoprak/bm25/BM25.js";
 
 test("BM25Retriever", async () => {
   const docs = [
@@ -25,3 +26,20 @@ test("BM25Retriever", async () => {
     "The quick brown fox jumps over the lazy dog"
   );
 });
+
+test("getTermFrequency escapes regex metacharacters", () => {
+  const corpus =
+    "**Version 1:** What is the country of origin for the person in question?";
+  const term = "**Version 1:**";
+
+  // Should not throw and should find at least one match
+  const freq = getTermFrequency(term, corpus);
+  expect(freq).toBeGreaterThanOrEqual(1);
+
+  // Also test other metacharacters
+  const corpus2 = "Does this match (maybe)? [yes] *stars* +plus+";
+  expect(getTermFrequency("(maybe)?", corpus2)).toBeGreaterThanOrEqual(1);
+  expect(getTermFrequency("[yes]", corpus2)).toBeGreaterThanOrEqual(1);
+  expect(getTermFrequency("*stars*", corpus2)).toBeGreaterThanOrEqual(1);
+  expect(getTermFrequency("+plus+", corpus2)).toBeGreaterThanOrEqual(1);
+});

libs/langchain-community/src/utils/@furkantoprak/bm25/BM25.ts

@@ -12,7 +12,10 @@ export const getWordCount = (corpus: string) => {
 
 /** Number of occurences of a word in a string. */
 export const getTermFrequency = (term: string, corpus: string) => {
-  return ((corpus || "").match(new RegExp(term, "g")) || []).length;
+  // Escape any RegExp metacharacters in the term so constructing a RegExp
+  // from user-provided or model-generated queries does not throw an error
+  const escaped = (term || "").replace(/[.*+?^${}()|[\]\\]/g, "\\$&");
+  return ((corpus || "").match(new RegExp(escaped, "g")) || []).length;
 };
 
 /** Inverse document frequency. */