From f5faf0c94fa2ca87db74244d8c1a970326905d2f Mon Sep 17 00:00:00 2001
From: Kevin Hu <kevinhu.sh@gmail.com>
Date: Tue, 25 Nov 2025 12:44:26 +0800
Subject: [PATCH] Feat: support operator in/not in for metadata filter.
 (#11503)

### What problem does this PR solve?

#11376 #11378

### Type of change

- [x] New Feature (non-breaking change which adds functionality)
---
 api/db/services/dialog_service.py |  2 ++
 rag/prompts/meta_filter.md        | 16 ++++++++++------
 2 files changed, 12 insertions(+), 6 deletions(-)

diff --git a/api/db/services/dialog_service.py b/api/db/services/dialog_service.py
index 0a09ea532..558ba1b0f 100644
--- a/api/db/services/dialog_service.py
+++ b/api/db/services/dialog_service.py
@@ -304,6 +304,8 @@ def meta_filter(metas: dict, filters: list[dict], logic: str = "and"):
             for conds in [
                 (operator == "contains", str(value).lower() in str(input).lower()),
                 (operator == "not contains", str(value).lower() not in str(input).lower()),
+                (operator == "in", str(input).lower() in str(value).lower()),
+                (operator == "not in", str(input).lower() not in str(value).lower()),
                 (operator == "start with", str(input).lower().startswith(str(value).lower())),
                 (operator == "end with", str(input).lower().endswith(str(value).lower())),
                 (operator == "empty", not input),
diff --git a/rag/prompts/meta_filter.md b/rag/prompts/meta_filter.md
index 7df3c4885..203291071 100644
--- a/rag/prompts/meta_filter.md
+++ b/rag/prompts/meta_filter.md
@@ -35,7 +35,7 @@ You are a metadata filtering condition generator. Analyze the user's question an
         - Value has no match in metadata
 
 5. **Example A**:
-   - User query: "上市日期七月份的有哪些商品，不要蓝色的"
+   - User query: "上市日期七月份的有哪些新品，不要蓝色的，只看鞋子和帽子"
    - Metadata: { "color": {...}, "listing_date": {...} }
    - Output: 
    {
@@ -43,19 +43,21 @@ You are a metadata filtering condition generator. Analyze the user's question an
         "conditions": [
           {"key": "listing_date", "value": "2025-07-01", "op": "≥"},
           {"key": "listing_date", "value": "2025-08-01", "op": "<"},
-          {"key": "color", "value": "blue", "op": "≠"}
+          {"key": "color", "value": "blue", "op": "≠"},
+          {"key": "category", "value": "shoes, hat", "op": "in"}
         ]
    }
 
 6. **Example B**:
-   - User query: "Both blue and red are acceptable."
-   - Metadata: { "color": {...}, "listing_date": {...} }
+   - User query: "It must be from China or India. Otherwise, it must not be blue or red."
+   - Metadata: { "color": {...}, "country": {...} }
+   - 
    - Output: 
    {
         "logic": "or",
         "conditions": [
-          {"key": "color", "value": "blue", "op": "="},
-          {"key": "color", "value": "red", "op": "="}
+          {"key": "color", "value": "blue, red", "op": "not in"},
+          {"key": "country", "value": "china, india", "op": "in"},
         ]
    }
 
@@ -94,6 +96,8 @@ You are a metadata filtering condition generator. Analyze the user's question an
             "enum": [
               "contains",
               "not contains",
+              "in",
+              "not in",
               "start with",
               "end with",
               "empty",