Add minimum score criteria for AI search results (#1417)

* Add minimum score criteria for AI search results

* Adjust input to support precise filtering in different search modes.

* Resolve comparison issue

* Update class style

* Fix parsing

* Add test

* Lint

* Format

* Fix tests

---------

Co-authored-by: Pamela Fox <pamela.fox@gmail.com>

app/backend/approaches/approach.py

@@ -123,6 +123,8 @@ async def search(
         vectors: List[VectorQuery],
         use_semantic_ranker: bool,
         use_semantic_captions: bool,
+        minimum_search_score: Optional[float],
+        minimum_reranker_score: Optional[float],
     ) -> List[Document]:
         # Use semantic ranker if requested and if retrieval mode is text or hybrid (vectors + text)
         if use_semantic_ranker and query_text:
@@ -161,7 +163,17 @@ async def search(
                         reranker_score=document.get("@search.reranker_score"),
                     )
                 )
-        return documents
+
+            qualified_documents = [
+                doc
+                for doc in documents
+                if (
+                    (doc.score or 0) >= (minimum_search_score or 0)
+                    and (doc.reranker_score or 0) >= (minimum_reranker_score or 0)
+                )
+            ]
+
+        return qualified_documents
 
     def get_sources_content(
         self, results: List[Document], use_semantic_captions: bool, use_image_citation: bool

app/backend/approaches/chatreadretrieveread.py

@@ -89,6 +89,9 @@ async def run_until_final_call(
         has_vector = overrides.get("retrieval_mode") in ["vectors", "hybrid", None]
         use_semantic_captions = True if overrides.get("semantic_captions") and has_text else False
         top = overrides.get("top", 3)
+        minimum_search_score = overrides.get("minimum_search_score", 0.0)
+        minimum_reranker_score = overrides.get("minimum_reranker_score", 0.0)
+
         filter = self.build_filter(overrides, auth_claims)
         use_semantic_ranker = True if overrides.get("semantic_ranker") and has_text else False
 
@@ -149,7 +152,16 @@ async def run_until_final_call(
         if not has_text:
             query_text = None
 
-        results = await self.search(top, query_text, filter, vectors, use_semantic_ranker, use_semantic_captions)
+        results = await self.search(
+            top,
+            query_text,
+            filter,
+            vectors,
+            use_semantic_ranker,
+            use_semantic_captions,
+            minimum_search_score,
+            minimum_reranker_score,
+        )
 
         sources_content = self.get_sources_content(results, use_semantic_captions, use_image_citation=False)
         content = "\n".join(sources_content)

app/backend/approaches/chatreadretrievereadvision.py

@@ -87,6 +87,8 @@ async def run_until_final_call(
         vector_fields = overrides.get("vector_fields", ["embedding"])
         use_semantic_captions = True if overrides.get("semantic_captions") and has_text else False
         top = overrides.get("top", 3)
+        minimum_search_score = overrides.get("minimum_search_score", 0.0)
+        minimum_reranker_score = overrides.get("minimum_reranker_score", 0.0)
         filter = self.build_filter(overrides, auth_claims)
         use_semantic_ranker = True if overrides.get("semantic_ranker") and has_text else False
 
@@ -134,7 +136,16 @@ async def run_until_final_call(
         if not has_text:
             query_text = None
 
-        results = await self.search(top, query_text, filter, vectors, use_semantic_ranker, use_semantic_captions)
+        results = await self.search(
+            top,
+            query_text,
+            filter,
+            vectors,
+            use_semantic_ranker,
+            use_semantic_captions,
+            minimum_search_score,
+            minimum_reranker_score,
+        )
         sources_content = self.get_sources_content(results, use_semantic_captions, use_image_citation=True)
         content = "\n".join(sources_content)
 

app/backend/approaches/retrievethenread.py

@@ -86,6 +86,8 @@ async def run(
 
         use_semantic_captions = True if overrides.get("semantic_captions") and has_text else False
         top = overrides.get("top", 3)
+        minimum_search_score = overrides.get("minimum_search_score", 0.0)
+        minimum_reranker_score = overrides.get("minimum_reranker_score", 0.0)
         filter = self.build_filter(overrides, auth_claims)
         # If retrieval mode includes vectors, compute an embedding for the query
         vectors: list[VectorQuery] = []
@@ -95,7 +97,16 @@ async def run(
         # Only keep the text query if the retrieval mode uses text, otherwise drop it
         query_text = q if has_text else None
 
-        results = await self.search(top, query_text, filter, vectors, use_semantic_ranker, use_semantic_captions)
+        results = await self.search(
+            top,
+            query_text,
+            filter,
+            vectors,
+            use_semantic_ranker,
+            use_semantic_captions,
+            minimum_search_score,
+            minimum_reranker_score,
+        )
 
         user_content = [q]
 

app/backend/approaches/retrievethenreadvision.py

@@ -89,6 +89,8 @@ async def run(
 
         use_semantic_captions = True if overrides.get("semantic_captions") and has_text else False
         top = overrides.get("top", 3)
+        minimum_search_score = overrides.get("minimum_search_score", 0.0)
+        minimum_reranker_score = overrides.get("minimum_reranker_score", 0.0)
         filter = self.build_filter(overrides, auth_claims)
         use_semantic_ranker = overrides.get("semantic_ranker") and has_text
 
@@ -107,7 +109,16 @@ async def run(
         # Only keep the text query if the retrieval mode uses text, otherwise drop it
         query_text = q if has_text else None
 
-        results = await self.search(top, query_text, filter, vectors, use_semantic_ranker, use_semantic_captions)
+        results = await self.search(
+            top,
+            query_text,
+            filter,
+            vectors,
+            use_semantic_ranker,
+            use_semantic_captions,
+            minimum_search_score,
+            minimum_reranker_score,
+        )
 
         image_list: list[ChatCompletionContentPartImageParam] = []
         user_content: list[ChatCompletionContentPartParam] = [{"text": q, "type": "text"}]

app/frontend/src/api/models.ts

@@ -23,6 +23,8 @@ export type ChatAppRequestOverrides = {
     exclude_category?: string;
     top?: number;
     temperature?: number;
+    minimum_search_score?: number;
+    minimum_reranker_score?: number;
     prompt_template?: string;
     prompt_template_prefix?: string;
     prompt_template_suffix?: string;

app/frontend/src/pages/ask/Ask.module.css

@@ -56,6 +56,8 @@
 }
 
 .askSettingsSeparator {
+    display: flex;
+    flex-direction: column;
     margin-top: 15px;
 }
 

app/frontend/src/pages/ask/Ask.tsx

@@ -22,6 +22,8 @@ export function Component(): JSX.Element {
     const [promptTemplatePrefix, setPromptTemplatePrefix] = useState<string>("");
     const [promptTemplateSuffix, setPromptTemplateSuffix] = useState<string>("");
     const [temperature, setTemperature] = useState<number>(0.3);
+    const [minimumRerankerScore, setMinimumRerankerScore] = useState<number>(0);
+    const [minimumSearchScore, setMinimumSearchScore] = useState<number>(0);
     const [retrievalMode, setRetrievalMode] = useState<RetrievalMode>(RetrievalMode.Hybrid);
     const [retrieveCount, setRetrieveCount] = useState<number>(3);
     const [useSemanticRanker, setUseSemanticRanker] = useState<boolean>(true);
@@ -92,6 +94,8 @@ export function Component(): JSX.Element {
                         exclude_category: excludeCategory.length === 0 ? undefined : excludeCategory,
                         top: retrieveCount,
                         temperature: temperature,
+                        minimum_reranker_score: minimumRerankerScore,
+                        minimum_search_score: minimumSearchScore,
                         retrieval_mode: retrievalMode,
                         semantic_ranker: useSemanticRanker,
                         semantic_captions: useSemanticCaptions,
@@ -134,6 +138,13 @@ export function Component(): JSX.Element {
         setTemperature(newValue);
     };
 
+    const onMinimumSearchScoreChange = (_ev?: React.SyntheticEvent<HTMLElement, Event>, newValue?: string) => {
+        setMinimumSearchScore(parseFloat(newValue || "0"));
+    };
+
+    const onMinimumRerankerScoreChange = (_ev?: React.SyntheticEvent<HTMLElement, Event>, newValue?: string) => {
+        setMinimumRerankerScore(parseFloat(newValue || "0"));
+    };
     const onRetrieveCountChange = (_ev?: React.SyntheticEvent<HTMLElement, Event>, newValue?: string) => {
         setRetrieveCount(parseInt(newValue || "3"));
     };
@@ -259,6 +270,25 @@ export function Component(): JSX.Element {
                     snapToStep
                 />
 
+                <SpinButton
+                    className={styles.askSettingsSeparator}
+                    label="Minimum search score"
+                    min={0}
+                    step={0.01}
+                    defaultValue={minimumSearchScore.toString()}
+                    onChange={onMinimumSearchScoreChange}
+                />
+
+                <SpinButton
+                    className={styles.askSettingsSeparator}
+                    label="Minimum reranker score"
+                    min={1}
+                    max={4}
+                    step={0.1}
+                    defaultValue={minimumRerankerScore.toString()}
+                    onChange={onMinimumRerankerScoreChange}
+                />
+
                 <SpinButton
                     className={styles.askSettingsSeparator}
                     label="Retrieve this many search results:"

app/frontend/src/pages/chat/Chat.module.css

@@ -97,6 +97,8 @@
 }
 
 .chatSettingsSeparator {
+    display: flex;
+    flex-direction: column;
     margin-top: 15px;
 }
 

app/frontend/src/pages/chat/Chat.tsx

@@ -33,6 +33,8 @@ const Chat = () => {
     const [isConfigPanelOpen, setIsConfigPanelOpen] = useState(false);
     const [promptTemplate, setPromptTemplate] = useState<string>("");
     const [temperature, setTemperature] = useState<number>(0.3);
+    const [minimumRerankerScore, setMinimumRerankerScore] = useState<number>(0);
+    const [minimumSearchScore, setMinimumSearchScore] = useState<number>(0);
     const [retrieveCount, setRetrieveCount] = useState<number>(3);
     const [retrievalMode, setRetrievalMode] = useState<RetrievalMode>(RetrievalMode.Hybrid);
     const [useSemanticRanker, setUseSemanticRanker] = useState<boolean>(true);
@@ -147,6 +149,8 @@ const Chat = () => {
                         exclude_category: excludeCategory.length === 0 ? undefined : excludeCategory,
                         top: retrieveCount,
                         temperature: temperature,
+                        minimum_reranker_score: minimumRerankerScore,
+                        minimum_search_score: minimumSearchScore,
                         retrieval_mode: retrievalMode,
                         semantic_ranker: useSemanticRanker,
                         semantic_captions: useSemanticCaptions,
@@ -212,6 +216,14 @@ const Chat = () => {
         setTemperature(newValue);
     };
 
+    const onMinimumSearchScoreChange = (_ev?: React.SyntheticEvent<HTMLElement, Event>, newValue?: string) => {
+        setMinimumSearchScore(parseFloat(newValue || "0"));
+    };
+
+    const onMinimumRerankerScoreChange = (_ev?: React.SyntheticEvent<HTMLElement, Event>, newValue?: string) => {
+        setMinimumRerankerScore(parseFloat(newValue || "0"));
+    };
+
     const onRetrieveCountChange = (_ev?: React.SyntheticEvent<HTMLElement, Event>, newValue?: string) => {
         setRetrieveCount(parseInt(newValue || "3"));
     };
@@ -395,6 +407,25 @@ const Chat = () => {
                         snapToStep
                     />
 
+                    <SpinButton
+                        className={styles.chatSettingsSeparator}
+                        label="Minimum search score"
+                        min={0}
+                        step={0.01}
+                        defaultValue={minimumSearchScore.toString()}
+                        onChange={onMinimumSearchScoreChange}
+                    />
+
+                    <SpinButton
+                        className={styles.chatSettingsSeparator}
+                        label="Minimum reranker score"
+                        min={1}
+                        max={4}
+                        step={0.1}
+                        defaultValue={minimumRerankerScore.toString()}
+                        onChange={onMinimumRerankerScoreChange}
+                    />
+
                     <SpinButton
                         className={styles.chatSettingsSeparator}
                         label="Retrieve this many search results:"

tests/test_chatapproach.py

@@ -1,10 +1,18 @@
 import json
 
 import pytest
+from azure.core.credentials import AzureKeyCredential
+from azure.search.documents.aio import SearchClient
 from openai.types.chat import ChatCompletion
 
 from approaches.chatreadretrieveread import ChatReadRetrieveReadApproach
 
+from .mocks import MockAsyncSearchResultsIterator
+
+
+async def mock_search(*args, **kwargs):
+    return MockAsyncSearchResultsIterator(kwargs.get("search_text"), kwargs.get("vector_queries"))
+
 
 @pytest.fixture
 def chat_approach():
@@ -297,3 +305,52 @@ def test_get_messages_from_history_few_shots(chat_approach):
     assert messages[4]["role"] == "assistant"
     assert messages[5]["role"] == "user"
     assert messages[5]["content"] == user_query_request
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize(
+    "minimum_search_score,minimum_reranker_score,expected_result_count",
+    [
+        (0, 0, 1),
+        (0, 2, 1),
+        (0.03, 0, 1),
+        (0.03, 2, 1),
+        (1, 0, 0),
+        (0, 4, 0),
+        (1, 4, 0),
+    ],
+)
+async def test_search_results_filtering_by_scores(
+    monkeypatch, minimum_search_score, minimum_reranker_score, expected_result_count
+):
+
+    chat_approach = ChatReadRetrieveReadApproach(
+        search_client=SearchClient(endpoint="", index_name="", credential=AzureKeyCredential("")),
+        auth_helper=None,
+        openai_client=None,
+        chatgpt_model="gpt-35-turbo",
+        chatgpt_deployment="chat",
+        embedding_deployment="embeddings",
+        embedding_model="text-",
+        sourcepage_field="",
+        content_field="",
+        query_language="en-us",
+        query_speller="lexicon",
+    )
+
+    monkeypatch.setattr(SearchClient, "search", mock_search)
+
+    filtered_results = await chat_approach.search(
+        top=10,
+        query_text="test query",
+        filter=None,
+        vectors=[],
+        use_semantic_ranker=True,
+        use_semantic_captions=True,
+        minimum_search_score=minimum_search_score,
+        minimum_reranker_score=minimum_reranker_score,
+    )
+
+    assert (
+        len(filtered_results) == expected_result_count
+    ), f"Expected {expected_result_count} results with minimum_search_score={minimum_search_score} and minimum_reranker_score={minimum_reranker_score}"