Skip to content

RAGFlow is an agent-native database: a Retrieval-Augmented Generation (RAG) engine that grew into context infrastructure.

Refactor: explictly use keywords_similarity_weight (#20379)

Wang Qif88c1d119ffe
diff --git a/docs/references/http_api_reference.md b/docs/references/http_api_reference.md
index 5a5dd374bc..88da403e30 100644
--- a/docs/references/http_api_reference.md
+++ b/docs/references/http_api_reference.md
@@ -2472,7 +2472,8 @@ Retrieves chunks from specified datasets.
- `"page"`: `integer`
- `"page_size"`: `integer`
- `"similarity_threshold"`: `float`
- - `"vector_similarity_weight"`: `float`
+ - `"keywords_similarity_weight"`: `float`
+ - `"vector_similarity_weight"`: `float` (legacy fallback)
- `"top_k"`: `integer` (deprecated; use `"knn_top_k"`)
- `"knn_top_k"`: `integer`
- `"knn_num_candidates"`: `integer`
@@ -2533,8 +2534,10 @@ curl --request POST \
The maximum number of chunks on each page. Defaults to `30`.
- `"similarity_threshold"`: (*Body parameter*)
The minimum similarity score. Defaults to `0.2`.
+- `"keywords_similarity_weight"`: (*Body parameter*), `float`
+ The weight of term similarity. Defaults to `0.7`; the vector cosine similarity weight is calculated as `1 - keywords_similarity_weight`.
- `"vector_similarity_weight"`: (*Body parameter*), `float`
- The weight of vector cosine similarity. Defaults to `0.3`. If x represents the weight of vector cosine similarity, then (1 - x) is the term similarity weight.
+ Legacy fallback used when `"keywords_similarity_weight"` is omitted. If both fields are provided, they must sum to `1`.
- `"top_k"`: (*Body parameter*), `integer`
**Deprecated.** An alias for `"knn_top_k"`. If both parameters are provided, `"knn_top_k"` takes precedence.
- `"knn_top_k"`: (*Body parameter*), `integer`
@@ -2713,7 +2716,8 @@ curl --request POST \
- `"serply_api_key"`: `string` The [Serply](https://serply.io) API key. Set `web_search_provider` to `"serply"` when using this field. See the [Serply documentation](https://serply.io/docs) for details.
- `"youcom_api_key"`: `string` The You.com API key. Set `web_search_provider` to `"youcom"` when using this field. Optional: You.com serves a rate-limited keyless endpoint, so `"youcom"` works with this field omitted, and a key lifts those limits.
- `"similarity_threshold"`: (*Body parameter*), `float`
-- `"vector_similarity_weight"`: (*Body parameter*), `float`
+- `"keywords_similarity_weight"`: (*Body parameter*), `float`
+- `"vector_similarity_weight"`: (*Body parameter*), `float`, legacy fallback. If both weights are provided, they must sum to `1`.
- `"top_n"`: (*Body parameter*), `int`
- `"top_k"`: (*Body parameter*), `int`
- `"rerank_id"`: (*Body parameter*), `string`
@@ -2761,6 +2765,7 @@ Success:
"rerank_id": "",
"similarity_threshold": 0.2,
"vector_similarity_weight": 0.3,
+ "keywords_similarity_weight": 0.7,
"top_n": 6,
"prompt_type": "simple",
"status": "1",
@@ -2828,7 +2833,7 @@ curl --request PUT \
"quote":true
},
"similarity_threshold":0.2,
- "vector_similarity_weight":0.3,
+ "keywords_similarity_weight":0.7,
"top_n":6,
"top_k":1024,
"rerank_id":""
@@ -2861,7 +2866,8 @@ curl --request PUT \
Similar to the presence penalty, this reduces the model's tendency to repeat the same words frequently. Defaults to `0.7`.
- `"prompt_config"`: (*Body parameter*), `object`
- `"similarity_threshold"`: (*Body parameter*), `float`
-- `"vector_similarity_weight"`: (*Body parameter*), `float`
+- `"keywords_similarity_weight"`: (*Body parameter*), `float`
+- `"vector_similarity_weight"`: (*Body parameter*), `float`, legacy fallback. If both weights are provided, they must sum to `1`.
- `"top_n"`: (*Body parameter*), `int`
- `"top_k"`: (*Body parameter*), `int`
- `"rerank_id"`: (*Body parameter*), `string`
@@ -2898,6 +2904,7 @@ Success: returns the full updated chat assistant object.
},
"similarity_threshold": 0.2,
"vector_similarity_weight": 0.3,
+ "keywords_similarity_weight": 0.7,
"top_n": 6,
"top_k": 1024,
"rerank_id": "",
@@ -2978,6 +2985,7 @@ Success:
"rerank_id": "",
"similarity_threshold": 0.2,
"vector_similarity_weight": 0.3,
+ "keywords_similarity_weight": 0.7,
"top_n": 6,
"status": "1",
"tenant_id": "69736c5e723611efb51b0242ac120007",
@@ -3260,6 +3268,7 @@ Success:
"rerank_id": "",
"similarity_threshold": 0.2,
"vector_similarity_weight": 0.3,
+ "keywords_similarity_weight": 0.7,
"top_n": 6,
"prompt_type": "simple",
"status": "1",
diff --git a/internal/handler/chat.go b/internal/handler/chat.go
index a7b11cd1e7..238c0af206 100644
--- a/internal/handler/chat.go
+++ b/internal/handler/chat.go
@@ -364,33 +364,34 @@ func (h *ChatHandler) GetChat(c *gin.Context) {
// Build response (same as Python _build_chat_response)
// The service already returns GetChatResponse with DatasetIDs and KBNames
result := map[string]interface{}{
- "id": chat.ID,
- "tenant_id": chat.TenantID,
- "name": chat.Name,
- "description": chat.Description,
- "icon": chat.Icon,
- "language": chat.Language,
- "llm_id": chat.LLMID,
- "llm_setting": chat.LLMSetting,
- "prompt_type": chat.PromptType,
- "prompt_config": chat.PromptConfig,
- "meta_data_filter": chat.MetaDataFilter,
- "similarity_threshold": chat.SimilarityThreshold,
- "vector_similarity_weight": chat.VectorSimilarityWeight,
- "top_n": chat.TopN,
- "rerank_candidates_count": chat.RerankCandidatesCount,
- "top_k": chat.TopK,
- "do_refer": chat.DoRefer,
- "rerank_id": chat.RerankID,
- "dataset_ids": chat.DatasetIDs,
- "kb_names": chat.KBNames,
- "status": chat.Status,
- "create_time": chat.CreateTime,
- "create_date": chat.CreateDate,
- "update_time": chat.UpdateTime,
- "update_date": chat.UpdateDate,
- "tenant_llm_id": chat.TenantLLMID,
- "tenant_rerank_id": chat.TenantRerankID,
+ "id": chat.ID,
+ "tenant_id": chat.TenantID,
+ "name": chat.Name,
+ "description": chat.Description,
+ "icon": chat.Icon,
+ "language": chat.Language,
+ "llm_id": chat.LLMID,
+ "llm_setting": chat.LLMSetting,
+ "prompt_type": chat.PromptType,
+ "prompt_config": chat.PromptConfig,
+ "meta_data_filter": chat.MetaDataFilter,
+ "keywords_similarity_weight": 1 - chat.VectorSimilarityWeight,
+ "similarity_threshold": chat.SimilarityThreshold,
+ "vector_similarity_weight": chat.VectorSimilarityWeight,
+ "top_n": chat.TopN,
+ "rerank_candidates_count": chat.RerankCandidatesCount,
+ "top_k": chat.TopK,
+ "do_refer": chat.DoRefer,
+ "rerank_id": chat.RerankID,
+ "dataset_ids": chat.DatasetIDs,
+ "kb_names": chat.KBNames,
+ "status": chat.Status,
+ "create_time": chat.CreateTime,
+ "create_date": chat.CreateDate,
+ "update_time": chat.UpdateTime,
+ "update_date": chat.UpdateDate,
+ "tenant_llm_id": chat.TenantLLMID,
+ "tenant_rerank_id": chat.TenantRerankID,
}
// Return success response
diff --git a/internal/handler/dataset.go b/internal/handler/dataset.go
index 8668c66416..2e09b00f46 100644
--- a/internal/handler/dataset.go
+++ b/internal/handler/dataset.go
@@ -1199,10 +1199,20 @@ func (h *DatasetsHandler) SearchDataset(c *gin.Context) {
}
func validateSearchDatasetsRequest(req *service.SearchDatasetsRequest) error {
+ vectorSimilarityWeight, err := service.ResolveVectorSimilarityWeight(req.KeywordsSimilarityWeight, req.VectorSimilarityWeight)
+ if err != nil {
+ return err
+ }
+ req.VectorSimilarityWeight = vectorSimilarityWeight
return validateSearchParams(req.Page, req.PageSize, req.Size, req.KNNTopK, req.TopK, req.KNNNumCandidates, req.SimilarityThreshold, req.VectorSimilarityWeight)
}
func validateSearchDatasetRequest(req *service.SearchDatasetRequest) error {
+ vectorSimilarityWeight, err := service.ResolveVectorSimilarityWeight(req.KeywordsSimilarityWeight, req.VectorSimilarityWeight)
+ if err != nil {
+ return err
+ }
+ req.VectorSimilarityWeight = vectorSimilarityWeight
return validateSearchParams(req.Page, req.PageSize, req.Size, req.KNNTopK, req.TopK, req.KNNNumCandidates, req.SimilarityThreshold, req.VectorSimilarityWeight)
}
diff --git a/internal/handler/dataset_search_test.go b/internal/handler/dataset_search_test.go
index 2ccc795a58..a7f83c4a6b 100644
--- a/internal/handler/dataset_search_test.go
+++ b/internal/handler/dataset_search_test.go
@@ -4,6 +4,7 @@ import (
"context"
"encoding/json"
"errors"
+ "math"
"net/http"
"net/http/httptest"
"strings"
@@ -50,7 +51,7 @@ func TestDatasetsHandlerSearchDataset(t *testing.T) {
h := &DatasetsHandler{searchDatasetService: fake}
rec := httptest.NewRecorder()
- req := httptest.NewRequest(http.MethodPost, "/api/v1/datasets/ds-1/search", strings.NewReader(`{"question":"hello","document_ids":["doc-1"],"page_size":9,"metadata_condition":{"logic":"and","conditions":[{"name":"author","comparison_operator":"=","value":"Luo"}]},"knn_top_k":7,"knn_num_candidates":14}`))
+ req := httptest.NewRequest(http.MethodPost, "/api/v1/datasets/ds-1/search", strings.NewReader(`{"question":"hello","document_ids":["doc-1"],"page_size":9,"metadata_condition":{"logic":"and","conditions":[{"name":"author","comparison_operator":"=","value":"Luo"}]},"knn_top_k":7,"knn_num_candidates":14,"keywords_similarity_weight":0.7}`))
req.Header.Set("Content-Type", "application/json")
c, _ := gin.CreateTestContext(rec)
c.Request = req
@@ -74,6 +75,9 @@ func TestDatasetsHandlerSearchDataset(t *testing.T) {
if fake.req.KNNTopK == nil || *fake.req.KNNTopK != 7 || fake.req.KNNNumCandidates == nil || *fake.req.KNNNumCandidates != 14 {
t.Fatalf("KNN parameters = (%v, %v), want (7, 14)", fake.req.KNNTopK, fake.req.KNNNumCandidates)
}
+ if fake.req.VectorSimilarityWeight == nil || math.Abs(*fake.req.VectorSimilarityWeight-0.3) > 1e-9 {
+ t.Fatalf("vector_similarity_weight = %v, want 0.3", fake.req.VectorSimilarityWeight)
+ }
if len(fake.req.ToSearchDatasetsRequest("ds-1").DatasetIDs) != 1 {
t.Fatal("request conversion failed")
}
diff --git a/internal/handler/searchbot.go b/internal/handler/searchbot.go
index 6822672106..85e2096e80 100644
--- a/internal/handler/searchbot.go
+++ b/internal/handler/searchbot.go
@@ -50,23 +50,24 @@ type SearchBotMindMapRequest struct {
// SearchBotRetrievalTestRequest is the request body for POST /api/v1/searchbots/retrieval_test.
type SearchBotRetrievalTestRequest struct {
- KbIDs common.StringSlice `json:"kb_ids" binding:"required"`
- Question string `json:"question" binding:"required"`
- Page *int `json:"page,omitempty"`
- Size *int `json:"size,omitempty"`
- RerankCandidatesCount *int `json:"rerank_candidates_count,omitempty"`
- DocIDs []string `json:"doc_ids,omitempty"`
- UseKG *bool `json:"use_kg,omitempty"`
- TopK *int `json:"top_k,omitempty"`
- CrossLanguages []string `json:"cross_languages,omitempty"`
- SearchID *string `json:"search_id,omitempty"`
- MetaDataFilter map[string]interface{} `json:"meta_data_filter,omitempty"`
- TenantRerankID *string `json:"tenant_rerank_id,omitempty"`
- RerankID *string `json:"rerank_id,omitempty"`
- Keyword *bool `json:"keyword,omitempty"`
- Highlight *bool `json:"highlight,omitempty"`
- SimilarityThreshold *float64 `json:"similarity_threshold,omitempty"`
- VectorSimilarityWeight *float64 `json:"vector_similarity_weight,omitempty"`
+ KbIDs common.StringSlice `json:"kb_ids" binding:"required"`
+ Question string `json:"question" binding:"required"`
+ Page *int `json:"page,omitempty"`
+ Size *int `json:"size,omitempty"`
+ RerankCandidatesCount *int `json:"rerank_candidates_count,omitempty"`
+ DocIDs []string `json:"doc_ids,omitempty"`
+ UseKG *bool `json:"use_kg,omitempty"`
+ TopK *int `json:"top_k,omitempty"`
+ CrossLanguages []string `json:"cross_languages,omitempty"`
+ SearchID *string `json:"search_id,omitempty"`
+ MetaDataFilter map[string]interface{} `json:"meta_data_filter,omitempty"`
+ TenantRerankID *string `json:"tenant_rerank_id,omitempty"`
+ RerankID *string `json:"rerank_id,omitempty"`
+ Keyword *bool `json:"keyword,omitempty"`
+ Highlight *bool `json:"highlight,omitempty"`
+ SimilarityThreshold *float64 `json:"similarity_threshold,omitempty"`
+ VectorSimilarityWeight *float64 `json:"vector_similarity_weight,omitempty"`
+ KeywordsSimilarityWeight *float64 `json:"keywords_similarity_weight,omitempty"`
}
// UnmarshalJSON accepts both kb_id (Python API) and kb_ids (Go compatibility).
@@ -194,6 +195,13 @@ func (h *SearchBotHandler) RetrievalTest(c *gin.Context) {
return
}
+ vectorSimilarityWeight, err := service.ResolveVectorSimilarityWeight(req.KeywordsSimilarityWeight, req.VectorSimilarityWeight)
+ if err != nil {
+ common.ResponseWithHttpCodeData(c, http.StatusBadRequest, common.CodeArgumentError, nil, err.Error())
+ return
+ }
+ req.VectorSimilarityWeight = vectorSimilarityWeight
+
applyRetrievalDefaults(&req)
if req.TopK != nil && *req.TopK <= 0 {
diff --git a/internal/service/chat.go b/internal/service/chat.go
index 1a488838d1..7e1b70a5f8 100644
--- a/internal/service/chat.go
+++ b/internal/service/chat.go
@@ -222,6 +222,9 @@ func (s *ChatService) Create(ctx context.Context, userID string, req map[string]
return nil, common.CodeDataError, err
}
req["name"] = name
+ if err := NormalizeSimilarityWeights(req); err != nil {
+ return nil, common.CodeDataError, err
+ }
if datasetIDsValue, ok := req["dataset_ids"]; ok {
kbIDs, err := s.validateCreateDatasetIDs(ctx, datasetIDsValue, userID)
@@ -597,6 +600,7 @@ func (s *ChatService) buildCreateChatResponse(ctx context.Context, chat *entity.
delete(data, "kb_ids")
data["kb_names"] = kbNames
data["meta_data_filter"] = normalizeMetaDataFilter(chat.MetaDataFilter)
+ data["keywords_similarity_weight"] = 1 - chat.VectorSimilarityWeight
return data, nil
}
@@ -850,6 +854,9 @@ func (s *ChatService) updateChatREST(ctx context.Context, userID, chatID string,
if !patch && isTruthy(req["tenant_id"]) {
return nil, errors.New("`tenant_id` must not be provided")
}
+ if err := NormalizeSimilarityWeights(req); err != nil {
+ return nil, err
+ }
if value, ok := req["name"]; ok {
name, shouldSet, err := validateRESTChatName(value, !patch)
@@ -1141,33 +1148,34 @@ func mergeJSONMap(base entity.JSONMap, patch map[string]interface{}) entity.JSON
func (s *ChatService) buildRESTChatResponse(ctx context.Context, chat *entity.Chat) map[string]interface{} {
kbNames, datasetIDs := s.getDatasetNamesAndIDs(ctx, chat.KBIDs)
return map[string]interface{}{
- "id": chat.ID,
- "tenant_id": chat.TenantID,
- "name": chat.Name,
- "description": chat.Description,
- "icon": chat.Icon,
- "language": chat.Language,
- "llm_id": chat.LLMID,
- "tenant_llm_id": chat.TenantLLMID,
- "llm_setting": chat.LLMSetting,
- "prompt_type": chat.PromptType,
- "prompt_config": chat.PromptConfig,
- "meta_data_filter": normalizeMetaDataFilter(chat.MetaDataFilter),
- "similarity_threshold": chat.SimilarityThreshold,
- "vector_similarity_weight": chat.VectorSimilarityWeight,
- "top_n": chat.TopN,
- "rerank_candidates_count": chat.RerankCandidatesCount,
- "top_k": chat.TopK,
- "do_refer": chat.DoRefer,
- "rerank_id": chat.RerankID,
- "tenant_rerank_id": chat.TenantRerankID,
- "dataset_ids": datasetIDs,
- "kb_names": kbNames,
- "status": chat.Status,
- "create_time": chat.CreateTime,
- "create_date": chat.CreateDate,
- "update_time": chat.UpdateTime,
- "update_date": chat.UpdateDate,
+ "id": chat.ID,
+ "tenant_id": chat.TenantID,
+ "name": chat.Name,
+ "description": chat.Description,
+ "icon": chat.Icon,
+ "language": chat.Language,
+ "llm_id": chat.LLMID,
+ "tenant_llm_id": chat.TenantLLMID,
+ "llm_setting": chat.LLMSetting,
+ "prompt_type": chat.PromptType,
+ "prompt_config": chat.PromptConfig,
+ "meta_data_filter": normalizeMetaDataFilter(chat.MetaDataFilter),
+ "similarity_threshold": chat.SimilarityThreshold,
+ "vector_similarity_weight": chat.VectorSimilarityWeight,
+ "top_n": chat.TopN,
+ "keywords_similarity_weight": 1 - chat.VectorSimilarityWeight,
+ "rerank_candidates_count": chat.RerankCandidatesCount,
+ "top_k": chat.TopK,
+ "do_refer": chat.DoRefer,
+ "rerank_id": chat.RerankID,
+ "tenant_rerank_id": chat.TenantRerankID,
+ "dataset_ids": datasetIDs,
+ "kb_names": kbNames,
+ "status": chat.Status,
+ "create_time": chat.CreateTime,
+ "create_date": chat.CreateDate,
+ "update_time": chat.UpdateTime,
+ "update_date": chat.UpdateDate,
}
}
diff --git a/internal/service/chat_rest_update_test.go b/internal/service/chat_rest_update_test.go
index 9d214e8f30..7ce8fbcadb 100644
--- a/internal/service/chat_rest_update_test.go
+++ b/internal/service/chat_rest_update_test.go
@@ -1,6 +1,7 @@
package service
import (
+ "math"
"strings"
"testing"
@@ -286,6 +287,45 @@ func TestChatServiceUpdateChatAcceptsMetaDataFilterObject(t *testing.T) {
}
}
+func TestChatServiceUpdateChatNormalizesKeywordsSimilarityWeight(t *testing.T) {
+ db := setupChatRESTUpdateServiceTestDB(t)
+ createChatRESTUpdateServiceTestChat(t, db, "chat-1", "user-1")
+
+ svc := NewChatService()
+ resp, err := svc.UpdateChat(t.Context(), "user-1", "chat-1", map[string]interface{}{
+ "name": "chat-chat-1",
+ "keywords_similarity_weight": 0.7,
+ })
+ if err != nil {
+ t.Fatalf("UpdateChat failed: %v", err)
+ }
+
+ chat, err := svc.chatDAO.GetByID(t.Context(), dao.DB, "chat-1")
+ if err != nil {
+ t.Fatalf("failed to fetch chat: %v", err)
+ }
+ if math.Abs(chat.VectorSimilarityWeight-0.3) > similarityWeightTolerance {
+ t.Fatalf("vector_similarity_weight = %v, want 0.3", chat.VectorSimilarityWeight)
+ }
+ if got, ok := resp["keywords_similarity_weight"].(float64); !ok || math.Abs(got-0.7) > similarityWeightTolerance {
+ t.Fatalf("keywords_similarity_weight = %#v, want 0.7", resp["keywords_similarity_weight"])
+ }
+}
+
+func TestChatServiceUpdateChatRejectsMismatchedSimilarityWeights(t *testing.T) {
+ db := setupChatRESTUpdateServiceTestDB(t)
+ createChatRESTUpdateServiceTestChat(t, db, "chat-1", "user-1")
+
+ _, err := NewChatService().UpdateChat(t.Context(), "user-1", "chat-1", map[string]interface{}{
+ "name": "chat-chat-1",
+ "keywords_similarity_weight": 0.7,
+ "vector_similarity_weight": 0.4,
+ })
+ if err == nil || !strings.Contains(err.Error(), "must sum to 1") {
+ t.Fatalf("expected weight sum error, got %v", err)
+ }
+}
+
func TestChatServiceUpdateChatBackfillsNilMetaDataFilter(t *testing.T) {
db := setupChatRESTUpdateServiceTestDB(t)
createChatRESTUpdateServiceTestChat(t, db, "chat-1", "user-1")
diff --git a/internal/service/dataset_types.go b/internal/service/dataset_types.go
index f869165ddf..21ad9c1f76 100644
--- a/internal/service/dataset_types.go
+++ b/internal/service/dataset_types.go
@@ -37,29 +37,30 @@ type EmbeddingCheckResponse struct {
// SearchDatasetsRequest is the request structure for searching chunks across datasets.
type SearchDatasetsRequest struct {
- DatasetIDs []string `json:"dataset_ids" binding:"required"`
- Question string `json:"question" binding:"required"`
- Page *int `json:"page,omitempty"`
- PageSize *int `json:"page_size,omitempty"`
- Size *int `json:"size,omitempty"`
- RerankCandidatesCount *int `json:"rerank_candidates_count,omitempty"`
- DocumentIDs []string `json:"document_ids,omitempty"`
- DocIDs []string `json:"doc_ids,omitempty"`
- UseKG *bool `json:"use_kg,omitempty"`
- KNNTopK *int `json:"knn_top_k,omitempty"`
- TopK *int `json:"top_k,omitempty"` // Legacy alias for knn_top_k.
- KNNNumCandidates *int `json:"knn_num_candidates,omitempty"`
- CrossLanguages []string `json:"cross_languages,omitempty"`
- SearchID *string `json:"search_id,omitempty"`
- MetadataCondition map[string]interface{} `json:"metadata_condition,omitempty"`
- MetadataFilter map[string]interface{} `json:"meta_data_filter,omitempty"`
- RerankID *string `json:"rerank_id,omitempty"`
- Keyword *bool `json:"keyword,omitempty"`
- Highlight *bool `json:"highlight,omitempty"`
- SimilarityThreshold *float64 `json:"similarity_threshold,omitempty"`
- VectorSimilarityWeight *float64 `json:"vector_similarity_weight,omitempty"`
- IncludeCompiledChunks *bool `json:"include_knowledge_compilation,omitempty"`
- ForceRefresh bool `json:"force_refresh"`
+ DatasetIDs []string `json:"dataset_ids" binding:"required"`
+ Question string `json:"question" binding:"required"`
+ Page *int `json:"page,omitempty"`
+ PageSize *int `json:"page_size,omitempty"`
+ Size *int `json:"size,omitempty"`
+ RerankCandidatesCount *int `json:"rerank_candidates_count,omitempty"`
+ DocumentIDs []string `json:"document_ids,omitempty"`
+ DocIDs []string `json:"doc_ids,omitempty"`
+ UseKG *bool `json:"use_kg,omitempty"`
+ KNNTopK *int `json:"knn_top_k,omitempty"`
+ TopK *int `json:"top_k,omitempty"` // Legacy alias for knn_top_k.
+ KNNNumCandidates *int `json:"knn_num_candidates,omitempty"`
+ CrossLanguages []string `json:"cross_languages,omitempty"`
+ SearchID *string `json:"search_id,omitempty"`
+ MetadataCondition map[string]interface{} `json:"metadata_condition,omitempty"`
+ MetadataFilter map[string]interface{} `json:"meta_data_filter,omitempty"`
+ RerankID *string `json:"rerank_id,omitempty"`
+ Keyword *bool `json:"keyword,omitempty"`
+ Highlight *bool `json:"highlight,omitempty"`
+ SimilarityThreshold *float64 `json:"similarity_threshold,omitempty"`
+ KeywordsSimilarityWeight *float64 `json:"keywords_similarity_weight,omitempty"`
+ VectorSimilarityWeight *float64 `json:"vector_similarity_weight,omitempty"`
+ IncludeCompiledChunks *bool `json:"include_knowledge_compilation,omitempty"`
+ ForceRefresh bool `json:"force_refresh"`
}
// SearchDatasetsResponse is the response structure for dataset search results.
@@ -72,26 +73,27 @@ type SearchDatasetsResponse struct {
// SearchDatasetRequest is the request structure for searching chunks within one dataset.
type SearchDatasetRequest struct {
- Question string `json:"question"`
- Page *int `json:"page,omitempty"`
- PageSize *int `json:"page_size,omitempty"`
- Size *int `json:"size,omitempty"`
- RerankCandidatesCount *int `json:"rerank_candidates_count,omitempty"`
- DocumentIDs []string `json:"document_ids,omitempty"`
- DocIDs []string `json:"doc_ids,omitempty"`
- UseKG *bool `json:"use_kg,omitempty"`
- KNNTopK *int `json:"knn_top_k,omitempty"`
- TopK *int `json:"top_k,omitempty"` // Legacy alias for knn_top_k.
- KNNNumCandidates *int `json:"knn_num_candidates,omitempty"`
- CrossLanguages []string `json:"cross_languages,omitempty"`
- SearchID *string `json:"search_id,omitempty"`
- MetadataCondition map[string]interface{} `json:"metadata_condition,omitempty"`
- MetadataFilter map[string]interface{} `json:"meta_data_filter,omitempty"`
- RerankID *string `json:"rerank_id,omitempty"`
- Keyword *bool `json:"keyword,omitempty"`
- SimilarityThreshold *float64 `json:"similarity_threshold,omitempty"`
- VectorSimilarityWeight *float64 `json:"vector_similarity_weight,omitempty"`
- IncludeCompiledChunks *bool `json:"include_knowledge_compilation,omitempty"`
+ Question string `json:"question"`
+ Page *int `json:"page,omitempty"`
+ PageSize *int `json:"page_size,omitempty"`
+ Size *int `json:"size,omitempty"`
+ RerankCandidatesCount *int `json:"rerank_candidates_count,omitempty"`
+ DocumentIDs []string `json:"document_ids,omitempty"`
+ DocIDs []string `json:"doc_ids,omitempty"`
+ UseKG *bool `json:"use_kg,omitempty"`
+ KNNTopK *int `json:"knn_top_k,omitempty"`
+ TopK *int `json:"top_k,omitempty"` // Legacy alias for knn_top_k.
+ KNNNumCandidates *int `json:"knn_num_candidates,omitempty"`
+ CrossLanguages []string `json:"cross_languages,omitempty"`
+ SearchID *string `json:"search_id,omitempty"`
+ MetadataCondition map[string]interface{} `json:"metadata_condition,omitempty"`
+ MetadataFilter map[string]interface{} `json:"meta_data_filter,omitempty"`
+ RerankID *string `json:"rerank_id,omitempty"`
+ Keyword *bool `json:"keyword,omitempty"`
+ SimilarityThreshold *float64 `json:"similarity_threshold,omitempty"`
+ KeywordsSimilarityWeight *float64 `json:"keywords_similarity_weight,omitempty"`
+ VectorSimilarityWeight *float64 `json:"vector_similarity_weight,omitempty"`
+ IncludeCompiledChunks *bool `json:"include_knowledge_compilation,omitempty"`
}
// ToSearchDatasetsRequest converts a single-dataset search request into the multi-dataset form.
@@ -100,27 +102,28 @@ func (req *SearchDatasetRequest) ToSearchDatasetsRequest(datasetID string) *Sear
return &SearchDatasetsRequest{DatasetIDs: []string{datasetID}}
}
return &SearchDatasetsRequest{
- DatasetIDs: []string{datasetID},
- Question: req.Question,
- Page: req.Page,
- PageSize: req.PageSize,
- Size: req.Size,
- RerankCandidatesCount: req.RerankCandidatesCount,
- DocumentIDs: req.DocumentIDs,
- DocIDs: req.DocIDs,
- UseKG: req.UseKG,
- KNNTopK: req.KNNTopK,
- TopK: req.TopK,
- KNNNumCandidates: req.KNNNumCandidates,
- CrossLanguages: req.CrossLanguages,
- SearchID: req.SearchID,
- MetadataCondition: req.MetadataCondition,
- MetadataFilter: req.MetadataFilter,
- RerankID: req.RerankID,
- Keyword: req.Keyword,
- SimilarityThreshold: req.SimilarityThreshold,
- VectorSimilarityWeight: req.VectorSimilarityWeight,
- IncludeCompiledChunks: req.IncludeCompiledChunks,
+ DatasetIDs: []string{datasetID},
+ Question: req.Question,
+ Page: req.Page,
+ PageSize: req.PageSize,
+ Size: req.Size,
+ RerankCandidatesCount: req.RerankCandidatesCount,
+ DocumentIDs: req.DocumentIDs,
+ DocIDs: req.DocIDs,
+ UseKG: req.UseKG,
+ KNNTopK: req.KNNTopK,
+ TopK: req.TopK,
+ KNNNumCandidates: req.KNNNumCandidates,
+ CrossLanguages: req.CrossLanguages,
+ SearchID: req.SearchID,
+ MetadataCondition: req.MetadataCondition,
+ MetadataFilter: req.MetadataFilter,
+ RerankID: req.RerankID,
+ Keyword: req.Keyword,
+ SimilarityThreshold: req.SimilarityThreshold,
+ KeywordsSimilarityWeight: req.KeywordsSimilarityWeight,
+ VectorSimilarityWeight: req.VectorSimilarityWeight,
+ IncludeCompiledChunks: req.IncludeCompiledChunks,
}
}
diff --git a/internal/service/search.go b/internal/service/search.go
index 871bf27385..9be3589618 100644
--- a/internal/service/search.go
+++ b/internal/service/search.go
@@ -458,8 +458,10 @@ func BuildAskStreamOptions(searchID string, searchConfig map[string]interface{})
if value, ok := floatFromSearchConfig(searchConfig["similarity_threshold"]); ok {
opts.SimilarityThreshold = &value
}
- if value, ok := floatFromSearchConfig(searchConfig["vector_similarity_weight"]); ok {
- opts.VectorSimilarityWeight = &value
+ keywordsSimilarityWeight, _ := similarityWeightFromMap(searchConfig, "keywords_similarity_weight")
+ vectorSimilarityWeight, _ := similarityWeightFromMap(searchConfig, "vector_similarity_weight")
+ if value, err := ResolveVectorSimilarityWeight(keywordsSimilarityWeight, vectorSimilarityWeight); err == nil && value != nil {
+ opts.VectorSimilarityWeight = value
}
if llmSetting, ok := searchConfigMapValue(searchConfig["llm_setting"]); ok {
opts.Temperature = generationFloat(llmSetting, "temperature", DefaultAskTemperature)
@@ -613,6 +615,9 @@ func (s *SearchService) UpdateSearch(ctx context.Context, userID string, searchI
return nil, fmt.Errorf("duplicated search name")
}
}
+ if err := NormalizeSimilarityWeights(req.SearchConfig); err != nil {
+ return nil, err
+ }
// Step 4: Merge search_config
// Python: req["search_config"] = {**current_config, **new_config}
diff --git a/internal/service/search_test.go b/internal/service/search_test.go
index 4a474dd613..1bc640462c 100644
--- a/internal/service/search_test.go
+++ b/internal/service/search_test.go
@@ -17,6 +17,7 @@
package service
import (
+ "math"
"strings"
"testing"
@@ -103,7 +104,7 @@ func TestSearchServiceCreateAndUpdateRoundTrip(t *testing.T) {
// The owner can update name + merge config.
req = &UpdateSearchRequest{
Name: "Updated Name",
- SearchConfig: map[string]interface{}{"summary": true},
+ SearchConfig: map[string]interface{}{"summary": true, "keywords_similarity_weight": 0.7},
}
updated, err := NewSearchService().UpdateSearch(ctx, "tenant-1", created.SearchID, req)
if err != nil {
@@ -115,6 +116,11 @@ func TestSearchServiceCreateAndUpdateRoundTrip(t *testing.T) {
if updated.SearchConfig["summary"] != true {
t.Fatalf("expected merged search_config, got %#v", updated.SearchConfig)
}
+ keywordsWeight, keywordsOK := updated.SearchConfig["keywords_similarity_weight"].(float64)
+ vectorWeight, vectorOK := updated.SearchConfig["vector_similarity_weight"].(float64)
+ if !keywordsOK || !vectorOK || math.Abs(keywordsWeight-0.7) > similarityWeightTolerance || math.Abs(vectorWeight-0.3) > similarityWeightTolerance {
+ t.Fatalf("similarity weights were not normalized: %#v", updated.SearchConfig)
+ }
persisted, err := dao.NewSearchDAO().GetByID(ctx, dao.DB, created.SearchID)
if err != nil {
diff --git a/internal/service/similarity_weight.go b/internal/service/similarity_weight.go
new file mode 100644
index 0000000000..7e1c4356c3
--- /dev/null
+++ b/internal/service/similarity_weight.go
@@ -0,0 +1,99 @@
+//
+// Copyright 2026 The InfiniFlow Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//
+
+package service
+
+import (
+ "encoding/json"
+ "fmt"
+ "math"
+)
+
+const similarityWeightTolerance = 1e-9
+
+// ResolveVectorSimilarityWeight validates the public keyword/vector weights and
+// returns the vector weight used by retrieval. Keyword weight takes precedence.
+func ResolveVectorSimilarityWeight(keywordsSimilarityWeight, vectorSimilarityWeight *float64) (*float64, error) {
+ if keywordsSimilarityWeight != nil && (*keywordsSimilarityWeight < 0 || *keywordsSimilarityWeight > 1) {
+ return nil, fmt.Errorf("keywords_similarity_weight must be between 0 and 1")
+ }
+ if vectorSimilarityWeight != nil && (*vectorSimilarityWeight < 0 || *vectorSimilarityWeight > 1) {
+ return nil, fmt.Errorf("vector_similarity_weight must be between 0 and 1")
+ }
+ if keywordsSimilarityWeight != nil && vectorSimilarityWeight != nil && math.Abs(*keywordsSimilarityWeight+*vectorSimilarityWeight-1) > similarityWeightTolerance {
+ return nil, fmt.Errorf("keywords_similarity_weight and vector_similarity_weight must sum to 1")
+ }
+ if keywordsSimilarityWeight != nil {
+ resolved := 1 - *keywordsSimilarityWeight
+ return &resolved, nil
+ }
+ return vectorSimilarityWeight, nil
+}
+
+// NormalizeSimilarityWeights validates similarity weights in a REST payload and
+// synchronizes both field names. Payloads that omit both fields are unchanged.
+func NormalizeSimilarityWeights(values map[string]interface{}) error {
+ keywordsSimilarityWeight, err := similarityWeightFromMap(values, "keywords_similarity_weight")
+ if err != nil {
+ return err
+ }
+ vectorSimilarityWeight, err := similarityWeightFromMap(values, "vector_similarity_weight")
+ if err != nil {
+ return err
+ }
+
+ resolvedVectorWeight, err := ResolveVectorSimilarityWeight(keywordsSimilarityWeight, vectorSimilarityWeight)
+ if err != nil {
+ return err
+ }
+ if resolvedVectorWeight == nil {
+ return nil
+ }
+
+ values["vector_similarity_weight"] = *resolvedVectorWeight
+ values["keywords_similarity_weight"] = 1 - *resolvedVectorWeight
+ return nil
+}
+
+func similarityWeightFromMap(values map[string]interface{}, key string) (*float64, error) {
+ value, ok := values[key]
+ if !ok || value == nil {
+ return nil, nil
+ }
+
+ var weight float64
+ switch typed := value.(type) {
+ case float64:
+ weight = typed
+ case float32:
+ weight = float64(typed)
+ case int:
+ weight = float64(typed)
+ case int32:
+ weight = float64(typed)
+ case int64:
+ weight = float64(typed)
+ case json.Number:
+ parsed, err := typed.Float64()
+ if err != nil {
+ return nil, fmt.Errorf("%s must be a number", key)
+ }
+ weight = parsed
+ default:
+ return nil, fmt.Errorf("%s must be a number", key)
+ }
+ return &weight, nil
+}
diff --git a/internal/service/similarity_weight_test.go b/internal/service/similarity_weight_test.go
new file mode 100644
index 0000000000..5f15cbbef9
--- /dev/null
+++ b/internal/service/similarity_weight_test.go
@@ -0,0 +1,85 @@
+//
+// Copyright 2026 The InfiniFlow Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//
+
+package service
+
+import (
+ "math"
+ "strings"
+ "testing"
+)
+
+func TestResolveVectorSimilarityWeight(t *testing.T) {
+ keywordWeight := 0.7
+ vectorWeight := 0.3
+
+ tests := []struct {
+ name string
+ keywords *float64
+ vector *float64
+ want *float64
+ wantErr string
+ }{
+ {name: "keyword weight is preferred", keywords: &keywordWeight, want: float64Pointer(0.3)},
+ {name: "vector weight remains supported", vector: &vectorWeight, want: &vectorWeight},
+ {name: "matching weights are accepted", keywords: &keywordWeight, vector: &vectorWeight, want: float64Pointer(0.3)},
+ {name: "missing weights stay missing"},
+ {name: "mismatched weights are rejected", keywords: &keywordWeight, vector: float64Pointer(0.4), wantErr: "must sum to 1"},
+ {name: "invalid keyword weight is rejected", keywords: float64Pointer(1.1), wantErr: "keywords_similarity_weight must be between 0 and 1"},
+ {name: "invalid vector weight is rejected", vector: float64Pointer(-0.1), wantErr: "vector_similarity_weight must be between 0 and 1"},
+ }
+
+ for _, tt := range tests {
+ t.Run(tt.name, func(t *testing.T) {
+ got, err := ResolveVectorSimilarityWeight(tt.keywords, tt.vector)
+ if tt.wantErr != "" {
+ if err == nil || !strings.Contains(err.Error(), tt.wantErr) {
+ t.Fatalf("error = %v, want containing %q", err, tt.wantErr)
+ }
+ return
+ }
+ if err != nil {
+ t.Fatalf("ResolveVectorSimilarityWeight failed: %v", err)
+ }
+ if tt.want == nil {
+ if got != nil {
+ t.Fatalf("weight = %v, want nil", *got)
+ }
+ return
+ }
+ if got == nil || math.Abs(*got-*tt.want) > similarityWeightTolerance {
+ t.Fatalf("weight = %v, want %v", got, *tt.want)
+ }
+ })
+ }
+}
+
+func TestNormalizeSimilarityWeights(t *testing.T) {
+ values := map[string]interface{}{"keywords_similarity_weight": 0.7}
+ if err := NormalizeSimilarityWeights(values); err != nil {
+ t.Fatalf("NormalizeSimilarityWeights failed: %v", err)
+ }
+ if got := values["vector_similarity_weight"].(float64); math.Abs(got-0.3) > similarityWeightTolerance {
+ t.Fatalf("vector_similarity_weight = %v, want 0.3", got)
+ }
+ if got := values["keywords_similarity_weight"].(float64); got != 0.7 {
+ t.Fatalf("keywords_similarity_weight = %v, want 0.7", got)
+ }
+}
+
+func float64Pointer(value float64) *float64 {
+ return &value
+}
diff --git a/web/src/components/similarity-slider/index.tsx b/web/src/components/similarity-slider/index.tsx
index 0900c2235f..61d26e97da 100644
--- a/web/src/components/similarity-slider/index.tsx
+++ b/web/src/components/similarity-slider/index.tsx
@@ -51,18 +51,10 @@ export const keywordsSimilarityWeightSchema = {
keywords_similarity_weight: z.number().min(0).max(1),
};
-export const vectorSimilarityWeightSchema = {
- vector_similarity_weight: z.number(),
-};
-
-export const initialVectorSimilarityWeightValue = {
- vector_similarity_weight: 0.3,
-};
-
export function SimilaritySliderFormField({
similarityName = 'similarity_threshold',
- similarityWeightName = 'vector_similarity_weight',
- similarityWeightType = 'vector',
+ similarityWeightName = 'keywords_similarity_weight',
+ similarityWeightType = 'keyword',
isTooltipShown,
numberInputClassName,
}: SimilaritySliderFormFieldProps) {
diff --git a/web/src/interfaces/database/chat.ts b/web/src/interfaces/database/chat.ts
index ae7a8df662..5d4a02db9e 100644
--- a/web/src/interfaces/database/chat.ts
+++ b/web/src/interfaces/database/chat.ts
@@ -81,7 +81,8 @@ export interface IDialog {
tenant_id: string;
update_date: string;
update_time: number;
- vector_similarity_weight: number;
+ keywords_similarity_weight?: number;
+ vector_similarity_weight?: number;
similarity_threshold: number;
top_k: number;
top_n: number;
diff --git a/web/src/interfaces/request/knowledge.ts b/web/src/interfaces/request/knowledge.ts
index b8b02b0df0..5715d9fb4f 100644
--- a/web/src/interfaces/request/knowledge.ts
+++ b/web/src/interfaces/request/knowledge.ts
@@ -3,7 +3,7 @@ export interface ITestRetrievalRequestBody {
page_size: number;
rerank_candidates_count: number;
similarity_threshold: number;
- vector_similarity_weight: number;
+ keywords_similarity_weight: number;
rerank_id?: string;
use_kg?: boolean;
highlight?: boolean;
diff --git a/web/src/pages/dataset/testing/testing-form.tsx b/web/src/pages/dataset/testing/testing-form.tsx
index adddae6601..b06cb85416 100644
--- a/web/src/pages/dataset/testing/testing-form.tsx
+++ b/web/src/pages/dataset/testing/testing-form.tsx
@@ -18,9 +18,9 @@ import { RerankFormFields } from '@/components/rerank';
import {
SimilaritySliderFormField,
initialSimilarityThresholdValue,
- initialVectorSimilarityWeightValue,
+ initialKeywordsSimilarityWeightValue,
similarityThresholdSchema,
- vectorSimilarityWeightSchema,
+ keywordsSimilarityWeightSchema,
} from '@/components/similarity-slider';
import { TopSelectFormItem } from '@/components/top-select';
import { ButtonLoading } from '@/components/ui/button';
@@ -63,7 +63,7 @@ export default function TestingForm({
message: t('knowledgeDetails.testTextPlaceholder'),
}),
...similarityThresholdSchema,
- ...vectorSimilarityWeightSchema,
+ ...keywordsSimilarityWeightSchema,
dataset_ids: z.array(z.string()).optional(),
...MetadataFilterSchema,
page_size: z.number().int().min(1).max(100),
@@ -78,7 +78,7 @@ export default function TestingForm({
resolver: zodResolver(formSchema),
defaultValues: {
...initialSimilarityThresholdValue,
- ...initialVectorSimilarityWeightValue,
+ ...initialKeywordsSimilarityWeightValue,
dataset_ids: [knowledgeBaseId],
page_size: 10,
rerank_candidates_count: 64,
diff --git a/web/src/pages/next-chats/chat/app-settings/chat-prompt-engine.tsx b/web/src/pages/next-chats/chat/app-settings/chat-prompt-engine.tsx
index 8992e255f4..d0dd1387ad 100644
--- a/web/src/pages/next-chats/chat/app-settings/chat-prompt-engine.tsx
+++ b/web/src/pages/next-chats/chat/app-settings/chat-prompt-engine.tsx
@@ -216,7 +216,11 @@ export function ChatPromptEngine({
<SimilaritySliderFormField
isTooltipShown
similarityName={prefixName(prefix, 'similarity_threshold')}
- similarityWeightName={prefixName(prefix, 'vector_similarity_weight')}
+ similarityWeightName={prefixName(
+ prefix,
+ 'keywords_similarity_weight',
+ )}
+ similarityWeightType="keyword"
></SimilaritySliderFormField>
<RerankCandidatesCountFormField
name={prefixName(prefix, 'rerank_candidates_count')}
diff --git a/web/src/pages/next-chats/chat/app-settings/chat-settings.tsx b/web/src/pages/next-chats/chat/app-settings/chat-settings.tsx
index 24bb7a801c..ca531a3973 100644
--- a/web/src/pages/next-chats/chat/app-settings/chat-settings.tsx
+++ b/web/src/pages/next-chats/chat/app-settings/chat-settings.tsx
@@ -89,7 +89,7 @@ export function ChatSettings({ hasSingleChatBox }: ChatSettingsProps) {
top_n: 8,
rerank_candidates_count: 64,
similarity_threshold: 0.2,
- vector_similarity_weight: 0.2,
+ keywords_similarity_weight: 0.8,
meta_data_filter: {
method: DatasetMetadata.Disabled,
manual: [],
@@ -134,6 +134,7 @@ export function ChatSettings({ hasSingleChatBox }: ChatSettingsProps) {
'update_date',
'id',
'top_k',
+ 'vector_similarity_weight',
]),
...nextValues,
},
@@ -152,8 +153,11 @@ export function ChatSettings({ hasSingleChatBox }: ChatSettingsProps) {
? { ...referenceMetadata, fields: undefined }
: referenceMetadata;
+ const keywordsSimilarityWeight =
+ data.keywords_similarity_weight ??
+ 1 - (data.vector_similarity_weight ?? 0.3);
const nextData = {
- ...omit(data, 'top_k'),
+ ...omit(data, ['top_k', 'vector_similarity_weight']),
prompt_config: {
...data.prompt_config,
// reset() skips undefined values, so fall back to '' to clear the field
@@ -161,6 +165,7 @@ export function ChatSettings({ hasSingleChatBox }: ChatSettingsProps) {
reference_metadata: normalizedReferenceMetadata,
},
...llmSettingEnabledValues,
+ keywords_similarity_weight: keywordsSimilarityWeight,
};
if (!isEmpty(data)) {
diff --git a/web/src/pages/next-chats/chat/app-settings/use-chat-setting-schema.tsx b/web/src/pages/next-chats/chat/app-settings/use-chat-setting-schema.tsx
index 08eae228be..0cb73987c9 100644
--- a/web/src/pages/next-chats/chat/app-settings/use-chat-setting-schema.tsx
+++ b/web/src/pages/next-chats/chat/app-settings/use-chat-setting-schema.tsx
@@ -7,7 +7,7 @@ import { rerankCandidatesCountSchema } from '@/components/rerank-candidates-coun
import { rerankFormSchema } from '@/components/rerank';
import {
similarityThresholdSchema,
- vectorSimilarityWeightSchema,
+ keywordsSimilarityWeightSchema,
} from '@/components/similarity-slider';
import { topnSchema } from '@/components/top-n-item';
import { WebSearchProvider } from '@/constants/chat';
@@ -81,7 +81,7 @@ export function useChatSettingSchema() {
llm_setting: z.object(LlmSettingFieldSchema),
...LlmSettingEnabledSchema,
llm_id: z.string().optional(),
- ...vectorSimilarityWeightSchema,
+ ...keywordsSimilarityWeightSchema,
...similarityThresholdSchema,
...topnSchema,
...rerankCandidatesCountSchema,
diff --git a/web/src/pages/next-chats/hooks/use-create-chat.ts b/web/src/pages/next-chats/hooks/use-create-chat.ts
index f6ae79b1f4..2a244999d0 100644
--- a/web/src/pages/next-chats/hooks/use-create-chat.ts
+++ b/web/src/pages/next-chats/hooks/use-create-chat.ts
@@ -42,7 +42,7 @@ export const useCreateChatDialog = () => {
tenant_llm_id: defaultModelDictionary?.llm_id,
llm_setting: {},
similarity_threshold: 0.2,
- vector_similarity_weight: 0.3,
+ keywords_similarity_weight: 0.7,
top_n: 8,
rerank_candidates_count: 64,
}),
diff --git a/web/src/pages/next-search/search-setting-hooks.ts b/web/src/pages/next-search/search-setting-hooks.ts
index 1fd8ef5709..b15e7ce555 100644
--- a/web/src/pages/next-search/search-setting-hooks.ts
+++ b/web/src/pages/next-search/search-setting-hooks.ts
@@ -38,7 +38,7 @@ export const SearchSettingFormSchema = z
description: z.string().optional(),
search_config: z.object({
kb_ids: z.array(z.string()).min(1, 'At least one dataset is required'),
- vector_similarity_weight: z.number().min(0).max(1),
+ keywords_similarity_weight: z.number().min(0).max(1),
web_search: z.boolean(),
similarity_threshold: z.number(),
use_kg: z.boolean(),
diff --git a/web/src/pages/next-search/search-setting.tsx b/web/src/pages/next-search/search-setting.tsx
index e74b49f1fe..143c3a2681 100644
--- a/web/src/pages/next-search/search-setting.tsx
+++ b/web/src/pages/next-search/search-setting.tsx
@@ -103,8 +103,9 @@ function SearchSetting({
description: data?.description || descriptionDefaultValue,
search_config: {
kb_ids: search_config?.kb_ids || [],
- vector_similarity_weight:
- search_config?.vector_similarity_weight ?? 0.3,
+ keywords_similarity_weight:
+ search_config?.keywords_similarity_weight ??
+ 1 - (search_config?.vector_similarity_weight ?? 0.3),
web_search: search_config?.web_search || false,
doc_ids: [],
similarity_threshold: search_config?.similarity_threshold ?? 0.2,
@@ -240,7 +241,7 @@ function SearchSetting({
void _maxTokensEnabled;
const {
llm_setting,
- vector_similarity_weight,
+ keywords_similarity_weight,
use_rerank,
rerank_id,
...other_config
@@ -274,7 +275,7 @@ function SearchSetting({
...other_config,
reference_metadata: normalizedReferenceMetadata,
chat_id: llm_setting.llm_id,
- vector_similarity_weight,
+ keywords_similarity_weight,
rerank_id: use_rerank ? rerank_id : '',
llm_setting: { ...llmSetting },
},
@@ -386,7 +387,8 @@ function SearchSetting({
<SimilaritySliderFormField
isTooltipShown
similarityName="search_config.similarity_threshold"
- similarityWeightName="search_config.vector_similarity_weight"
+ similarityWeightName="search_config.keywords_similarity_weight"
+ similarityWeightType="keyword"
numberInputClassName="rounded-sm"
></SimilaritySliderFormField>
<RerankCandidatesCountFormField
diff --git a/web/src/pages/next-searches/hooks.ts b/web/src/pages/next-searches/hooks.ts
index 6489086405..5087005cbf 100644
--- a/web/src/pages/next-searches/hooks.ts
+++ b/web/src/pages/next-searches/hooks.ts
@@ -215,7 +215,8 @@ export interface ISearchAppDetailProps {
top_k: number;
rerank_candidates_count: number;
use_kg: boolean;
- vector_similarity_weight: number;
+ keywords_similarity_weight?: number;
+ vector_similarity_weight?: number;
web_search: boolean;
chat_settingcross_languages: string[];
meta_data_filter?: {
diff --git a/web/src/services/knowledge-service.ts b/web/src/services/knowledge-service.ts
index 78e6e53df9..ec67beb4dd 100644
--- a/web/src/services/knowledge-service.ts
+++ b/web/src/services/knowledge-service.ts
@@ -196,7 +196,7 @@ const chunkService = {
page: rest.page,
page_size: rest.page_size ?? rest.size,
similarity_threshold: rest.similarity_threshold,
- vector_similarity_weight: rest.vector_similarity_weight,
+ keywords_similarity_weight: rest.keywords_similarity_weight,
top_k: rest.top_k,
knn_top_k: rest.knn_top_k,
knn_num_candidates: rest.knn_num_candidates,