Hybrid Search Alpha Parameter (#714)

backend/danswer/configs/app_configs.py

@@ -211,6 +211,8 @@
     EDIT_KEYWORD_QUERY = os.environ.get("EDIT_KEYWORD_QUERY", "").lower() == "true"
 else:
     EDIT_KEYWORD_QUERY = not os.environ.get("DOCUMENT_ENCODER_MODEL")
+# Weighting factor between Vector and Keyword Search, 1 for completely vector search
+HYBRID_ALPHA = max(0, min(1, float(os.environ.get("HYBRID_ALPHA") or 0.6)))
 
 
 #####

backend/danswer/document_index/interfaces.py

@@ -108,6 +108,7 @@ def hybrid_retrieval(
         filters: IndexFilters,
         favor_recent: bool,
         num_to_retrieve: int,
+        hybrid_alpha: float | None = None,
     ) -> list[InferenceChunk]:
         raise NotImplementedError
 

backend/danswer/document_index/vespa/app_config/schemas/danswer_chunk.sd

@@ -110,6 +110,7 @@ schema danswer_chunk {
             expression: max(if(isNan(attribute(doc_updated_at)) == 1, 7890000, now() - attribute(doc_updated_at)) / 31536000, 0)
         }
 
+        # Document score decays from 1 to 0.5 as age of last updated time increases
         function inline recency_bias() {
             # Cap the loss at 50% score reduction
             expression: max(1 / (1 + query(decay_factor) * document_age), 0.5)
@@ -150,7 +151,7 @@ schema danswer_chunk {
         }
 
         global-phase {
-            expression: (normalize_linear(closeness(field, embeddings)) + normalize_linear(bm25(content))) / 2 * document_boost * recency_bias
+            expression: ((query(alpha) * normalize_linear(closeness(field, embeddings))) + ((1 - query(alpha)) * normalize_linear(bm25(content)))) * document_boost * recency_bias
             rerank-count: 1000
         }
 

backend/danswer/document_index/vespa/index.py

@@ -19,6 +19,7 @@
 from danswer.configs.app_configs import DOCUMENT_INDEX_NAME
 from danswer.configs.app_configs import EDIT_KEYWORD_QUERY
 from danswer.configs.app_configs import FAVOR_RECENT_DECAY_MULTIPLIER
+from danswer.configs.app_configs import HYBRID_ALPHA
 from danswer.configs.app_configs import NUM_RETURNED_HITS
 from danswer.configs.app_configs import VESPA_DEPLOYMENT_ZIP
 from danswer.configs.app_configs import VESPA_HOST
@@ -432,7 +433,7 @@ def _vespa_hit_to_inference_chunk(hit: dict[str, Any]) -> InferenceChunk:
     )
 
 
-def _query_vespa(query_params: Mapping[str, str | int]) -> list[InferenceChunk]:
+def _query_vespa(query_params: Mapping[str, str | int | float]) -> list[InferenceChunk]:
     if "query" in query_params and not cast(str, query_params["query"]).strip():
         raise ValueError("No/empty query received")
     response = requests.get(SEARCH_ENDPOINT, params=query_params)
@@ -669,6 +670,7 @@ def hybrid_retrieval(
         filters: IndexFilters,
         favor_recent: bool,
         num_to_retrieve: int,
+        hybrid_alpha: float | None = HYBRID_ALPHA,
         distance_cutoff: float | None = SEARCH_DISTANCE_CUTOFF,
         edit_keyword_query: bool = EDIT_KEYWORD_QUERY,
     ) -> list[InferenceChunk]:
@@ -690,11 +692,14 @@ def hybrid_retrieval(
             " ".join(remove_stop_words(query)) if edit_keyword_query else query
         )
 
-        params: dict[str, str | int] = {
+        params: dict[str, str | int | float] = {
             "yql": yql,
             "query": query_keywords,
             "input.query(query_embedding)": str(query_embedding),
             "input.query(decay_factor)": str(DOC_TIME_DECAY * decay_multiplier),
+            "input.query(alpha)": hybrid_alpha
+            if hybrid_alpha is not None
+            else HYBRID_ALPHA,
             "hits": num_to_retrieve,
             "offset": 0,
             "ranking.profile": "hybrid_search",

backend/danswer/search/search_runner.py

@@ -8,6 +8,7 @@
 from sentence_transformers import SentenceTransformer  # type: ignore
 from sqlalchemy.orm import Session
 
+from danswer.configs.app_configs import HYBRID_ALPHA
 from danswer.configs.model_configs import ASYM_QUERY_PREFIX
 from danswer.configs.model_configs import CROSS_ENCODER_RANGE_MAX
 from danswer.configs.model_configs import CROSS_ENCODER_RANGE_MIN
@@ -104,21 +105,33 @@ def chunks_to_search_docs(chunks: list[InferenceChunk] | None) -> list[SearchDoc
 
 @log_function_time()
 def doc_index_retrieval(
-    query: SearchQuery, document_index: DocumentIndex
+    query: SearchQuery,
+    document_index: DocumentIndex,
+    hybrid_alpha: float = HYBRID_ALPHA,
 ) -> list[InferenceChunk]:
     if query.search_type == SearchType.KEYWORD:
         top_chunks = document_index.keyword_retrieval(
-            query.query, query.filters, query.favor_recent, query.num_hits
+            query=query.query,
+            filters=query.filters,
+            favor_recent=query.favor_recent,
+            num_to_retrieve=query.num_hits,
         )
 
     elif query.search_type == SearchType.SEMANTIC:
         top_chunks = document_index.semantic_retrieval(
-            query.query, query.filters, query.favor_recent, query.num_hits
+            query=query.query,
+            filters=query.filters,
+            favor_recent=query.favor_recent,
+            num_to_retrieve=query.num_hits,
         )
 
     elif query.search_type == SearchType.HYBRID:
         top_chunks = document_index.hybrid_retrieval(
-            query.query, query.filters, query.favor_recent, query.num_hits
+            query=query.query,
+            filters=query.filters,
+            favor_recent=query.favor_recent,
+            num_to_retrieve=query.num_hits,
+            hybrid_alpha=hybrid_alpha,
         )
 
     else:
@@ -282,6 +295,7 @@ def apply_boost(
 def search_chunks(
     query: SearchQuery,
     document_index: DocumentIndex,
+    hybrid_alpha: float = HYBRID_ALPHA,  # Only applicable to hybrid search
     retrieval_metrics_callback: Callable[[RetrievalMetricsContainer], None]
     | None = None,
     rerank_metrics_callback: Callable[[RerankMetricsContainer], None] | None = None,
@@ -293,7 +307,9 @@ def _log_top_chunk_links(search_flow: str, chunks: list[InferenceChunk]) -> None
         ]
         logger.info(f"Top links from {search_flow} search: {', '.join(top_links)}")
 
-    top_chunks = doc_index_retrieval(query=query, document_index=document_index)
+    top_chunks = doc_index_retrieval(
+        query=query, document_index=document_index, hybrid_alpha=hybrid_alpha
+    )
 
     if not top_chunks:
         logger.info(

deployment/docker_compose/docker-compose.dev.yml

@@ -32,6 +32,10 @@ services:
       - DISABLE_GENERATIVE_AI=${DISABLE_GENERATIVE_AI:-}
       - NOTION_CONNECTOR_ENABLE_RECURSIVE_PAGE_LOOKUP=${NOTION_CONNECTOR_ENABLE_RECURSIVE_PAGE_LOOKUP:-}
       - DISABLE_LLM_FILTER_EXTRACTION=${DISABLE_LLM_FILTER_EXTRACTION:-}
+      # Recency Bias for search results, decay at 1 / (1 + DOC_TIME_DECAY * x years)
+      - DOC_TIME_DECAY=${DOC_TIME_DECAY:-}
+      # Hybrid Search Alpha (0 for entirely keyword, 1 for entirely vector)
+      - HYBRID_ALPHA=${HYBRID_ALPHA:-}
       # Don't change the NLP model configs unless you know what you're doing
       - DOCUMENT_ENCODER_MODEL=${DOCUMENT_ENCODER_MODEL:-}
       - NORMALIZE_EMBEDDINGS=${NORMALIZE_EMBEDDINGS:-}
@@ -86,6 +90,8 @@ services:
       - NOTIFY_SLACKBOT_NO_ANSWER=${NOTIFY_SLACKBOT_NO_ANSWER:-}
       # Recency Bias for search results, decay at 1 / (1 + DOC_TIME_DECAY * x years)
       - DOC_TIME_DECAY=${DOC_TIME_DECAY:-}
+      # Hybrid Search Alpha (0 for entirely keyword, 1 for entirely vector)
+      - HYBRID_ALPHA=${HYBRID_ALPHA:-}
       # Don't change the NLP model configs unless you know what you're doing
       - DOCUMENT_ENCODER_MODEL=${DOCUMENT_ENCODER_MODEL:-}
       - NORMALIZE_EMBEDDINGS=${NORMALIZE_EMBEDDINGS:-}