[Feat] 하이브리드 방식 인덱스 개선 및 비교

alembic/versions/20260504_0013_add_regulation_chunk_search_tsvector.py

@@ -0,0 +1,49 @@
+"""add regulation chunk search tsvector
+
+Revision ID: 20260504_0013
+Revises: 20260429_0012
+Create Date: 2026-05-04 00:00:00.000000
+"""
+
+from alembic import op
+import sqlalchemy as sa
+from sqlalchemy.dialects import postgresql
+
+
+revision = "20260504_0013"
+down_revision = "20260429_0012"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.add_column(
+        "regulation_chunk",
+        sa.Column("search_tsvector", postgresql.TSVECTOR(), nullable=True),
+    )
+
+    op.execute(
+        """
+        UPDATE regulation_chunk AS rc
+        SET search_tsvector = to_tsvector(
+            'simple',
+            COALESCE(rc.chunk_text, '') || ' ' ||
+            COALESCE(rd.content, '') || ' ' ||
+            COALESCE(rc.keywords::text, '')
+        )
+        FROM regulation_document AS rd
+        WHERE rd.regulation_document_id = rc.regulation_document_id
+        """
+    )
+
+    op.create_index(
+        "idx_regulation_chunk_search_tsvector",
+        "regulation_chunk",
+        ["search_tsvector"],
+        postgresql_using="gin",
+    )
+
+
+def downgrade() -> None:
+    op.drop_index("idx_regulation_chunk_search_tsvector", table_name="regulation_chunk")
+    op.drop_column("regulation_chunk", "search_tsvector")

app/db/models/regulation_chunk.py

@@ -5,6 +5,7 @@
 from pgvector.sqlalchemy import Vector
 from sqlalchemy import BigInteger, Boolean, DateTime, ForeignKey, Index, Integer, String, Text
 from sqlalchemy.dialects.postgresql import JSONB
+from sqlalchemy.dialects.postgresql import TSVECTOR
 from sqlalchemy import func
 from sqlalchemy import text
 from sqlalchemy.orm import Mapped, mapped_column
@@ -20,6 +21,7 @@ class RegulationChunk(Base):
         Index("idx_regulation_chunk_document_version", "document_version"),
         Index("idx_regulation_chunk_chunk_id", "chunk_id"),
         Index("idx_regulation_chunk_is_active", "is_active"),
+        Index("idx_regulation_chunk_search_tsvector", "search_tsvector", postgresql_using="gin"),
     )
 
     regulation_chunk_id: Mapped[int] = mapped_column(BigInteger, primary_key=True, autoincrement=True)
@@ -33,6 +35,7 @@ class RegulationChunk(Base):
     chunk_index: Mapped[int] = mapped_column(Integer, nullable=False)
     chunk_text: Mapped[Optional[str]] = mapped_column(Text, nullable=True)
     keywords: Mapped[Optional[list[str]]] = mapped_column(JSONB, nullable=True)
+    search_tsvector: Mapped[Optional[str]] = mapped_column(TSVECTOR, nullable=True)
     embedding: Mapped[Optional[List[float]]] = mapped_column(Vector(1536), nullable=True)
     chunk_hash: Mapped[Optional[str]] = mapped_column(String(255), nullable=True)
     embedding_model: Mapped[Optional[str]] = mapped_column(String(100), nullable=True)

app/repositories/regulation_chunk_repository.py

@@ -53,11 +53,45 @@ def create_regulation_chunks_for_document(
         created_chunks.append(regulation_chunk)
 
     db.flush()
+    refresh_search_vectors_for_chunks(
+        db,
+        [
+            regulation_chunk.regulation_chunk_id
+            for regulation_chunk in created_chunks
+            if regulation_chunk.regulation_chunk_id is not None
+        ],
+    )
     for regulation_chunk in created_chunks:
         db.refresh(regulation_chunk)
     return created_chunks
 
 
+def refresh_search_vectors_for_chunks(db: Session, regulation_chunk_ids: list[int]) -> int:
+    """저장된 청크 검색 텍스트를 tsvector 컬럼에 반영합니다."""
+
+    if not regulation_chunk_ids:
+        return 0
+
+    result = db.execute(
+        text(
+            """
+            UPDATE regulation_chunk AS rc
+            SET search_tsvector = to_tsvector(
+                'simple',
+                COALESCE(rc.chunk_text, '') || ' ' ||
+                COALESCE(rd.content, '') || ' ' ||
+                COALESCE(rc.keywords::text, '')
+            )
+            FROM regulation_document AS rd
+            WHERE rd.regulation_document_id = rc.regulation_document_id
+              AND rc.regulation_chunk_id = ANY(:regulation_chunk_ids)
+            """
+        ),
+        {"regulation_chunk_ids": regulation_chunk_ids},
+    )
+    return result.rowcount or 0
+
+
 def deactivate_chunks_for_document(db: Session, regulation_document_id: int) -> int:
     statement = (
         update(RegulationChunk)
@@ -392,23 +426,13 @@ def _search_hybrid_chunks(
                 rd.dormitory,
                 1 - (rc.embedding <=> CAST(:embedding AS vector)) AS vector_similarity,
                 ts_rank_cd(
-                    to_tsvector(
-                        'simple',
-                        COALESCE(rc.chunk_text, '') || ' ' ||
-                        COALESCE(rd.content, '') || ' ' ||
-                        COALESCE(rc.keywords::text, '')
-                    ),
+                    rc.search_tsvector,
                     websearch_to_tsquery('simple', :query_text)
                 ) AS keyword_score,
                 NULL::bigint AS vector_rank,
                 ROW_NUMBER() OVER (
                     ORDER BY ts_rank_cd(
-                        to_tsvector(
-                            'simple',
-                            COALESCE(rc.chunk_text, '') || ' ' ||
-                            COALESCE(rd.content, '') || ' ' ||
-                            COALESCE(rc.keywords::text, '')
-                        ),
+                        rc.search_tsvector,
                         websearch_to_tsquery('simple', :query_text)
                     ) DESC
                 ) AS keyword_rank
@@ -419,12 +443,7 @@ def _search_hybrid_chunks(
               AND rc.is_active = TRUE
               AND rd.is_active = TRUE
               AND rc.embedding IS NOT NULL
-              AND to_tsvector(
-                    'simple',
-                    COALESCE(rc.chunk_text, '') || ' ' ||
-                    COALESCE(rd.content, '') || ' ' ||
-                    COALESCE(rc.keywords::text, '')
-                  ) @@ websearch_to_tsquery('simple', :query_text)
+              AND rc.search_tsvector @@ websearch_to_tsquery('simple', :query_text)
             ORDER BY keyword_score DESC
             LIMIT :candidate_k
         ),

app/services/chat_service.py

@@ -25,14 +25,16 @@
 from app.repositories.regulation_chunk_repository import search_hybrid_chunks
 from app.repositories.regulation_chunk_repository import search_hybrid_chunks_all_dormitories
 from app.repositories.regulation_chunk_repository import search_hybrid_chunks_for_dormitories
+from app.repositories.regulation_chunk_repository import search_similar_chunks
+from app.repositories.regulation_chunk_repository import search_similar_chunks_for_dormitories
+from app.repositories.regulation_chunk_repository import search_similar_chunks_all_dormitories
 from app.schemas.chat import ChatRequest
 from app.schemas.chat import ChatResponse
 from app.services.embeddings import create_query_embedding
 from app.services.generator import AnswerGenerationResult
 from app.services.generator import generate_answer
 from app.services.validator import validate_question
 from app.services.query_rewriter import expand_query_for_retrieval
-from app.repositories.regulation_chunk_repository import search_similar_chunks_all_dormitories
 from app.services.room_floor_resolver import resolve_room_floor_question
 
 ERROR_TYPE_TIMEOUT = "TIMEOUT"
@@ -274,10 +276,13 @@ def _answer_single_dormitory_chat(
         # 전체 생활관 fallback까지 했는데도 답변을 못 만들면
         # LLM query expansion으로 검색용 질의를 확장한 뒤 재검색
         if _is_no_answer(answer_result.answer):
-            expanded_query = expand_query_for_retrieval(
-                question=question,
-                dormitory=dormitory,
-            )
+            if rewritten_query != question:
+                expanded_query = rewritten_query
+            else:
+                expanded_query = expand_query_for_retrieval(
+                    question=question,
+                    dormitory=dormitory,
+                )
 
             if expanded_query != question:
                 expanded_query_embedding = create_query_embedding(expanded_query)
@@ -442,10 +447,13 @@ def _answer_unspecified_dormitory_chat(
         # 비로그인/생활관 미지정 상태에서 원문 검색으로 답변을 못 만들면
         # query expansion으로 검색용 질의를 확장한 뒤 전체 생활관 대상으로 재검색
         if _is_no_answer(answer_result.answer):
-            expanded_query = expand_query_for_retrieval(
-                question=question,
-                dormitory=None,
-            )
+            if rewritten_query != question:
+                expanded_query = rewritten_query
+            else:
+                expanded_query = expand_query_for_retrieval(
+                    question=question,
+                    dormitory=None,
+                )
 
 
             if expanded_query != question: