feat: add knowledge retrieval service

backend/apps/knowledge/services.py

@@ -1,7 +1,8 @@
 import hashlib
 import math
 
-from apps.knowledge.models import KnowledgeChunk, KnowledgeDocument, KnowledgeItem
+from apps.knowledge.models import KnowledgeBase, KnowledgeChunk, KnowledgeDocument, KnowledgeItem
+from apps.solutions.models import Solution
 
 DEFAULT_CHUNK_SIZE = 600
 
@@ -23,6 +24,18 @@ def embed_text(content: str, dimensions: int = 16) -> list[float]:
     return [round(value / length, 6) for value in values]
 
 
+def keyword_score(query: str, content: str) -> float:
+    normalized_query = query.lower()
+    terms = [term for term in normalized_query.split() if term]
+    terms.extend(character for character in normalized_query if not character.isspace())
+    lowered = content.lower()
+    return sum(1 for term in terms if term in lowered)
+
+
+def vector_score(left: list[float], right: list[float]) -> float:
+    return sum(a * b for a, b in zip(left, right, strict=False))
+
+
 def rebuild_item_chunks(item: KnowledgeItem) -> list[KnowledgeChunk]:
     item.chunks.all().delete()
     content = "\n".join(
@@ -85,3 +98,70 @@ def ingest_document_text(
             },
         )
     return document
+
+
+def index_published_solution(solution: Solution) -> list[KnowledgeChunk]:
+    base, _ = KnowledgeBase.objects.get_or_create(
+        name="解决方案知识库",
+        defaults={"description": "从已发布解决方案生成的检索知识", "scope": "solutions"},
+    )
+    KnowledgeChunk.objects.filter(metadata__solution_id=solution.id).delete()
+    content = "\n".join(
+        [
+            solution.title,
+            solution.problem_signature,
+            solution.related_system,
+            solution.environment_scope,
+            solution.root_cause,
+            solution.solution,
+            solution.risk_note,
+        ]
+    )
+    chunks = []
+    for index, chunk_text in enumerate(split_text(content)):
+        chunks.append(
+            KnowledgeChunk.objects.create(
+                knowledge_base=base,
+                chunk_index=index,
+                content=chunk_text,
+                embedding=embed_text(chunk_text),
+                metadata={
+                    "source_type": "solution",
+                    "title": solution.title,
+                    "solution_id": solution.id,
+                },
+            )
+        )
+    return chunks
+
+
+def retrieve_knowledge(
+    query: str,
+    knowledge_base_id: int | None = None,
+    limit: int = 5,
+) -> list[dict]:
+    query_embedding = embed_text(query)
+    queryset = KnowledgeChunk.objects.select_related("knowledge_base", "item", "document")
+    if knowledge_base_id:
+        queryset = queryset.filter(knowledge_base_id=knowledge_base_id)
+    scored = []
+    for chunk in queryset[:1000]:
+        score = keyword_score(query, chunk.content) * 10 + vector_score(
+            query_embedding,
+            chunk.embedding,
+        )
+        if score > 0:
+            scored.append((score, chunk))
+    scored.sort(key=lambda pair: pair[0], reverse=True)
+    return [
+        {
+            "id": chunk.id,
+            "knowledge_base_id": chunk.knowledge_base_id,
+            "title": chunk.metadata.get("title", chunk.knowledge_base.name),
+            "source_type": chunk.metadata.get("source_type", "chunk"),
+            "content": chunk.content,
+            "score": round(score, 4),
+            "metadata": chunk.metadata,
+        }
+        for score, chunk in scored[:limit]
+    ]