feat: 重构图谱抽取 — 支持建筑规范文档的 8 种实体类型

- 实体类型改为：构件/设备/设施/指标/场所/材料/措施/条款
- 中文 prompt + 建筑规范领域 Few-shot 示例
- 前端图谱颜色、筛选项同步更新
- QA agent 系统提示词适配建筑规范领域

backend/pipeline/entity_extractor.py

@@ -19,30 +19,61 @@ DEEPSEEK_BASE_URL = os.getenv("DEEPSEEK_BASE_URL", "https://api.deepseek.com")
 MODEL_ID = "deepseek-chat"
 
 PROMPT_DESCRIPTION = (
-    "Extract named entities from the text in order of appearance. "
+    "从建筑规范文本中按出现顺序提取命名实体。"
-    "Entity types: TECHNOLOGY (software, algorithms, models, tools), "
+    "实体类型："
-    "ORGANIZATION (companies, research groups, institutions), "
+    "构件 — 建筑构件和部位（防火墙、防火门、楼梯间、电梯井、管道井、变形缝、外墙、屋面、楼板、梁、柱、承重墙、隔墙、门窗、幕墙、雨篷、阳台、走廊、前室、竖井等）；"
-    "PERSON (individual people), "
+    "设备 — 消防设备和系统（消火栓、灭火器、喷淋系统、报警系统、防烟排烟系统、应急照明、疏散指示、防火卷帘、消防电梯、消防水箱、消防水泵、气体灭火系统、火灾自动报警系统、电气火灾监控系统等）；"
-    "LOCATION (places, geographic entities), "
+    "设施 — 建筑设施和配套（停车场、锅炉房、变配电室、发电机房、储油间、空调机房、通风机房、电梯机房、消防控制室、水泵房等）；"
-    "CONCEPT (technical concepts, methodologies, frameworks)."
+    "指标 — 技术参数和量化要求（耐火等级、防火间距、疏散宽度、最大面积、高度限值、距离要求、时间要求、容量要求、温度限值、压力要求等）；"
+    "场所 — 建筑分类和空间类型（高层建筑、地下建筑、公共建筑、住宅建筑、工业建筑、商业建筑、医疗建筑、教育建筑、娱乐场所、仓库、厂房、中庭、避难层、避难走道、敞开楼梯间等）；"
+    "材料 — 建筑材料和产品（不燃材料、难燃材料、可燃材料、防火涂料、防火玻璃、防火密封件、保温材料、装饰材料、钢结构防火保护材料等）；"
+    "措施 — 防火措施和策略（防火分隔、防火封堵、自然排烟、机械加压送风、机械排烟、安全疏散、防火保护、消防供电、消防水源、灭火救援等）；"
+    "条款 — 规范条文引用（强制性条文、推荐性条文、术语定义、一般规定、基本要求等）。"
 )
 
 EXAMPLES = [
     lx.data.ExampleData(
         text=(
-            "LangChain is a framework created by Harrison Chase for building "
+            "建筑高度大于100m的公共建筑应设置避难层。避难层应采用耐火极限不低于3.00h的防火墙和甲级防火门进行分隔。"
-            "LLM applications. It integrates with OpenAI models and Pinecone "
+            "建筑内的消防控制室应采用耐火极限不低于2.00h的防火隔墙与乙级防火门与其他部位分隔。"
-            "vector database for semantic search."
+            "疏散走道两侧的隔墙应采用不燃材料，其耐火极限不应低于1.00h。"
         ),
         extractions=[
-            lx.data.Extraction(extraction_class="TECHNOLOGY", extraction_text="LangChain"),
+            lx.data.Extraction(extraction_class="场所", extraction_text="公共建筑"),
-            lx.data.Extraction(extraction_class="PERSON", extraction_text="Harrison Chase"),
+            lx.data.Extraction(extraction_class="构件", extraction_text="避难层"),
-            lx.data.Extraction(extraction_class="CONCEPT", extraction_text="LLM applications"),
+            lx.data.Extraction(extraction_class="指标", extraction_text="耐火极限不低于3.00h"),
-            lx.data.Extraction(extraction_class="TECHNOLOGY", extraction_text="OpenAI models"),
+            lx.data.Extraction(extraction_class="构件", extraction_text="防火墙"),
-            lx.data.Extraction(extraction_class="TECHNOLOGY", extraction_text="Pinecone"),
+            lx.data.Extraction(extraction_class="构件", extraction_text="甲级防火门"),
-            lx.data.Extraction(extraction_class="CONCEPT", extraction_text="semantic search"),
+            lx.data.Extraction(extraction_class="设施", extraction_text="消防控制室"),
+            lx.data.Extraction(extraction_class="指标", extraction_text="耐火极限不低于2.00h"),
+            lx.data.Extraction(extraction_class="构件", extraction_text="防火隔墙"),
+            lx.data.Extraction(extraction_class="构件", extraction_text="乙级防火门"),
+            lx.data.Extraction(extraction_class="构件", extraction_text="疏散走道"),
+            lx.data.Extraction(extraction_class="构件", extraction_text="隔墙"),
+            lx.data.Extraction(extraction_class="材料", extraction_text="不燃材料"),
+            lx.data.Extraction(extraction_class="指标", extraction_text="耐火极限不应低于1.00h"),
         ],
-    )
+    ),
+    lx.data.ExampleData(
+        text=(
+            "住宅建筑室内消火栓设计流量不应小于5L/s。高位消防水箱有效容积不应小于18m3，"
+            "且应设置在建筑最高部位。消防电梯前室应设置室内消火栓。"
+            "公共建筑内每个防火分区的安全出口不应少于2个。"
+        ),
+        extractions=[
+            lx.data.Extraction(extraction_class="场所", extraction_text="住宅建筑"),
+            lx.data.Extraction(extraction_class="设备", extraction_text="室内消火栓"),
+            lx.data.Extraction(extraction_class="指标", extraction_text="不应小于5L/s"),
+            lx.data.Extraction(extraction_class="设备", extraction_text="高位消防水箱"),
+            lx.data.Extraction(extraction_class="指标", extraction_text="不应小于18m3"),
+            lx.data.Extraction(extraction_class="构件", extraction_text="消防电梯前室"),
+            lx.data.Extraction(extraction_class="设备", extraction_text="室内消火栓"),
+            lx.data.Extraction(extraction_class="场所", extraction_text="公共建筑"),
+            lx.data.Extraction(extraction_class="构件", extraction_text="防火分区"),
+            lx.data.Extraction(extraction_class="构件", extraction_text="安全出口"),
+            lx.data.Extraction(extraction_class="指标", extraction_text="不应少于2个"),
+        ],
+    ),
 ]
 
 

backend/pipeline/qa_agent.py

@@ -83,17 +83,14 @@ def make_tools(G: nx.Graph) -> list:
     def get_entities_by_type(entity_type: str) -> str:
         """List all entities of a specific type.
         Args:
-            entity_type: TECHNOLOGY, CONCEPT, PERSON, ORGANIZATION, or LOCATION.
+            entity_type: 构件, 设备, 设施, 指标, 场所, 材料, 措施, 条款, etc.
         """
-        t_upper = entity_type.strip().upper()
+        t = entity_type.strip()
-        valid = {"TECHNOLOGY", "CONCEPT", "PERSON", "ORGANIZATION", "LOCATION"}
+        matches = [d for _, d in G.nodes(data=True) if d.get("type","") == t]
-        if t_upper not in valid:
-            present = sorted({d.get("type","") for _, d in G.nodes(data=True)})
-            return f"Unknown type '{entity_type}'. Present: {present}"
-        matches = [d for _, d in G.nodes(data=True) if d.get("type","") == t_upper]
         if not matches:
-            return f"No {t_upper} entities found."
+            present = sorted({d.get("type","") for _, d in G.nodes(data=True)})
-        lines = [f"Found {len(matches)} {t_upper} entities:"]
+            return f"No '{entity_type}' entities found. Present types: {present}"
+        lines = [f"Found {len(matches)} {t} entities:"]
         for m in matches[:30]:
             lines.append(f"  \"{m['name']}\" (page={m.get('page',0)}, id={m['id']})")
         if len(matches) > 30:
@@ -148,15 +145,17 @@ def run_qa(
     )
 
     system_prompt = (
-        "You are a helpful assistant with access to a knowledge graph (KG) built from the user's documents.\n"
+        "You are a helpful assistant with access to a knowledge graph (KG) built from building code documents.\n"
+        "The KG contains entities extracted from building fire protection codes (建筑防火规范).\n"
+        "\n"
+        "Entity types include: 构件 (building components), 设备 (fire equipment), 设施 (facilities),\n"
+        "指标 (technical parameters), 场所 (building types/spaces), 材料 (materials),\n"
+        "措施 (fire measures), 条款 (regulation clauses).\n"
         "\n"
         "Guidelines:\n"
-        "- If the question is clearly unrelated to the KG (greetings, math, general knowledge, etc.), "
+        "- Use tools to search and explore the KG before answering.\n"
-        "answer directly WITHOUT using any tools.\n"
+        "- Cite the entity names and types you found.\n"
-        "- If the question might be answered by the KG (topics related to entities in the documents), "
+        "- If the KG has no relevant information, say so honestly.\n"
-        "use the tools to search and explore before answering.\n"
-        "- When you DO use the KG, cite the entity names and types you found.\n"
-        "- If the KG has no relevant information, say so honestly and answer from general knowledge if possible.\n"
         "\n"
         "Available tools: search entities by name, get neighbors, list entities by type, get graph overview."
     )

frontend/src/app/components/pages/KGExplorer.tsx

@@ -5,7 +5,7 @@ import { ZoomIn, ZoomOut, Maximize2, Search, Download, Image, X, MessageSquare,
 import { useAppState, type KGNode } from '../../store';
 import { TYPE_COLORS } from '../../mock-data';
 
-const ENTITY_TYPES = ['TECHNOLOGY', 'CONCEPT', 'PERSON', 'ORGANIZATION', 'LOCATION'] as const;
+const ENTITY_TYPES = ['构件', '设备', '设施', '指标', '场所', '材料', '措施', '条款', 'TECHNOLOGY', 'CONCEPT', 'PERSON', 'ORGANIZATION', 'LOCATION'] as const;
 const CONFIDENCE_LEVELS = ['match_exact', 'match_greater', 'match_lesser', 'match_fuzzy'] as const;
 
 export function KGExplorer() {

frontend/src/app/components/pages/SearchPage.tsx

@@ -6,7 +6,7 @@ import { useAppState, mapApiNode, mapApiEdge, type KGNode } from '../../store';
 import { api, ApiError } from '../../api';
 import { TYPE_COLORS } from '../../mock-data';
 
-const ENTITY_TYPES_OPTIONS = ['全部类型', 'TECHNOLOGY', 'CONCEPT', 'PERSON', 'ORGANIZATION', 'LOCATION'];
+const ENTITY_TYPES_OPTIONS = ['全部类型', '构件', '设备', '设施', '指标', '场所', '材料', '措施', '条款', 'TECHNOLOGY', 'CONCEPT', 'PERSON', 'ORGANIZATION', 'LOCATION'];
 
 export function SearchPage() {
   const { nodes, edges, getNeighbors } = useAppState();

frontend/src/app/mock-data.ts

@@ -3,7 +3,7 @@
 export interface KGNode {
   id: string;
   name: string;
-  type: 'TECHNOLOGY' | 'CONCEPT' | 'PERSON' | 'ORGANIZATION' | 'LOCATION';
+  type: '构件' | '设备' | '设施' | '指标' | '场所' | '材料' | '措施' | '条款' | 'TECHNOLOGY' | 'CONCEPT' | 'PERSON' | 'ORGANIZATION' | 'LOCATION';
   page: number;
   confidence: 'match_exact' | 'match_greater' | 'match_lesser' | 'match_fuzzy';
   degree: number;
@@ -66,6 +66,14 @@ export interface HistoryItem {
 
 // Entity type colors
 export const TYPE_COLORS: Record<string, string> = {
+  '构件': '#58a6ff',
+  '设备': '#ff7b72',
+  '设施': '#d29922',
+  '指标': '#3fb950',
+  '场所': '#bc8cff',
+  '材料': '#ffa657',
+  '措施': '#f0f6fc',
+  '条款': '#8b949e',
   TECHNOLOGY: '#58a6ff',
   CONCEPT: '#bc8cff',
   PERSON: '#3fb950',

frontend/src/app/store.tsx

@@ -6,7 +6,7 @@ import { api, type ApiDoc, type ApiKGNode, type ApiKGEdge, ApiError } from './ap
 export interface KGNode {
   id: string;
   name: string;
-  type: 'TECHNOLOGY' | 'CONCEPT' | 'PERSON' | 'ORGANIZATION' | 'LOCATION';
+  type: '构件' | '设备' | '设施' | '指标' | '场所' | '材料' | '措施' | '条款' | 'TECHNOLOGY' | 'CONCEPT' | 'PERSON' | 'ORGANIZATION' | 'LOCATION';
   page: number;
   confidence: 'match_exact' | 'match_greater' | 'match_lesser' | 'match_fuzzy';
   degree: number;
@@ -169,9 +169,9 @@ const DEFAULT_STATS: StatsData = { kg_nodes: 0, kg_edges: 0, documents: 0, queri
 
 const SUGGESTED_PROMPTS = [
   '给我一个知识图谱的概览',
-  '列出所有 TECHNOLOGY 实体',
+  '列出所有 构件 实体',
-  'GraphRAG 与知识图谱有什么关系？',
+  '列出所有 设备 实体',
-  '什么是检索增强生成？',
+  '文档中有哪些防火措施？',
 ];
 
 // ─── Provider ─────────────────────────────────────────────────────────────────