YG-Soft/X-Financial
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat: 新增风险规则生成引擎与知识图谱可视化

Browse Source 
后端新增风险规则自动生成和模板执行服务，支持从规则资产
批量生成并持久化风险规则文件；知识库入库日志增强图谱
查询和本地 RAG 回退，前端审计页面增加风险规则模型和流
程图组件，知识入库面板拆分为图谱可视化子组件，报销创
建页面增加引导式流程模型，更新知识库索引数据。
This commit is contained in:
caoxiaozhu
2026-05-23 19:54:42 +08:00
parent 5b388d08c0
commit 575f093c74
63 changed files with 35497 additions and 1517 deletions
Download Patch File Download Diff File Expand all files Collapse all files

153
server/tests/test_knowledge_rag_service.py
View File

@@ -8,8 +8,10 @@ from app.services.knowledge_ingest_log import (
build_document_graph_summary,
build_ingest_document_summary,
build_ingest_status_summary,
enrich_knowledge_ingest_route_json,
)
from app.services.knowledge_rag import KnowledgeRagService
from app.services.knowledge_rag_local import query_local_text_chunks
def test_build_hits_prioritizes_structured_table_evidence_for_standard_queries() -> None:
@@ -82,6 +84,84 @@ def test_build_hits_prioritizes_answer_clue_appendix_for_rule_queries() -> None:
assert [item["candidate_id"] for item in hits] == ["clue-1", "plain-1"]
def test_query_local_text_chunks_prioritizes_relevant_policy_chunk(tmp_path) -> None:
workspace = tmp_path / "knowledge" / ".lightrag" / "x_financial_knowledge"
workspace.mkdir(parents=True)
(workspace / "kv_store_text_chunks.json").write_text(
json.dumps(
{
"chunk-travel": {
"_id": "chunk-travel",
"full_doc_id": "doc-1",
"chunk_order_index": 1,
"file_path": "/tmp/doc-1__差旅费管理办法.pdf",
"content": (
"第十三条 差旅费。酒店住宿限额标准如下：其他员工直辖市350元、"
"省会城市300元、其他地区250元。确因紧急公务、特别情形等事项"
"导致住宿超过规定标准时，超标20%以内由部门负责人审批，"
"超标20%以上需分管领导审批。"
),
},
"chunk-office": {
"_id": "chunk-office",
"full_doc_id": "doc-2",
"chunk_order_index": 1,
"file_path": "/tmp/doc-2__办公用品管理办法.pdf",
"content": "办公用品采购应遵循预算和验收流程。",
},
}
),
encoding="utf-8",
)
result = query_local_text_chunks(
lightrag_root=tmp_path / "knowledge" / ".lightrag",
workspace="x_financial_knowledge",
query="住宿费超过标准审批依据是什么？",
limit=2,
)
assert result.confident is True
assert result.hits[0]["candidate_id"] == "chunk-travel"
assert "住宿超过规定标准" in result.hits[0]["content"]
def test_query_knowledge_uses_local_chunks_before_lightrag_runtime(tmp_path, monkeypatch) -> None:
workspace = tmp_path / "knowledge" / ".lightrag" / "x_financial_knowledge"
workspace.mkdir(parents=True)
(workspace / "kv_store_text_chunks.json").write_text(
json.dumps(
{
"chunk-1": {
"_id": "chunk-1",
"full_doc_id": "doc-1",
"chunk_order_index": 1,
"file_path": "/tmp/doc-1__公司支出管理办法.pdf",
"content": (
"第八条 支出报销申请时限。公司各类支出报销结算申请时限为三个月。"
"逾期需说明原因，经分管领导审批后方可报销。"
),
}
}
),
encoding="utf-8",
)
def fail_if_runtime_is_used(_self):
raise AssertionError("local high-confidence queries should not initialize LightRAG")
monkeypatch.setattr(KnowledgeRagService, "_get_runtime", fail_if_runtime_is_used)
payload = KnowledgeRagService(storage_root=tmp_path).query_knowledge(
"费用发生后多久内必须报销？超过三个月还能不能报？",
limit=3,
)
assert payload["record_count"] == 1
assert payload["metadata"]["retrieval_strategy"] == "local_text_chunks"
assert "三个月" in payload["hits"][0]["content"]
def test_build_hits_demotes_chapter_navigation_for_specific_rule_queries() -> None:
hits = KnowledgeRagService._build_hits_from_query_data(
query="探亲差旅归哪个部门管理？",
@@ -227,6 +307,46 @@ def test_build_document_graph_summary_reads_lightrag_storage(tmp_path) -> None:
),
encoding="utf-8",
)
(workspace / "kv_store_entity_chunks.json").write_text(
json.dumps(
{
"远光软件": {"chunk_ids": ["chunk-1", "chunk-missing"]},
"支出管理": {"chunk_ids": ["chunk-2"]},
}
),
encoding="utf-8",
)
(workspace / "graph_chunk_entity_relation.graphml").write_text(
"""<?xml version="1.0" encoding="UTF-8"?>
<graphml xmlns="http://graphml.graphdrawing.org/xmlns">
<key id="n0" for="node" attr.name="entity_id" attr.type="string" />
<key id="n1" for="node" attr.name="entity_type" attr.type="string" />
<key id="n2" for="node" attr.name="description" attr.type="string" />
<key id="n3" for="node" attr.name="created_at" attr.type="string" />
<key id="e0" for="edge" attr.name="weight" attr.type="double" />
<key id="e1" for="edge" attr.name="description" attr.type="string" />
<key id="e2" for="edge" attr.name="keywords" attr.type="string" />
<graph edgedefault="undirected">
<node id="远光软件">
<data key="n0">远光软件</data>
<data key="n1">ORGANIZATION</data>
<data key="n2">公司主体&lt;SEP&gt;费用制度适用公司</data>
<data key="n3">2026-05-23</data>
</node>
<node id="支出管理">
<data key="n0">支出管理</data>
<data key="n1">TOPIC</data>
<data key="n2">规范费用支出、预算和审批。</data>
</node>
<edge source="远光软件" target="支出管理">
<data key="e0">2.5</data>
<data key="e1">远光软件通过支出管理制度约束费用审批。</data>
<data key="e2">制度&lt;SEP&gt;审批</data>
</edge>
</graph>
</graphml>""",
encoding="utf-8",
)
summary = build_document_graph_summary(
tmp_path,
@@ -235,10 +355,39 @@ def test_build_document_graph_summary_reads_lightrag_storage(tmp_path) -> None:
)
assert summary["entity_count"] == 2
assert summary["entities"] == ["远光软件", "支出管理"]
assert [item["name"] for item in summary["entities"]] == ["远光软件", "支出管理"]
assert summary["entities"][0]["type"] == "ORGANIZATION"
assert summary["entities"][0]["descriptions"][0] == "公司主体"
assert summary["relation_count"] == 1
assert summary["relations"] == [{"source": "远光软件", "target": "支出管理", "type": "关联"}]
assert summary["relations"][0]["source"] == "远光软件"
assert summary["relations"][0]["target"] == "支出管理"
assert summary["relations"][0]["description"] == "远光软件通过支出管理制度约束费用审批。"
assert summary["relations"][0]["keywords"] == ["制度", "审批"]
assert summary["relations"][0]["weight"] == 2.5
assert [item["id"] for item in summary["chunks"]] == ["chunk-1", "chunk-2"]
assert summary["chunks"][0]["excerpt"].startswith("第一条")
assert summary["entity_chunks"] == [
{"entity": "远光软件", "chunk_ids": ["chunk-1"]},
{"entity": "支出管理", "chunk_ids": ["chunk-2"]},
]
enriched_route = enrich_knowledge_ingest_route_json(
{
"lightrag_workspace": "test_workspace",
"knowledge_ingest": {
"graph": {
"entities": ["远光软件"],
"relations": [
{"source": "远光软件", "target": "支出管理", "type": "关联"}
],
}
},
},
storage_root=tmp_path,
)
enriched_entities = enriched_route["knowledge_ingest"]["graph"]["entities"]
assert [item["name"] for item in enriched_entities] == ["远光软件", "支出管理"]
assert enriched_entities[1]["type"] == "TOPIC"
def test_build_ingest_document_summary_extracts_sections() -> None:

106
server/tests/test_risk_rule_generation.py Normal file
View File

@@ -0,0 +1,106 @@
from __future__ import annotations
import json
from sqlalchemy import create_engine
from sqlalchemy.orm import Session, sessionmaker
from sqlalchemy.pool import StaticPool
from app.core.agent_enums import AgentAssetDomain, AgentAssetStatus
from app.db.base import Base
from app.models.agent_asset import AgentAsset
from app.schemas.agent_asset import AgentAssetRiskRuleGenerateRequest
from app.services.agent_asset_rule_library import AgentAssetRuleLibraryManager
from app.services.agent_asset_spreadsheet import RISK_RULES_LIBRARY
from app.services.risk_rule_flow_diagram import RiskRuleFlowDiagramRenderer, RiskRuleFlowDiagramSpec
from app.services.risk_rule_generation import RiskRuleGenerationService
class NullRuntimeChatService:
def complete(self, *args, **kwargs) -> None:
return None
def build_session() -> Session:
engine = create_engine(
"sqlite+pysqlite:///:memory:",
connect_args={"check_same_thread": False},
poolclass=StaticPool,
)
Base.metadata.create_all(bind=engine)
session_factory = sessionmaker(bind=engine, autoflush=False, autocommit=False)
return session_factory()
def test_generate_risk_rule_asset_creates_draft_json_rule(tmp_path) -> None:
with build_session() as db:
service = RiskRuleGenerationService(
db,
rule_library_manager=AgentAssetRuleLibraryManager(rule_root=tmp_path / "rules"),
runtime_chat_service=NullRuntimeChatService(),
)
asset_id = service.generate_rule_asset(
AgentAssetRiskRuleGenerateRequest(
business_domain=AgentAssetDomain.EXPENSE,
risk_level="high",
natural_language="住宿城市必须出现在本次差旅行程城市中，否则提示高风险。",
),
actor="pytest",
)
asset = db.get(AgentAsset, asset_id)
assert asset is not None
assert asset.status == AgentAssetStatus.DRAFT.value
assert asset.config_json["detail_mode"] == "json_risk"
assert asset.config_json["evaluator"] == "template_rule"
assert asset.current_version == "v0.1.0"
file_name = asset.config_json["rule_document"]["file_name"]
rule_path = tmp_path / "rules" / RISK_RULES_LIBRARY / file_name
payload = json.loads(rule_path.read_text(encoding="utf-8"))
assert payload["rule_code"] == asset.code
assert payload["outcomes"]["fail"]["severity"] == "high"
assert payload["template_key"] == "field_compare_v1"
assert payload["metadata"]["natural_language"].startswith("住宿城市")
assert payload["inputs"]["fields"]
assert payload["flow_diagram_svg"].startswith("<svg")
assert 'width="760" height="280"' in payload["flow_diagram_svg"]
assert 'data-risk-flow-style="review-node-only"' in payload["flow_diagram_svg"]
assert "RULE FLOW" in payload["flow_diagram_svg"]
assert "进入复核" in payload["flow_diagram_svg"]
assert "否" in payload["flow_diagram_svg"]
assert "是" in payload["flow_diagram_svg"]
assert "#dc2626" in payload["flow_diagram_svg"]
assert "#fecaca" in payload["flow_diagram_svg"]
assert "#10a37f" not in payload["flow_diagram_svg"]
assert "#f97316" not in payload["flow_diagram_svg"]
assert "feDropShadow" not in payload["flow_diagram_svg"]
def test_risk_rule_flow_diagram_uses_risk_level_palette() -> None:
renderer = RiskRuleFlowDiagramRenderer()
def render(severity: str, label: str) -> str:
return renderer.render(
RiskRuleFlowDiagramSpec(
title="测试规则",
domain_label="报销",
severity=severity,
severity_label=label,
fields=(),
start="业务单据提交",
evidence="读取规则字段",
decision="判断是否命中风险",
basis="根据规则字段判断",
pass_text="未命中风险，继续流转",
fail_text=f"命中{label}，进入复核",
)
)
assert "#2563eb" in render("low", "低风险")
assert "#f97316" in render("medium", "中风险")
high_svg = render("high", "高风险")
assert "#dc2626" in high_svg
assert high_svg.count("#dc2626") == 1
assert "#10a37f" not in high_svg