AI 知識萃取

Human → AI

AI 從文件中提取實體並建構知識圖譜，領域專家進行驗證。

5 個節點 · 5 條連接enterprise

agenthumandb

視覺化

上傳文件human

團隊上傳報告、Wiki、會議記錄與操作手冊。

↓sequential→ AI 實體提取

AI 實體提取agent

從文本中識別人物、概念、流程與關係。

↓sequential→ 知識圖譜建構

知識圖譜建構db

將提取的實體與邊關係寫入圖形資料庫。

↓sequential→ 專家審查

專家審查human

領域專家驗證實體準確性與關係標籤。

↓conditional→ 發佈至 Wiki

↓loop→ AI 實體提取

發佈至 Wikiapi

將已驗證的知識圖譜匯出至內部 Wiki 與搜尋索引。

在編輯器開啟

uc-knowledge-extraction.osop.yaml

osop_version: "1.0"
id: "ai-knowledge-extraction"
name:"AI 知識萃取"
description:"AI 從文件中提取實體並建構知識圖譜，領域專家進行驗證。"

nodes:
  - id: "upload_docs"
    type: "human"
    subtype: "input"
    name: "上傳文件"
    description: "團隊上傳報告、Wiki、會議記錄與操作手冊。"

  - id: "entity_extract"
    type: "agent"
    subtype: "llm"
    name: "AI 實體提取"
    description: "從文本中識別人物、概念、流程與關係。"
    timeout_sec: 180

  - id: "graph_build"
    type: "db"
    subtype: "query"
    name: "知識圖譜建構"
    description: "將提取的實體與邊關係寫入圖形資料庫。"

  - id: "expert_review"
    type: "human"
    subtype: "review"
    name: "專家審查"
    description: "領域專家驗證實體準確性與關係標籤。"
    security:
      approval_gate: true

  - id: "publish_wiki"
    type: "api"
    subtype: "rest"
    name: "發佈至 Wiki"
    description: "將已驗證的知識圖譜匯出至內部 Wiki 與搜尋索引。"

edges:
  - from: "upload_docs"
    to: "entity_extract"
    mode: "sequential"
  - from: "entity_extract"
    to: "graph_build"
    mode: "sequential"
  - from: "graph_build"
    to: "expert_review"
    mode: "sequential"
  - from: "expert_review"
    to: "publish_wiki"
    mode: "conditional"
    when: "review.approved == true"
  - from: "expert_review"
    to: "entity_extract"
    mode: "loop"
    label: "Re-extract with corrected labels"