OpenAIを使ったRAG応答の精度評価：ドキュメント・ベースライン指標
上級
これはEngineering, AI分野の自動化ワークフローで、25個のノードを含みます。主にSet, Evaluation, HttpRequest, ManualTrigger, Agentなどのノードを使用、AI技術を活用したスマート自動化を実現。 OpenAIによるRAG応答の精度評価：ドキュメントベースライン指標
前提条件
- •ターゲットAPIの認証情報が必要な場合あり
- •OpenAI API Key
使用ノード (25)
ワークフロープレビュー
ノード接続関係を可視化、ズームとパンをサポート
ワークフローをエクスポート
以下のJSON設定をn8nにインポートして、このワークフローを使用できます
{
"meta": {
"instanceId": "408f9fb9940c3cb18ffdef0e0150fe342d6e655c3a9fac21f0f644e8bedabcd9",
"templateCredsSetupCompleted": true
},
"nodes": [
{
"id": "976ded9c-2080-484e-ad64-eb3c6e671961",
"name": "ワークフロー実行時",
"type": "n8n-nodes-base.manualTrigger",
"position": [
-760,
-320
],
"parameters": {},
"typeVersion": 1
},
{
"id": "348dfa43-be32-48a8-a9e1-a0641852517e",
"name": "データシート取得",
"type": "n8n-nodes-base.httpRequest",
"position": [
-540,
-320
],
"parameters": {
"url": "https://bitcoin.org/bitcoin.pdf",
"options": {}
},
"typeVersion": 4.2
},
{
"id": "e4854cab-8bc9-4e63-911f-a9a4f5993855",
"name": "Simple Vector Store",
"type": "@n8n/n8n-nodes-langchain.vectorStoreInMemory",
"position": [
-300,
-320
],
"parameters": {
"mode": "insert",
"memoryKey": "evaluations_document_groundness",
"clearStore": true
},
"typeVersion": 1.1
},
{
"id": "f69492e2-c85a-451c-9e02-f5902f214da6",
"name": "Embeddings OpenAI",
"type": "@n8n/n8n-nodes-langchain.embeddingsOpenAi",
"position": [
-320,
-100
],
"parameters": {
"options": {}
},
"credentials": {
"openAiApi": {
"id": "8gccIjcuf3gvaoEr",
"name": "OpenAi account"
}
},
"typeVersion": 1.2
},
{
"id": "65131cfb-ac44-48fa-a648-033c0611e8ea",
"name": "Default Data Loader",
"type": "@n8n/n8n-nodes-langchain.documentDefaultDataLoader",
"position": [
-200,
-100
],
"parameters": {
"options": {
"metadata": {
"metadataValues": [
{
"name": "origin",
"value": "evaluations"
}
]
}
},
"dataType": "binary"
},
"typeVersion": 1
},
{
"id": "0800667d-7b33-4c8b-bc61-d12ccf94a640",
"name": "Recursive Character Text Splitter",
"type": "@n8n/n8n-nodes-langchain.textSplitterRecursiveCharacterTextSplitter",
"position": [
-120,
100
],
"parameters": {
"options": {}
},
"typeVersion": 1
},
{
"id": "04a8a558-07cf-41f4-85a9-37e79be90b52",
"name": "AI Agent",
"type": "@n8n/n8n-nodes-langchain.agent",
"position": [
-300,
620
],
"parameters": {
"options": {
"systemMessage": "You are a helpful reference assistant who fully grounds their answer using retrieved knowledgebase document. It is important not to give any information from other sources as the goal is to be faithful to the contents of the knowledgebase.",
"returnIntermediateSteps": true
}
},
"typeVersion": 2
},
{
"id": "6f6961f7-73c3-4e16-942e-e2da096fec94",
"name": "OpenAI Chat Model",
"type": "@n8n/n8n-nodes-langchain.lmChatOpenAi",
"position": [
-320,
845
],
"parameters": {
"model": {
"__rl": true,
"mode": "list",
"value": "gpt-4o-mini"
},
"options": {}
},
"credentials": {
"openAiApi": {
"id": "8gccIjcuf3gvaoEr",
"name": "OpenAi account"
}
},
"typeVersion": 1.2
},
{
"id": "a063f080-9227-471f-8330-3da574e275d1",
"name": "Simple Vector Store1",
"type": "@n8n/n8n-nodes-langchain.vectorStoreInMemory",
"position": [
-200,
847.5
],
"parameters": {
"mode": "retrieve-as-tool",
"toolName": "bitcoin_whitepaper",
"memoryKey": "evaluations_document_groundness",
"toolDescription": "Call this tool to query over the bitcoin whitepaper to answer technical questions about bitcoin."
},
"typeVersion": 1.1
},
{
"id": "85584b59-0970-4bd9-ac96-f31a0046f20e",
"name": "Embeddings OpenAI1",
"type": "@n8n/n8n-nodes-langchain.embeddingsOpenAi",
"position": [
-120,
980
],
"parameters": {
"options": {}
},
"credentials": {
"openAiApi": {
"id": "8gccIjcuf3gvaoEr",
"name": "OpenAi account"
}
},
"typeVersion": 1.2
},
{
"id": "a99b3732-f8ac-45d3-a2f5-7a1177dc5742",
"name": "チャットメッセージ受信時",
"type": "@n8n/n8n-nodes-langchain.chatTrigger",
"position": [
-540,
725
],
"webhookId": "638886a2-37cc-4edd-9b9f-95b09b255ffb",
"parameters": {
"options": {}
},
"typeVersion": 1.1
},
{
"id": "c21ee596-bc9c-4bb3-8cac-f243a2d5102b",
"name": "評価",
"type": "n8n-nodes-base.evaluation",
"position": [
240,
620
],
"parameters": {
"operation": "checkIfEvaluating"
},
"typeVersion": 4.6
},
{
"id": "64104870-1796-4fcc-8c0d-cc4302de5167",
"name": "No Operation, 何もしない",
"type": "n8n-nodes-base.noOp",
"position": [
460,
720
],
"parameters": {},
"typeVersion": 1
},
{
"id": "91e0a42e-6afc-463b-8f25-7117b057159d",
"name": "ドキュメント取得",
"type": "n8n-nodes-base.set",
"position": [
460,
480
],
"parameters": {
"options": {},
"assignments": {
"assignments": [
{
"id": "a1e75813-76a9-49ba-a8c1-1f2cd62d647c",
"name": "output",
"type": "string",
"value": "={{ $json.output }}"
},
{
"id": "94f08596-6060-4fb6-9998-ac1b3f0e730d",
"name": "documents",
"type": "array",
"value": "={{\n$json.intermediateSteps\n .find(step => step.action.tool === 'bitcoin_whitepaper')\n .observation\n}}"
}
]
}
},
"typeVersion": 3.4
},
{
"id": "8fba17b7-c24b-4a26-a998-e3d2f7acc481",
"name": "データセット行取得時",
"type": "n8n-nodes-base.evaluationTrigger",
"position": [
-760,
525
],
"parameters": {
"sheetName": {
"__rl": true,
"mode": "list",
"value": 633230352,
"cachedResultUrl": "https://docs.google.com/spreadsheets/d/1YOnu2JJjlxd787AuYcg-wKbkjyjyZFgASYVV0jsij5Y/edit#gid=633230352",
"cachedResultName": "RAG Document Groundedness"
},
"documentId": {
"__rl": true,
"mode": "list",
"value": "1YOnu2JJjlxd787AuYcg-wKbkjyjyZFgASYVV0jsij5Y",
"cachedResultUrl": "https://docs.google.com/spreadsheets/d/1YOnu2JJjlxd787AuYcg-wKbkjyjyZFgASYVV0jsij5Y/edit?usp=drivesdk",
"cachedResultName": "96. Evaluations Test"
}
},
"credentials": {
"googleSheetsOAuth2Api": {
"id": "XHvC7jIRR8A2TlUl",
"name": "Google Sheets account"
}
},
"typeVersion": 4.6
},
{
"id": "611a6e12-f52b-4b3c-b74c-238a4f75f612",
"name": "入力再マッピング",
"type": "n8n-nodes-base.set",
"position": [
-540,
525
],
"parameters": {
"options": {},
"assignments": {
"assignments": [
{
"id": "cc14c8c8-5a61-46db-b1c7-6d9ddfb2e4c8",
"name": "chatInput",
"type": "string",
"value": "={{ $json.input }}"
}
]
}
},
"typeVersion": 3.4
},
{
"id": "0c1e25b8-9145-443e-8c9f-34eb62aa7192",
"name": "OpenAI Chat Model1",
"type": "@n8n/n8n-nodes-langchain.lmChatOpenAi",
"position": [
660,
660
],
"parameters": {
"model": {
"__rl": true,
"mode": "list",
"value": "gpt-4.1-mini",
"cachedResultName": "gpt-4.1-mini"
},
"options": {}
},
"credentials": {
"openAiApi": {
"id": "8gccIjcuf3gvaoEr",
"name": "OpenAi account"
}
},
"typeVersion": 1.2
},
{
"id": "a9d8a911-9519-4497-9127-0a2e3da9ae1b",
"name": "Structured Output Parser",
"type": "@n8n/n8n-nodes-langchain.outputParserStructured",
"position": [
860,
660
],
"parameters": {
"jsonSchemaExample": "{\n \"rating\": 1,\n \"reason\": \"The date of birth of Einstein is mentioned clearly in the context.\"\n}"
},
"typeVersion": 1.2
},
{
"id": "038f408d-51f0-4625-b39d-e326f24850b4",
"name": "Document Grounding",
"type": "@n8n/n8n-nodes-langchain.chainLlm",
"position": [
660,
480
],
"parameters": {
"text": "=# Document and AI-generated Response\n## Documents\n{{ $json.documents\n .map(doc => `* ${doc.text.parseJson().pageContent.replaceAll('\\n', ' ')}`)\n .join('\\n')\n}}\n\n## AI-generated Response\n{{ $('Get Documents').first().json.output }}",
"batching": {},
"messages": {
"messageValues": [
{
"message": "=# Instruction\nYou are an expert evaluator. Your task is to evaluate the quality of the responses generated by AI models.\nWe will provide you with the user input and an AI-generated responses.\nYou should first read the user input carefully for analyzing the task, and then evaluate the quality of the responses based on the criteria provided in the Evaluation section below.\nYou will assign the response a rating following the Rating Rubric and Evaluation Steps. Give step-by-step explanations for your rating, and only choose ratings from the Rating Rubric.\n\n# Evaluation\n## Metric Definition\nYou will be assessing groundedness, which measures the ability to provide or reference information included only in the user prompt.\n\n## Criteria\nGroundedness: The response contains information included only in the document. The response does not reference any outside information.\n\n## Rating Rubric\n1: (Fully grounded). All aspects of the response are attributable to the context.\n0: (Not fully grounded). The entire response or a portion of the response is not attributable to the context provided by the documents.\n\n## Evaluation Steps\nSTEP 1: Assess the response in aspects of Groundedness. Identify any information in the response not present in the documents and provide assessment according to the criterion. \nSTEP 2: Score based on the rating rubric. Give a brief rationale to explain your evaluation considering Groundedness."
}
]
},
"promptType": "define",
"hasOutputParser": true
},
"typeVersion": 1.7
},
{
"id": "e867ed3e-df63-40e2-a3f6-0d19bbace13a",
"name": "出力設定",
"type": "n8n-nodes-base.evaluation",
"position": [
1020,
480
],
"parameters": {
"outputs": {
"values": [
{
"outputName": "output",
"outputValue": "={{ $('Get Documents').item.json.output }}"
},
{
"outputName": "documents",
"outputValue": "={{ $('Get Documents').item.json.documents.map(doc => doc.text).join('\\n') }}"
},
{
"outputName": "score",
"outputValue": "={{ $json.output.rating }}"
},
{
"outputName": "score_reason",
"outputValue": "={{ $json.output.reason }}"
}
]
},
"sheetName": {
"__rl": true,
"mode": "list",
"value": 633230352,
"cachedResultUrl": "https://docs.google.com/spreadsheets/d/1YOnu2JJjlxd787AuYcg-wKbkjyjyZFgASYVV0jsij5Y/edit#gid=633230352",
"cachedResultName": "RAG Document Groundedness"
},
"documentId": {
"__rl": true,
"mode": "list",
"value": "1YOnu2JJjlxd787AuYcg-wKbkjyjyZFgASYVV0jsij5Y",
"cachedResultUrl": "https://docs.google.com/spreadsheets/d/1YOnu2JJjlxd787AuYcg-wKbkjyjyZFgASYVV0jsij5Y/edit?usp=drivesdk",
"cachedResultName": "96. Evaluations Test"
}
},
"credentials": {
"googleSheetsOAuth2Api": {
"id": "XHvC7jIRR8A2TlUl",
"name": "Google Sheets account"
}
},
"typeVersion": 4.6
},
{
"id": "7350d6c6-1a9d-4663-b509-de1c9d73588a",
"name": "メトリクス設定",
"type": "n8n-nodes-base.evaluation",
"position": [
1220,
480
],
"parameters": {
"metrics": {
"assignments": [
{
"id": "e3f04944-ab59-4e2f-83f8-6efa36816671",
"name": "score",
"type": "number",
"value": "={{ $json.output.rating }}"
}
]
},
"operation": "setMetrics"
},
"typeVersion": 4.6
},
{
"id": "e5313dae-6eff-4b71-83b1-0c89ff319662",
"name": "付箋",
"type": "n8n-nodes-base.stickyNote",
"position": [
-840,
-500
],
"parameters": {
"color": 7,
"width": 960,
"height": 780,
"content": "## 1. Ready your RAG Vector Store\n[Read more about the Simple Vector Store](https://docs.n8n.io/integrations/builtin/cluster-nodes/root-nodes/n8n-nodes-langchain.vectorstoreinmemory/)\n\nFor this exercise, we'll use the Bitcoin Whitepaper as a source of documents for our evaluation."
},
"typeVersion": 1
},
{
"id": "36da4904-2256-4d4a-9a1d-17ad3af939d4",
"name": "付箋1",
"type": "n8n-nodes-base.stickyNote",
"position": [
-840,
300
],
"parameters": {
"color": 7,
"width": 960,
"height": 900,
"content": "## 2. Setup Your AI Workflow to Use Evaluations\n[Learn more about the Evaluations Trigger](https://docs.n8n.io/integrations/builtin/?utm_source=n8n_app&utm_medium=node_settings_modal-credential_link&utm_campaign=n8n-nodes-base.evaluationTrigger)\n\nThe Evaluations Trigger is a separate execution which does not affect your production workflow in any way. It is manually triggered and automatically pulled datasets from the assigned Google Sheet."
},
"typeVersion": 1
},
{
"id": "d892b15f-126d-4fac-8458-167a9f4bf1b1",
"name": "付箋2",
"type": "n8n-nodes-base.stickyNote",
"position": [
160,
300
],
"parameters": {
"color": 7,
"width": 1260,
"height": 640,
"content": "## 3. Document Groundedness: Is the AI response based on the retrieved documents?\n[Learn more about the Evaluations Node](https://docs.n8n.io/integrations/builtin/?utm_source=n8n_app&utm_medium=node_settings_modal-credential_link&utm_campaign=n8n-nodes-base.evaluation)\n\nFor this evaluation, we simply want to check if the Agent's answer was grounded in any of the documents retrieved from our vector store.\nA higher score represents a greater alignment between the retrieved information and the expected output, indicating that the retriever is effectively sourcing relevant and accurate content to aid the generator in producing contextually appropriate responses."
},
"typeVersion": 1
},
{
"id": "c500b17e-0c27-4121-9f89-40531b83f48e",
"name": "付箋3",
"type": "n8n-nodes-base.stickyNote",
"position": [
-1280,
-500
],
"parameters": {
"width": 380,
"height": 880,
"content": "## Try It Out!\n### This n8n template demonstrates how to calculate the evaluation metric \"RAG document groundedness\" which in this scenario, measures the ability to provide or reference information included only in retrieved vector store documents.\n\nThe scoring approach is adapted from [https://cloud.google.com/vertex-ai/generative-ai/docs/models/metrics-templates#pointwise_groundedness](https://cloud.google.com/vertex-ai/generative-ai/docs/models/metrics-templates#pointwise_groundedness)\n\n### How it works\n* This evaluation works best for an agent that requires document retrieval from a vector store or similar source.\n* For our scoring, we need to collect the agent's response and the documents retrieved and use an LLM to assess if the former is based off the latter.\n* A key factor is to look out information in the response which is not mentioned in the documents.\n* A high score indicates LLM adherence and alignment whereas a low score could signal inadequate prompt or model hallucination.\n\n### Requirements\n* n8n version 1.94+\n* Check out this Google Sheet for a sample data [https://docs.google.com/spreadsheets/d/1YOnu2JJjlxd787AuYcg-wKbkjyjyZFgASYVV0jsij5Y/edit?usp=sharing](https://docs.google.com/spreadsheets/d/1YOnu2JJjlxd787AuYcg-wKbkjyjyZFgASYVV0jsij5Y/edit?usp=sharing)\n\n\n### Need Help?\nJoin the [Discord](https://discord.com/invite/XPKeKXeB7d) or ask in the [Forum](https://community.n8n.io/)!\n\nHappy Hacking!"
},
"typeVersion": 1
}
],
"pinData": {},
"connections": {
"04a8a558-07cf-41f4-85a9-37e79be90b52": {
"main": [
[
{
"node": "c21ee596-bc9c-4bb3-8cac-f243a2d5102b",
"type": "main",
"index": 0
}
]
]
},
"c21ee596-bc9c-4bb3-8cac-f243a2d5102b": {
"main": [
[
{
"node": "91e0a42e-6afc-463b-8f25-7117b057159d",
"type": "main",
"index": 0
}
],
[
{
"node": "64104870-1796-4fcc-8c0d-cc4302de5167",
"type": "main",
"index": 0
}
]
]
},
"611a6e12-f52b-4b3c-b74c-238a4f75f612": {
"main": [
[
{
"node": "04a8a558-07cf-41f4-85a9-37e79be90b52",
"type": "main",
"index": 0
}
]
]
},
"e867ed3e-df63-40e2-a3f6-0d19bbace13a": {
"main": [
[
{
"node": "7350d6c6-1a9d-4663-b509-de1c9d73588a",
"type": "main",
"index": 0
}
]
]
},
"348dfa43-be32-48a8-a9e1-a0641852517e": {
"main": [
[
{
"node": "e4854cab-8bc9-4e63-911f-a9a4f5993855",
"type": "main",
"index": 0
}
]
]
},
"91e0a42e-6afc-463b-8f25-7117b057159d": {
"main": [
[
{
"node": "038f408d-51f0-4625-b39d-e326f24850b4",
"type": "main",
"index": 0
}
]
]
},
"f69492e2-c85a-451c-9e02-f5902f214da6": {
"ai_embedding": [
[
{
"node": "e4854cab-8bc9-4e63-911f-a9a4f5993855",
"type": "ai_embedding",
"index": 0
}
]
]
},
"6f6961f7-73c3-4e16-942e-e2da096fec94": {
"ai_languageModel": [
[
{
"node": "04a8a558-07cf-41f4-85a9-37e79be90b52",
"type": "ai_languageModel",
"index": 0
}
]
]
},
"038f408d-51f0-4625-b39d-e326f24850b4": {
"main": [
[
{
"node": "e867ed3e-df63-40e2-a3f6-0d19bbace13a",
"type": "main",
"index": 0
}
]
]
},
"85584b59-0970-4bd9-ac96-f31a0046f20e": {
"ai_embedding": [
[
{
"node": "a063f080-9227-471f-8330-3da574e275d1",
"type": "ai_embedding",
"index": 0
}
]
]
},
"0c1e25b8-9145-443e-8c9f-34eb62aa7192": {
"ai_languageModel": [
[
{
"node": "038f408d-51f0-4625-b39d-e326f24850b4",
"type": "ai_languageModel",
"index": 0
}
]
]
},
"65131cfb-ac44-48fa-a648-033c0611e8ea": {
"ai_document": [
[
{
"node": "e4854cab-8bc9-4e63-911f-a9a4f5993855",
"type": "ai_document",
"index": 0
}
]
]
},
"e4854cab-8bc9-4e63-911f-a9a4f5993855": {
"main": [
[]
]
},
"a063f080-9227-471f-8330-3da574e275d1": {
"ai_tool": [
[
{
"node": "04a8a558-07cf-41f4-85a9-37e79be90b52",
"type": "ai_tool",
"index": 0
}
]
]
},
"a9d8a911-9519-4497-9127-0a2e3da9ae1b": {
"ai_outputParser": [
[
{
"node": "038f408d-51f0-4625-b39d-e326f24850b4",
"type": "ai_outputParser",
"index": 0
}
]
]
},
"a99b3732-f8ac-45d3-a2f5-7a1177dc5742": {
"main": [
[
{
"node": "04a8a558-07cf-41f4-85a9-37e79be90b52",
"type": "main",
"index": 0
}
]
]
},
"8fba17b7-c24b-4a26-a998-e3d2f7acc481": {
"main": [
[
{
"node": "611a6e12-f52b-4b3c-b74c-238a4f75f612",
"type": "main",
"index": 0
}
]
]
},
"0800667d-7b33-4c8b-bc61-d12ccf94a640": {
"ai_textSplitter": [
[
{
"node": "65131cfb-ac44-48fa-a648-033c0611e8ea",
"type": "ai_textSplitter",
"index": 0
}
]
]
},
"976ded9c-2080-484e-ad64-eb3c6e671961": {
"main": [
[
{
"node": "348dfa43-be32-48a8-a9e1-a0641852517e",
"type": "main",
"index": 0
}
]
]
}
}
}よくある質問
このワークフローの使い方は？
上記のJSON設定コードをコピーし、n8nインスタンスで新しいワークフローを作成して「JSONからインポート」を選択、設定を貼り付けて認証情報を必要に応じて変更してください。
このワークフローはどんな場面に適していますか？
上級 - エンジニアリング, 人工知能
有料ですか？
このワークフローは完全無料です。ただし、ワークフローで使用するサードパーティサービス（OpenAI APIなど）は別途料金が発生する場合があります。
関連ワークフロー
AIエージェントの応答正確性をOpenAIとRAGASメソッドで評価
OpenAIとRAGASメソッドを使用してAIエージェントの応答の正確性を評価する
Set
Code
Merge
+
Set
Code
Merge
27 ノードJimleuk
エンジニアリング
AIエージェントの応答関連性をOpenAIとコサイン類似度で評価
OpenAIとコサイン類似度を使用してAIエージェントの応答の関連性を評価する
Set
Code
Evaluation
+
Set
Code
Evaluation
20 ノードJimleuk
エンジニアリング
評価指標例：RAGドキュメント関連性
評価指標サンプル：RAGドキュメント関連性
Set
Evaluation
Google Sheets
+
Set
Evaluation
Google Sheets
26 ノードDavid Roberts
エンジニアリング
評価指標: 応答の類似性
評価指標: 応答の類似性
Set
Code
Merge
+
Set
Code
Merge
21 ノードJimleuk
エンジニアリング
評価指標: 概要
評価指標: 概要
Set
Webhook
Evaluation
+
Set
Webhook
Evaluation
17 ノードJimleuk
エンジニアリング
AI スマートアシスタント: Supabase ストレージと Google Drive ファイルとの対話
AIワンチャットボット：SupabaseストレージとGoogle Driveのファイルと対話
If
Set
Wait
+
If
Set
Wait
62 ノードMark Shcherbakov
エンジニアリング
ワークフロー情報
難易度
上級
ノード数25
カテゴリー2
ノードタイプ16
作成者
Jimleuk
@jimleukFreelance consultant based in the UK specialising in AI-powered automations. I work with select clients tackling their most challenging projects. For business enquiries, send me an email at hello@jimle.uk LinkedIn: https://www.linkedin.com/in/jimleuk/ X/Twitter: https://x.com/jimle_uk
外部リンク
n8n.ioで表示 →
このワークフローを共有