Overview
The Document Relevance evaluator determines whether a retrieved document contains information relevant to answering a specific question. This is essential for evaluating RAG (Retrieval-Augmented Generation) systems where document quality directly impacts response quality.
When to Use
Use the Document Relevance evaluator when you need to:
- Evaluate RAG retrieval quality - Assess whether your retrieval system is returning useful documents
- Debug poor RAG responses - Identify if issues stem from retrieval vs generation
- Compare retrieval strategies - Test different embedding models, chunking strategies, or search algorithms
- Monitor retrieval in production - Track document relevance over time
This evaluator assesses individual document relevance to a query. For evaluating whether a response is faithful to its context, use the Faithfulness evaluator instead. Supported Levels
The level of an evaluator determines the scope of the evaluation in OpenTelemetry terms. Some evaluations are applicable to individual spans, some to full traces or sessions, and some are applicable at multiple levels.
| Level | Supported | Notes |
|---|
| Span | Yes | Best for retriever spans. Evaluate each retrieved document individually. |
| Field | Type | Description |
|---|
input | string | The user’s query or question |
document_text | string | The document text to evaluate for relevance |
In TypeScript, the field is named documentText (camelCase) instead of document_text (snake_case).
- Evaluate one document at a time - Run the evaluator separately for each retrieved document
- Use the full document chunk - Include the complete text that was retrieved, not just snippets
- Include metadata if helpful - Document titles or sources can provide useful context
Output Interpretation
The evaluator returns a Score object with the following properties:
| Property | Value | Description |
|---|
label | "relevant" or "unrelated" | Classification result |
score | 1.0 or 0.0 | Numeric score (1.0 = relevant, 0.0 = unrelated) |
explanation | string | LLM-generated reasoning for the classification |
direction | "maximize" | Higher scores are better |
metadata | object | Additional information such as the model name. When tracing is enabled, includes the trace_id for the evaluation. |
- Relevant (1.0): The document contains information that can help answer the question
- Unrelated (0.0): The document does not contain relevant information for the question
Usage Examples
from phoenix.evals import LLM
from phoenix.evals.metrics import DocumentRelevanceEvaluator
# Initialize the LLM client
llm = LLM(provider="openai", model="gpt-4o")
# Create the evaluator
relevance_eval = DocumentRelevanceEvaluator(llm=llm)
# Inspect the evaluator's requirements
print(relevance_eval.describe())
# Evaluate a single document
eval_input = {
"input": "What is the capital of France?",
"document_text": "Paris is the capital and largest city of France."
}
scores = relevance_eval.evaluate(eval_input)
print(scores[0])
# Score(name='document_relevance', score=1.0, label='relevant', ...)
import { createDocumentRelevanceEvaluator } from "@arizeai/phoenix-evals";
import { openai } from "@ai-sdk/openai";
// Create the evaluator
const relevanceEvaluator = createDocumentRelevanceEvaluator({
model: openai("gpt-4o"),
});
// Evaluate a document
const result = await relevanceEvaluator.evaluate({
input: "What is the capital of France?",
documentText: "Paris is the capital and largest city of France.",
});
console.log(result);
// { score: 1, label: "relevant", explanation: "..." }
Evaluating Multiple Documents
To evaluate all documents returned by a retriever, iterate over each document:
from phoenix.evals import LLM
from phoenix.evals.metrics import DocumentRelevanceEvaluator
llm = LLM(provider="openai", model="gpt-4o")
relevance_eval = DocumentRelevanceEvaluator(llm=llm)
query = "What are the symptoms of COVID-19?"
retrieved_documents = [
"COVID-19 symptoms include fever, cough, and fatigue.",
"The history of coronaviruses dates back to the 1960s.",
"Treatment options for COVID-19 include antiviral medications."
]
# Evaluate each document
for i, doc in enumerate(retrieved_documents):
scores = relevance_eval.evaluate({
"input": query,
"document_text": doc
})
print(f"Document {i+1}: {scores[0].label} ({scores[0].score})")
# Document 1: relevant (1.0)
# Document 2: unrelated (0.0)
# Document 3: unrelated (0.0)
import { createDocumentRelevanceEvaluator } from "@arizeai/phoenix-evals";
import { openai } from "@ai-sdk/openai";
const relevanceEvaluator = createDocumentRelevanceEvaluator({
model: openai("gpt-4o"),
});
const query = "What are the symptoms of COVID-19?";
const retrievedDocuments = [
"COVID-19 symptoms include fever, cough, and fatigue.",
"The history of coronaviruses dates back to the 1960s.",
"Treatment options for COVID-19 include antiviral medications.",
];
// Evaluate each document
for (const [i, doc] of retrievedDocuments.entries()) {
const result = await relevanceEvaluator.evaluate({
input: query,
documentText: doc,
});
console.log(`Document ${i + 1}: ${result.label} (${result.score})`);
}
from phoenix.evals import LLM
from phoenix.evals.metrics import DocumentRelevanceEvaluator
llm = LLM(provider="openai", model="gpt-4o")
relevance_eval = DocumentRelevanceEvaluator(llm=llm)
# Example with different field names
eval_input = {
"query": "What is machine learning?",
"chunk": "Machine learning is a subset of AI that enables systems to learn from data."
}
# Use input mapping to match expected field names
input_mapping = {
"input": "query",
"document_text": "chunk"
}
scores = relevance_eval.evaluate(eval_input, input_mapping)
import { bindEvaluator, createDocumentRelevanceEvaluator } from "@arizeai/phoenix-evals";
import { openai } from "@ai-sdk/openai";
const relevanceEvaluator = createDocumentRelevanceEvaluator({
model: openai("gpt-4o"),
});
// Bind with input mapping for different field names
const boundEvaluator = bindEvaluator(relevanceEvaluator, {
inputMapping: {
input: "query",
documentText: "chunk",
},
});
const result = await boundEvaluator.evaluate({
query: "What is machine learning?",
chunk: "Machine learning is a subset of AI that enables systems to learn from data.",
});
Configuration
For LLM client configuration options, see Configuring the LLM.
Viewing and Modifying the Prompt
You can view the latest versions of our prompt templates on GitHub. The evaluators are designed to work well in a variety of contexts, but we highly recommend modifying the prompt to be more specific to your use case. Feel free to adapt them.
from phoenix.evals.metrics import DocumentRelevanceEvaluator
from phoenix.evals import LLM, ClassificationEvaluator
llm = LLM(provider="openai", model="gpt-4o")
evaluator = DocumentRelevanceEvaluator(llm=llm)
# View the prompt template
print(evaluator.prompt_template)
# Create a custom evaluator based on the built-in template
custom_evaluator = ClassificationEvaluator(
name="document_relevance",
prompt_template=evaluator.prompt_template, # Modify as needed
llm=llm,
choices={"relevant": 1.0, "unrelated": 0.0},
direction="maximize",
)
import { DOCUMENT_RELEVANCE_CLASSIFICATION_EVALUATOR_CONFIG, createDocumentRelevanceEvaluator } from "@arizeai/phoenix-evals";
import { openai } from "@ai-sdk/openai";
// View the prompt template
console.log(DOCUMENT_RELEVANCE_CLASSIFICATION_EVALUATOR_CONFIG.template);
// Create a custom evaluator with a modified template
const customEvaluator = createDocumentRelevanceEvaluator({
model: openai("gpt-4o"),
promptTemplate: DOCUMENT_RELEVANCE_CLASSIFICATION_EVALUATOR_CONFIG.template, // Modify as needed
});
Using with Phoenix
Evaluating Traces
Run evaluations on traces collected in Phoenix and log results as annotations:
Running Experiments
Use the Document Relevance evaluator in Phoenix experiments:
API Reference
