Struct IntentClassifier 

pub struct IntentClassifier { /* private fields */ }

Intent classifier for query analysis and refusal detection.

Uses pattern matching (keywords and regex) to classify query intent and determine whether queries should be answered or refused. Supports detection of inappropriate content, ambiguous queries, and provides reformulation suggestions.

Classification Process

1. Check for inappropriate content (if enabled)
2. Score query against each intent pattern
3. Detect ambiguity if secondary intents have high confidence
4. Determine if query should be refused based on thresholds
5. Generate reformulation suggestions if needed

Supported Intents

• Factual, Definitional, Relational
• Temporal, Causal, Comparative
• Exploratory, Summary
• Inappropriate, Ambiguous (special cases)

Example

use graphrag_core::rograg::IntentClassifier;

let classifier = IntentClassifier::new()?;
let result = classifier.classify("What is X?")?;

if result.should_refuse {
    println!("Refusing: {}", result.refusal_reason.unwrap());
} else {
    println!("Intent: {:?}", result.primary_intent);
}

Implementations

impl IntentClassifier

pub fn new() -> Result<IntentClassifier, GraphRAGError>

Create a new intent classifier with default configuration.

Initializes pattern databases for all supported intent types.

Returns

Returns an IntentClassifier with default thresholds and all detection features enabled, or an error if pattern initialization fails.

Errors

Returns an error if regex pattern compilation fails during initialization.

pub fn with_config( config: IntentClassificationConfig, ) -> Result<IntentClassifier, GraphRAGError>

Create a new intent classifier with custom configuration.

Arguments

• config - Custom configuration for classification behavior

Returns

Returns an IntentClassifier with the specified configuration, or an error if pattern initialization fails.

Errors

Returns an error if regex pattern compilation fails during initialization.

pub fn classify(&self, query: &str) -> Result<IntentResult, GraphRAGError>

Classify the intent of a query.

Analyzes the query to determine its primary intent, detect ambiguity, identify inappropriate content, and decide whether to refuse answering.

Arguments

• query - The query text to classify

Returns

Returns an IntentResult containing:

• Primary and secondary intents with confidence scores
• Whether to refuse answering
• Refusal reason (if applicable)
• Suggested reformulation (if configured)
• Query complexity score

Example

let result = classifier.classify("What is entity X?")?;
println!("Intent: {:?}, Confidence: {:.2}",
         result.primary_intent, result.confidence);

pub fn get_config(&self) -> &IntentClassificationConfig

Get configuration

pub fn update_config( &mut self, config: IntentClassificationConfig, ) -> Result<(), GraphRAGError>

Update configuration

pub fn get_statistics(&self) -> IntentClassificationStats

Get intent statistics