Skip to main content

​
Overview

The Intent Classification metric evaluates the intent behind a given text to determine how well it aligns with one or more predefined expected intents.
IntentClassificationMetric leverages the classifyIntent function to perform this evaluation.

​
Methods

​
classifyIntent Function

This function classifies the intent behind a given text and compares it against expected intents.
  • input: The text to be classified.
  • expectedIntents: A list of expected intents to evaluate against.
It uses a transformer model to determine the intent and assesses the match against the expected intents based on confidence and similarity scores. The function returns a promise that resolves to an object containing the detected intent, its confidence score, and a similarity score with the expected intents.

​
IntentClassificationMetric Class

IntentClassificationMetric classifies and evaluates the intent behind text based on predefined expectations.
  • input: The text to be classified.
  • expectedIntents: The expected intents against which the text will be evaluated.
The evaluateSteps method invokes classifyIntent and returns a detailed result, including a score and reasons. The score reflects how closely the detected intent matches the expected intents, with detailed reasons explaining the similarities and differences.

​
Example

import { evaluate, IntentClassificationMetric } from '@evalkit/core';

evaluate({
    // The text input for intent classification
    input: "Can you help me find a nearby pharmacy?",
    // The expected intent(s) for the classification
    expectedIntents: ["LocateService", "HealthInquiry"],
}, [IntentClassificationMetric])

// outputs
{
  passed: true,
  // The coherence score is 0.9, which is calculated based on the percentage of coherent statements from all statements in the text.
  score: 0.5,
  Reasons: [
    'Detected intents: LocateService, HealthInquiry',
    'Matching intents: 1 out of 2'
  ]
}
⌘I