liveBook · Manning

Golden intent

As explained in Effective Conversational AI

A golden intent is the authoritative intent label assigned to a conversational input for evaluation or testing.

A golden intent is the authoritative intent label assigned to a conversational input for evaluation or testing. It belongs on an utterance only when the user’s goal is clear and relevant to the domain. section 4.4.2

When no golden intent applies

A banking chatbot utterance such as “account” could mean opening, closing, or checking an account, so it is usually too ambiguous for classifier accuracy testing. “Pop culture trivia” is intelligible but unrelated to an electric-utility domain and should instead support negative testing. section 4.4.2

Treat an utterance as unsuitable for golden-intent labeling when it is:

  • Unrelated to the bot’s domain or intended purpose. section 4.4.2
  • Too ambiguous to express a clear goal, such as “driver license,” “that one,” or “2.” section 4.4.2
  • A flow response that lacks standalone meaning, unless it aligns with an intent used within that flow. section 4.4.2

How Effective Conversational AI covers it

How deep it goes
Distinguishes assignable utterances from unrelated or ambiguous inputs, but stops short of describing a labeling workflow.
Examples
Uses “account,” “driver license,” “that one,” “2,” and off-domain pop-culture questions as screening examples.
What it assumes
Assumes a defined domain, bot purpose, and intent taxonomy; clarifies why unclear goals cannot receive labels.
What you can do after
Evaluate classifier performance by separating assignable test utterances from negative or unsuitable inputs.
How it connects
Ties golden intent to classifier test sets, negative testing, ambiguous utterances, and information-gathering flows.

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Effective Conversational AI. The section links point to the parts of the book it draws on.