Confidence scores play a key role in generating chatbot responses but function differently depending on the knowledge source.

This article describes the role that the confidence score plays with Q&A pairs as the knowledge source and intents or triggers in a conversation workflow.

<aside> 💡

If you are using an Azure AI Search and OpenAI configuration, please refer to the articles listed below to learn how confidence scores serve as a retrieval factor when leveraging these Azure services:

</aside>

How to View the RAG Raw Results

Evaluate Zammo.ai

Technical Definition

In machine learning, a Confidence Score is a number between zero and one that represents the likelihood that the output of a machine learning model is correct and will satisfy a user’s request.

Translating that to Conversational AI, Confidence Score represents the likelihood that your bot is returning the correct response to an utterance. Zammo translates the confidence score into a percentage between zero and one hundred percent.

Working with Confidence Scores

Answers that do not meet the confidence score threshold will not be returned by the bot. In addition to finding no matches or a weak match to an utterance, answers may have a low confidence score if too many matches are found. Therefore, it may be useful to lower the confidence score threshold rather than adding more question alternatives or new question-answer pairs.

Follow the step-by-step directions HERE to adjust the confidence score threshold for Q&A pairs.

Please see this article to adjust the confidence score threshold for intents and triggers: Intent Confidence Score Threshold.