Fuzzy Questions

Overview

Each Mainstay understanding has a "Summary Question" that represents the main idea this understanding is about. But someone might not ask "Where is the dining hall?" They might say, "What's the location of the cafeteria?" - completely different words, but same intent. Fuzzy Questions are similar sentences that mean the same things, to anticipate different ways someone might ask, so our AI can direct them all to the right answer.

Global Fuzzy Questions

When your organization is first onboarded, the Knowledge Base will generally include many understandings that already have "Global" Fuzzy Questions. (See the FAQ section for more details.) These can be individually turned off if they are not applicable to your institution.

Organization Fuzzy Questions

You can also add your own fuzzy questions individually by clicking the + Add Question button on the Understanding panel.

You can also use generative AI to add many fuzzy questions quickly:

1. Navigate to the Understanding and clicking the Generate 10 Questions button.
   • 
2. You will see a loading state. Generating these questions may take a few seconds.
   • 
3. The new fuzzy questions will be outlined with a dark border. You can still edit these or delete them as desired.
   • 

Note: If "Generate Fuzzy Questions for new Understandings" is enabled, then all newly created Understandings will automatically have 10 fuzzy questions added via generative AI.

Fuzzy Question Matching in Conversations

When a learner sends in a message, Mainstay first checks if this learner is in a Live Chat or an interactive Script (like a Campaign or #Command). If not, the AI attempts to match their incoming message to a fuzzy question. On the Conversations page, you can see what fuzzy question was matched to, and what understanding it's part of. 

Sometimes the incoming message exactly matches a fuzzy question; other times, it's just very similar, such as with synonyms or typos. There are also cases where the match is based on a text pattern, such as for sensitive words or #Commands:

FAQ

Do I need to add Fuzzy Questions to existing understandings?

Adding custom fuzzy questions is NOT required for "global" understandings, because these typically have hundreds of variations already. Custom fuzzy questions can provide more coverage to help our MatchMaker AI better understand messages from your learners, especially if you use specific terminology. (For example, if your school library is called the "Glenn G. Bartle Library", maybe learners won't ask "When is the library opened?" but instead "When does Bartle open?")

For new, custom understandings, it is very important to add fuzzy questions, or the MatchMaker AI will be very unlikely to match to these. Mainstay recommends at least 10-20 for each understanding.

Where do Mainstay Fuzzy Questions come from?

There are thousands and thousands of Mainstay Fuzzy Questions throughout the Knowledge Base which have been generated over time by the Mainstay knowledge building team. When an incoming message did not match to any understanding, it would be reviewed and added to the right understanding's fuzzy questions list to improving matching to similar messages in the future. 

To learn more about how the AI works in Mainstay's chatbots, check out this blog post. 

Why do some of the Mainstay Fuzzy Questions contain words beginning with an "@" symbol?

Mainstay Questions occasionally include words beginning with an “@” symbol, like @college or @orientation, which are generic representations of an institution’s information and naming conventions These "entities" allow the Mainstay Fuzzy Questions to be applied across different bots’ Knowledge Bases; behind the scenes, each institution has a list of keywords that we expect to receive from learners that really map to those entities.

Does a Fuzzy Question have to be present in an Understanding verbatim for my bot to be able to identify it?

No! Good thing, because that would be a lot of questions.

Mainstay leverages AI to match incoming messages to the best response from the Knowledge Base. Simply put, your bot's AI uses a complex set of processes to compare each incoming message to every fuzzy question, and identifies the ones that are most similar. If the closest match is above the confidence threshold, the bot will send that understanding's answer (or, in the case of personalized responses, the specific answer this learner is eligible for).