Knowledge Base

Overview

The Knowledge Base is a repository of information that your learners can access through chatting with the Mainstay bot. When a learner sends in a message (outside of a Live Chat or Script), our MatchMaker AI looks through your Knowledge Base to find the best match.

How does matching work?

A learner's incoming message is passed through a few different systems to ensure we send the best response available.

1. First, we check for sensitive keywords to identify emergency situations. We highly recommend setting up email escalations for these so you can be notified immediately.
2. We also check for #command keywords to see if the learner intends to trigger a #Command.
3. Next, we check for entities, like different names for your institution (like "Mainstay University" vs. "The University of Mainstay"). Learners might use different terms that effectively refer to the same thing. This helps simplify the incoming message for our AI matching system.
4. Next, we embed the incoming message in a vector space, turning words into numbers that represent concepts. This means that similar concepts are "close together" even if the words themselves aren't that similar, like "instructor" and "professor". At this point, we can look for fuzzy questions in the same region of that vector space.
   • In July 2024, we upgraded this system to use a more advanced GPT model to better capture synonyms, slang terms, institution-specific terminology, and typos. This is rolling out to partners in waves over the course of July.
5. If there are any matches found, we select the closest match where the understanding has an approved answer.
   • If it has a personalized response for which the learner meets the criteria, then they'll receive that, or else we send the default response.
6. If there are no matches found above the matching threshold, the learner will instead receive a fallback response. Note that there are different fallbacks for different situations.

Search, Sort, and Filter

When searching for an understanding, you can search by topic or question. You can sort the table by topic, summary question, # responses, and # fuzzy questions.

You can also use the sidebar to filter your view:

1. Select "Understandings" to view the entire Knowledge Base.
2.  Filter by approval status:
   • Approved: understandings available for your bot to send
   • Not Approved: understandings not available for your bot to send
   • No Answer: understandings with no answer 
3. Filter by understanding topic

Understandings

The Understandings table includes information about each understanding in that section:

• Topic Area: the primary organization structure for your Knowledge Base. The topics are available as filters in the sidebar. Every understanding has a unique topic/subtopic combination.
• Summary Question: the main idea of the understanding.
• Response: the first approved default answer.
  • If no default answer is approved, it will display the first default answer.
  • If there are no default answers, it will display a Mainstay suggested answer, if available.
• # Responses: the number of responses attached to this understanding, including personalized responses.
• # Fuzzy Questions: the number of fuzzy questions attached to this understanding.

For more information on creating and editing understandings, visit Knowledge Base Understandings.

Other Knowledge Base tools

The sidebar includes many other features:

• #Commands
• Fallback Responses
• Organization Attributes
• Knowledge Maintenance
• Knowledge Scraping / Knowledge Sources