Knowledge Base Understandings

Overview

Your bot's Knowledge Base is made up of understandings. You could think of each one as a kernel of knowledge that your bot has learned. Each understanding is a collection of question synonyms and the answer(s) that have been crafted for your bot to use when its AI matches an incoming message to that understanding. 

Most understandings in your bot's Knowledge Base come from Mainstay's collection of Understandings, built from all the questions and messages that have come into our bots in the past. Those understandings cover a lot, but by no means do they cover everything. When you encounter something that's missing or that your bot isn't able to answer, you can add it to your bot's Knowledge Base by creating a new understanding. Below is the "anatomy" of an understanding:

Responses

• A Default Response is the general answer to the understanding's questions. 
• A Personalized Response is a more specific response you can customize based on specific criteria if the answers to your questions can vary depending on the scenario. 
• An Interactive Response is a response that utilizes a script to clarify the intent of a question, and connect the learner with important contextual information. (This can be personalized or a default.)
• Other tools to leverage in Responses:
  • Organization Attributes
  • Contact Fields
  • Shortened links (mainst.ai)
  • Approval Status

Fuzzy Questions

• The Summary Question represents the intent or central meaning of an understanding. It should be unique and simple.
• Organization-specific Questions are synonyms or close variations of the original Summary Question. These are unique to each bot/institution and are added by your team.
• Mainstay Global Questions are synonyms or close variations of the original Summary Question. These are automatically added to all default understandings we provide.

Creating an Understanding

To get started, click + New Understanding button at the top left of the Knowledge Base 

Summary Question

1. Enter the Summary Question that captures the intent or core meaning of the Understanding you want to create.
2. Click Search.
3. You will see a list of the Understandings your bot thinks are closest in meaning to your Summary Question. Click the arrow, >, to the left of the question to see a preview of your bot’s answer for that question.
4. If any of the search results show an existing understanding that matches the one you are planning to create, click the arrow next to that result, then click Edit this Understanding to be redirected to that Understanding. 
5. No good matches? If you do not see a fitting existing understanding among the search results, click Add New Understanding at the bottom of the results.

Topic and Subtopic

1. In the “Create New Understanding” panel, you will see the Summary Question from the previous step. It’s also added automatically to the Fuzzy Questions.
2. You can edit the Summary Question if needed. As in the first step, the search will show similar understandings to help you verify that you are not creating a duplicate. 
3. Next, add Topics that reflect the content of your understanding. These will be used to organize your understanding within the Knowledge Base. The categories in the first dropdown correspond to the ones listed on the left side of the Knowledge Base. The categories in the second dropdown correspond to the sub-categories already present in your Knowledge Base. You can use select an existing sub-category or create a new one.
4. Click Next on the bottom to go to the Edit Understanding screen.

Editing an Understanding

Responses

After creating a new understanding, or when clicking into an existing understanding, you are able to add one or more responses. These can be:

• Default Responses
• Personalized Responses
• Interactive Responses

Fuzzy Questions

To ensure the MatchMaker AI is able to surface these responses when a learner asks a similar question, Mainstay uses Fuzzy Questions - different ways someone might phrase this idea.

Deleting an Understanding

• Understandings
  • Use a reactive strategy when adding understandings to your bot's Knowledge Base. Rather than trying to proactively anticipate questions your bot might be asked, focus your efforts on adding new understandings based on the messages that come in from contacts, or via testing.
  • Understandings should be as unique as possible. Overlap in subject areas may cause unintended bot mismatching.
  • Your bot does best answering short, specific questions. Paragraph long questions will confuse the bot. Keep your fuzzy questions short and detailed.
• Responses
  • We recommend providing more general answers (with a linked resource) in response to very specific questions.
  • Answers should be kept around 160 characters. 
  • General answers will make sure the bot is answering all variations of the question.
  • Responses should not begin with a "yes" or "no". Learners can ask yes-or-no questions in different ways, and answers should focus on the information that is most important to learners.
  • Make sure to write your answers in the voice of the bot.
  • Include emoji when appropriate, but avoid ending your message with an emoji, as this sometimes gets flagged as spam by phone carriers.
  • Avoid starting or ending messages with links to prevent messages from being blocked as spam. Also, be careful about putting links next to emojis - on some devices, the emoji could become part of the link, so it would not lead to the right page.
  • Use a custom link shorter like Rebrandly or our built-in mainst.ai Link Shortener to avoid long links. (Public shorteners like bit.ly are frequently blocked by phone carriers.)
• Fuzzy Questions
  • Each custom understanding should have at least 10-20 fuzzy questions to help the AI effectively match to understandings.
  • Fuzzy questions should be representative of simple ways that someone could ask a question. Long, multi-sentence questions should be avoided, as well as overlapping content with other understandings/questions.
  • The maximum length of a fuzzy question should be around 140-160 characters.
  • All three of these questions have the same intent and could live on the same understanding; avoid making too many separate understandings when they would all have the same answer, and instead add those as fuzzy questions:
    • "is there a campus security office"
    • "is there not a campus security office"
    • "can you tell me about the campus security office"
      
• Testing
  • Use Test the Bot when creating understandings to double check your work and verify the KB actions you're taking.