What is a Knowledge Base?

Overview

The Knowledge Base is your chatbot’s brain. It is a collection of stored responses that is available for the chatbot to draw from when answering questions. Similar to a human brain, the Knowledge Base will develop and grow over time with your help. 

Simply put, the bot matches to the best, approved answer in your Knowledge Base by using the closest Mainstay and Organization synonym question.  This matched response will be provided to the contact who is asking the question.

Knowledge Base Size

We measure the size of your Knowledge Base by the number of understandings you have created, and not by the number of questions. The number of questions that can be mapped to a particular answer is infinite and ever-growing as your chatbot learns to better recognize human language.

When you view your Knowledge Base, you will only see a single question mapped to a response. This question is just a representation of the dozens of question synonyms your chatbot can process and understand.

Knowledge Base Organization 

The Knowledge Base is structured this way in order to capture as many questions as possible. Each contact could ask the same question slightly differently. For example, “What is the deadline for the FAFSA?” and “When is the FAFSA due?” are mapped to the same answer about the FAFSA deadline. We refer to these questions as “synonyms.” Together, each group of "fuzzy" (synonym) questions and their matching answer is called an understanding.

The Knowledge Base consists of questions and answer pairings called "Understandings".  On your Knowledge Base page, you will see one representative question and its associated answer, which you can edit. If you click on that understanding, you will also be able to see all of the synonym or "fuzzy" questions that are also mapped to that answer. These are the questions used in the bots matching process. 

How does the Knowledge Base grow?

The Knowledge Base grows by increasing answers or increasing number of question synonyms mapped to those responses, via two distinct processes:

1. Seeding: direct loading of answers into your chatbot's Knowledge Base. 

2. Triage: reviewing questions sent to the chatbot and giving it feedback on how to answer them. Triage is done when you're training the chatbot before it launches all the way to when you've launched it and are talking to students.

When a question is received by your chatbot, it immediately processes that question via its AI algorithm and produces a best-match response. When the bot his unsure of its response, it will flag the question for review by our team. Each of these responses is then reviewed by human staff to ensure that your bot is processing things correctly and learning when it needs to. If the answer provided by the bot was incorrect or an “I don’t know,” the staff member can train your bot to match this question to its correct answer in the future or teach it a completely new answer.

We expose your chatbot to both of these processes via each of the Knowledge Base development steps: seeding, training, reviewing, testing, and student exposure. Each of these steps takes your Knowledge Base to a new stage of development.

During the seeding period, a prioritized set of answers is crafted and loaded into your Knowledge Base based on information found on your website or other documents. (See Knowledge Seeding for more information) While in training mode, your training team sends in questions for your chatbot to process and learn. The review step allows you to review all of these responses for accuracy and tone. (See Knowledge Base Review) Testing provides a final opportunity to help your bot recognize questions better and learn relevant information before its exposed to students. (See Knowledge Base Testing) Once you launch your bot, your Knowledge Base will continue to grow as your chatbot is exposed to new student questions. You should also check out: How Your Chatbot Learns