Knowledge Base Training

Overview

Training your chatbot is one of the most exciting steps in the development of your Knowledge Base. The training process will expand your chatbot’s knowledge from a starter list of answers to a collection of answers that is more relevant to your contact’s needs at the time of launch.

How to Train Your Chatbot

The best way of training a chatbot new information is by asking it questions that are common amongst your target population. During training, your bot training crew will send questions that you expect contacts will ask when they first gain access to the chatbot. The Mainstay team will review each of these questions and teach the chatbot how to respond to any new ones that have not been covered by the initial seeding process.

Bot Training Team

When you begin working with Mainstay, identify a group of students who will participate in the training process. Remember: their job will be to message questions to the chatbot for one day. It is a fun responsibility, but they should be comfortable interacting with the bot’s limited knowledge at this stage (see “Training Expectations" below). A good-sized group typically ranges between 5-10 people, but it can be as large as desired. You should also check out Assembling Your Chatbot Development Team for more information.

Training Expectations

Keep in mind, your chatbot will always be learning, so the training never stops! However, the official training period for your training crew is typically one day.

While we encourage trainers to have fun with the bot and send in questions to test the conversational feel of the bot, we do want to keep the questions as realistic as possible for what we might expect from your target group of students at launch. For example, if our target group is prospective students, it is unlikely that they will ask a question about the graduation request process.

It's important to remember that, during this stage of Knowledge Base development, your chatbot is learning how to respond to each of the questions you are sending, so you will notice it often says “I don’t know” or gives you an incorrect response. It’s important for your group of trainers to be aware of this, so they don’t expect to interact with a launch-ready chatbot at this stage.

Bot Learning

After your bot training crew sends in their questions, our team will begin the reviewal process to teach the bot to better respond to those questions in the future. This process, known as "triage", allows our team to help the bot learn so that next time it receives a similar question, it can confidently provide the correct response. 

When the bot is confident in its responses, it is automatically approved by the AI system. If the bot doesn’t feel confident but gives a response,  we will review the response and either approve it as correct or update it to a better, correct response. If the bot does not know the answer at all, that question will be flagged so that we can teach it an appropriate response. Check out the above flow chart to better understand how the bot learns. You can also check out How Your Chatbot Learns.