6 Training your assistant

 

This chapter covers

  • Training an assistant to recognize intents from user input
  • Collecting a viable training data set for your AI assistant
  • Analyzing how training data affects an assistant’s performance

Training is an important part of building an AI assistant. Good training increases the intelligence of the assistant and enables it to be more useful to your users. Training influences almost every single success metric and is important no matter which type of assistant you are building. This chapter will teach you what you need to start training your assistant.

Fictitious Inc. prototyped its first conversational AI using sample data from its AI provider. This data trained the assistant to be proficient at recognizing generic customer service intents. But what if a suitable sample data set did not exist? How could Fictitious Inc. build a training data set and use it to teach its conversational AI? Let’s explore a scenario in which Fictitious Inc. was not given a precreated training set and had to build its own.

6.1 Training an AI assistant

6.2 Finding training data

6.2.1 Production logs

6.2.2 A mock user interface

6.2.3 Subject-matter experts

6.2.4 Organizing training data into intents

6.3 Assessing whether you have the right training data

6.3.1 Training data variety

6.3.2 Training data volume

6.3.3 Training data veracity

Summary