This article will give you a brief introduction to the formalized approach that can be incorporated into the project planning for Intent training, particularly from the standpoint of time management.

Introduction

With great success, many of our customers have used the Intent feedback analyzer to continually enhance their NLU models. However, we are frequently asked about a more systematic approach to Intent training in projects, particularly with regard to time management within the project. The purpose of this article is to present a formalized approach that can be implemented into the project planning process.

Note: It is assumed that the intent definition and scoping phases have already been completed, and we are just concerned with getting the recognition to the desired level.

Overview

The following three phases can be used to outline the intent creation/training process:

1. Improving Intent recognition
   1. Machine learning
   2. Cognigy Script
2. Resolving Intent conflicts
   1. Moving Example Sentences to the better-suited Intent
   2. Moving Intents into a hierarchy
   3. Merging multiple Intents into one
   4. Outsourcing Intents to a separate Flow
   5. Adjusting the NLU settings
3. Reducing false positives
   1. Machine learning / Reject Intent
   2. Adjusting the NLU settings

You can find all of the showcased Flows and Lexicons in the attached package.

We will go into the details of each step in the next sections. First, let's start by learning various ways of achieving Intent recognition and understand how these methods affect model performance in different ways.

Cognigy NLU Training: Improving Intent recognition