Testing a Model on a Training Object

This topic describes part of the functionality of Genesys Content Analyzer.

It includes:

• Setting Up Training Objects for Testing

• Procedure: Creating Two New Training Objects

• Procedure: Extracting Random Text Objects

• Procedure: Testing a Model on a Training Object

You can test a model on a training object. This process applies the model to the texts in the training object and compares the resulting classification with the classification in the training object itself. The training object must use the same category tree as the model you are testing on it.

Important

It is not advisable to test a model on the training object that generated it: the results will be unrealistically favorable.

Setting Up Training Objects for Testing

You may want to create a new training object just for the purpose of testing. Use the same category tree but different text objects. There are two ways to do this:

• Starting with a collection of categorized e-mails, create two new training objects. See the figure "Creating two new training objects" on this page.
• Starting with an existing training object, create a second training object using randomly-chosen text objects from the first. In more detail:
  1. Move five percent of the text objects, randomly selected, from one training object to another.
  2. Train a model on the first training object.
  3. Test the model on the second training object.
  This is similar to using cross-validation with two partitions. Procedure: Extracting random text objects describes this process in detail.