There are four ways to test a model:
If you select cross-validation when you schedule training, Training Server produces accuracy ratings along with the model.
You can test the model on a training object, also producing accuracy ratings.
You can test the model to see how it classifies text that you compose.
You can apply the model to the uncategorized messages of a training object. This method does not produce accuracy ratings.
On the Training or Training Schedule tab, select Schedule Model Testing from the Tools menu. The Model Testing Schedule dialog box appears.
Select the testing object (that is, select the training object to test the model on).
Select the model to test.
Enter a start time.
On the Models tab, go to the left-hand pane and select the root Models node.
On the right-hand pane, select the model that you want to test, then right-click and select Test from the shortcut menu. A dialog box appears, titled Model: <modelname>. It contains a Subject box, a Text box, and a box for displaying the test results.
Enter text in either or both of the Subject and Text boxes, then click Classification Test.
Results display in the lowest box.
The results are in the form of a list of categories, each category preceded by the rating of the confidence with which the system assigns the test text to that category.
On the Models tab, go to the left-hand pane and select the root Models node.
On the right-hand pane, right-click the model that you want to test, then select Test Uncategorized messages from TO.
In the resulting window, select a training object from the drop-down list. Be sure to select a training object that contains a good number of uncategorized messages.
Click Test Uncategorized texts.
The results show, for each confidence level:
% of All Texts: The percentage of texts that were classified above this level of confidence.
% in Confidence Interval: The percentage of texts that were classified with a level of confidence between this level and the next higher level on the scale.
These results tell you how well the model does, according to its own internal metric, at assigning new texts to some category or other. They do not evaluate the accuracy of these category assignments.
See also "Testing Models" in the "Genesys Knowledge Management: Content Analyzer" chapter of the Multimedia 7.6 User's Guide.