Analyzing Model Recall (Studio)

About Analyzing Model Recall

One of the ways to evaluate the effectiveness of a categorization model is to measure its recall. Recall is the percentage of documents that were categorized into a particular model.

To analyze model recall, you should compare the percentage of documents categorized into the model with the number of documents not categorized. To do this, you can create 2 metrics: one for categorized data, the other for uncategorized data.

Qtip: Interested in exploring your data further? See our pages on Filtering by an Entire Category Model and Exploring Uncategorized Data (global other).

Creating a Percent Categorized Metric

Create a metric.

Select Filtered metric.

---

×

close

Under Based On, select % Total.

Under Conditions, select the category model you’re interested in.

Select categorized.

Save your metric.

Qtip: If a model has root-level rules, the “categorized” condition returns all documents that match the root-level rule. However, no other categories within that model are taken into account. To learn more about these results, see How Root-Level Rules Affect the Categorized Filter.

Creating a Percent Uncategorized Metric

Create a metric.

Select Custom metric.

---

×

close

In the equation box, type 1.

Click the minus sign ( – )

Use the Metric dropdown to select the Percent Categorized metric you created earlier.

Go to the Format tab.

---

×

close

Add % as the Suffix.

Set the conversion to X100.

Save your metric.

Displaying the Results

Once you’ve built your metrics, you can display them in widgets like any other metric (for example, displaying 1 value in a metric widget, or displaying both values in a pie widget). These metrics can be used interchangeably depending on your analysis.

---

×

close