Training Exclusions

Use training exclusions to build and manage a list of words that will be ignored in any training set that is used for creating or improving a knowledge base.

Text strings in spend data often include common words that can appear in a wide variety of unrelated spend records, making them poor predictors for assigning or updating category codes. Building up a set of training word exclusions ensures that these terms don't interfere with category predictions and improves knowledge base performance. Spend Classification machine learning ignores words in the training exclusion list while classifying the spend data.

Here’s how you get started:

1. Go to the Training Exclusions page in the Configuration section of Spend Classification.
2. Start by reviewing the sample exclusions available within the solution. You can download these exclusions on to a spreadsheet using the Download Sample Exclusions action.
3. Modify the list of words as required.
4. Click Upload Exclusions to import your changes back into the application.

   Alternatively, you can make changes directly on the Training Exclusions page. Add words to be excluded one at a time in the Excluded Word field, click the ‘X’ against a word to remove it from the exclusion list.

5. Incorporate the training exclusions that you have uploaded into your classification process.
   • When you create a new knowledge base, the list of exclusions if available, is incorporated during knowledge base creation.
   • For an existing knowledge base, you must improve the knowledge base using the Improve Knowledge Base action to incorporate the latest training exclusions.
   Note: If no additional training is required, use only the training sets that were initially used to create the knowledge base.
6. When the knowledge base is ready, use it to classify the spend transactions to improve your results.

The training exclusions list evolves over time as it undergoes multiple iterations. To download and review the exclusion list that was last applied to any given knowledge base, go to the Knowledge Base page, and click the Download Exclusions action for that knowledge base. In the spreadsheet, you’ll find these lists:

• Current exclusions: Words currently excluded from the training set.
• Canceled exclusions: Words that were removed from the list of training exclusions since the knowledge base was created or last improved.
• Added exclusions: Words that were added to the list of training exclusions since the knowledge base was created or last improved.

Use this list to determine if a knowledge base should be rebuilt or improved to incorporate more training exclusions, or if training exclusions should be modified to add the words previously removed.