Batches

Create a Classification Batch

Here's how you classify a data set and generate a batch:

Navigate to the Data Sets page.
Select the data set and from the options and select Classify.
In the Classify Data Set window, select the knowledge base you want to use to classify the data set.
Enter a name for the batch.

It's important to give a meaningful name to the batch. Something to the represent the data and filter conditions used to create it.
Select a business unit.
Specify transactions from and to dates such that the number of transactions available for classification is less than 100,000. For easy analysis of classification results, a batch can have a maximum of 100,000 transactions.
Click Get Transactions to retrieve the number of transactions available for classification. Create multiple batches in case there’s a large number of unclassified transactions.
Click Classify to initiate the classification process.

You can also use the Start Spend Classification scheduled process to schedule the classification of spend data. If there are over 100,000 transactions in a batch, the scheduled process automatically distributes the transactions into the required number of batches and processes all the batches at the same time. Find the scheduled process in the Scheduled Processes work area and define the same parameters that are required when you create a classification batch manually.

Review Classification Results

Go to the Batch Summary page for the list of batches, status of batches, and classification trends such as high, medium, or low.

Click on a batch name to go to the batch details page. Here you can review the status, knowledge base used, data set from which the batch was created, taxonomy, classification date, first and the last transaction dates in the batch, and spend amount. Use filters and filter suggestions to prioritize your review of the classification results.

Click on the Analytics option on the Batch Details page to go to the Analytics page for that batch. Batch analytics include detailed visualizations and search options that help you narrow down the transactions that may need your attention. You can review the results of the classification process and compare the categorization of transactions before and after the classification process. Major deviations may require review and intervention.

Click on a transaction ID on the batch details page to go to the transaction details page where you can view individual transaction details such as predicted category name and description, manual category name and description, transaction type, spend amount, and so on. You can view the predicted categories and the list of keywords in the keyword analysis.

The analytics for the batch are organized into four sections: Classification Performance, Classification Spread, Positive Influencers, and Negative Influencers. Click on each section to view the detailed charts and related metrics.

Classification Performance: Review the category-wise classification results for the batch based on the system confidence level for the predictions that have been made. The default view is for low confidence transactions.
The Classification Performance chart shows you how many of the transactions predicted for a category were originally assigned to that category and how many were assigned to a different category. This split is shown by amount and by the number of transactions and can highlight the areas where there may be issues with the predictions being generated by Spend Classification. For example, if the volume of transactions predicted in a category is low, but they represent a significant proportion of the total spend for that category, this would warrant a detailed investigation.
The Classification Performance chart makes this easy because you can drill down to the actual transactions for any of the regions in the chart. For example, if you click the Prediction Varies from Original bar within the Total Spend Amount for a specific category in this chart, you’ll be redirected to the batch details page that is filtered to list only those transactions included in that spend amount.
Classification Spread: The metric in this section allows you to understand how many of your categories include predicted spend. This helps judge the relevance of your taxonomy or notice significant issues with your Knowledge Base. For example, you may have Training Set Data for a significant number of categories that are receiving no predicted spend.
You can compare the transactions categorization before and after classification process to see the breakup of this spread in terms of transaction amount or the number of transactions per category. Major deviations may warrant intervention and review. As with the transactions by category chart, you can use these charts to drill down to the batch details and make changes if required.
Positive Influencers: This section within the Analytics dashboard shows you the words in the training data that have greater significance and a positive impact on classification or category prediction. Review the positive keywords to assess whether you can make improvements to the training data by refining existing keywords or augmenting them with additional similar terms. These changes help you ensure that the training data is adequate, and the results are impacted by positive influencers.
Negative Influencers: This section shows the words in the training data that either have no impact or have a negative impact on classification or category prediction. Review negative keywords and remove them from training data to improve the classification results. You can add these negative keywords to training exclusions to improve the classification results for a training set. See Training Exclusions for more information on how to exclude words in training sets.

View Transaction Details

In the batch details page, click on the transaction ID to view the transaction details. In the transaction details page you can view the transaction description such as predicted category code, manual category code, transaction type, amount, and so on. You can view the predicted categories and the list of keywords in the keyword analysis.

Review and Improve Classification Results

Classification is done in batches of data created while classifying a data set. The Batch Summary page provides you with trend details to help understand whether successive batch processing is yielding better or worse results. Key processing metrics are also provided to highlight the average time to classify, identify the most recent and most frequently used knowledge base and taxonomy, and to indicate when the most frequently used knowledge base was last improved.

For any given batch, you can drill into the batch details to review all of the individual transactions with the option to enter manual changes to the predicted category for any spend record. You can filter the batch details using the high, medium, and low classification statuses to help prioritize activity within your review.

The classification status is determined by comparing the confidence at each level of the predicted category against the classification threshold percentage entered in classification controls.

You can make corrections to the category predictions on the Batch Details page or using a downloaded batch file. On the Batch Details page, select a category name for a spend record from its drop-down list in the Manual Category Name column and click Save. To perform bulk manual classifications using batch download, follow these steps:

In the Batch Details page, click the Download action and save the batch file.
Review the changes needed in the spreadsheet and add the category names to the Manual Category Name column.
In the Rules Applied column, you can see the names of the classification rules that were used to improve category predictions. These rules could have influenced the category prediction or modified the values of some of the attributes. You can use this information to determine how the rules influenced the predictions and opt to revise the rules or add new ones.
Delete all the other transactions where you have not made any changes.
Tip: Filter out the blank entries in the Manual Category Name column and delete those rows.
Use the Upload action to upload the batch file.

When you assess classification predictions for individual spend records and make corrections, you can view overall improvements to your batch results in real time.

Manage Batches

You can also perform these actions on a batch:

Approval - After you’ve completed the improvements to the category classifications, you can approve a batch. The batch becomes available on the Performance page.
Reset - To reclassify a batch using the same taxonomy, reset the batch
Create a data set using manual corrections - Incorporate the feedback using manual corrections to improve the knowledge base using the training data set created by taking this action. This data set contains all transactions classified manually along with their updated category predictions.
Delete - Manage the number of batches created by deleting them as required.

Troubleshooting

Use this section to troubleshoot the following issues: Import data sets:

Ensure to follow the data preparation instructions as mentioned in FBDI template. Date format in the upload file should be YYYY/MM/DD.
Change the data set name in upload file, while uploading training set.
Remove any comma "," from the data descriptions in the training data set file, as it interferes with the upload process.

To update taxonomy:

Download the latest categories.
Delete the categories in the relevant taxonomy.
Upload the updated file

Manual corrections data set can be created only from approve batches.

Performance page shows data for approved batches only.