Support Article
Text Analysis model result varies from Pega 7.2.1 to Pega 7.4
SA-62226
Summary
Post upgrade from Pega 7.2.1 to Pega 7.4, Text Analysis model result is different. The result varies from Pega 7.2 and Pega 7.4.
Error Messages
Not Applicable
Steps to Reproduce
Execute Text Analysis model in Pega 7.2.1 and Pega 7.4. Compare the results.
Root Cause
An issue in the custom application code or rules.
In Pega 7.2.1, a radial kernel was used for the Support Vector Machine (SVM). In Pega 7.4, the system shifted to a linear kernel for efficient memory usage and to reduce the overall process time for SVM. A shrinking heuristics was added with the kernel. This lead to the change in accuracy from Pega 7.2.1 to Pega 7.4. The accuracy is expected to be the same while using the Maximum Entropy algorithm (MaXENT).
Resolution
Perform the following local-change if accuracy similar to Pega 7.2.1 is desired after upgrade to Pega 7.4:
Use the Maximum Entropy (MaXENT) algorithm instead of Support Vector Machine (SVM) for the Text Analysis.
Published February 13, 2019 - Updated December 2, 2021
Have a question? Get answers now.
Visit the Collaboration Center to ask questions, engage in discussions, share ideas, and help others.