Using model-based topic detection to route emails

1. Download and extract the TrainingData training set.
2. Log in as Data Scientist with user name CASysAdmin and password rules.
3. In the header, click Application: Customer Service > Channels and interfaces.
   
4. Scroll down and click UBankSupport to open the email channel configuration.
5. Click the Behavior tab.
6. In the Text Analyzer section, click Open text analyzer rule.
7. In the top right, click Actions > Run.
8. In the Sample text field, write a message concerning an address change and a transaction dispute.

9. In the top right, click Run.
10. In the Topic detection section, notice that both the Account Address Change topic and the Dispute Transaction topic are detected with a confidence score of 1.
    
    Note: The confidence score of a topic is either 0 or 1 in rule-based methodology, depending on the presence or absence of keywords in a message.

Close the Run Text Analyzer: UBankSupport window.

1. In the upper left corner, switch from DEV Studio to Prediction Studio.
2. In the navigation pane on the left, click Models.
3. Click New > Text categorization.
4. In the New text categorization model dialog box, in the Name field, enter UBankSupport_Trained.
5. In the Save to IVA channel list, select UBankSupport.
6. In the Language list, select English.
   
7. Confirm that the Methodology is set to Use machine learning.
8. Click Continue.
9. In the Source selection section, click Choose File and select the TrainingData.csv file.
10. Click Next.
11. In the Sample construction section, select User defined sampling to construct training and test sets.
12. Click Next.
13. In the Taxonomy review section, expand the Action list and examine the number of records and words counts.
    
14. Click Next.
15. In the Model creation section, ensure that the Maximum Entropy model is selected.
16. Click Next.
    
    Note: The F-score is used as the metric for the expected performance of categorical models.

17. Click Next.
18. In the model selection section, ensure that the Maximum Entropy model is selected.
19. Click Create.
20. In the upper left, click Test. Enter a message about an address change and click Test:

21. Scroll down and notice the topic with the highest Confidence score.
    
22. Enter a message about disputing a transaction and click Test:

23. Scroll down and notice the topic with the highest Confidence score.
    
24. Close the Test panel.
25. Click Save.

1. In the upper left corner, switch from Prediction Studio to DEV Studio.
2. On the Text Analyzer: UBankSupport tab, in the upper right, click Check out.
3. In the Topic model field, enter UbankSupport_Trained.
4. In the Topic preference section, select Use model-based topics if available.
   
   Note: This option ensures that the topic detection looks at available models first and then applies the rule-based topic detection only if there are no models.

5. In the Text extraction section, select Enable text extraction.
6. Inspect the entities that are detected by the installed extraction models and the Ruta script.
7. On the Advanced tab, in the Topic settings section, select Sentence level granularity.
   
8. In the top right, click Check in and enter appropriate check-in comments.