Question
Areteans
GB
Last activity: 26 Jan 2022 8:13 EST
Pega email BOT Not scanning entire email
Hi all,
We are facing an issue after configuring a pega email channel. (Pega platform 8.1.4 , Pega customer Service - 8.1) We have configured the necessary text analysis rules for topic detection, entity extraction etc. and all of those working fine.
But below to issues are observed :
a. Not entire email body is being scanned for information, only the first email text in a trail email is being searched for information. The requirement here is to search for entire email body.
b. also we have configured in the email channel to enable training data but we don't see any training data on email channel -> training data tab after triage cases are resolved.
These issues are being observed in dev environment, we are yet to dpeloy this to higher envs.
Any help on these issues will be highly appreciated.
Accenture
IT
Hi, i am following this discussion to as i am interested in understanding how this works.
BTW, can you share me references on how to configure an email bot? thank you in advance.
Pegasystems Inc.
IN
Entire email body not analyzed -
Only latest reply is taken into consideration for the purpose of analysis and routing. This is by design to keep the UI clean and analysis current and consistent. Entire email body contains lot of information which skews 'Topic' detection towards lower confidence. Reply parsing happens in 2 stages -
- pyParseLatestReplyMail - Extracts latest reply using regular expression. This will remove everything after first occurrence of , for instance, 'From:' for the purpose of NLP analysis and display
- pyEmailSplitter - Machine learning model to remove greeting, signature and disclaimers from email.
You may turn both of these off but there are design costs related to UI and Analysis.
Record training data
Training data is sent back to NLP model when 'service case' is created manually from triaging screen. In case of auto-resolve, model already understands the topic and there is no need to retrain.
Areteans
GB
Thanks, Vikas for the detailed explanation. I would like to understand the record training data feature in a bit more detail. For example, as you mentioned the triaged email content is pushed to NLP model after manually creating a service case, to which model this training data is sent ?
As per the structure of Text analysis, the text-analyzer rule contains the model rules for topic identification, intent detection and entity extraction, so is the triaged content sent individually to all these models?
And does the training recording only happens when manually creating a service case ? what if we are creating a service case though a service call in pega (for a triage case), in that case, is there any api to mimic the service case creation though a triage case?
Thanks for your cooperation.
Updated: 2 Sep 2020 5:17 EDT
Pegasystems Inc.
IN
- As per the structure of Text analysis, the text-analyzer rule contains the model rules for topic identification, intent detection and entity extraction, so is the triaged content sent individually to all these models?
Yes, the content is sent to all models individually. Feedback is sent to entites only when you have entity property mapping and agent changes the automatic suggestions. What you see on Training data tab is feedback to topic model. Feedback to entity models is not shown on Training data tab.
- And does the training recording only happens when manually creating a service case ? what if we are creating a service case though a service call in pega (for a triage case), in that case, is there any api to mimic the service case creation though a triage case?
I don't follow this entirely. For automatic case creation there is no feedback. You can use pxCaptureTAFeedback API to send feedback programatically. https://community.pega.com/knowledgebase/articles/decision-management-overview/providing-feedback-text-analytics-models