The need for "Big data" to get accurate Categorization and Score

Hi,

This is great stuff and really cool. I tested the automatically categorization feature on my DEMO environment,

The complete numbers of request is about 160. The "score" and result was poor ( orange) See image:

 

 

Do we have any lower limit and recommendation regarding the latter? I would expect that the larger amout of data will be more accurate.

 2: I see that it takes some time from the ticket is imported to Service to I see the suggested AI features on the ticket. ( Both translation, sentiment and categorization). I woul think that this is due to the fact tha all AI services are handled after the ticket is imported and based on amount etc this is the delay?

Please advise.

 

RE: The need for "Big data" to get accurate Categorization and Score

Hi Jan,

Your accuracy score is not only dependent on amount of tickets, but also distribution across categories. Machine learning is about looking for patterns, to reach a higher accuracy you could need a larger dataset but it could also help to restructure categories.

Currently AI functions are executed after ticket/messages are stored by a background process, this process runs on each tenant every minute so normally delay should not be very noticable. 

By: Michel Krohn-Dale 12 Mar 2021

RE: The need for "Big data" to get accurate Categorization and Score

Hi

I would think that more data is better.

I did a selection for 2 quarters on our real data, many cases have a several mails going back and forth. I'm not sure if this is the cause, but it seems like that drains the system pretty hard on ressources. 

I would think that the auto categorization would only look at the first message, as soon as the message is read, I would think that it is categorized by the user. 

 

Best regards

Ole

 

By: Ole Ejlersen 23 Mar 2021

RE: The need for "Big data" to get accurate Categorization and Score

Thank you for  your feedback!

As informed in Categorization questions / curiosities we're planning to release a fix for the load issue tonight.

On your comment on suggested categorization on "first message" - this was reported a few days ago - and we will change "suggested categorization" from 'message(s)' to 'ticket' - so it will only be one suggested categorization on each ticket from 9.2 R7 (scheduled 30.03.2021)

By: Frode Berntsen 24 Mar 2021