Set Up Ticket Learning
What is Ticket Learning?
Ticket Learning is a method to understand the types of issues end users face by analyzing historical tickets. It identifies key intents and groups them based on similarity.
Here's a breakdown of the key components:
Understood Intents by Category:
A treemap view displays the top categories of intents that Aisera has learned over time.
The green treemap shows understood tickets organized by categories, meaning AI is confident in analyzing and classifying these tickets.
Not Understood Intents by Category:
Another treemap view shows categories of intents that were not understood from the tickets.
Even if the intent isn't found, Aisera will suggest related phrases, variables, entities, and entity classes for further review.
Top Understood & Not Understood Intents:
This view helps users identify high-priority intents that are either understood or not understood, making it easier to review and address them.
Top Entities in Understood & Not Understood Tickets:
Displays the top entities understood from the tickets by validating against the local ontology.
Helps identify missing fulfillments and manually attach fulfillments to intents.
Useful KPIs
Analyzed Tickets: Total tickets analyzed during a Ticket Learning job.
Understood Tickets: Total count of tickets matched with the intents.
Not Understood Tickets: Total count of tickets not matched with the intents. Aisera will recommend related intents based on the analysis.
Understood Tickets MTTR (Days): Mean time to resolve understood tickets by agents without Aisera.
Tickets Fulfilled by Knowledge: Total understood tickets fulfilled by knowledge flow.
Tickets Fulfilled by Workflow: Total understood tickets fulfilled by action flow.
Discovered Entity Synonyms: Count of synonyms extracted from the tickets, shown when the user chooses to generate synonyms by selecting 'Discover ontology synonyms (Entities)'.
Step-by-step Guide to Run Ticket Learning
The ticket learning job can be found Gen AI Learning -> Ticket Learning tab. For more controls, add ‘?system’ in the URL. Once you have finalized the relevant set of tickets, we will initiate the ticket learning job by configuring the parameters.
Step 1 - Configuring “Conditions” Parameters
Go to Actions → Configuration
Add the relevant conditions of the field name of the interest. These fields are self explanatory and users can easily relate to the fields in their respective ticketing system.
Notes:
It is recommended that you set Max Ticket Count to 60,000.
60,000 ticket count is a healthy sample size. If we don’t limit ticket count we could possibly ingest a large number of tickets causing the job to run for multiple days and delay results.
The conditions that we put here are later converted into a SQL query to fetch the filtered data from the Tickets database.
Multiple selections for a single field will be of “OR” equivalent while multiple field name conditions will be of “AND” type. E.g., Conditions on Data Source and Creation Date are included by “AND” operator while Gap Prod - Incidents and Gap Prod - Ticket Learning data sources will be incorporated as “OR” if both selected.
Step 2 - Configuring “Preprocessor” Parameters
Sometimes despite scanning through tickets and assessing their quality, we might find ourselves in a situation where we have some unnecessary tickets which we should exclude.
We can choose what fields we want TL to be performed on, i.e., Title or Description or Both from Fields to Process
In the Inclusion/Exclusion Rules section, we can add specific conditions on Title/Description to filter tickets based on certain keywords.
Substitutions help you to substitute certain keywords in the text before passing it through the TL pipeline.
There are several other preprocessing options also provided which can be used pending any special requirement.
Step 3 - Configuring “Scoring” Parameters
Skip ICM : This option will provide the coverage of intents from Tickets. Check this option, if the intention is only to generate the ontology ,to eliminate the unnecessary data processing steps. However there is no harm in running this option either.
Join Title and Description : If checked, ticket title and description will be considered during the scoring process.
Use Global Ontology (Entities): Check if you want to use the Ontology from the content pack.
Discover Ontology Synonyms(Entities): Extract the synonyms from KB for the entities generated from Entity Discovery and validate against tickets, if usage found in the tickets those are extracted and attached to the entities
Use Global Taxonomy (Intents): Check if you want to use the Taxonomy (Intents) from the content pack
Text Field char limit: Maximum number of characters to consider in title and description
Text Field word limit: Maximum number of words consider in the title and description
Step 4 - Kick off Ticket Learning job
Click on ‘Start Learning Job’ once the conditions and configurations are set, the user will get an acknowledgement saying the job has triggered and is in progress. We can monitor the status by clicking on ‘view job’ in the acknowledgement or by navigating to
Settings-> System jobs -> Generic jobs -> Search for Ticket Learning.
Learn Interactions
The Aisera platform supports a content type called, Interaction. It is a child content type for Ticket, like Problems, and Alerts. However, it will not be automatically ingested when the data source function is set to LearnTickets, but only when it is set to LearnInteractions.
You can now filter the Interactions under the Tickets Tab.
Last updated