This video shows how to use
natural language understanding workbench to refine the model for ITSM Issue
Auto Resolution using the ITSM IAR Model can help users resolve their own incidents faster. The prebuilt ITSM Issue Auto Resolution
Model maps the short descriptions in ITSM incidents
to specific intents, like resetting a password
or printer issue. IAR uses
those intents to trigger virtual agent topics to help users solve their issues. NLU Workbench helps you refine the model
to perform at its best. From the IAR admin console. We can go straight to NLU workbench
by clicking tune in NLU work bench. That takes us to the home page
for the ITSM IAR model. Refining our model is an iterative process. First we get feedback on the model's
predictions. Then we analyze the model's performance
to see if it's ready to publish or if we need to give more feedback. We repeat those two steps
till the model is performing well. A match rate of 80% is a good target. When it's ready,
we publish it to start using it with IAR. In the first step, we give feedback on how
well the model predicted the correct intents from the short
descriptions of recent incidents. The data for this feedback
comes from incidents from the last seven days as defined in the task
configuration tab in IAR admin console. Let's look at the short descriptions
that need reviewing for printer issues for the description issue
reported on printing, the model predicted the printer issues intent. That seems like a good prediction. So we'll choose match. Here for the description. Can't access my account. The model predicted the software access
request intent. That doesn't seem right, so we'll choose
mismatch and select reset password instead. When we're done, we save our feedback. If we don't save
our feedback will be discarded. Now that we provided new feedback, we'll analyze our model
to check its performance. And here are the results. These columns show the model's accuracy
before and after our feedback. Notice that some of the intents
are not mapped to virtual agent topics. You only need to provide feedback
for these intents if you're preparing to use them later. We can also tune our model
for greater precision in matching descriptions
to intents or automation to direct more issues to virtual
agent or a balance between the two. The last step is to publish our updated model
to make it available to IAR. Now IAR is using our updated model. The intents that are supplied with IAR
are already mapped to virtual agent topics,
so we don't need to do that ourselves. We can change and activate those mappings
in the intent to topic map in the IAR admin console. When your model is in use, you'll repeat
this feedback process regularly. To keep your model performing well, you should refine your model once a week
for the first month, your model is in use and every two or three months
after that, with NLU workbench, you can refine your model to help users
resolve their own incidents. For more information, see our product documentation
or knowledge base or ask a question. In the ServiceNow community.