on - edited on by
Victor Chen
Step 3: Enabling NLU In Your Conversations
At work, we deal with language constantly. Whether it's granting access to a server, requesting time off, or delivering a new monitor to someone. We're able to read each other's text, interpret it, and respond appropriately.
Natural Language Understanding can be enabled to improve the overall experience for end-users. Natural Language Understanding, or NLU for short, allows end-users to engage with the Virtual Agent using natural sentences. NLU not only improves the Virtual Agent's user experience, but it also increases the Virtual Agent's accuracy when presenting topics.
How does NLU Work?
To understand how NLU works, you don't need a degree in Machine Learning or Data Science. All you need to do is familiarize yourself with the following three terms:
| Intent |
In NLU, an Intent represents what end-users hope to accomplish. In our world, it's an action someone could take in the system. That could be anything from opening an IT ticket to checking on the status of one.
|
| Utterance |
Utterances are the different ways of expressing an Intent. Something Critical is broken. Can IT help ASAP? Or New crucial incident...these are all utterances that are intending the same thing: Open an IT ticket.
|
| Entity | Entities are the objects and values within the Utterance that provide additional context for the Intent. Critical, ASAP, crucial...these entities are all expressing a sense of urgency. If an IT ticket does get open on behalf of the user, it should be opened with high urgency. |
Now that you understand key NLU terms, let's talk a bit about NLU Models. An NLU Model is made up of a collection of Utterance examples and their associated Intent and Entities. The NLU Workbench allows you to easily create, train, and publish your very own NLU models.
Activating NLU
As of the Utah release, NLU is enabled by default. You want to make sure you update the "NLU Workbench - Advanced Features" plugin too. You can then view this in the Conversational Interfaces Home Page, under Settings > Virtual Agent.
After activating NLU, navigate to NLU Workbench > Models. You should see a list view similar to the screenshot below. (Click on the image to see a larger version).
| Exploring the NLU Workbench |
|
The NLU Workbench is simple and intuitive. Here you can create and manage your NLU models, intents, and utterances. First, let's explore an NLU model that is already provided to you. Step 1: Navigate to NLU Workbench > Models, and find NLU model: ITSM NLU Model for Virtual Agent. Click on a language, e.g. "English (Primary)" to open the model.
Step 2: Click on View phase for "Manage your model content". Browse through the list of Intents and select OpenITTicket
Step 3: Test a few utterances by clicking on "Try model". You'll be prompted with the option to enter an utterance. Test the following utterances and note their respective prediction scores.
The prediction score tells you how close the tested utterance is to the list of utterances within an Intent. The higher the predicted score (max 100%), the closer the tested utterance matches against the utterances within that Intent.
Step 4: If you added or changed utterances to your intent or model, click the "Train model" button on the top right so that your model takes the new utterances into account. Try your model again to ensure that it performs with your modified utterances in mind. Then return to the model's main page and click on View phase under "Test and publish your model". Here you can either run a test set against your model, or you can publish you model.
Advanced Features The NLU Workbench also has testing and performance measurement capabilities such as Conflict Review and Batch Testing. Use these tools to tune and optimize your NLU models. If you do not see them under "NLU Workbench", you can manually install the plugin: NLU Workbench - Advanced Features (sn_nlu_workbench). |
| Binding NLU to a Virtual Agent topic |
|
Now let's bind an NLU intent to a Virtual Agent topic, so that when a user types an utterance recognized by NLU, it will run that topic. Start off by opening Virtual Agent Designer and select a topic, e.g. Check IT Ticket Status. In the Properties tab, choose the NLU model and intent associated with the Virtual Agent topic. You can also create a new NLU intent straight from the "Associated intent" drop-down. Once you selected your associated intent, the "NLU Intent" tab at the top will show you the intent utterances. After you save and publish the topic, try testing out a few utterances in Virtual Agent such as "check ticket status" or "my incident status." |
Learn more about NLU with our FAQ and Virtual Agent Academy sessions
- NLU FAQ, Best Practices, and Troubleshooting.
- Maintain your Virtual Agent health with advanced NLU tools
- Improve your NLU performance with Model Optimization
- Expand your NLU vocabulary for effective Virtual Agent conversations
| What's Next? |
| Now that you have a basic Virtual Agent up and running with NLU, there are some more things you can do: Extend an ITSM OOTB NLU Model |
