Improving your dialog skill

Picking a data source

The term data source refers to the logs compiled from conversations between customers and the assistant or custom application by which a dialog skill was deployed.

When you open the Analytics page, metrics are shown that were generated by user interactions with the current dialog skill. No metrics are shown if the current skill isn't deployed for customers to use.

To populate the metrics with message data from a dialog skill or skill version that was added to a different assistant or custom application, one that interacted with customers, complete these steps:

1. Click the Data source field to see a list of assistants with log data that you might want to use.

   The list includes assistants that are deployed and to which you have access. Or you can choose to show a list of other deployments. For more information about other types of deployments, see Show deployment IDs explained.

2. Choose a data source.

Statistical information for the selected data source is displayed.

Notice that the list does not include skill versions. To get data that is associated with a specific skill version, you must know the time frame during which a specific skill version was used by a deployed assistant. You can select the assistant as the data source, and then filter the metrics by the appropriate dates to see log data that was generated by the assistant only while it was using the skill version.

Show deployment IDs explained

Applications that use the older v1 runtime API must specify a deployment ID in each message that is sent by using the /message API. This ID identifies the deployed app that the call was made from. The Analytics page can use this deployment ID to retrieve and display logs that are associated with a specific live application.

For assistants or custom apps that use the v2 version of the API, your assistant automatically includes an assistant ID with each /message call, so you can choose a data source by assistant name instead of using a deployment ID.

To add the deployment ID, v1 API users include the deployment property inside the metadata of the context, as in this example:

"context" : {
  "metadata" : {
       "deployment": "HelpDesk-Production"
  }
}

For an Enterprise with Data Isolation plan, you can ask IBM to configure your instances so you can access log data from deployed applications across different instances. Each instance must use the v1 API and specify a deployment ID with each /message call. (If your instances use the v2 API, you cannot get log data from across different instances.)

If log sharing is enabled, only someone with Manager service access to the current instance can view logs. The person can view logs from all of the instances that are being shared. Logs from across all of the participating instances are displayed, regardless of the current user's service level access to the other instances. Similarly, when someone with Manager service access to an instance sends a GET request to the v1 /logs API endpoint, logs from all instances that participate in log sharing are returned, regardless of the user's service level access to each instance.

Correcting an intent

1. To correct an intent, click Edit icon.

2. From the list provided, select the correct intent for this input.

   • Begin typing in the entry field and the list of intents is filtered.
   • You can also choose Mark as irrelevant from this menu. (For more information, see Teaching your assistant about topics to ignore.) Or, you can choose Do not train on intent, which does not save this message as an example for training.

3. Select Save.

   The Watson Assistant service supports adding user input as an example to an intent as-is. If you are using @entity references as examples in your intent training data, and a user message that you want to save contains an entity value or synonym from your training data, then you must edit the message later. After you save it, edit the message from the Intents page to replace the entity that it references. For more information, see Directly referencing an @Entity as an intent example.

Adding an entity value or synonym

1. To add an entity value or synonym, click the Edit icon for the chosen entity.

2. Select Add entity.

3. Select a word or phrase in the underlined user input.

4. Choose an entity to which the highlighted phrase is added as a value.

   • Begin typing in the entry field and the list of entities and values is filtered.
   • To add the highlighted phrase as a synonym for an existing value, choose the @entity:value from the list.

5. Select Save.

Teaching your assistant about topics to ignore

It is important to help your assistant stay focused on the types of customer questions and business transactions that you designed it to handle. You can use information that is collected from real customer conversations to highlight subjects that you do not want your assistant to even attempt to address.

For more information, see Defining what's irrelevant.