Review the Intent History

The Intent History page of the developer console displays aggregated, anonymized frequent utterances and the resolved intents. You can use this to learn how users interact with your skill to identify improvements you may want to make to your interaction model. You can also map utterances to your intents and slots from this page.

Review intent history in the developer console
Review intent history in the developer console

Available Data

The data displayed on the Intent History page is aggregated and anonymized. This means you will not see all utterances sent to your skill:

  • A locale must have at least 10 unique users in a day for any data from that locale and day to be available. For example, suppose the English (US) version of a skill had 20 unique users on June first, and the English (UK) version had only 9 on that same day. Data from June first would be available for English (US), but not English (UK).

  • The utterances shown are a subset of utterances sent to your skill, filtered to show only frequent utterances. This means that there might not be any utterances shown, even if the daily user threshold has been met.

The data is available for both development and live skills, and for skills in a beta test. Noted that the 10-unique users per day requirement applies to all skills, so you may not see much data for skills still in development.

The data is shown on two tabs:

  • Unresolved Utterances displays utterances that you have not yet reviewed.
  • Resolved Utterances displays utterances that you have reviewed and resolved. To resolve an utterance, you can map it to an intent or slot, or just mark it resolved to remove it from the unresolved list. See Review unresolved utterances and update your interaction model.

The page displays the following columns.

Column Description

Utterance

The utterance the user spoke to your skill.

Confidence

The confidence that the utterance resolved to the correct intent. This can be HIGH. MEDIUM, or LOW. Note that LOW confidence requests are not sent to your skill as an intent request, but trigger a reprompt instead.

Intent

The intent in your skill that the utterance matched. For a LOW confidence request, this represents the best guess match, but the intent was not sent to your skill.

Slots

A list of slots that were filled by the utterance.

Dialog Act

If the utterance is part of a dialog, displays the dialog action triggered by the utterance. This can be Elicit Slot, Confirm Slot, or Confirm Intent.

Interaction Type

Indicates if the utterance was run as a ONE_SHOT interaction or MODAL interaction.

  • ONE_SHOT: The user invokes the skill and states their intent in a single phrase.
  • MODAL: The user first invokes the skill and then states their intent.

Action

For unresolved utterances, displays the Map button to map the utterance to an intent or slot.

For resolved utterances, shows the action taken:

  • For utterances that have been mapped, this shows the intent/slot it was mapped to, along with the final utterance if you edited the text.
  • For utterances that were resolved without mapping, this just shows "No action."

See Review unresolved utterances and update your interaction model.

Search and Filter the Results

You can search and filter the Intent History results. For a free-for search, enter the search terms in the Search Utterances box. To filter based on set criteria, select the criteria:

  • Confidence level
  • Stage (live or development)
  • Resolved status (resolved or unresolved)
  • A specific Built-in intent
  • A specific Custom Intent

Click Apply to see the results of the search or filter.

You can also use the Export option in the upper right to export the results.

Review unresolved utterances and update your interaction model

The Intent History page displays the frequent utterances in two tabs, Unresolved Utterances and Resolved Utterances. This lets you review the utterances, update your interaction model to account for phrases that were not routed correctly, and mark utterances as resolved.

For example, suppose you see a particular utterance that was sent to AMAZON.FallbackIntent, but it is actually a reasonable phrase that should trigger one of your custom intents. You can map that utterance directly to that intent and update your interaction model right from the Intent History page.

For utterances that were already mapped correctly, mark them as resolved to remove them from the unresolved list without changing your interaction model.

An utterance remains on the Unresolved Utterances tab until you either map it to an intent or slot, or mark it as resolved.

Map an utterance to an intent or slot

To map a single utterance to an intent or slot, navigate to the Intent History page and click the Unresolved Utterances tab. Note that the list defaults to the first 10 utterances across both the unresolved and resolved tabs, so you may need click Show…more to display additional utterances.

You can map an utterance to any intent in your interaction model. This adds the utterance to that intent's set of sample utterances.

You can map an utterance to any intent slot that is defined as a required slot in your dialog model. This adds the utterance to the user utterances used to elicit a value for the slot.

You can edit the text of the utterance if necessary. Use curly brackets ({ }) to specify slot names in the utterance.

To map an utterance to an intent or slot:

  1. Find the utterance in the table and click the Map button.
  2. Do one of the following:
    • To map to an intent, select the intent from the drop-down list, then edit the utterance if necessary.
    • To map to a slot, select the Slot tab, then select both the intent and the slot. Note that you can select only from slots that are configured as required in the dialog model.
  3. Click Map Utterance.
  4. Save and build the interaction model.

You can also map multiple utterances to the same intent or slot at the same time. Note that you cannot update the utterance text in this case – each selected utterance is added as-is to the intent or slot utterances.

To map multiple utterances at once:

  1. Select the check box for each utterance to map.
  2. Above the table of utterances, click Bulk Map.
  3. Select an intent from the drop-down list, or switch to the Slot tab and select both the intent and the slot. Note that you can select only from slots that are configured as required in the dialog model.
  4. Click Map Utterance.
  5. Save and build the interaction model.

After you map an utterance to a different intent or slot, the original utterance is moved to the Resolved Utterances tab. When the revised utterance is used frequently enough in the future to be included in the intent history, it is shown as a new utterance on the Unresolved Utterances tab.

Manually move an utterance to the resolved tab

When you manually resolve an utterance without mapping it to a different intent, it is moved to the Resolved Utterances tab.

To resolve one or more utterances, select the check box for each utterance, then click Resolve.

Once an utterance is resolved in this way, it remains on the Resolved Utterances tab. If the same utterance meets the criteria to be included in the intent history in the future, it is still considered resolved and is displayed on the Resolved Utterances tab.

Intent History and AMAZON.FallbackIntent

The AMAZON.FallbackIntent built-in intent is triggered when the user's utterance does not match any of your skill's intents. This intent displays just like the others on the Intent History page, so you can see frequent utterances that did not match your interaction model.

This is a useful way to see functionality users may be expecting, or common misspoken phrases that you may want to add to your skill. For example, if many users ask your "plan my trip" skill about the weather in a destination city, you may want to add this functionality to the skill.

Be sure to include AMAZON.FallbackIntent in your interaction model so that you can see these results.

Was this page helpful?

Last updated: Jan 26, 2024