Amazon Lex Model Building Service Update: Amazon Lex supports the option to enable accuracy improvements and specify an intent classification confidence score threshold.

Author: AWS

.changes/next-release/feature-AmazonLexModelBuildingService-4077aca.json

@@ -0,0 +1,5 @@
+{
+    "type": "feature",
+    "category": "Amazon Lex Model Building Service",
+    "description": "Amazon Lex supports the option to enable accuracy improvements and specify an intent classification confidence score threshold."
+}

services/lexmodelbuilding/src/main/resources/codegen-resources/service-2.json

@@ -952,6 +952,11 @@
       },
       "documentation":"<p>Specifies a Lambda function that verifies requests to a bot or fulfills the user's request to a bot..</p>"
     },
+    "ConfidenceThreshold":{
+      "type":"double",
+      "max":1,
+      "min":0
+    },
     "ConflictException":{
       "type":"structure",
       "members":{
@@ -1086,6 +1091,10 @@
           "shape":"Boolean",
           "documentation":"<p>For each Amazon Lex bot created with the Amazon Lex Model Building Service, you must specify whether your use of Amazon Lex is related to a website, program, or other application that is directed or targeted, in whole or in part, to children under age 13 and subject to the Children's Online Privacy Protection Act (COPPA) by specifying <code>true</code> or <code>false</code> in the <code>childDirected</code> field. By specifying <code>true</code> in the <code>childDirected</code> field, you confirm that your use of Amazon Lex <b>is</b> related to a website, program, or other application that is directed or targeted, in whole or in part, to children under age 13 and subject to COPPA. By specifying <code>false</code> in the <code>childDirected</code> field, you confirm that your use of Amazon Lex <b>is not</b> related to a website, program, or other application that is directed or targeted, in whole or in part, to children under age 13 and subject to COPPA. You may not specify a default value for the <code>childDirected</code> field that does not accurately reflect whether your use of Amazon Lex is related to a website, program, or other application that is directed or targeted, in whole or in part, to children under age 13 and subject to COPPA.</p> <p>If your use of Amazon Lex relates to a website, program, or other application that is directed in whole or in part, to children under age 13, you must obtain any required verifiable parental consent under COPPA. For information regarding the use of Amazon Lex in connection with websites, programs, or other applications that are directed or targeted, in whole or in part, to children under age 13, see the <a href=\"https://aws.amazon.com/lex/faqs#data-security\">Amazon Lex FAQ.</a> </p>"
         },
+        "enableModelImprovements":{
+          "shape":"Boolean",
+          "documentation":"<p>Indicates whether the bot uses the new natural language understanding (NLU) model or the original NLU. True indicates that the bot is using the new model, otherwise, false.</p>"
+        },
         "detectSentiment":{
           "shape":"Boolean",
           "documentation":"<p>Indicates whether utterances entered by the user should be sent to Amazon Comprehend for sentiment analysis.</p>"
@@ -1759,6 +1768,14 @@
           "shape":"IntentList",
           "documentation":"<p>An array of <code>intent</code> objects. For more information, see <a>PutBot</a>.</p>"
         },
+        "enableModelImprovements":{
+          "shape":"Boolean",
+          "documentation":"<p>Indicates whether the bot uses the new natural language understanding (NLU) model or the original NLU. True indicates that the bot is using the new model, otherwise, false.</p>"
+        },
+        "nluIntentConfidenceThreshold":{
+          "shape":"ConfidenceThreshold",
+          "documentation":"<p>The score that determines where Amazon Lex inserts the <code>AMAZON.FallbackIntent</code>, <code>AMAZON.KendraSearchIntent</code>, or both when returning alternative intents in a <a href=\"https://docs.aws.amazon.com/lex/latest/dg/API_runtime_PostContent.html\">PostContent</a> or <a href=\"https://docs.aws.amazon.com/lex/latest/dg/API_runtime_PostText.html\">PostText</a> response. <code>AMAZON.FallbackIntent</code> and <code>AMAZON.KendraSearchIntent</code> are only inserted if they are configured for the bot.</p>"
+        },
         "clarificationPrompt":{
           "shape":"Prompt",
           "documentation":"<p>The message Amazon Lex uses when it doesn't understand the user's request. For more information, see <a>PutBot</a>. </p>"
@@ -2940,6 +2957,14 @@
           "shape":"IntentList",
           "documentation":"<p>An array of <code>Intent</code> objects. Each intent represents a command that a user can express. For example, a pizza ordering bot might support an OrderPizza intent. For more information, see <a>how-it-works</a>.</p>"
         },
+        "enableModelImprovements":{
+          "shape":"Boolean",
+          "documentation":"<p>Set to <code>true</code> to enable the use of a new natural language understanding (NLU) model. Using the new NLU may improve the performance of your bot. </p> <p>When you set the <code>enableModelImprovements</code> parameter to <code>true</code> you can use the <code>nluIntentConfidenceThreshold</code> parameter to configure confidence scores. For more information, see <a href=\"https://docs.aws.amazon.com/lex/latest/dg/confidence-scores.html\">Confidence Scores</a>.</p> <p>You can only set the <code>enableModelImprovements</code> parameter in certain Regions. If you set the parameter to <code>true</code>, your bot will use the new NLU. If you set the parameter to <code>false</code>, your bot will continue to use the original NLU. If you set the parameter to <code>false</code> after setting it to <code>true</code>, your bot will return to the original NLU.</p> <p>The Regions where you can set the <code>enableModelImprovements</code> parameter to <code>true</code> are:</p> <ul> <li> <p>US East (N. Virginia) (us-east-1)</p> </li> <li> <p>US West (Oregon) (us-west-2)</p> </li> <li> <p>Asia Pacific (Sydney) (ap-southeast-2)</p> </li> <li> <p>EU (Ireland) (eu-west-1)</p> </li> </ul> <p>In other Regions, the <code>enableModelImprovements</code> parameter is set to <code>true</code> by default. In these Regions setting the parameter to <code>false</code> throws a <code>ValidationException</code> exception.</p> <ul> <li> <p>Asia Pacific (Singapore) (ap-southeast-1)</p> </li> <li> <p>Asia Pacific (Tokyo) (ap-northeast-1)</p> </li> <li> <p>EU (Frankfurt) (eu-central-1)</p> </li> <li> <p>EU (London) (eu-west-2)</p> </li> </ul>"
+        },
+        "nluIntentConfidenceThreshold":{
+          "shape":"ConfidenceThreshold",
+          "documentation":"<p>Determines the threshold where Amazon Lex will insert the <code>AMAZON.FallbackIntent</code>, <code>AMAZON.KendraSearchIntent</code>, or both when returning alternative intents in a <a href=\"https://docs.aws.amazon.com/lex/latest/dg/API_runtime_PostContent.html\">PostContent</a> or <a href=\"https://docs.aws.amazon.com/lex/latest/dg/API_runtime_PostText.html\">PostText</a> response. <code>AMAZON.FallbackIntent</code> and <code>AMAZON.KendraSearchIntent</code> are only inserted if they are configured for the bot.</p> <p>You must set the <code>enableModelImprovements</code> parameter to <code>true</code> to use confidence scores.</p> <p>For example, suppose a bot is configured with the confidence threshold of 0.80 and the <code>AMAZON.FallbackIntent</code>. Amazon Lex returns three alternative intents with the following confidence scores: IntentA (0.70), IntentB (0.60), IntentC (0.50). The response from the <code>PostText</code> operation would be:</p> <ul> <li> <p>AMAZON.FallbackIntent</p> </li> <li> <p>IntentA</p> </li> <li> <p>IntentB</p> </li> <li> <p>IntentC</p> </li> </ul>"
+        },
         "clarificationPrompt":{
           "shape":"Prompt",
           "documentation":"<p>When Amazon Lex doesn't understand the user's intent, it uses this message to get clarification. To specify how many times Amazon Lex should repeat the clarification prompt, use the <code>maxAttempts</code> field. If Amazon Lex still doesn't understand, it sends the message in the <code>abortStatement</code> field. </p> <p>When you create a clarification prompt, make sure that it suggests the correct response from the user. for example, for a bot that orders pizza and drinks, you might create this clarification prompt: \"What would you like to do? You can say 'Order a pizza' or 'Order a drink.'\"</p> <p>If you have defined a fallback intent, it will be invoked if the clarification prompt is repeated the number of times defined in the <code>maxAttempts</code> field. For more information, see <a href=\"https://docs.aws.amazon.com/lex/latest/dg/built-in-intent-fallback.html\"> AMAZON.FallbackIntent</a>.</p> <p>If you don't define a clarification prompt, at runtime Amazon Lex will return a 400 Bad Request exception in three cases: </p> <ul> <li> <p>Follow-up prompt - When the user responds to a follow-up prompt but does not provide an intent. For example, in response to a follow-up prompt that says \"Would you like anything else today?\" the user says \"Yes.\" Amazon Lex will return a 400 Bad Request exception because it does not have a clarification prompt to send to the user to get an intent.</p> </li> <li> <p>Lambda function - When using a Lambda function, you return an <code>ElicitIntent</code> dialog type. Since Amazon Lex does not have a clarification prompt to get an intent from the user, it returns a 400 Bad Request exception.</p> </li> <li> <p>PutSession operation - When using the <code>PutSession</code> operation, you send an <code>ElicitIntent</code> dialog type. Since Amazon Lex does not have a clarification prompt to get an intent from the user, it returns a 400 Bad Request exception.</p> </li> </ul>"
@@ -3001,6 +3026,14 @@
           "shape":"IntentList",
           "documentation":"<p>An array of <code>Intent</code> objects. For more information, see <a>PutBot</a>.</p>"
         },
+        "enableModelImprovements":{
+          "shape":"Boolean",
+          "documentation":"<p>Indicates whether the bot uses the new natural language understanding (NLU) model or the original NLU. True indicates that the bot is using the new model, otherwise, false.</p>"
+        },
+        "nluIntentConfidenceThreshold":{
+          "shape":"ConfidenceThreshold",
+          "documentation":"<p>The score that determines where Amazon Lex inserts the <code>AMAZON.FallbackIntent</code>, <code>AMAZON.KendraSearchIntent</code>, or both when returning alternative intents in a <a href=\"https://docs.aws.amazon.com/lex/latest/dg/API_runtime_PostContent.html\">PostContent</a> or <a href=\"https://docs.aws.amazon.com/lex/latest/dg/API_runtime_PostText.html\">PostText</a> response. <code>AMAZON.FallbackIntent</code> and <code>AMAZON.KendraSearchIntent</code> are only inserted if they are configured for the bot.</p>"
+        },
         "clarificationPrompt":{
           "shape":"Prompt",
           "documentation":"<p> The prompts that Amazon Lex uses when it doesn't understand the user's intent. For more information, see <a>PutBot</a>. </p>"