Gerar respostas inteligentes com o Kit de ML no Android
Mantenha tudo organizado com as coleções
Salve e categorize o conteúdo com base nas suas preferências.
O Kit de ML pode gerar respostas curtas a mensagens usando um modelo no dispositivo.
Para gerar respostas inteligentes, transmita ao Kit de ML um registro das mensagens recentes de uma
conversa. Se o Kit de ML determinar que a conversa está em inglês e que
ela não trata de um assunto potencialmente confidencial, o Kit vai
gerar até três respostas, que podem ser sugeridas ao usuário.
O modelo é vinculado de forma estática ao app no momento da criação.
O modelo é baixado dinamicamente pelo Google Play Services.
Impacto do tamanho do app
Aumento de cerca de 5,7 MB.
Aumento de cerca de 200 KB.
Tempo de inicialização
O modelo fica disponível imediatamente.
Talvez seja necessário esperar o download do modelo antes do primeiro uso.
Faça um teste
Teste o app de exemplo para
conferir um exemplo de uso dessa API.
Antes de começar
No arquivo build.gradle no nível do projeto, inclua o repositório
Maven do Google nas seções buildscript e allprojects.
Adicione as dependências das bibliotecas do Kit de ML para Android ao arquivo Gradle
no nível do app do módulo, que geralmente é app/build.gradle. Escolha uma das dependências a seguir com base nas suas necessidades:
Para agrupar o modelo e o app:
dependencies{// ...// Use this dependency to bundle the model with your appimplementation'com.google.mlkit:smart-reply:17.0.4'}
Se você optar por usar o modelo no Google Play Services, poderá configurar
seu app para fazer o download automático do modelo no dispositivo depois que ele for
instalado na Play Store. Adicione a declaração abaixo ao
arquivo AndroidManifest.xml do app:
Também é possível verificar explicitamente a disponibilidade do modelo e solicitar o download pela
API ModuleInstallClient do Google Play Services.
Se você não ativar os downloads do modelo no momento da instalação ou solicitar um download explícito,
o modelo será transferido na primeira vez em que você executar o gerador de respostas inteligentes.
As solicitações feitas antes da conclusão do download não produzem resultados.
1. Criar um objeto de histórico de conversas
Para gerar respostas inteligentes, transmita para o Kit de ML uma List de objetos TextMessage em ordem cronológica, com o carimbo de data/hora mais antigo primeiro.
Sempre que o usuário enviar uma mensagem, adicione a mensagem e o carimbo de data/hora ao histórico de conversas:
Kotlin
conversation.add(TextMessage.createForLocalUser("heading out now",System.currentTimeMillis()))
Java
conversation.add(TextMessage.createForLocalUser("heading out now",System.currentTimeMillis()));
Sempre que o usuário receber uma mensagem, adicione a mensagem, o carimbo de data/hora e o ID do usuário do remetente ao histórico de conversas. O ID do usuário pode ser qualquer string que
identifique exclusivamente o remetente na conversa. O ID do usuário não precisa
corresponder a nenhum dado do usuário e não precisa estar consistente
entre conversas ou invocações do gerador de respostas inteligentes.
Kotlin
conversation.add(TextMessage.createForRemoteUser("Are you coming back soon?",System.currentTimeMillis(),userId))
Java
conversation.add(TextMessage.createForRemoteUser("Are you coming back soon?",System.currentTimeMillis(),userId));
Um objeto de histórico de conversas é parecido com o exemplo a seguir:
Carimbo de data/hora
userID
isLocalUser
Mensagem
Thu Feb 21 13:13:39 PST 2019
verdadeiro
are you on your way?
Thu Feb 21 13:15:03 PST 2019
FRIEND0
falso
Running late, sorry!
O Kit de ML sugere respostas para a última mensagem em um histórico de conversas. A última mensagem
precisa ser de um usuário não local. No exemplo acima, a última mensagem na conversa
é do usuário não local FRIEND0. Quando você usa o Kit de ML para transmitir esse registro, ele sugere
respostas à mensagem de FRIENDO: "Estou atrasado, desculpe!"
2. Receber respostas de mensagens
Para gerar respostas inteligentes a uma mensagem, receba uma instância de SmartReplyGenerator
e transmita o histórico de conversas para o método suggestReplies():
Kotlin
valsmartReplyGenerator=SmartReply.getClient()smartReply.suggestReplies(conversation).addOnSuccessListener{result->if(result.getStatus()==SmartReplySuggestionResult.STATUS_NOT_SUPPORTED_LANGUAGE){// The conversation's language isn't supported, so// the result doesn't contain any suggestions.}elseif(result.getStatus()==SmartReplySuggestionResult.STATUS_SUCCESS){// Task completed successfully// ...}}.addOnFailureListener{// Task failed with an exception// ...}
Java
SmartReplyGeneratorsmartReply=SmartReply.getClient();smartReply.suggestReplies(conversation).addOnSuccessListener(newOnSuccessListener(){@OverridepublicvoidonSuccess(SmartReplySuggestionResultresult){if(result.getStatus()==SmartReplySuggestionResult.STATUS_NOT_SUPPORTED_LANGUAGE){// The conversation's language isn't supported, so// the result doesn't contain any suggestions.}elseif(result.getStatus()==SmartReplySuggestionResult.STATUS_SUCCESS){// Task completed successfully// ...}}}).addOnFailureListener(newOnFailureListener(){@OverridepublicvoidonFailure(@NonNullExceptione){// Task failed with an exception// ...}});
Se a operação for bem-sucedida, um objeto SmartReplySuggestionResult será transmitido para
o gerenciador de sucesso. Esse objeto contém uma lista de até três respostas sugeridas,
que você pode apresentar ao usuário:
Talvez o Kit de ML não retorne resultados se o modelo não estiver confiante na
relevância das respostas sugeridas, se a conversa de entrada não estiver em
inglês ou se o modelo detectar um assunto confidencial.
[null,null,["Última atualização 2025-08-29 UTC."],[[["\u003cp\u003eML Kit's Smart Reply API generates up to three relevant reply suggestions for English conversations using an on-device model.\u003c/p\u003e\n"],["\u003cp\u003eYou can integrate Smart Reply by either bundling the model with your app (larger size) or dynamically downloading it (smaller size, requires Google Play Services).\u003c/p\u003e\n"],["\u003cp\u003eTo use the API, provide a conversation history as input, and ML Kit will suggest replies to the last message if it's from a non-local user.\u003c/p\u003e\n"],["\u003cp\u003eThe suggested replies are returned only if the conversation is in English, does not contain sensitive content, and the model is confident in their relevance.\u003c/p\u003e\n"]]],["ML Kit generates up to three smart replies to messages in English conversations, excluding sensitive content. This is done by passing a chronologically ordered list of `TextMessage` objects to the `suggestReplies()` method. The API can use a bundled model (5.7 MB increase) or an unbundled model (200 KB increase) via Google Play Services. The unbundled model may have a delay before the first use, and may not include any results. Implementation requires adding the appropriate library dependency and building the conversation history.\n"],null,["ML Kit can generate short replies to messages using an on-device model.\n\nTo generate smart replies, you pass ML Kit a log of recent messages in a\nconversation. If ML Kit determines the conversation is in English, and that\nthe conversation doesn't have potentially sensitive subject matter, ML Kit\ngenerates up to three replies, which you can suggest to your user.\n\n\u003cbr /\u003e\n\n| This API is available using either an unbundled library that must be downloaded before use or a bundled library that increases your app size. See [this guide](/ml-kit/tips/installation-paths) for more information on the differences between the two installation options.\n\n| | Bundled | Unbundled |\n|-------------------------|-------------------------------------------------------|------------------------------------------------------------|\n| **Library name** | `com.google.mlkit:smart-reply` | `com.google.android.gms:play-services-mlkit-smart-reply` |\n| **Implementation** | Model is statically linked to your app at build time. | Model is dynamically downloaded via Google Play Services. |\n| **App size impact** | About 5.7 MB size increase. | About 200 KB size increase. |\n| **Initialization time** | Model is available immediately. | Might have to wait for model to download before first use. |\n\n| **Note:** The unbundled version of Smart Reply is currently offered in beta, which means it might be changed in backward-incompatible ways and is not subject to any SLA or deprecation policy.\n\nTry it out\n\n- Play around with [the sample app](https://github.com/googlesamples/mlkit/tree/master/android/smartreply) to see an example usage of this API.\n\nBefore you begin This API requires Android API level 21 or above. Make sure that your app's build file uses a `minSdkVersion` value of 21 or higher.\n\n1. In your project-level `build.gradle` file, make sure to include Google's\n Maven repository in both your `buildscript` and `allprojects` sections.\n\n2. Add the dependencies for the ML Kit Android libraries to your module's\n app-level gradle file, which is usually `app/build.gradle`. Choose one of\n the following dependencies based on your needs:\n\n - To bundle the model with your app:\n\n dependencies {\n // ...\n // Use this dependency to bundle the model with your app\n implementation 'com.google.mlkit:smart-reply:17.0.4'\n }\n\n - To use the model in Google Play Services:\n\n dependencies {\n // ...\n // Use this dependency to use the dynamically downloaded model in Google Play Services\n implementation 'com.google.android.gms:play-services-mlkit-smart-reply:16.0.0-beta1'\n }\n\n If you choose to use the model in Google Play Services, you can configure\n your app to automatically download the model to the device after your app is\n installed from the Play Store. By adding the following declaration to your\n app's `AndroidManifest.xml` file: \n\n \u003capplication ...\u003e\n ...\n \u003cmeta-data\n android:name=\"com.google.mlkit.vision.DEPENDENCIES\"\n android:value=\"smart_reply\" \u003e\n \u003c!-- To use multiple models: android:value=\"smart_reply,model2,model3\" --\u003e\n \u003c/application\u003e\n\n You can also explicitly check the model availability and request download through\n Google Play services [ModuleInstallClient API](https://developers.google.com/android/guides/module-install-apis).\n\n If you don't enable install-time model downloads or request explicit download,\n the model is downloaded the first time you run the smart reply generator.\n Requests you make before the download has completed produce no results.\n\n\n 1. Create a conversation history object\n\n To generate smart replies, you pass ML Kit a chronologically-ordered `List`\n of `TextMessage` objects, with the earliest timestamp first.\n\n Whenever the user sends a message, add the message and its timestamp to the\n conversation history: \n\n Kotlin \n\n ```kotlin\n conversation.add(TextMessage.createForLocalUser(\n \"heading out now\", System.currentTimeMillis()))\n ```\n\n Java \n\n ```java\n conversation.add(TextMessage.createForLocalUser(\n \"heading out now\", System.currentTimeMillis()));\n ```\n\n Whenever the user receives a message, add the message, its timestamp, and the\n sender's user ID to the conversation history. The user ID can be any string that\n uniquely identifies the sender within the conversation. The user ID doesn't need\n to correspond to any user data, and the user ID doesn't need to be consistent\n between conversation or invocations of the smart reply generator. \n\n Kotlin \n\n ```kotlin\n conversation.add(TextMessage.createForRemoteUser(\n \"Are you coming back soon?\", System.currentTimeMillis(), userId))\n ```\n\n Java \n\n ```java\n conversation.add(TextMessage.createForRemoteUser(\n \"Are you coming back soon?\", System.currentTimeMillis(), userId));\n ```\n\n A conversation history object looks like the following example:\n\n | Timestamp | userID | isLocalUser | Message |\n |------------------------------|---------|-------------|----------------------|\n | Thu Feb 21 13:13:39 PST 2019 | | true | are you on your way? |\n | Thu Feb 21 13:15:03 PST 2019 | FRIEND0 | false | Running late, sorry! |\n\n ML Kit suggests replies to the last message in a conversation history. The last message\n should be from a non-local user. In the example above, the last message in the conversation\n is from the non-local user FRIEND0. When you use pass ML Kit this log, it suggests\n replies to FRIENDO's message: \"Running late, sorry!\"\n\n 2. Get message replies\n\n To generate smart replies to a message, get an instance of `SmartReplyGenerator`\n and pass the conversation history to its `suggestReplies()` method: \n\n Kotlin \n\n ```kotlin\n val smartReplyGenerator = SmartReply.getClient()\n smartReply.suggestReplies(conversation)\n .addOnSuccessListener { result -\u003e\n if (result.getStatus() == SmartReplySuggestionResult.STATUS_NOT_SUPPORTED_LANGUAGE) {\n // The conversation's language isn't supported, so\n // the result doesn't contain any suggestions.\n } else if (result.getStatus() == SmartReplySuggestionResult.STATUS_SUCCESS) {\n // Task completed successfully\n // ...\n }\n }\n .addOnFailureListener {\n // Task failed with an exception\n // ...\n }\n ```\n\n Java \n\n ```java\n SmartReplyGenerator smartReply = SmartReply.getClient();\n smartReply.suggestReplies(conversation)\n .addOnSuccessListener(new OnSuccessListener() {\n @Override\n public void onSuccess(SmartReplySuggestionResult result) {\n if (result.getStatus() == SmartReplySuggestionResult.STATUS_NOT_SUPPORTED_LANGUAGE) {\n // The conversation's language isn't supported, so\n // the result doesn't contain any suggestions.\n } else if (result.getStatus() == SmartReplySuggestionResult.STATUS_SUCCESS) {\n // Task completed successfully\n // ...\n }\n }\n })\n .addOnFailureListener(new OnFailureListener() {\n @Override\n public void onFailure(@NonNull Exception e) {\n // Task failed with an exception\n // ...\n }\n });\n ```\n\n If the operation succeeds, a `SmartReplySuggestionResult` object is passed to\n the success handler. This object contains a list of up to three suggested replies,\n which you can present to your user: \n\n Kotlin \n\n ```kotlin\n for (suggestion in result.suggestions) {\n val replyText = suggestion.text\n }\n ```\n\n Java \n\n ```java\n for (SmartReplySuggestion suggestion : result.getSuggestions()) {\n String replyText = suggestion.getText();\n }\n ```\n\n Note that ML Kit might not return results if the model isn't confident in\n the relevance of the suggested replies, the input conversation isn't in\n English, or if the model detects sensitive subject matter."]]