Extract entities with ML Kit on Android

  • This is a beta API for extracting entities like dates, flight numbers, and money from text using machine learning.

  • The API requires Android API level 21 or higher and has an app size impact of up to ~5.6MB.

  • You can configure the entity extractor with options like language and filter, then call annotate() to get the entities.

  • Models are downloaded dynamically, and you can manage them explicitly using the model management API.

  • This API is offered in beta and is subject to change, without an SLA or deprecation policy.

To analyze a piece of text and extract the entities in it, invoke the annotate() method and pass it either the text string or an instance of EntityExtractionParams which can specify additional options such as a reference time, timezone, or a filter to limit the search for a subset of entity types. The API returns a list of EntityAnnotation objects containing information about each entity.

SDK Nameentity-extraction
ImplementationAssets for base detector are statically linked to your app at build time.
Asset Size ImpactEntity extraction has an app size impact of up to ~5.6MB.

Try it out

  • Play around with the sample app to see an example usage of this API.

Before you begin

  1. In your project-level build.gradle file, make sure that Google's Maven repository is included in both the buildscript and allprojects sections.

  2. Add the dependency for the ML Kit entity extraction library to your module's app-level gradle file, which is usually named app/build.gradle:

    dependencies{
// ...
implementation'com.google.mlkit:entity-extraction:16.0.0-beta6'
}

Extract entities

Create an EntityExtractor object, and configure it with EntityExtractorOptions

Kotlin

valentityExtractor=
EntityExtraction.getClient(
EntityExtractorOptions.Builder(EntityExtractorOptions.ENGLISH)
.build())

Java

EntityExtractorentityExtractor=
EntityExtraction.getClient(
newEntityExtractorOptions.Builder(EntityExtractorOptions.ENGLISH)
.build());

EntityExtractorOptions also accepts a user-defined Executor if you need, otherwise it will use default Executor in ML Kit

Ensure the required model has been downloaded to the device.

Kotlin

entityExtractor
.downloadModelIfNeeded()
.addOnSuccessListener{_->
/* Model downloading succeeded, you can call extraction API here. */
}
.addOnFailureListener{_->/* Model downloading failed. */}

Java

entityExtractor
.downloadModelIfNeeded()
.addOnSuccessListener(
aVoid->{
// Model downloading succeeded, you can call the extraction API here. 
})
.addOnFailureListener(
exception->{
// Model downloading failed.
});

After you confirm the model has been downloaded, pass a string or EntityExtractionParams to annotate(). Don’t call annotate() until you know the model is available.

Kotlin

valparams=
EntityExtractionParams.Builder("My flight is LX373, please pick me up at 8am tomorrow.")
.setEntityTypesFilter((/* optional entity type filter */)
.setPreferredLocale(/* optional preferred locale */)
.setReferenceTime(/* optional reference date-time */)
.setReferenceTimeZone(/* optional reference timezone */)
.build()
entityExtractor
.annotate(params)
.addOnSuccessListener{
// Annotation process was successful, you can parse the EntityAnnotations list here.
}
.addOnFailureListener{
// Check failure message here.
}

Java

EntityExtractionParamsparams=newEntityExtractionParams
.Builder("My flight is LX373, please pick me up at 8am tomorrow.")
.setEntityTypesFilter(/* optional entity type filter */)
.setPreferredLocale(/* optional preferred locale */)
.setReferenceTime(/* optional reference date-time */)
.setReferenceTimeZone(/* optional reference timezone */)
.build();
entityExtractor
.annotate(params)
.addOnSuccessListener(newOnSuccessListener<List<EntityAnnotation>>(){
@Override
publicvoidonSuccess(List<EntityAnnotation>entityAnnotations){
// Annotation process was successful, you can parse the EntityAnnotations list here.
}
})
.addOnFailureListener(newOnFailureListener(){
@Override
publicvoidonFailure(@NonNullExceptione){
// Check failure message here.
}
});

PreferredLocale, ReferenceTime and ReferenceTimeZone will only influence DateTime entities. If these are not explicitly set, they default to values from the user's device.

Loop over the annotation results to retrieve information about the recognized entities.

Kotlin

for(entityAnnotationinentityAnnotations){
valentities:List<Entity>=entityAnnotation.entities
Log.d(TAG,"Range: ${entityAnnotation.start} - ${entityAnnotation.end}")
for(entityinentities){
when(entity){
isDateTimeEntity->{
Log.d(TAG,"Granularity: ${entity.dateTimeGranularity}")
Log.d(TAG,"TimeStamp: ${entity.timestampMillis}")
}
isFlightNumberEntity->{
Log.d(TAG,"Airline Code: ${entity.airlineCode}")
Log.d(TAG,"Flight number: ${entity.flightNumber}")
}
isMoneyEntity->{
Log.d(TAG,"Currency: ${entity.unnormalizedCurrency}")
Log.d(TAG,"Integer part: ${entity.integerPart}")
Log.d(TAG,"Fractional Part: ${entity.fractionalPart}")
}
else->{
Log.d(TAG," $entity")
}
}
}
}

Java

List<EntityAnnotation>entityAnnotations=/* Get from EntityExtractor */;
for(EntityAnnotationentityAnnotation:entityAnnotations){
List<Entity>entities=entityAnnotation.getEntities();
Log.d(TAG,String.format("Range: [%d, %d)",entityAnnotation.getStart(),entityAnnotation.getEnd()));
for(Entityentity:entities){
switch(entity.getType()){
caseEntity.TYPE_DATE_TIME:
DateTimeEntitydateTimeEntity=entity.asDateTimeEntity();
Log.d(TAG,"Granularity: "+dateTimeEntity.getDateTimeGranularity());
Log.d(TAG,"Timestamp: "+dateTimeEntity.getTimestampMillis());
caseEntity.TYPE_FLIGHT_NUMBER:
FlightNumberEntityflightNumberEntity=entity.asFlightNumberEntity();
Log.d(TAG,"Airline Code: "+flightNumberEntity.getAirlineCode());
Log.d(TAG,"Flight number: "+flightNumberEntity.getFlightNumber());
caseEntity.TYPE_MONEY:
MoneyEntitymoneyEntity=entity.asMoneyEntity();
Log.d(TAG,"Currency: "+moneyEntity.getUnnormalizedCurrency());
Log.d(TAG,"Integer Part: "+moneyEntity.getIntegerPart());
Log.d(TAG,"Fractional Part: "+moneyEntity.getFractionalPart());
caseEntity.TYPE_UNKNOWN:
default:
Log.d(TAG,"Entity: "+entity);
}
}
}

Call the close() method when you no longer need the EntityExtractor object. If you are using EntityExtractor in a Fragment or AppCompatActivity, you can call LifecycleOwner.getLifecycle() on the Fragment or AppCompatActivity, and then Lifecycle.addObserver. For example:

Kotlin

valoptions=...
valextractor=EntityExtraction.getClient(options);
getLifecycle().addObserver(extractor);

Java

EntityExtractorOptionsoptions=...
EntityExtractorextractor=EntityExtraction.getClient(options);
getLifecycle().addObserver(extractor);

Explicitly manage entity extraction models

When you use the entity extraction API as described above, ML Kit automatically downloads language specific models to the device as required (when you call downloadModelIfNeeded()). You can also explicitly manage the models you want available on the device by using ML Kit’s model management API. This can be useful if you want to download models ahead of time. The API also allows you to delete models that are no longer required.

Kotlin

valmodelManager=RemoteModelManager.getInstance()
// Get entity extraction models stored on the device.
modelManager.getDownloadedModels(EntityExtractionRemoteModel::class.java)
.addOnSuccessListener{
// ...
}
.addOnFailureListener({
// Error.
})

// Delete the German model if it's on the device.
valgermanModel=
EntityExtractionRemoteModel.Builder(EntityExtractorOptions.GERMAN).build()
modelManager.deleteDownloadedModel(germanModel)
.addOnSuccessListener({
// Model deleted.
})
.addOnFailureListener({
// Error.
})

// Download the French model.
valfrenchModel=
EntityExtractionRemoteModel.Builder(EntityExtractorOptions.FRENCH).build()
valconditions=
DownloadConditions.Builder()
.requireWifi()
.build()
modelManager.download(frenchModel,conditions)
.addOnSuccessListener({
// Model downloaded.
})
.addOnFailureListener({
// Error.
})

Java

// Get entity extraction models stored on the device.
modelManager.getDownloadedModels(EntityExtractionRemoteModel.class)
.addOnSuccessListener(newOnSuccessListener<Set<EntityExtractionRemoteModel>>(){
@Override
publicvoidonSuccess(Set<EntityExtractionRemoteModel>models){
// ...
}
})
.addOnFailureListener(newOnFailureListener(){
@Override
publicvoidonFailure(@NonNullExceptione){
// Error.
}
});
// Delete the German model if it's on the device.
EntityExtractionRemoteModelgermanModel=newEntityExtractionRemoteModel.Builder(EntityExtractorOptions.GERMAN).build();
modelManager.deleteDownloadedModel(germanModel)
.addOnSuccessListener(newOnSuccessListener<Void>(){
@Override
publicvoidonSuccess(Voidv){
// Model deleted.
}
})
.addOnFailureListener(newOnFailureListener(){
@Override
publicvoidonFailure(@NonNullExceptione){
// Error.
}
});
// Download the French model.
EntityExtractionRemoteModelfrenchModel=newEntityExtractionRemoteModel.Builder(EntityExtractorOptions.FRENCH).build();
DownloadConditionsconditions=newDownloadConditions.Builder()
.requireWifi()
.build();
modelManager.download(frenchModel,conditions)
.addOnSuccessListener(newOnSuccessListener<Void>(){
@Override
publicvoidonSuccess(Voidv){
// Model downloaded.
}
})
.addOnFailureListener(newOnFailureListener(){
@Override
publicvoidonFailure(@NonNullExceptione){
// Error.
}
});

Except as otherwise noted, the content of this page is licensed under the Creative Commons Attribution 4.0 License, and code samples are licensed under the Apache 2.0 License. For details, see the Google Developers Site Policies. Java is a registered trademark of Oracle and/or its affiliates.

Last updated 2026年09月01日 UTC.