Magnolia 6.1 reached end of life on March 31, 2021. This branch is no longer supported, see End-of-life policy.
The Text Classification module uses the Amazon Comprehend service to analyze and tag your text content. Amazon Comprehend is a natural language processing (NLP) service that uses machine learning to find insights and relationships in text. Magnolia uses the AWS Key Phrases service (BatchDetectKeyPhrases) to detect key phrases in your content during the classification process.
A key phrase is a string containing a noun phrase that describes a particular thing. It generally consists of a noun and the modifiers that distinguish it. For example, "day" is a noun; "a beautiful day" is a noun phrase that includes an article ("a") and an adjective ("beautiful"). Each key phrase includes a score that indicates the level of confidence that Amazon Comprehend has that the string is a noun phrase. You can use the score to determine if the detection has high enough confidence for your application.
artifactID | Description |
---|---|
| Parent reactor. |
| Provides the text classification module and service. |
| Provides an API to classify text. |
| Provides functionality to classify text via Amazon Comprehend. |
| Provides functionality to integrate content tags and the text classification service using decorations in the Pages app. |
Maven is the easiest way to install the modules. Add the following dependencies to your webapp. All other necessary dependencies will be brought in automatically:
<dependency>
<groupId>info.magnolia.ai.text</groupId>
<artifactId>magnolia-text-classification</artifactId>
<version>1.0.1</version>
</dependency>
<dependency>
<groupId>info.magnolia.ai.text</groupId>
<artifactId>magnolia-text-classification-api</artifactId>
<version>1.0.1</version>
</dependency>
<dependency>
<groupId>info.magnolia.ai.text</groupId>
<artifactId>magnolia-amazon-text-classification</artifactId>
<version>1.0.1</version>
</dependency>
<dependency>
<groupId>info.magnolia.ai.text</groupId>
<artifactId>magnolia-pages-content-tags-integration</artifactId>
<version>1.0.1</version>
</dependency>
If you don't use Maven, note that you may have to add these dependencies manually:
When using our out-of-the-box solution:
pages-content-tags-integration
submodule brings the content-tags functionality to the Pages app and handles aggregating text from the
website
workspace.magnolia-amazon-text-classification
submodule provides an out-of-the-box implementation to use Amazon Comprehend.This solution is straightforward to configure:
aggregateDefinition
for the Pages app (website
workspace) to specify:minConfidence
property to change the classification confidence score.If you so require, you can also write:
First, make sure that you have acquired appropriate permissions for the service in the Amazon IAM Management Console.
You need an AWS secret access key to make secure REST requests to the Amazon Comprehend API. Access keys consist of two parts:
Generate the key in the security credentials section of the Amazon IAM Management Console. (In the navigation bar on the upper right, choose your user name, and then choose My Security Credentials.)
Add the two parts of the key to your Magnolia instance in the Password manager app using the following names:
| |
| |
|
For more information about the key, see Understanding and Getting Your Security Credentials.
Under /amazon-text-classification/config.yaml
, you must configure the following properties for the classification service:
region: name: your_aws_region_name languageCode: en minConfidence: 0.85
Name | Description |
---|---|
| required Label designating a regional endpoint to which the text classification service connects, such as You must set a region name to configure the Amazon Comprehend service in Magnolia. To reduce data latency, AWS offers several regional endpoints. Each of the endpoints can be referred to in service configurations by a region name, for example For a list of available regions and labels, see https://docs.aws.amazon.com/general/latest/gr/rande.html#comprehend_region. |
languageCode | required, default is The language of the input documents. You can specify any of the primary languages supported by Amazon Comprehend: German ("de"), English ("en"), Spanish ("es"), French ("fr"), Italian ("it"), or Portuguese ("pt"). All documents must be in the same language. Amazon Comprehend can perform text analysis on English, French, German, Italian, Portuguese, and Spanish texts. |
minConfidence
| required, default is The confidence score of the classification. A decimal value between 0 and 1. The filter drops the tags with a confidence score lower than the value of this property. The Amazon Comprehend solution returns a confidence score for each key phrase tag. Tags with a confidence score lower than the value of the minConfidence property are dropped. Setting the value higher usually results in fewer key phrase tags being returned for your content. A higher confidence score means that the tag more correctly describes the text. |
The pages-content-tags-integration
module brings the content-tags functionality to the Pages app and handles aggregating text from the website
workspace out-of-the-box.
Text aggregators collect and aggregate the content that the classification service analyzes and generates tags from. You can specify from which field types content should be taken in the text aggregator configuration.
By default, the text aggregator for the Pages app gathers text from text, rich text, composite and switchable field types.
aggregateDefinition: fieldTypes: [text, richText, composite, switchable]
You can blacklist the terms you do not want to appear in your tags. For example, you may want to exclude your company name.
To do so, go to the Resource files app, under /text-classification/config.yaml
and add comma-separated terms to the excludedTerms
list. In this example, the words ACME, corporation and coyote are excluded:
termFilteringDefinition: excludedTerms: [ACME, corporation, coyote]
Note that the blacklist is case insensitive.
If you want to run text classification on a custom content app, you must write your own text aggregator implementation.
To do so:
TextAggregator
interface. @Multibinding
and add it to the module descriptor as a component for injection. For example, see pages-content-tags-integration/src/main/resources/META-INF/magnolia/pages-content-tags-integration.xml
.Decorate the text-classification configuration file, for example:
workspaceClassificationConfigurations: website: textAggregatorClassName: info.magnolia.ai.text.YourTextAggregator workspace: yourworkspace nodeType: mgnl:yournodetype
workspaceClassificationConfigurations | required |
| required Arbitrary, unique name for the decoration configuration. |
| required Fully qualified classname for your text aggregator. Example: |
| required The workspace where the content to be analyzed is stored. |
| required The name of the JCR node type for storing an item of the given content type. Example: |
The magnolia-amazon-text-classification
submodule provides an out-of-the-box implementation to use Amazon Comprehend.
However, if you want to use another third-party text classification service to classify and tag your content, you can write your own custom text classifier implementation.
Before configuring the text classifier, make sure you have administrator access to your third-party classification service, including the API documentation.
To create a custom text classifier you must implement the info.magnolia.ai.text.TextClassifier
interface.
Note that you can inject the TextClassifier
interface as a component in any running instance of Magnolia.
/** * Commons interface to classify text. */ public interface TextClassifier { /** * Takes a {@link String text} as parameter and returns a {@link Collection collection} * of {@link TextLabel Text label}s as output. * * <p> * Returns empty collection for the cases below: * <li>Upon exception</li> * <li>Text couldn't be classified</li> * </p> */ Collection<TextLabel> classify(String text); /** * Takes a collection containing the text of the input documents as a parameter. * * @param texts * A collection containing the text of the input documents. * @return Returns a {@link Map map} where keys are input texts, values are {@link Collection collection}s of detected {@link TextLabel Text label}s * for the input text or empty collections if an error occurs while processing the input text. * The returned map preserves the order of the texts in the input collection. * * <p> * Returns an empty map for the cases below: * <li>Input {@link Collection collection} is null or empty</li> * <li>All documents in input {@link Collection collection} are processed with an error</li> * </p> */ default Map<String, Collection<TextLabel>> classify(Collection<String> texts) { if (CollectionUtils.isNotEmpty(texts)) { return texts.stream() .collect(Collectors.toMap(mapper -> mapper, this::classify)); } return Collections.emptyMap(); }
Only one TextClassifier
should be used in a Magnolia instance. Remove the out-of-the-box AmazonTextClassifier
if you choose to implement your own.
If you have more than one module that specifies the TextClassifier implementation in the module class, the TextClassifier from the module that was started last is used.
See the following files for an example implementation:
info.magnolia.ai.text.amazon.AmazonTextClassifier
META-INF/magnolia/amazon-text-classification.xml
The text classification and tagging action are executed during the startup of the author instance. You can also trigger the action manually in the Pages app by selecting one or more pages and clicking the Run classification action.
Pages that have already been tagged are marked as such using a JCR property called lastTaggingAttemptDateByTextClassifier
. Executing the manual classification action forces a new tag to be set even if the content was previously tagged.
The text classification feature is available only on author instances.
Once a page has been tagged, you can remove some or all of the tags by selecting the page and clicking the Modify tags action in the Pages app.
In the dialog box that opens, you can remove individual tags or click Remove all tags.
Note that content tagging currently has an issue when creating tags of words with accented characters. For example, Genève is tagged as Gen-ve
. This means that searching for the tag Geneve or Genève will not return any results. The issue is being tracked here:
-
CONTTAGS-69Getting issue details...
STATUS