Package com.redis.vl.extensions.summarization

Class ExtractiveSelector

java.lang.Object

com.redis.vl.extensions.summarization.ExtractiveSelector

public class ExtractiveSelector
extends Object

BERT-based extractive summarization using sentence clustering.

This class selects the most representative sentences from a document by embedding sentences with BERT, clustering them with k-means, and selecting the sentence closest to each cluster centroid.

Key Feature: Preserves original text exactly, which is critical for SubEM (Substring Exact Match) evaluation where paraphrasing fails.

Example Usage:

 SentenceTransformersVectorizer vectorizer = SentenceTransformersVectorizer.builder()
     .modelName("all-MiniLM-L6-v2")
     .build();

 ExtractiveSelector selector = new ExtractiveSelector(vectorizer);
 SentenceSplitter splitter = new SentenceSplitter();

 String document = "Long document text...";
 List<String> sentences = splitter.split(document);
 List<String> keySentences = selector.selectKeySentences(sentences, 10);

 // keySentences contains the 10 most representative sentences
 // in their original order, with exact original text preserved
 

Nested Class Summary

Nested Classes

Modifier and Type	Class	Description
static class	ExtractiveSelector.Builder

Constructor Summary

Constructors

Constructor	Description
ExtractiveSelector(SentenceTransformersVectorizer embedder)	Create an extractive selector with default settings.
ExtractiveSelector(SentenceTransformersVectorizer embedder, int defaultNumSentences)	Create an extractive selector with custom number of sentences.
ExtractiveSelector(SentenceTransformersVectorizer embedder, int defaultNumSentences, int maxIterations)	Create an extractive selector with full configuration.

Method Summary

Modifier and Type	Method	Description
static ExtractiveSelector.Builder	builder(SentenceTransformersVectorizer embedder)	Builder for ExtractiveSelector.
List<String>	selectKeySentences(List<String> sentences)	Select the most representative sentences using the default count.
List<String>	selectKeySentences(List<String> sentences, int k)	Select the k most representative sentences from the input.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Details

ExtractiveSelector

public ExtractiveSelector(SentenceTransformersVectorizer embedder)

Create an extractive selector with default settings.

Parameters:

embedder - The sentence transformer vectorizer for embeddings

ExtractiveSelector

public ExtractiveSelector(SentenceTransformersVectorizer embedder,
 int defaultNumSentences)

Create an extractive selector with custom number of sentences.

Parameters:

embedder - The sentence transformer vectorizer for embeddings

defaultNumSentences - Default number of sentences to select

ExtractiveSelector

public ExtractiveSelector(SentenceTransformersVectorizer embedder,
 int defaultNumSentences,
 int maxIterations)

Create an extractive selector with full configuration.

Parameters:

embedder - The sentence transformer vectorizer for embeddings

defaultNumSentences - Default number of sentences to select

maxIterations - Maximum k-means iterations

Method Details

selectKeySentences

public List<String> selectKeySentences(List<String> sentences)

Select the most representative sentences using the default count.

Parameters:

sentences - List of sentences to select from

Returns:

Selected sentences in original order

selectKeySentences

public List<String> selectKeySentences(List<String> sentences,
 int k)

Select the k most representative sentences from the input.

Algorithm:

1. Embed all sentences using BERT
2. Cluster embeddings using k-means++
3. For each cluster, select the sentence closest to the centroid
4. Return sentences in their original order

Parameters:

sentences - List of sentences to select from

k - Number of sentences to select

Returns:

Selected sentences in original order (preserves exact text)

builder

public static ExtractiveSelector.Builder builder(SentenceTransformersVectorizer embedder)

Builder for ExtractiveSelector.