Package dev.langchain4j.data.document.splitter

Class DocumentByWordSplitter

java.lang.Object

dev.langchain4j.data.document.splitter.HierarchicalDocumentSplitter

dev.langchain4j.data.document.splitter.DocumentByWordSplitter

All Implemented Interfaces:

DocumentSplitter

public class DocumentByWordSplitter
extends HierarchicalDocumentSplitter

Splits the provided Document into words and attempts to fit as many words as possible into a single TextSegment, adhering to the limit set by maxSegmentSize.

The maxSegmentSize can be defined in terms of characters (default) or tokens. For token-based limit, a Tokenizer must be provided.

Word boundaries are detected by a minimum of one space (" "). Any additional whitespaces before or after are ignored. So, the following examples are all valid word separators: " ", " ", "\n", and so on.

If multiple words fit within maxSegmentSize, they are joined together using a space (" ").

Although this should not happen, if a single word is too long and exceeds maxSegmentSize, the subSplitter (DocumentByCharacterSplitter by default) is used to split it into smaller parts and place them into multiple segments. Such segments contain only the parts of the split long word.

Each TextSegment inherits all metadata from the Document and includes an "index" metadata key representing its position within the document (starting from 0).

Field Summary

Fields inherited from class dev.langchain4j.data.document.splitter.HierarchicalDocumentSplitter

maxOverlapSize, maxSegmentSize, subSplitter, tokenizer

Constructor Summary

Constructors

Constructor	Description
DocumentByWordSplitter(int maxSegmentSizeInChars, int maxOverlapSizeInChars)
DocumentByWordSplitter(int maxSegmentSizeInChars, int maxOverlapSizeInChars, DocumentSplitter subSplitter)
DocumentByWordSplitter(int maxSegmentSizeInTokens, int maxOverlapSizeInTokens, Tokenizer tokenizer)
DocumentByWordSplitter(int maxSegmentSizeInTokens, int maxOverlapSizeInTokens, Tokenizer tokenizer, DocumentSplitter subSplitter)

Method Summary

Modifier and Type	Method	Description
protected DocumentSplitter	defaultSubSplitter()	The default sub-splitter to use when a single segment is too long.
String	joinDelimiter()	Delimiter string to use to re-join the parts.
String[]	split(String text)	Splits the provided text into parts.

Methods inherited from class dev.langchain4j.data.document.splitter.HierarchicalDocumentSplitter

split

Methods inherited from class java.lang.Object

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

Methods inherited from interface dev.langchain4j.data.document.DocumentSplitter

splitAll

Constructor Details

DocumentByWordSplitter

public DocumentByWordSplitter(int maxSegmentSizeInChars,
 int maxOverlapSizeInChars)

DocumentByWordSplitter

public DocumentByWordSplitter(int maxSegmentSizeInChars,
 int maxOverlapSizeInChars,
 DocumentSplitter subSplitter)

DocumentByWordSplitter

public DocumentByWordSplitter(int maxSegmentSizeInTokens,
 int maxOverlapSizeInTokens,
 Tokenizer tokenizer)

DocumentByWordSplitter

public DocumentByWordSplitter(int maxSegmentSizeInTokens,
 int maxOverlapSizeInTokens,
 Tokenizer tokenizer,
 DocumentSplitter subSplitter)

Method Details

split

public String[] split(String text)

Description copied from class: HierarchicalDocumentSplitter

Splits the provided text into parts. Implementation API.

Specified by:

split in class HierarchicalDocumentSplitter

Parameters:

text - The text to be split.

Returns:

An array of parts.

joinDelimiter

public String joinDelimiter()

Description copied from class: HierarchicalDocumentSplitter

Delimiter string to use to re-join the parts.

Specified by:

joinDelimiter in class HierarchicalDocumentSplitter

Returns:

The delimiter.

defaultSubSplitter

protected DocumentSplitter defaultSubSplitter()

Description copied from class: HierarchicalDocumentSplitter

The default sub-splitter to use when a single segment is too long.

Specified by:

defaultSubSplitter in class HierarchicalDocumentSplitter

Returns:

The default sub-splitter.