Skip to content

GH-2123: Add chunkOverlap support to TokenTextSplitter #4054

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
Open

GH-2123: Add chunkOverlap support to TokenTextSplitter #4054

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
115 changes: 77 additions & 38 deletions ...-commons/src/main/java/org/springframework/ai/transformer/splitter/TokenTextSplitter.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -33,11 +33,14 @@
* @author Raphael Yu
* @author Christian Tzolov
* @author Ricken Bazolo
* @author Seunghwan Jung
*/
public class TokenTextSplitter extends TextSplitter {

private static final int DEFAULT_CHUNK_SIZE = 800;

private static final int DEFAULT_CHUNK_OVERLAP = 50;

private static final int MIN_CHUNK_SIZE_CHARS = 350;

private static final int MIN_CHUNK_LENGTH_TO_EMBED = 5;
Expand All @@ -46,13 +49,17 @@ public class TokenTextSplitter extends TextSplitter {

private static final boolean KEEP_SEPARATOR = true;


private final EncodingRegistry registry = Encodings.newLazyEncodingRegistry();

private final Encoding encoding = this.registry.getEncoding(EncodingType.CL100K_BASE);

// The target size of each text chunk in tokens
private final int chunkSize;

// The overlap size of each text chunk in tokens
private final int chunkOverlap;

// The minimum size of each text chunk in characters
private final int minChunkSizeChars;

Expand All @@ -65,16 +72,18 @@ public class TokenTextSplitter extends TextSplitter {
private final boolean keepSeparator;

public TokenTextSplitter() {
this(DEFAULT_CHUNK_SIZE, MIN_CHUNK_SIZE_CHARS, MIN_CHUNK_LENGTH_TO_EMBED, MAX_NUM_CHUNKS, KEEP_SEPARATOR);
this(DEFAULT_CHUNK_SIZE, DEFAULT_CHUNK_OVERLAP, MIN_CHUNK_SIZE_CHARS, MIN_CHUNK_LENGTH_TO_EMBED, MAX_NUM_CHUNKS, KEEP_SEPARATOR);
}

public TokenTextSplitter(boolean keepSeparator) {
this(DEFAULT_CHUNK_SIZE, MIN_CHUNK_SIZE_CHARS, MIN_CHUNK_LENGTH_TO_EMBED, MAX_NUM_CHUNKS, keepSeparator);
this(DEFAULT_CHUNK_SIZE, DEFAULT_CHUNK_OVERLAP, MIN_CHUNK_SIZE_CHARS, MIN_CHUNK_LENGTH_TO_EMBED, MAX_NUM_CHUNKS, keepSeparator);
}

public TokenTextSplitter(int chunkSize, int minChunkSizeChars, int minChunkLengthToEmbed, int maxNumChunks,
boolean keepSeparator) {
public TokenTextSplitter(int chunkSize, int chunkOverlap, int minChunkSizeChars, int minChunkLengthToEmbed, int maxNumChunks,
boolean keepSeparator) {
Assert.isTrue(chunkOverlap < chunkSize, "chunk overlap must be less than chunk size");
this.chunkSize = chunkSize;
this.chunkOverlap = chunkOverlap;
this.minChunkSizeChars = minChunkSizeChars;
this.minChunkLengthToEmbed = minChunkLengthToEmbed;
this.maxNumChunks = maxNumChunks;
Expand All @@ -87,57 +96,80 @@ public static Builder builder() {

@Override
protected List<String> splitText(String text) {
return doSplit(text, this.chunkSize);
return doSplit(text, this.chunkSize, this.chunkOverlap);
}

protected List<String> doSplit(String text, int chunkSize) {
protected List<String> doSplit(String text, int chunkSize, int chunkOverlap) {
if (text == null || text.trim().isEmpty()) {
return new ArrayList<>();
}

List<Integer> tokens = getEncodedTokens(text);
List<String> chunks = new ArrayList<>();
int num_chunks = 0;
while (!tokens.isEmpty() && num_chunks < this.maxNumChunks) {
List<Integer> chunk = tokens.subList(0, Math.min(chunkSize, tokens.size()));
String chunkText = decodeTokens(chunk);

// Skip the chunk if it is empty or whitespace
if (chunkText.trim().isEmpty()) {
tokens = tokens.subList(chunk.size(), tokens.size());
continue;
}
// If text is smaller than chunk size, return as a single chunk
if (tokens.size() <= chunkSize) {
String processedText = this.keepSeparator ? text.trim() :
text.replace(System.lineSeparator(), " ").trim();

// Find the last period or punctuation mark in the chunk
int lastPunctuation = Math.max(chunkText.lastIndexOf('.'), Math.max(chunkText.lastIndexOf('?'),
Math.max(chunkText.lastIndexOf('!'), chunkText.lastIndexOf('\n'))));

if (lastPunctuation != -1 && lastPunctuation > this.minChunkSizeChars) {
// Truncate the chunk text at the punctuation mark
chunkText = chunkText.substring(0, lastPunctuation + 1);
if (processedText.length() > this.minChunkLengthToEmbed) {
return List.of(processedText);
}
return new ArrayList<>();
}
List<String> chunks = new ArrayList<>();

String chunkTextToAppend = (this.keepSeparator) ? chunkText.trim()
: chunkText.replace(System.lineSeparator(), " ").trim();
if (chunkTextToAppend.length() > this.minChunkLengthToEmbed) {
chunks.add(chunkTextToAppend);
int position = 0;
int num_chunks = 0;
while (position < tokens.size() && num_chunks < this.maxNumChunks) {
int chunkEnd = Math.min(position + chunkSize, tokens.size());

// Extract tokens for this chunk
List<Integer> chunkTokens = tokens.subList(position, chunkEnd);
String chunkText = decodeTokens(chunkTokens);

// Apply sentence boundary optimization
String finalChunkText = optimizeChunkBoundary(chunkText);
int finalChunkTokenCount = getEncodedTokens(finalChunkText).size();
int advance = Math.max(1, finalChunkTokenCount - chunkOverlap);
position += advance;

// Format according to keepSeparator setting
String formattedChunk = this.keepSeparator ? finalChunkText.trim() :
finalChunkText.replace(System.lineSeparator(), " ").trim();

// Add chunk if it meets minimum length
if (formattedChunk.length() > this.minChunkLengthToEmbed) {
chunks.add(formattedChunk);
num_chunks++;
}
}

// Remove the tokens corresponding to the chunk text from the remaining tokens
tokens = tokens.subList(getEncodedTokens(chunkText).size(), tokens.size());
return chunks;
}

num_chunks++;
private String optimizeChunkBoundary(String chunkText) {
if (chunkText.length() <= this.minChunkSizeChars) {
return chunkText;
}

// Handle the remaining tokens
if (!tokens.isEmpty()) {
String remaining_text = decodeTokens(tokens).replace(System.lineSeparator(), " ").trim();
if (remaining_text.length() > this.minChunkLengthToEmbed) {
chunks.add(remaining_text);
// Look for sentence endings: . ! ? \n
int bestCutPoint = -1;

// Check in reverse order to find the last sentence ending
for (int i = chunkText.length() - 1; i >= this.minChunkSizeChars; i--) {
char c = chunkText.charAt(i);
if (c == '.' || c == '!' || c == '?' || c == '\n') {
bestCutPoint = i + 1; // Include the punctuation
break;
}
}

return chunks;
// If we found a good cut point, use it
if (bestCutPoint > 0) {
return chunkText.substring(0, bestCutPoint);
}

// Otherwise return the original chunk
return chunkText;
}

private List<Integer> getEncodedTokens(String text) {
Expand All @@ -156,6 +188,8 @@ public static final class Builder {

private int chunkSize = DEFAULT_CHUNK_SIZE;

private int chunkOverlap = DEFAULT_CHUNK_OVERLAP;

private int minChunkSizeChars = MIN_CHUNK_SIZE_CHARS;

private int minChunkLengthToEmbed = MIN_CHUNK_LENGTH_TO_EMBED;
Expand All @@ -172,6 +206,11 @@ public Builder withChunkSize(int chunkSize) {
return this;
}

public Builder withChunkOverlap(int chunkOverlap) {
this.chunkOverlap = chunkOverlap;
return this;
}

public Builder withMinChunkSizeChars(int minChunkSizeChars) {
this.minChunkSizeChars = minChunkSizeChars;
return this;
Expand All @@ -193,7 +232,7 @@ public Builder withKeepSeparator(boolean keepSeparator) {
}

public TokenTextSplitter build() {
return new TokenTextSplitter(this.chunkSize, this.minChunkSizeChars, this.minChunkLengthToEmbed,
return new TokenTextSplitter(this.chunkSize, this.chunkOverlap, this.minChunkSizeChars, this.minChunkLengthToEmbed,
this.maxNumChunks, this.keepSeparator);
}

Expand Down
Loading