RecursiveCharacterTextSplitter in LangChain

Introduction

The RecursiveCharacterTextSplitter helps break long text into smaller pieces. It does this by splitting text step-by-step using different characters, making sure the pieces are easy to handle.

When you have a long document and want to split it into smaller parts for easier processing.

When you want to keep sentences or paragraphs intact while splitting text.

When you need to prepare text for language models that have input size limits.

When you want to split text by paragraphs first, then by sentences, and finally by words if needed.

When you want to avoid cutting text in the middle of words or sentences.

Syntax

LangChain

from langchain.text_splitter import RecursiveCharacterTextSplitter

text_splitter = RecursiveCharacterTextSplitter(
    chunk_size=1000,
    chunk_overlap=200,
    separators=["\n\n", "\n", ".", "!", "?", ",", " ", ""]
)

chunks = text_splitter.split_text(long_text)

The chunk_size sets the maximum size of each text piece.

The chunk_overlap controls how much text overlaps between chunks to keep context.

Examples

Splits text into chunks of 500 characters with 50 characters overlapping.

LangChain

text_splitter = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=50)
chunks = text_splitter.split_text(long_text)

Splits text by new lines, then sentences, then spaces without overlap.

LangChain

text_splitter = RecursiveCharacterTextSplitter(
    chunk_size=300,
    chunk_overlap=0,
    separators=["\n", ".", " "]
)
chunks = text_splitter.split_text(long_text)

Splits text by paragraphs first, then by sentences, with some overlap.

LangChain

text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=100, separators=["\n\n", "."])
chunks = text_splitter.split_text(long_text)

Sample Program

This program splits a text with three paragraphs into smaller chunks of max 50 characters. It overlaps 10 characters between chunks to keep context. It tries to split by paragraphs first, then sentences, then words.

LangChain

from langchain.text_splitter import RecursiveCharacterTextSplitter

long_text = (
    "This is the first paragraph. It has two sentences.\n\n"
    "Here is the second paragraph! It also has sentences? Yes, it does.\n\n"
    "Finally, the third paragraph is here."
)

text_splitter = RecursiveCharacterTextSplitter(
    chunk_size=50,
    chunk_overlap=10,
    separators=["\n\n", ".", "!", "?", ",", " ", ""]
)

chunks = text_splitter.split_text(long_text)

print("Number of chunks:", len(chunks))
for i, chunk in enumerate(chunks, 1):
    print(f"Chunk {i}:", repr(chunk))

Important Notes

The splitting tries separators in order, so order matters for best results.

Time complexity depends on text length and number of separators but is generally efficient for normal documents.

Common mistake: setting chunk_size too small can create many tiny chunks.

Summary

RecursiveCharacterTextSplitter breaks text into manageable chunks using multiple separators.

It keeps context by overlapping parts of chunks.

Useful for preparing text for language models or any tool with input size limits.

Practice

1. What is the main purpose of the RecursiveCharacterTextSplitter in langchain?

easy

A. To split long text into smaller chunks using multiple separators while keeping context.

B. To merge multiple texts into one large string.

C. To translate text into different languages.

D. To encrypt text for secure storage.

Solution

1. Step 1: Understand the function of RecursiveCharacterTextSplitter

   This tool is designed to break down long text into smaller pieces using various separators like paragraphs, sentences, or characters.

2. Step 2: Recognize the benefit of overlapping chunks

   It keeps some overlap between chunks to maintain context, which helps language models understand the text better.

3. Final Answer:

   To split long text into smaller chunks using multiple separators while keeping context. -> Option A

4. Quick Check:

   Splitting text with context = A [OK]

Hint: Remember: It splits text into chunks with overlap [OK]

Common Mistakes:

• Thinking it merges text instead of splitting
• Confusing it with translation or encryption tools
• Ignoring the overlap feature

2. Which of the following is the correct way to create a RecursiveCharacterTextSplitter with a chunk size of 100 and chunk overlap of 20?

easy

A. RecursiveCharacterTextSplitter(chunkSize=100, chunkOverlap=20)

B. RecursiveCharacterTextSplitter(size=100, overlap=20)

C. RecursiveCharacterTextSplitter(chunk=100, overlap=20)

D. RecursiveCharacterTextSplitter(chunk_size=100, chunk_overlap=20)

Solution

1. Step 1: Recall the parameter names for RecursiveCharacterTextSplitter

   The correct parameters are chunk_size and chunk_overlap to set chunk length and overlap.

2. Step 2: Match the correct syntax

   RecursiveCharacterTextSplitter(chunk_size=100, chunk_overlap=20) uses the exact parameter names and values correctly.

3. Final Answer:

   RecursiveCharacterTextSplitter(chunk_size=100, chunk_overlap=20) -> Option D

4. Quick Check:

   Correct parameter names = D [OK]

Hint: Use exact parameter names: chunk_size and chunk_overlap [OK]

Common Mistakes:

• Using incorrect parameter names like size or chunkSize
• Mixing camelCase and snake_case
• Omitting required parameters

3. Given the following code snippet, what will be the length of the first chunk if the text is "Hello world! This is a test." and chunk_size=12, chunk_overlap=4?

splitter = RecursiveCharacterTextSplitter(chunk_size=12, chunk_overlap=4)
chunks = splitter.split_text("Hello world! This is a test.")
print(len(chunks[0]))

medium

A. 12

B. 16

C. 8

D. 20

Solution

1. Step 1: Understand chunk size and overlap effect

   The chunk size is 12, so each chunk will be at most 12 characters long. Overlap affects next chunks, not the first chunk length.

2. Step 2: Determine first chunk length

   The first chunk will be exactly 12 characters long, as it starts from the beginning and takes up to chunk_size.

3. Final Answer:

   12 -> Option A

4. Quick Check:

   First chunk length = chunk_size = 12 [OK]

Hint: First chunk length equals chunk_size, overlap affects next chunks [OK]

Common Mistakes:

• Adding overlap to first chunk length
• Assuming first chunk is shorter than chunk_size
• Confusing chunk size with total text length

4. What is the error in the following code snippet using RecursiveCharacterTextSplitter?

splitter = RecursiveCharacterTextSplitter(chunk_size=50, chunk_overlap=60)
chunks = splitter.split_text("Some long text here...")

medium

A. Missing import statement for RecursiveCharacterTextSplitter

B. split_text method does not exist

C. chunk_overlap cannot be larger than chunk_size

D. chunk_size must be a string, not an integer

Solution

1. Step 1: Check parameter constraints

   chunk_overlap must be less than or equal to chunk_size to avoid logical errors in splitting.

2. Step 2: Identify the error in given values

   chunk_overlap=60 is greater than chunk_size=50, which is invalid and will cause an error.

3. Final Answer:

   chunk_overlap cannot be larger than chunk_size -> Option C

4. Quick Check:

   Overlap ≤ chunk size rule = B [OK]

Hint: Overlap must never exceed chunk size [OK]

Common Mistakes:

• Ignoring parameter value relationships
• Assuming method names are incorrect without checking
• Confusing data types for parameters

5. You want to split a large document into chunks of 200 characters with 50 characters overlapping, prioritizing splitting first by paragraphs, then sentences, then words. Which setup of RecursiveCharacterTextSplitter correctly achieves this?

hard

A. RecursiveCharacterTextSplitter(chunk_size=200, chunk_overlap=50, separators=[" ", ". ", "\n\n"])

B. RecursiveCharacterTextSplitter(chunk_size=200, chunk_overlap=50, separators=["\n\n", ". ", " "])

C. RecursiveCharacterTextSplitter(chunk_size=200, chunk_overlap=50, separators=[". ", "\n\n", " "])

D. RecursiveCharacterTextSplitter(chunk_size=200, chunk_overlap=50, separators=[" ", "\n\n", ". "])

Solution

1. Step 1: Understand separator priority order

   The splitter tries separators in order; to prioritize paragraphs, then sentences, then words, the list must start with paragraph separator "\n\n", then sentence ". ", then space " ".

2. Step 2: Match the correct separators list

   RecursiveCharacterTextSplitter(chunk_size=200, chunk_overlap=50, separators=["\n\n", ". ", " "]) correctly lists separators in the desired priority order.

3. Final Answer:

   RecursiveCharacterTextSplitter(chunk_size=200, chunk_overlap=50, separators=["\n\n", ". ", " "]) -> Option B

4. Quick Check:

   Separator order matches priority = A [OK]

Hint: List separators from largest to smallest text unit [OK]

Common Mistakes:

• Reversing separator order
• Using wrong separator strings
• Ignoring overlap or chunk size settings