Metadata preservation during splitting in LangChain - Practice Problems & Coding Challenges

Challenge - 5 Problems

❓ component_behavior

intermediate

What happens to metadata after splitting?

Given a document with metadata, what will be the metadata of the first chunk after splitting using a typical LangChain text splitter?

LangChain

from langchain.schema import Document
from langchain.text_splitter import RecursiveCharacterTextSplitter

doc = Document(page_content="Hello world! This is a test.", metadata={"source": "test_source", "page": 1})
splitter = RecursiveCharacterTextSplitter(chunk_size=12, chunk_overlap=0)
chunks = splitter.split_documents([doc])
first_chunk_metadata = chunks[0].metadata

A {"source": "test_source", "page": 1}

B {}

C {"source": "test_source"}

D None

💡 Hint

Think about whether the splitter copies metadata or discards it.

LangChain's RecursiveCharacterTextSplitter preserves the metadata of the original document in each chunk it creates. So each chunk retains the same metadata dictionary.

❓ state_output

intermediate

How many chunks retain metadata after splitting?

If you split a single document with metadata into 3 chunks using LangChain's RecursiveCharacterTextSplitter, how many chunks will have the original metadata preserved?

LangChain

from langchain.schema import Document
from langchain.text_splitter import RecursiveCharacterTextSplitter

doc = Document(page_content="A long text that will be split into three parts.", metadata={"author": "Alice"})
splitter = RecursiveCharacterTextSplitter(chunk_size=20, chunk_overlap=0)
chunks = splitter.split_documents([doc])
count_with_metadata = sum(1 for c in chunks if c.metadata == {"author": "Alice"})

A 1

B 3

C 0

D Depends on chunk content

💡 Hint

Consider if metadata is copied or removed during splitting.

Each chunk created by the splitter retains the original document's metadata, so all 3 chunks have the metadata.

📝 Syntax

advanced

Which code correctly preserves metadata during splitting?

Which of the following code snippets correctly preserves the metadata of documents after splitting using LangChain's RecursiveCharacterTextSplitter?

A chunks = splitter.split_documents([doc])

B chunks = splitter.split_text(doc.page_content)

C chunks = [Document(page_content=chunk) for chunk in splitter.split_text(doc.page_content)]

D chunks = [Document(page_content=chunk, metadata=doc.metadata) for chunk in splitter.split_text(doc.page_content)]

💡 Hint

Look for the method that returns Document objects with metadata.

split_documents returns Document objects preserving metadata. split_text returns strings without metadata, so manual wrapping is needed to preserve metadata.

🔧 Debug

advanced

Why is metadata lost after splitting?

A developer uses splitter.split_text(doc.page_content) and then creates new Document objects from the chunks but forgets to assign metadata. What is the result for metadata in the chunks?

LangChain

chunks_text = splitter.split_text(doc.page_content)
chunks = [Document(page_content=chunk) for chunk in chunks_text]
metadata_list = [chunk.metadata for chunk in chunks]

A None # metadata attribute missing

B [{"source": "original"}, ...] # original metadata preserved

C [{}, {}, ...] # empty metadata dictionaries

D Error: Document constructor requires metadata

💡 Hint

Check what happens if metadata is not passed when creating Document objects.

If metadata is not passed, Document objects have empty metadata dictionaries by default.

🧠 Conceptual

expert

How to ensure custom metadata updates during splitting?

You want to split documents but also add a new metadata field 'chunk_index' to each chunk indicating its order. Which approach correctly achieves this while preserving original metadata?

LangChain

from langchain.schema import Document
from langchain.text_splitter import RecursiveCharacterTextSplitter

doc = Document(page_content="Some long text here.", metadata={"source": "book"})
splitter = RecursiveCharacterTextSplitter(chunk_size=10, chunk_overlap=0)
chunks = splitter.split_documents([doc])
# Add chunk_index metadata here

A chunks = splitter.split_text(doc.page_content); for i, text in enumerate(chunks): chunks[i] = Document(page_content=text, metadata={'chunk_index': i})

B for i, c in enumerate(chunks): c.metadata['chunk_index'] = i

C chunks = [Document(page_content=c.page_content, metadata={'chunk_index': i}) for i, c in enumerate(chunks)]

D chunks = [Document(page_content=c.page_content, metadata={**c.metadata, 'chunk_index': i}) for i, c in enumerate(chunks)]

💡 Hint

Think about preserving original metadata and adding new keys immutably.

Option D creates new Document objects merging original metadata with the new 'chunk_index' key, preserving all original metadata fields.