Text splitters in Prompt Engineering / GenAI - Model Metrics & Evaluation

Metrics & Evaluation - Text splitters

Which metric matters for Text Splitters and WHY

Text splitters break long text into smaller parts. The key metric is chunk quality, which means how well the text is split without losing meaning or context. We want splits that keep sentences whole and keep related ideas together. This helps models understand text better.

Confusion matrix or equivalent visualization

Example of text splitter evaluation:

Original text length: 1000 characters
Split into chunks:
  Chunk 1: 300 chars
  Chunk 2: 350 chars
  Chunk 3: 350 chars

Evaluation:
- Overlap between chunks: 20 chars (good for context)
- Sentence breaks inside chunks: 0 (ideal)
- Meaning preserved: 95% (human score)

No confusion matrix applies directly, but chunk overlap and sentence boundary accuracy are key.
    

Precision vs Recall tradeoff with examples

For text splitters, think of precision as how often splits happen at the right place (not breaking sentences). Recall is how many important split points are found (like paragraph ends).

High precision, low recall: Splits only at perfect points but misses some natural breaks. Result: chunks may be too big.

High recall, low precision: Splits at many points, including bad ones. Result: chunks may be too small or cut sentences.

Good text splitters balance both to keep chunks meaningful and manageable.

What "good" vs "bad" metric values look like for Text Splitters

• Good: Sentence boundary accuracy > 95%, chunk overlap 10-30 chars, chunk size consistent, meaning preserved > 90%
• Bad: Sentence breaks inside chunks > 20%, chunk overlap 0 or very large (losing context), chunks too uneven or too small, meaning preserved < 70%

Common pitfalls in Text Splitter metrics

• Ignoring sentence boundaries causes chunks that confuse models.
• Too little overlap loses context between chunks.
• Too much overlap wastes space and slows processing.
• Evaluating only chunk size without meaning can mislead.
• Using only automatic metrics without human checks misses quality issues.

Self-check question

Your text splitter creates chunks with 98% sentence boundary accuracy but only 10 characters overlap between chunks. Is this good?

Answer: It is mostly good because sentence boundaries are respected, which keeps meaning clear. However, 10 characters overlap might be too small to keep enough context between chunks. Increasing overlap slightly can help models understand connections better.

Key Result

For text splitters, high sentence boundary accuracy and balanced chunk overlap are key to preserving meaning and context.

Practice

1. What is the main purpose of a text splitter in AI applications?

easy

A. To translate text into different languages

B. To generate new text from a prompt

C. To break long text into smaller, manageable pieces

D. To summarize text into a single sentence

Solution

1. Step 1: Understand the role of text splitters

   Text splitters are designed to divide long text into smaller parts for easier processing.

2. Step 2: Compare options to the definition

   Only To break long text into smaller, manageable pieces describes breaking text into smaller pieces, which matches the purpose of text splitters.

3. Final Answer:

   To break long text into smaller, manageable pieces -> Option C

4. Quick Check:

   Text splitter purpose = break text [OK]

Hint: Text splitters cut text into chunks for easier handling [OK]

Common Mistakes:

• Confusing splitting with translation
• Thinking splitters summarize text
• Assuming splitters generate new text

2. Which of the following is the correct way to set chunk size and overlap in a text splitter?

easy

A. chunk_size='100', overlap=20

B. chunkSize=100, overlap=20

C. chunk_size=100, overlap=twenty

D. chunk_size=100, overlap=20

Solution

1. Step 1: Identify correct parameter names and types

   Parameters should be named with underscores and numeric values for size and overlap.

2. Step 2: Check each option for syntax and type correctness

   chunk_size=100, overlap=20 uses correct parameter names and numeric values; others have wrong names or types.

3. Final Answer:

   chunk_size=100, overlap=20 -> Option D

4. Quick Check:

   Correct param names and numeric values = chunk_size=100, overlap=20 [OK]

Hint: Use underscores and numbers for chunk size and overlap [OK]

Common Mistakes:

• Using camelCase instead of snake_case
• Passing string instead of number for overlap
• Misspelling parameter names

3. Given the following Python code using a text splitter:

text = "Hello world! This is a test of text splitting."
chunk_size = 12
overlap = 4
splitter = TextSplitter(chunk_size=chunk_size, overlap=overlap)
chunks = splitter.split(text)
print(chunks)

What is the expected output?

medium

A. ["Hello world!", "world! This is", "This is a test", "a test of text", "of text splitting."]

B. ["Hello world! This", "This is a test of", "test of text splitting."]

C. ["Hello world! This is a test of text splitting."]

D. ["Hello", "world!", "This", "is", "a", "test", "of", "text", "splitting."]

Solution

1. Step 1: Understand chunk size and overlap effect

   Chunk size 12 means each piece has up to 12 characters; overlap 4 means next chunk starts 4 characters before previous ends.

2. Step 2: Apply splitting logic to the text

   Chunks are: "Hello world!" (12 chars), then start 8 chars in (12-4=8) at "world! This is", and so on, producing the listed chunks in ["Hello world!", "world! This is", "This is a test", "a test of text", "of text splitting."].

3. Final Answer:

   ["Hello world!", "world! This is", "This is a test", "a test of text", "of text splitting."] -> Option A

4. Quick Check:

   Chunk size 12 + overlap 4 = overlapping chunks [OK]

Hint: Chunk size limits length; overlap repeats last part [OK]

Common Mistakes:

• Ignoring overlap and making chunks non-overlapping
• Using wrong chunk sizes
• Returning entire text as one chunk

4. Consider this code snippet that tries to split text but raises an error:

text = "Sample text for splitting."
splitter = TextSplitter(chunk_size='10', overlap=3)
chunks = splitter.split(text)

What is the most likely cause of the error?

medium

A. chunk_size should be an integer, not a string

B. overlap cannot be less than 5

C. TextSplitter requires a minimum chunk_size of 20

D. The text variable is too short to split

Solution

1. Step 1: Check parameter types

   chunk_size is given as a string '10' instead of an integer 10, which causes a type error.

2. Step 2: Validate other options

   Overlap 3 is valid; no minimum chunk size of 20 is required; text length is sufficient.

3. Final Answer:

   chunk_size should be an integer, not a string -> Option A

4. Quick Check:

   Parameter type mismatch = chunk_size should be an integer, not a string [OK]

Hint: Use numbers, not strings, for chunk size and overlap [OK]

Common Mistakes:

• Passing chunk_size as string
• Assuming overlap minimum is 5
• Thinking text length causes error

5. You have a very long document and want to split it for an AI model that can only process 500 tokens at a time. You want some context overlap to keep meaning. Which approach best balances chunk size and overlap?

hard

A. Set chunk_size to 600 tokens and overlap to 0 tokens

B. Set chunk_size to 500 tokens and overlap to 100 tokens

C. Set chunk_size to 400 tokens and overlap to 200 tokens

D. Set chunk_size to 100 tokens and overlap to 50 tokens

Solution

1. Step 1: Understand model token limit and overlap purpose

   The model can process 500 tokens max; overlap adds repeated context to help understanding.

2. Step 2: Evaluate options for chunk size and overlap

   Set chunk_size to 500 tokens and overlap to 100 tokens uses chunk size 500 (max allowed) and overlap 100 (reasonable context). Others exceed limit or have too small chunks.

3. Final Answer:

   Set chunk_size to 500 tokens and overlap to 100 tokens -> Option B

4. Quick Check:

   Chunk size ≤ 500 with overlap for context = Set chunk_size to 500 tokens and overlap to 100 tokens [OK]

Hint: Keep chunk size at max limit, add moderate overlap [OK]

Common Mistakes:

• Exceeding model token limit
• Setting overlap too large or zero
• Using very small chunk sizes unnecessarily