chunkBench(data): A Systematic Investigation of Document Chunking Strategies and Embedding Sensitivity
Published: 20 October 2026| Version 1 | DOI: 10.17632/c8w2gcm83y.1
Contributor:
Description
The data provides insights on how different chunking strategies affect retrieval quality. By comparing LLM-generated evaluation scores across strategies, the dataset highlights: - The relative performance of each chunking approach. - The trade-off between larger chunks (which may preserve more context) and smaller chunks (which may improve retrieval precision). - The influence of chunk overlap and segmentation methods on retrieval effectiveness. - Variations in the quality and relevance of retrieved content across different chunking configurations.
Files
Institutions
- University of CanberraAustralian Capital Territory, Canberra
Categories
Information Retrieval Evaluation, Generative Artificial Intelligence
