Researchers propose multi-split boundary decision to lower LLM document segmentation cost
Scanned mail, PDFs and bundled attachments often arrive as continuous page streams that need to be broken into separate documents before classification, extraction or routing. The paper introduces Multi‑Split Boundary Decision (MSBD), a zero‑shot approach that lets a large language model predict several document boundaries inside a single page window, cutting the number of inference calls…
Key points
- MSBD predicts multiple document boundaries per LLM call, reducing inference requests
- Experiments show MSBD keeps high segmentation accuracy while boosting efficiency within a model‑ and corpus‑dependent window range
- Accuracy declines sharply when window sizes grow too large, with over‑ and under‑segmentation varying by model
The authors test MSBD on a variety of language models, document collections, input types and window sizes. Results show that, within a model‑ and corpus‑specific window range, MSBD retains strong segmentation accuracy while markedly improving inference efficiency. Larger windows, however, cause a rapid drop in performance, with distinct patterns of over‑ and under‑segmentation that differ across models. The study demonstrates that choosing an appropriate window size lets practitioners achieve a favorable accuracy‑efficiency trade‑off for zero‑shot page‑stream segmentation.
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Research
All →- Study finds LLM judge consensus overstates evidence because judges share errors · 1 src
- Study shows LLM deliberation improves accuracy across domains · 1 src
- Study finds Gemini 3.1 Flash-Lite deficit recurs on fresh data under historical configuration · 1 src
- Researchers propose Goal-driven variant categorization using LLM · 1 src
- Study shows AI agents select fewer papers when they see others' choices · 1 src
Comments
via GitHub Discussions