Toward General Semantic Chunking: A Discriminative Framework for Ultra-Long Documents

TL;DR

This paper introduces a discriminative model based on Qwen3-0.6B for ultra-long document segmentation, supporting single-pass inputs of up to 13k tokens, with improved accuracy and efficiency over existing generative models.

Contribution

The paper presents a novel discriminative segmentation framework that handles ultra-long documents efficiently and accurately, with a new vector fusion method for downstream retrieval.

Findings

Outperforms generative models in macro-averaged F1 score.

Achieves two orders of magnitude faster inference.

Supports single-pass processing of up to 13,000 tokens.

Abstract

Long-document topic segmentation plays an important role in information retrieval and document understanding, yet existing methods still show clear shortcomings in ultra-long text settings. Traditional discriminative models are constrained by fixed windows and cannot model document-level semantics; generative large language models can output paragraph boundaries, but inference is expensive and long inputs are difficult to support. To address these issues, we propose a discriminative segmentation model based on Qwen3-0.6B. On top of the backbone network, we add a cross-window context fusion layer and a boundary classification head, and combine them with an overlapping sliding-window strategy. Our model supports single-pass inputs of up to 13k tokens and can be extended to ultra-long documents for paragraph boundary detection. To further enhance downstream retrieval efficiency, we derive a vector fusion method with scalar correction, which compresses the representation of ultra-long segments into a single vector without semantic loss. Experiments on the Wikipedia long-document topic segmentation dataset WIKI-727K show that, compared with three generative models based on Qwen2-0.5B released by Jina, our method achieves a better macro-averaged F1 and delivers two orders of magnitude faster inference, substantially improving the practicality and scalability of long-document processing.