Saved in:
Bibliographic Details
Main Authors: Huang, Xinhao, Ren, Zhibo, Yu, Yipeng, Zhou, Ying, Chen, Zulong, Wen, Zeyi
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2508.20778
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916927312494592
author Huang, Xinhao
Ren, Zhibo
Yu, Yipeng
Zhou, Ying
Chen, Zulong
Wen, Zeyi
author_facet Huang, Xinhao
Ren, Zhibo
Yu, Yipeng
Zhou, Ying
Chen, Zulong
Wen, Zeyi
contents In long structured document retrieval, existing methods typically fine-tune pre-trained language models (PLMs) using contrastive learning on datasets lacking explicit structural information. This practice suffers from two critical issues: 1) current methods fail to leverage structural features and element-level semantics effectively, and 2) the lack of datasets containing structural metadata. To bridge these gaps, we propose \our, a novel contrastive learning framework. It leverages structure-aware learning to preserve semantic hierarchies and masked element alignment for fine-grained semantic discrimination. Furthermore, we release \dataset, a long structured document retrieval dataset with rich structural annotations. Extensive experiments on both released and industrial datasets across various modern PLMs, along with online A/B testing, demonstrate consistent performance improvements, boosting NDCG@10 from 73.96\% to 77.84\% on BGE-M3. The resources are available at https://github.com/xinhaoH/SEAL.
format Preprint
id arxiv_https___arxiv_org_abs_2508_20778
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle SEAL: Structure and Element Aware Learning to Improve Long Structured Document Retrieval
Huang, Xinhao
Ren, Zhibo
Yu, Yipeng
Zhou, Ying
Chen, Zulong
Wen, Zeyi
Information Retrieval
Machine Learning
In long structured document retrieval, existing methods typically fine-tune pre-trained language models (PLMs) using contrastive learning on datasets lacking explicit structural information. This practice suffers from two critical issues: 1) current methods fail to leverage structural features and element-level semantics effectively, and 2) the lack of datasets containing structural metadata. To bridge these gaps, we propose \our, a novel contrastive learning framework. It leverages structure-aware learning to preserve semantic hierarchies and masked element alignment for fine-grained semantic discrimination. Furthermore, we release \dataset, a long structured document retrieval dataset with rich structural annotations. Extensive experiments on both released and industrial datasets across various modern PLMs, along with online A/B testing, demonstrate consistent performance improvements, boosting NDCG@10 from 73.96\% to 77.84\% on BGE-M3. The resources are available at https://github.com/xinhaoH/SEAL.
title SEAL: Structure and Element Aware Learning to Improve Long Structured Document Retrieval
topic Information Retrieval
Machine Learning
url https://arxiv.org/abs/2508.20778