Language model-based stakeholder-specific policy texts classification of GBA construction industry cross-regional collaboration

Purpose Construction industry-related regional policies are direct factors affecting cross-regional collaboration. Different stakeholders have varying demands for policy knowledge. This study aims to address the inefficiency of traditional manual management of policy knowledge among diverse stakeholders in the construction industry within the Guangdong–Hong Kong–Macao Greater Bay Area (GBA) urban agglomeration. Design/methodology/approach To optimize the inefficiency problem of traditional manual management of specific policy knowledge for stakeholders, given the advantages of large-scale pre-trained language models in text understanding, this study proposes the LFGNBert model, which optimizes traditional Bidirectional Encoder Representations from Transformers (BERT) models by fusing multi-attention mechanisms and graph networks to achieve intelligent classification of construction industry cross-regional collaboration policy texts. Findings Experimental results show that the LFGNBert model achieves weighted precision, recall, and F1 scores of 0.9723, 0.9717, and 0.9716, representing approximately a 5% improvement over traditional BERT models. Based on model inference results, a further constructed Large Language Model (LLM)-based intelligent consulting system is evaluated under a retrieval-augmented generation (RAG) setting, yielding a 5%–9% improvement in answer correctness across four question formats over retrieval from unprocessed documents, offering a validated practical reference for knowledge management of GBA construction industry cross-regional collaboration. Originality/value In addition to its topical novelty, this paper also makes methodological contributions by integrating multi-attention mechanisms and a graph network into a BERT-based model for classifying text clauses. A domain-specific LLM agent based on the classified clauses through RAG also effectively improves the efficiency of developing a domain-specific LLM agent. The proposed framework provides a reference for similar research on policy-clause knowledge management and its specific LLM development in different domains.

Authors

Institutions

Publication Details

Journal
Engineering Construction & Architectural Management
Published
2026-11-16
DOI
https://doi.org/10.1108/ecam-10-2025-1620
Primary Topic
Topic Modeling
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Language model-based stakeholder-specific policy texts classification of GBA construction industry cross-regional collaboration

Qiping Shen, Dawei He, Binwei Gao, Ruiyang Ma et al.
2 links
Engineering Construction & Architectural Management
Topic Modeling
article

Language model-based stakeholder-specific policy texts classification of GBA construction industry cross-regional collaboration

Qiping Shen, Dawei He, Binwei Gao, Ruiyang Ma, Jieru Miao, Liang Xiao, Rubing Wang
article en

Abstract

Purpose Construction industry-related regional policies are direct factors affecting cross-regional collaboration. Different stakeholders have varying demands for policy knowledge. This study aims to address the inefficiency of traditional manual management of policy knowledge among diverse stakeholders in the construction industry within the Guangdong–Hong Kong–Macao Greater Bay Area (GBA) urban agglomeration. Design/methodology/approach To optimize the inefficiency problem of traditional manual management of specific policy knowledge for stakeholders, given the advantages of large-scale pre-trained language models in text understanding, this study proposes the LFGNBert model, which optimizes traditional Bidirectional Encoder Representations from Transformers (BERT) models by fusing multi-attention mechanisms and graph networks to achieve intelligent classification of construction industry cross-regional collaboration policy texts. Findings Experimental results show that the LFGNBert model achieves weighted precision, recall, and F1 scores of 0.9723, 0.9717, and 0.9716, representing approximately a 5% improvement over traditional BERT models. Based on model inference results, a further constructed Large Language Model (LLM)-based intelligent consulting system is evaluated under a retrieval-augmented generation (RAG) setting, yielding a 5%–9% improvement in answer correctness across four question formats over retrieval from unprocessed documents, offering a validated practical reference for knowledge management of GBA construction industry cross-regional collaboration. Originality/value In addition to its topical novelty, this paper also makes methodological contributions by integrating multi-attention mechanisms and a graph network into a BERT-based model for classifying text clauses. A domain-specific LLM agent based on the classified clauses through RAG also effectively improves the efficiency of developing a domain-specific LLM agent. The proposed framework provides a reference for similar research on policy-clause knowledge management and its specific LLM development in different domains.

Engineering Construction & Architectural Management
Tongji University (CN), Hong Kong Polytechnic University (HK), Chongqing University (CN), Hefei University of Technology (CN), Nanjing University (CN)
Industry, innovation and infrastructure
Topic Modeling
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon