|
|
| |
|
|
| 本文已被：浏览 119次 下载 360次 |
码上扫一扫！ |
| 资源型地区转型的研究演进趋势与关键条件识别
——基于大语言模型的文本挖掘 |
|
李博1, 周慧敏2, 范亚娟3, 余建辉4
|
|
1.天津理工大学管理学院，教授;2.天津理工大学管理学院，博士研究生;3.天津理工大学管理学院，硕士研究生;4.（通讯作者）：中国科学院地理科学与资源研究所，副研究员，yujh@igsnrr.ac.cn
|
|
| 摘要: |
| 资源型地区转型是中国区域高质量
发展的战略任务，其驱动机制的复杂性亟待
系统识别。本研究利用大语言模型结构化抽
取与多算法机器学习协同识别对1 240 篇中
文文献进行分析。研究结果表明现有文献关
注焦点高度集中于政策环境和产业结构两大
维度，反映了政府主导的顶层设计地位。研
究趋势已从政策驱动的绝对集中，转向技术
环境关注度的提升及数字化、能源革命等前
沿议题的聚焦。五种典型模式的关键驱动条
件存在显著差异，即绿色转型、文化旅游转
型、产业多元化转型、区域一体化转型四种
模式均以政策环境为核心但驱动侧重不同；
而平台依托型转型的关键驱动条件为技术环
境平台建设和市场环境活力激发。研究有效
弥补了传统方法难以进行字段级深层挖掘的
不足，为转型模式的理论研究与政策制定提
供了科学参考。 |
| 关键词: 资源型地区 转型模式 研究演
进 驱动条件 大语言模型 |
| DOI：10.13791/j.cnki.hsfwest.20251127004 |
| 分类号: |
| 基金项目:国家自然科学基金项目（42171290）；教育部人文社会科学研究规划基金项目（25YJAZH077） |
|
| Evolutionary trends and key condition identification in resource-based areatransformation: A large language model based on text mining approach |
|
LI Bo,ZHOU Huimin,FAN Yajuan,YU Jianhui
|
| Abstract: |
| The transformation of resource-based areas represents a critical strategic task in China’s
pursuit of high-quality regional development, particularly under the combined pressures of resource
depletion, ecological constraints, and structural economic adjustment. Historically dependent on the
extraction and primary processing of natural resources such as coal, minerals, and petroleum, these
areas have long played a foundational role in national industrialization. However, persistent path
dependence and the well-documented “resource curse” have increasingly exposed them to economic
stagnation, environmental degradation, and social imbalance. In this context, a systematic
understanding of how academic research on resource-based area transformation has evolved, as well
as a clear identification of the key driving conditions underlying different transformation pathways, is
essential for advancing theory and informing policy practice.Previous studies on resource-based area
transformation can generally be grouped into three methodological streams: qualitative case-based
analyses, quantitative econometric research, and bibliometric or knowledge-mapping studies. While
each has generated valuable insights, their limitations are also evident. Qualitative case studies offer
rich contextual understanding but often lack generalizability. Quantitative econometric approaches
typically focus on selected explanatory variables and outcomes, making it difficult to capture the
internal mechanisms and multidimensional interactions involved in transformation processes.
Bibliometric analyses, relying on tools such as citation networks and keyword co-occurrence, are
effective in identifying research hotspots and collaboration patterns, yet they remain constrained to
surface-level information and are insufficient for uncovering mechanism-oriented semantic structures
and their evolution over time. As a result, a methodological gap persists in conducting large-scale,
fine-grained, and mechanism-focused synthesis of the transformation literature.To address this gap,
this study develops a hybrid analytical framework that integrates large language models (LLMs) with
multi-algorithm machine learning (ML) techniques to enable structured text mining and robust
identification of key driving conditions. The analysis is based on a corpus of 1,240 Chinese-language
academic articles published in authoritative journals indexed in CNKI that address resource-based
area transformation. Using an LLM-based structured extraction approach supported by carefully
designed prompt engineering, unstructured full-text content is transformed into a standardized dataset.
This process enables field-level semantic judgment across 61 predefined indicators, encompassing
seven dimensions of driving conditions—policy environment, market environment, technological
environment, regional environment, industrial structure, factor structure, and urban–rural structure—
as well as five representative transformation modes: green transformation, cultural tourism
transformation, industrial diversification, platform-dependent transformation, and regional integration.
Building on the structured dataset, the study constructs a multi-algorithm key condition identification
framework to enhance the stability and reliability of variable selection. This framework integratesensemble tree models, regularized linear models, and statistical testing methods, and synthesizes their outputs through a consensus-based frequency approach.
By emphasizing agreement across different algorithmic assumptions, the framework reduces the bias associated with any single model and highlights the most
stable and influential driving conditions associated with each transformation mode.The results indicate that the existing literature exhibits a highly concentrated
attention structure, with the policy environment and industrial structure consistently receiving the highest levels of scholarly attention. This concentration
reflects the prominent role of government-led top-level design in China’s resource-based area transformation, where institutional arrangements, policy
instruments, and strategic planning are widely recognized as core drivers. At the same time, a clear evolutionary trend can be observed. Early research was
overwhelmingly policy-oriented, whereas more recent studies show a marked increase in attention to the technological environment, particularly in relation to
digital transformation, innovation platforms, and the energy transition under carbon constraints. Although market mechanisms, regional coordination, and urbanrural
integration remain relatively underrepresented, their presence in the literature has gradually expanded, indicating a slow but discernible broadening of
analytical perspectives.Further analysis reveals substantial heterogeneity in the configurations of key driving conditions across different transformation modes.
Green transformation, cultural tourism transformation, industrial diversification, and regional integration all position the policy environment as a central driver,
yet each emphasizes different complementary mechanisms. Green transformation is characterized by the combined effects of policy enforcement and
endogenous industrial restructuring; cultural tourism transformation relies on policy support alongside spatial restructuring and urban-rural integration;
industrial diversification emphasizes strategic policy guidance coupled with industrial upgrading and market expansion; and regional integration highlights the
joint role of regional connectivity and policy coordination. In contrast, platform-dependent transformation follows a distinct logic, in which the construction of
technological platforms, such as innovation parks, research infrastructure, and collaborative innovation systems—and the activation of market vitality emerge
as the primary driving forces, with policy support playing a secondary but enabling role. From a methodological perspective, the study demonstrates that
integrating LLM-based deep semantic understanding with multi-algorithm ML analysis provides an effective way to overcome the limitations of traditional
bibliometric approaches. This integrated framework enables large-scale, field-level, and mechanism-oriented knowledge discovery, offering a more nuanced
understanding of transformation drivers. In fact, the findings contribute a differentiated and empirically grounded account of resource-based area
transformation, providing a rigorous reference for future theoretical research and for the formulation of more balanced and resilient transformation policies that
move beyond excessive reliance on policy intervention toward a coordinated role of markets, technology, and regional collaboration. |
| Key words: resource-based area transformation model research evolution driving conditions large language model |
|
|
|
|
