Methodology of a Knowledge Graph for the Changes of China??s Administrative Divisions

LIU Jimeng¹  WANG Shu^2,3*  DAI Xiaoliang^2,3  WANG Chunling^2,3  GE Shuangshuang⁴  HAN Baomin¹  ZHU Yunqiang^2,5*

1. School of Civil Engineering and Geomatics, Shandong University of Technology, Zibo 255000, China;

2. State Key Laboratory of Resources and Environmental Information System, Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing 100101, China;

3. University of Chinese Academy of Sciences, Beijing 100049, China;

4. College of Surveying and Mapping Engineering, Heilongjiang Institute of Technology, Harbin 150000, China;

5. College of Ecology, College of Aerospace Intelligence, Hainan University, Haikou 570228, China

Abstract: The changes of administrative divisions is a critical foundation for studying national governance, regional development, and changes in spatial patterns. However, traditional datasets related to administrative divisions, such as government gazettes, local chronicles, statistical yearbooks, and geographical spatial data, which record administrative regions and their spatiotemporal evolution, are usually stored in fragmented forms. They lack dynamic association and semantic integration capabilities, making it difficult to support complex queries and change analysis. Given the characteristics of administrative division datasets, such as diverse sources, complex change types, and close spatiotemporal feature coupling, this paper proposes a method for constructing a knowledge graph of China??s administrative divisions evolution. At the knowledge modeling level, by constructing an ontology of administrative divisions evolution, the semantic expression of entities, attributes, and their evolutionary relationships in administrative divisions and their spatiotemporal changes is standardized. At the knowledge extraction level, a multi-path extraction framework is designed for multi-source heterogeneous data: for structured data, rule mapping and GIS spatial topological relationship calculation methods are used to extract spatial and evolutionary information of administrative divisions. For unstructured announcement text, a joint extraction is carried out by combining domain ontology and large language models to realize the structured expression of administrative division nodes, attributes, and evolutionary relationships. At the knowledge fusion level, entity alignment and fusion are carried out by comprehensively utilizing the overlapping characteristics and semantic similarity of entities in the spatial dimension. At the application level, the fused data is archived in the Neo4j graph database to construct a knowledge graph of China??s administrative division evolution.

Keywords: administrative divisions; knowledge graph; evolution of divisions; evolution analysis; China

DOI: https://doi.org/10.3974/geodp.2026.01.08

1 Introduction

Administrative division, regional politics and their evolution refer to the long-term and sequential dynamic adjustment process of regional division carried out by the country or regions in different historical periods according to changes in factors such as political, economic, ethnic conditions, and historical and geographical conditions^[1,2]. Through the adjustment of administrative levels, regional reorganization, relocation of administrative seats, etc., the spatial structure of regions is optimized^[3,4], the level of national governance is improved^[5,6], social and economic development is promoted^[7], and thus the high-quality and sustainable development of regions and the country is promoted^[8,9]. The evolution data of administrative divisions and the laws contained therein are of important reference value for coordinating and formulating regional development strategies and national modern governance.

Data of China??s administrative divisions evolution are mainly recorded and used in 4 forms: First, official gazettes and decrees, which record the evolution of administrative divisions and related information in the form of public documents and reports, such as the annual adjustment announcements issued by the Ministry of Civil Affairs (MCA)^[10], local yearbooks, and chronicles^[11], etc.; Second, map images, which use maps or images as carriers and adopt the form of annual compilation to record administrative division boundaries and related information, such as Collection of Administrative Division Maps of China^[12,13], Handbook of Administrative Divisions of China^[14], Collection of Evolution of Administrative Division Maps of China^[15], and Collection of Changes in Administrative Division Maps of China (1980?C2017)^[16]. Third, geospatial data, which record administrative boundaries and associated attributes with high precision on a yearly basis using structured vector and raster data, such as the Global Administrative Areas (GADM) dataset^[17], OpenStreetMap (OSM) data^[18], and the Tianditu administrative division basemap service^[19], etc.; Fourth, research papers, under the drive of different research objectives, the changes of administrative division systems, data or systems, such as the evolution of Yangzhou administrative divisions from the perspective of events^[20], the analysis of the evolution of China??s administrative division pattern^[21], the knowledge graph of administrative divisions and their evolution^[22], etc.

However, existing knowledge graph construction methods for administrative division evolution still face limitations when processing massive, multi-source, and heterogeneous data. At present, the ability of the model to express the process of change is limited, and the spatiotemporal relationship between administrative divisions and the spatial attributes of administrative divisions cannot be deeply described in terms of attributes and characteristics, and it is difficult to fully present the nature and dynamics of the complex structure between changes. In addition, for the above-mentioned multi-source heterogeneous data extraction, alignment and integration, there is still a lack of integrated administrative division information methods, and it is difficult to accurately mine and integrate.

In view of this, this paper proposes a systematic construction method for the knowledge graph of China??s administrative division evolution, aiming to address the issue that existing data cannot support in-depth, long-term spatiotemporal evolution analysis. The study first constructs an administrative division ontology to establish a standardized expression framework for administrative division evolution. Subsequently, considering the characteri­stics of multi-source heterogeneous data, it proposes a set of differentiated knowledge extraction methods, including Large Language Model (LLM), rule mapping, and GIS topology computation, to achieve knowledge transformation from datasets such as official gazettes and decrees and geospatial data to dynamic evolution logic. On this basis, through spatiotemporal topology constraints and semantic alignment techniques, it completes the fusion of multi-source entities and, relying on the Neo4j graph database, constructs a knowledge graph of China??s administrative division evolution with precise boundaries, semantic associations, and continuous evolution.

2 Construction Method of Knowledge Graph for the Evolution of China??s Administrative Divisions

2.1 Overall Framework

To address the issues of multi-source heterogeneity and complex spatiotemporal evolution in China??s administrative division historical data, this paper proposes an overall framework for constructing a knowledge graph of administrative division historical evolution (Figure 1). This framework takes the ontology model of administrative division historical evolution as the semantic constraint core, integrating knowledge extraction and entity alignment methods based on multi-source data to achieve the unification of administrative division historical knowledge.

Figure 1  Overall method framework for constructing the knowledge graph of administrative division evolution

In the design stage of ontology model, the existing ontology, government announcement, scientific and technological literature, and other relevant data are comprehensively analyzed. The concepts, attributes and relationships involved in the process of administrative division evolution are systematically sorted out. Then the ontology model covering the administrative division level, attribute system, and evolution relationship is constructed.

In the knowledge extraction phase, different methods are employed to extract admi­nistrative division entities based on the structure of data sources. For unstructured texts such as official announcements, semantic extraction is performed using ontology constraints and LLMs, supplemented by manual verification to enhance data accuracy. For geographic spatial data, GIS topological calculations and attribute tables are utilized to extract spatial relationships and attribute information. Structured data is processed through rule mapping to derive structured knowledge within the framework of ontology constraints.

During entity alignment, discrepancies in attribute descriptions may arise due to varying data sources. For instance, both Beijing City and Changchun City in Jilin Province have administrative divisions named ??Chaoyang District?? (same name but different geographical locations). This necessitates aligning entities extracted from different data sources. Our study employs a method that combines the consistency of administrative division identifiers with the similarity of attribute information to achieve entity alignment.

Finally, Python code and the APOC (Awesome Procedures on Cypher) library are used to extract and align knowledge, then store it in the Neo4j graph database according to the ontology model??s conceptual, attribute, and relational storage structure to achieve visualization.

2.2 Method of Modeling the Evolution of Administrative Division

The construction adopts a top-down ontology modeling method, starting from the top-level concepts of the domain, and defines the core and most general classifications and logical frameworks within the domain. Specifically, first, based on existing ontologies in the domain, government gazettes, scientific and technical documents, and other materials, the ??skeleton?? of the ontology is determined. The administrative division entity is established as the core of the ontology; Second, on the basis of the top-level administrative division concept, subclasses and subordinate levels are gradually subdivided downward, and the entity is further subdivided into ??provincial-level??, ??prefectural-level??, and ??county- level?? units for construction; Finally, after the top-level framework is built, specific attributes and mutual relationships are defined for each level.

2.3 Knowledge Extraction Method

During the knowledge extraction instance stage, to address potential inconsistencies in data source formats??for example, geospatial data are often stored in vector formats, while public notices exist as unstructured text??different knowledge extraction methods are adopted, including: ontology-semantic-driven LLM knowledge extraction method, attribute extraction method based on rule mapping and field analysis, administrative division relationship identification method based on spatial topological operators, and administrative division evolution relationship change determination method based on attribute field changes.

For structured knowledge, for unstructured text data such as public notices, a knowledge extraction method based on ontology semantics and driven by large language models (LLMs) is adopted. This approach transforms the ??entity-attribute-relationship?? structure of the ontology model into LLM recognition logic, enabling the large language model to perform targeted parsing and matching based on the ontology content. In the specific extraction process, prompt engineering is employed to construct a structured extraction template (Prompt) through multi-round dialogue and instructions, requiring the LLM to identify the elements of ??subject (S), predicate (P), object (O), and time (T)??, and output them as a standardized quadruples. The extraction results are then manually verified in combination with the constraints of the ontology to eliminate content that does not conform to the ontology semantic specifications.

For long-sequence, multi-source GIS-based administrative division information spatial vector data, it is necessary to standardize the extraction of entity attributes and the construction of geometric information. Specifically, an attribute extraction method based on rule mapping and field parsing is adopted: First, establish a mapping rule library between ontological attributes and GIS fields, and perform standard conversion on the fields of the attribute table; then use the open-source geographic information processing tool GeoPandas[1] to read the spatial vector data of administrative divisions, jointly process the attribute fields and geometric objects, calculate and extract the spatiotemporal extents; At the same time, obtain the corresponding year through file names parsing, uniformly assign it to the entity attributes, and finally form the entity attribute dataset of administrative divisions.

To identify the spatial relationships and evolutionary relationships of administrative divisions, a method for identifying spatial administrative division relationships based on spatial topological operators is adopted. The core principle of this method is to abstract vector boundaries into geometric objects. Through topological judgment and spatial overlay calculation, static relationships such as adjacency and inclusion of spatial administrative divisions, as well as dynamic evolutionary relationships such as merger, split, establishment, and revocation, are identified. In the same year, the DE-9IM matrix^[23] is used to describe the topological relationship between two administrative division geometric objects A and B, which can be expressed as 9 intersections between their interiors, boundaries, and exteriors, as shown in Equation 1:

             (1)

By judging the combination of the value patterns of each element in the matrix, the type of topological relationship between two spatial objects can be accurately characterized. For example, when and there is at least one boundary intersection in the matrix that is not, the two administrative divisions are determined to be adjacent. When and , the two administrative divisions are determined to be in an inclusion relationship.

In the process of evolution relationship extraction, the GIS overlay analysis method^[24]is employed to calculate the spatial intersection between the administrative division geometric setsandat adjacent time nodes t and t+1, thereby determining whether administrative divisions have persisted or disappeared over time^[25]. For cases where spatial intersections exist, the area coverage model^[26] is introduced to further classify the evolution type. The spatial coverage of administrative divisions is calculated using the following Equation 2:

                                                                  (2)

First, set a coverage threshold. When the coverage is close to 100%, it indicates that the main body of the administrative division remains basically unchanged. When multiple have a coverage of the sameexceeding the set threshold, it is determined that a merger has occurred; when a singlehas a coverage of multipleexceeding the set threshold, it is determined that the administrative division has been split.

The evolution of administrative divisions encompasses not only spatial boundary changes through mergers and splits, but also relationships involving name changes, administrative level adjustments, and administrative affiliation modifications. To identify these relationships, an attribute-based evolution analysis method is employed. The process begins by screening adjacent years?? administrative division pairs that show high spatial overlap and coverage consistency through overlay analysis, forming a candidate set for attribute comparison. Subsequently, key attributes including Chinese names, division codes, administrative levels, affiliated regions, and government locations are systematically compared. For instance, when spatial boundaries remain unchanged but Chinese names vary, it is classified as a ??name change?? relationship. Similarly, when division codes stay consistent but affiliations change, it is categorized as an ??affiliation adjustment?? relationship.

2.4 Entity Alignment Method

Entity alignment refers to the process of identifying and resolving instances from multiple data sources that point to the same real-world entity, while performing heterogeneous disambiguation to achieve unified representation. In this study, data sources included official announcements, GIS vector data, and other structured or semi-structured data. Due to variations in naming conventions and attribute completeness across different sources, the same administrative division may be inconsistently described. To address this, we developed 2 alignment methods: entity identifier consistency alignment and attribute similarity alignment, which are tailored to varying levels of entity identification completeness.

For entities with standardized identifiers or standardized names, this study employs the identifier consistency alignment method^[27]. By utilizing the standardized Chinese full names and unique administrative division codes of administrative entities, precise entity anchoring is performed under a unified time benchmark T. First, the entities in two data sources are determined to be on the same time slice. If an entity possesses identical six-digit administrative division codes, it is identified as the same entity. If the division code is missing, the standardized Chinese full name and its affiliated administrative division are compared. The Equation is expressed as follows:
               (3)

where, (e₁,e₂) denotes entities from distinct data sources S₁ and S₂. represents the administrative division code, while Name(e) indicates the full Chinese name of the administrative division. Parent(e) denotes the full Chinese name of the parent administrative division, and T represents the year of the data section. The consistency score  is assigned when =1, indicating the entity??s identity.

For cases where administrative division codes are incomplete or Chinese names are non-standardized, this study employs an attribute similarity alignment method. By calculating the similarity between multiple attributes of entities, it determines whether they belong to the same entity^[27]. Let the attribute sets of administrative division entities andbe denoted as, and the comprehensive attribute similarity calculation method is as follows:

                                                                    (4)

where,  denotes the weight of attribute , with ??=1. () represents the similarity function for the corresponding attribute. When aligning administrative division entities, we first set the decision threshold ??, using Chinese names and administrative affiliation attributes as core evaluation metrics. Chinese names are assigned higher weight, while administrative affiliation serves as auxiliary judgment. The comprehensive similarity score is calculated through weighted summation of these formulas, then compared with the preset threshold ??. If ,the two entities are determined to represent the same administrative division.

3 Knowledge Graph Construction and Validation of Administrative Division Evolution from 1949 to 2023

3.1 Data Sources

The data for constructing the knowledge graph of administrative division evolution mainly come from China??s administrative division vector data of different years, China??s provincial statistical yearbooks, and announcements of administrative division changes at or above the county level. For specific sources and explanations, please refer to the Table 1.

Table 1  Data sources

Note: The scope of China??s administrative division vector data and China??s provincial statistical yearbook data covers the three levels of administrative divisions (provinces, cities, and counties) in China??s mainland and Hong Kong, Macao, excluding Taiwan.

The administrative division vector data of China used in this paper consists of 2 parts. The vector data from 1949 to 2014 is sourced from the Geographic Information Database of the University of California, Berkeley, while the vector data from 2014 to 2023 is cited from the China Temporal Sequence Administrative Map (CTAmap) multi-temporal administrative division database published by Cheng Rui, et al. (2025)^[28]. CTAmap uses the 2020 administrative division pattern of China as the benchmark, integrating authoritative data such as the basic geographic information data from the National Geospatial Information Data Center (NGCC) and the administrative division codes from the Ministry of Civil Affairs. By combining forward temporal inference with backward retrospective methods, it reconstructs the spatial boundaries, administrative codes, and hierarchical affiliations of provincial, municipal, and county-level administrative divisions from 2009 to 2023. Prior to the study, the spatial boundaries of the database data were cross-checked against standard administrative division maps, and key attribute information was verified against the official administrative division data published by the Ministry of Civil Affairs over the years. The results showed that the data is highly consistent with authoritative sources in terms of spatial morphology and attribute information, making it suitable as the data foundation for constructing the knowledge graph dataset of the evolution of China??s administrative divisions.

3.2 Ontology Model of China??s Administrative Division Evolution

The administrative division evolution ontology model developed in this study comprises 2 core categories of concepts, and their attributes, as well as 4 major types of relationships. The top-level core concepts include administrative divisions (or administrative units) and administrative centers. Administrative divisions refer to the areas that a country divides for administrative management, which are the regions governed by local administrative authorities. They include four levels: national-level administrative divisions, provincial-level administrative divisions, prefecture-level administrative divisions, and county-level administrative divisions. Administrative centers denote the locations of a nation??s central or local governments, including capital cities and regional administrative seats. The attributes of administrative divisions primarily consist of: identification information describing and marking administrative divisions; management information standardizing hierarchical structures; spatial information defining geographic boundaries; and temporal information recording lifecycle changes. Specifically, administrative division entities contain 2 major categories of attributes: core attributes and extended attributes. Core attributes exhibit strong correlation with administrative division evolution, as any changes in divisions inevitably involve alterations in at least 1 core attribute. Extended attributes provide further descriptive details about administrative divisions without direct connection to division changes, though they often evolve with division transformations??for instance, adjustments to spatial boundaries may lead to corresponding GDP changes. Administrative center attributes mainly include: identification attributes clarifying the center??s identity; spatial attributes precisely defining its location and scope; temporal attributes recording its existence duration; and management attributes specifying its authority and organizational structure. For detailed conceptual frameworks and attribute classifications, refer to Table 2.

The ontological model of administrative division evolution developed in this study comprises 4 fundamental relationship types: spatial, hierarchical, and evolutionary and association relationships. Spatial relationships delineate the geographical positions of administrative entities, while hierarchical relationships define their hierarchical subordinations. Evolutionary relationships, functioning as temporal attributes, document administrative changes across different periods. Association relationships are used to link administrative division entities with their economic, natural, and other indicators. Detailed relationship types and examples are presented in Table 3.

Table 2  Ontology concepts and attribute classification

Note: The core and extended attributes of administrative divisions include the common attributes of national-level, provincial-level, prefectural-level, and county-level administrative regions. Administrative centers share attributes with both the capital and local government seats.

3.3 Knowledge Extraction and Alignment

This experiment targets the various data sources mentioned in Section 3.1, completing the structured acquisition and organization of administrative division entities, attributes, and relationships using the knowledge extraction and alignment methods described in Chapter 2. During the knowledge extraction phase, for the vector data of China??s administrative divisions, this paper employed rule mapping and field parsing methods to extract static attributes such as Chinese names, administrative codes, and affiliations from the vector data

Table 3  Ontology part relationship type and instance data table 

attribute table. Simultaneously, GeoPandas was utilized to parse spatial geometric objects, serializing administrative boundaries as spatial range attributes. Based on this, DE-9IM spatial topology operators and overlay analysis methods were introduced to horizontally identify adjacent relationships between administrative divisions in the same year, and vertically determine the evolution relationships of administrative divisions (establishment, abolition, merger, and division) through comparisons of spatial objects in adjacent years. For announcements of administrative division changes at or above the county level, an ontology- driven large language model extraction method was adopted to identify information such as subjects, evolution types, time, and approval documents in administrative division change events, corresponding to the evolution relationships designed in the ontology. Finally, for statistical indicators in China??s provincial statistical yearbooks, they were extracted as extended attribute information and established as attribute associations with corresponding administrative division entities.

During the knowledge alignment phase, the system sequentially applied 2 methods to integrate administrative entities extracted from different data sources: identifier consistency alignment and attribute similarity alignment. First, precise alignment was performed using administrative codes, Chinese names, and temporal information from GIS vector data, statistical yearbooks, and official announcements. For entities with missing codes or name discrepancies, the attribute similarity alignment method was employed to comprehensively compare key attributes such as Chinese names and administrative affiliations, thereby achieving entity merging and disambiguation. At the relational level, the aligned entity identifiers were unified to specify spatial relationships and their evolutionary connections.

3.4 Knowledge Graph Storage

This study employed the Neo4j graph database for storing and managing the knowledge graph of administrative division evolution. As a native graph database, Neo4j stores data directly in the form of nodes, attributes, and relationships, allowing entities and their semantic connections to be naturally mapped into a graph structure. This facilitates efficient semantic reasoning and path analysis. A connection pool between Python and Neo4j was established using py2neo, and Neo4j??s APOC batch processing technology was utilized to store the nodes, attributes, and relationships obtained from Section 3.3 in batches. Finally, the constructed knowledge graph of China??s administrative division evolution includes 233,243 administrative division nodes and 89,073 attribute nodes, with specific node types and corresponding quantities shown in Table 4; The knowledge graph of China??s administrative division evolution mainly covers 1,352,945 spatial relationships, 268,689 hierarchical relationships 35,033 evolutionary relationships, and 344,275 association relationships for association expansion. Examples of relationship types and quantities are provided in Table 5.

Table 4  Examples of node types and node quantities in the knowledge graph of China??s administrative division evolution

Table 5  Examples of relationship types and relationship quantities in the knowledge graph of China??s administrative division evolution

Figure 2 presents an example of knowledge graph nodes and relationships. The knowledge graph not only shows the current administrative boundaries but also includes historical administrative entities from various periods, each corresponding to a specific timeframe. These rich attribute details reveal the evolution of administrative divisions. Taking Chengdu as an example (Figure 3), the knowledge graph displays ??Chengdu?? nodes from different periods: Chengdu (2016) and Chengdu (2017) are connected through an evolution relationship (regional merger), forming a chronological chain. Similarly, Ziyang City (2016) is linked to Chengdu (2017?C2023) via an evolution relationship (regional division). Overall, this indicates that in 2016, Chengdu underwent an administrative transformation, merging its original territory with parts of Ziyang City to form the new Chengdu.

Figure 2  Example diagram of nodes and relationships in the knowledge graph of
China??s administrative division change

3.5 Feasibility Test and Quality Change of Knowledge Graph Construction Method

This paper conducts a comparative experiment between the evolution data recorded in the knowledge graph of administrative division changes and the county-level and above administrative division changes published by the Ministry of Civil Affairs, to verify the feasibility and quality of the systematic construction method of the China administrative division evolution knowledge graph. Specific steps are: First, collect the announcement materials of county-level and above administrative division changes released by the Ministry of Civil Affairs. Second, randomly select the administrative division change records from different years between 2015 and 2023, and extract the change types, administrative divisions before and after the changes, as well as the affiliation information before and after the changes as the validation set. Finally, the knowledge graph was compared with the announcement of administrative division changes at or above the county level issued by the Ministry of Civil Affairs, focusing on 4 key aspects of information:

First, the revocation and establishment of administrative divisions. In the knowledge graph, this primarily manifests as whether nodes persist in the subsequent year after revocation, and whether nodes are successfully generated after establishment. Second, the accuracy of administrative division names. In the knowledge graph, this is reflected in whether node attribute information aligns with official records. Third, the accuracy of administrative division affiliation information recorded in announcements. In the knowledge graph, this is represented by the hierarchical relationships between nodes. Finally, the accuracy of change records, which in the knowledge graph indicates whether specific evolutionary relationships exist between administrative division nodes before and after changes. Example comparative results are shown in Table 6.

Table 6  Example of comparison between the knowledge graph of China??s administrative division change and the announcement of the Ministry of Civil Affairs

The discrepancies primarily stem from 3 factors: Firstly, the knowledge graph only covers provincial, prefectural, and county-level administrative divisions, excluding special institutions such as administrative committees, resulting in missing records. Secondly, certain details (e.g., government locations) lack systematic inclusion in the original data, making them hard to visualize in the knowledge graph. Thirdly, some administrative adjustments in the data source are recorded with time discrepancies, causing inconsistencies with the official announcement dates.

The validation set was then used to query the knowledge graph of administrative division evolution, with recall rate, accuracy rate, and F1-score employed to evaluate its query performance. Comparative analysis revealed that the recorded administrative division information in the knowledge graph consistently exceeded 95% in average recall rate, accuracy rate, and F1-score when compared to the administrative division changes published by the Ministry of Civil Affairs, with overall records highly consistent with official data, demonstrating the feasibility of the construction method system proposed in this paper for the knowledge graph of the evolution of China??s administrative divisions. Detailed performance comparison results are presented in Table 7.

4 Summary and Future Work

To address the issues of scattered data storage, difficulties in cross-dataset association, and the complexity of evolutionary queries caused by diverse sources and heterogeneous formats in traditional administrative division datasets, this paper proposes and implements a systematic construction method for the knowledge graph of China??s administrative division evolution, covering 4 key steps: ontology modeling, multi-source knowledge extraction, entity alignment and fusion, graph database organization. First, an ontological model of administrative division evolution is constructed, with administrative division entities as the core and spatial relationships, hierarchical relationships, and evolutionary relationships as the primary relationship types. Second, differentiated knowledge extraction methods are proposed for vector spatial data, statistical yearbooks, and announcement texts. Then, an entity alignment method based on identifier consistency and attribute similarity is designed to resolve issues such as duplication, homonymy, and ambiguity of administrative division entities in multi-source data. Finally, by utilizing the batch processing mechanism of Neo4j graph database and APOC library, the storage and visualization of the knowledge graph of China??s administrative division evolution were achieved, and the quality of the graph was also verified.

Future research can be expanded in multiple directions. First, extending the current time frame to a longer historical period, we can construct a knowledge graph of administrative division evolution covering extended cycles to reveal broader evolutionary patterns. Second, by integrating finer-grained socio-economic data, policy texts, and historical documents, we can enrich the node attributes and relationship types of the knowledge graph, thereby exploring the underlying driving mechanisms of administrative division changes. Finally, combining with large models, we can leverage their powerful language understanding and generation capabilities to achieve more intelligent query and analysis functions for administrative division evolution, reducing user barriers and enabling more intuitive and convenient exploration and utilization of this knowledge.

Author Contributions

Liu, J. M., Wang, S., Zhu, Y. Q., and Han, B. M. contributed to the overall design of the method; Liu, J. M., Wang, C. L., and Ge, S. S. carried out method research and conducted method practice, wrote the paper; Dai, X. L. collected the data sources for the construction of China??s administrative division evolution knowledge graph; Liu, J. M., Ge, S. S., Wang, S., and Zhu, Y. Q. participated in the revision of the paper.

Conflicts of Interest

The authors declare no conflicts of interest.

References

[1]        Ren, X. R., Ren, F., Chen, H. P., et al. Design and realization of display system about historical evolution of Hubei Province??s administrative divisions [J]. Journal of Geomatics, 2017, 42(3): 6.

[2]        Zhao, Y. C., Wang, K. Y., Zhao, B., et al. Spatio-temporal process and pattern of the establishment of county-level administrative divisions in China in the past 2200 years [J]. Acta Geographica Sinica, 2024, 79(4): 890?C908.

[3]        Feng, R. D., Wang, K. Y. The direct and lag effects of administrative division adjustment on urban expansion patterns in Chinese mega-urban agglomerations [J]. Land Use Policy, 2022, 112: 105805.

[4]        Feng, R. D., Wang, K. Y. Spatiotemporal effects of administrative division adjustment on urban expansion in China [J]. Land Use Policy, 2021, 101: 105143.

[5]        Wang, F. L., Liu, Y. G. China??s urban planning and administrative urbanization: case of Ordos [J]. Urban design and planning, 2014, 167(5): 196?C208.

[6]        Chen, Y. L., Yu, P. H., Wang, L., et al. Polycentric urban development with state-led administrative division adjustment: a policy insight for urban spatial transformation [J]. Journal of Geographical Sciences, 2023, 33(12): 2400?C2424.

[7]        Feng, R. D., Wang, K. Y., Wang, F. Y. Quantifying influences of administrative division adjustment on PM_2.5 pollution in China??s mega-urban agglomerations [J]. Journal of environmental management, 2022, 302: 113993.

[8]        Wei, S., Zheng, W., Wang, L. Understanding the configuration of bus networks in urban China from the perspective of network types and administrative division effect [J]. Transport Policy, 2021, 104: 1?C17.

[9]        Zhu, J. H., Chen, X., Chen, T. Spheres of urban influence and factors in Beijing-Tianjin-Hebei Metropolitan Region based on viewpoint of administrative division adjustment [J]. Chinese Geographical Science, 2017, 27(5): 709?C721.

[10]     Ministry of Civil Affairs of P. R. China. Changes in administrative divisions above the county level in China [EB/OL]. (2023-04-03) [2025-09-16]. http://xzqh.mca.gov.cn/description?dcpid=2023.

[11]     Duan, B. R. Beijing Gazetteer ?? Statistics Chronicle [M]. Beijing: Beijing Publishing House, 2016.

[12]     Dai, J. L., Bai, B. Atlas of Administrative Divisions of P. R. China [M]. Beijing: Sinomap Press, 2005.

[13]     Ministry of Civil Affairs of P. R. China, National Geomatics Center of China. Atlas of Administrative Divisions of China [M]. Beijing: Sinomap Press, 2005.

[14]     Dai, J. L. Handbook of Administrative Divisions of China [M]. Beijing: China Social Science Press, 2009.

[15]     Chen, H. L. Atlas of Administrative Division Evolution of China (Deluxe Edition) [M]. Beijing: Sinomap Press, 2003.

[16]     Yang, Y. P. Atlas of Administrative Division Changes of China (1980?C2017) [M]. Beijing: Sinomap Press, 2016.

[17]     Food and Agriculture Organization of the United Nations (FAO). Administrative boundaries (level 1)- GADM 3.6 [EB/OL]. (2024-7-30) [2025-09-16]. https://data.apps.fao.org/catalog/dataset/aecbbc85-2a46- 498b-83b4-beca24178f71.

[18]     Geofabrik. China OpenStreetMap dataset [EB/OL]. (2025-09-15) [2025-09-16]. https://download.geofabrik. de/asia/china.html.

[19]     Map World (Tianditu). China administrative division dataset [EB/OL]. (2024-05-01) [2025-09-16]. https:// cloudcenter.tianditu.gov.cn/administrativeDivision.

[20]     Lu, Y. X., Zhang, X. Y., Zhang, C. J. Construction method of knowledge graph for administrative division evolution from an event-oriented perspective [J]. Journal of Geo-information Science, 2025, 27(10): 2440?C2452.

[21]     Zhu, J. H., Chen, T., Wang, K. Y., et al. Spatial pattern evolution and driving force analysis of administrative division in China since the reform and opening-up [J]. Geographical Research, 2015, 34(2): 247?C258.

[22]     Chen, S. H. Research on the construction method of knowledge graph of administrative division evolution since the founding of China [D]. Nanjing: Nanjing Normal University, 2022.

[23]     Clementini, E., Di Felice, P., van Oosterom, P. A small set of formal topological relationships suitable for end-user interaction [J]. Springer Berlin Heidelberg, 2005, 277?C295.

[24]     Tai, Y. Y., Wang, Q., Sun, K. Algorithm on polygon overlaying based on topological information in GIS [J]. Journal of Southeast University (Natural Science Edition), 2006, 36(3): 442?C445. DOI: 10.3969/j.issn.1001-0505.2006.03.023.

[25]     Wang, C. L., Zhu, Y. Q., Wang, S., et al. Construction of administrative division knowledge graph considering spatio-temporal characteristics and evolution relationships [J]. Journal of Geo-information Science, 2026, 28(1): 89?C104 https://doi.org/10.12082/dqxxkx.2026.250471.

[26]     Weng, J. C., Ge, Y., Wang, C., et al. Transit Travel Index and analysis models for transit service evaluation [J]. Journal of Highway and Transportation Research and Development, 2016, 33(1): 130?C134 https://doi.org/10.3969/j.issn.1002-0268.2016.01.020.

[27]     Zhuang, Y., Li, G. L., Feng, J. H. A review of knowledge base entity alignment techniques [J]. Journal of Computer Research and Development, 2016, 53(1): 165?C192.

[28]     Rui, C., Zhang, H. F., Chen, B. Z. China temporal administrative map: a multitemporal database for Chinese historical administrative divisions (2009?C2023) [C]. Third International Conference on Environmental Remote Sensing and Geographic Information Technology (ERSGIT 2024). SPIE, 2025, 13565: 497?C507. https://doi.org/10.1117/12.3059430.