A natural resource archive retrieval and utilization method combining a large model and a knowledge graph technology

Abstract

The application provides a natural resource archive retrieval and utilization method combining a large model and a knowledge graph technology, relates to the technical field of data processing, and comprises the following steps: identifying business entities in the field of natural resources and semantic correlations thereof, and constructing RDF triples with attribute constraints; performing geometric feature maintaining type coordinate simplification on a land entity vector boundary coordinate point sequence through a Douglas-Pok compression algorithm to generate a set of key coordinate points after noise reduction; performing vector geometry superposition operation on the polygon boundaries of two land entities based on the set of key coordinate points, calculating the area proportion of the overlapping area according to the intersection-union ratio, and performing entity space alignment when the intersection-union ratio value exceeds a preset threshold value to obtain a set of spatially aligned triples; and fusing the set of spatially aligned triples to construct a knowledge graph in the field of natural resources. The application effectively reduces the interference of redundant information.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method for retrieving and utilizing natural resource archives that combines large-scale models and knowledge graph technology. Background Technology

[0002] Constructing domain-specific knowledge graphs enables the associative organization of archival information. For example, some existing technologies extract triple data (entity-relationship-entity) from archives to construct natural resource knowledge graphs that include spatiotemporal and business relationships, providing a semantic foundation for archive retrieval. Simultaneously, the advantages of large language models in text understanding and information extraction have been introduced into this field. Pre-trained models enhance the accuracy of specialized terminology recognition and entity attribute extraction, aiding in the construction and completion of knowledge graphs.

[0003] However, existing technologies still have some limitations in combining large models and knowledge graphs to achieve efficient utilization of natural resource archives:

[0004] For example, natural resource archives often have complex formats (such as scanned documents, PDFs, and CAD drawings). Existing technologies have limited ability to automatically parse document directory levels and nested structures, making it difficult to accurately break them down into sub-documents according to the "document object model topology" and "directory hierarchy rules," resulting in coarse-grained semantic analysis. Traditional knowledge extraction methods sometimes rely on rule matching or shallow machine learning models, which lack sufficient understanding of the professional semantics implicit in the archives (such as land ownership change logic and planning policy relevance). Although some technologies introduce large models, they do not design dedicated processing mechanisms for the characteristics of the natural resource field (such as boundary point coordinates and land use coding), resulting in insufficient completeness and accuracy of the extracted structured metadata. Summary of the Invention

[0005] The technical problem to be solved by this invention is to provide a method for retrieving and utilizing natural resource archives that combines large model and knowledge graph technology, which effectively reduces redundant information interference, makes deep semantic understanding more targeted, and greatly improves the automation and accuracy of archive preprocessing.

[0006] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:

[0007] A method for retrieving and utilizing natural resource archives that combines large-scale models and knowledge graph technology, the method comprising:

[0008] Step 1: Obtain the target natural resource archive set, which contains at least one natural resource archive to be processed;

[0009] Step 2: Based on the target natural resource archive set obtained in Step 1, parse the document object model topology and generate corresponding sub-document sets according to the directory hierarchy rules;

[0010] Step 3: Using a pre-defined large model of natural resources, perform deep semantic understanding and structured information extraction on the sub-document set generated in Step 2, and output structured metadata describing the core elements of the archive.

[0011] Step 4: Based on the structured metadata output in Step 3, identify business entities in the natural resources domain and their semantic relationships, and construct RDF triples with attribute constraints; wherein, the identified business entities include land parcel entities with vector boundary coordinate point sequences;

[0012] Step 5: The sequence of boundary coordinate points of the land parcel entity vector extracted in Step 4 is simplified by geometric feature preservation using the Douglas-Puk compression algorithm to generate a set of denoised key coordinate points.

[0013] Step 6: Based on the key coordinate point set after noise reduction in Step 5, perform vector geometric superposition operation on the polygonal boundaries of the two land parcel entities, calculate the area ratio of the overlapping area according to the intersection-union ratio, and when the intersection-union ratio exceeds the preset threshold, perform entity spatial alignment to obtain a set of spatially aligned triplets.

[0014] Step 7: Merge the spatially aligned triples from Step 6 to construct a knowledge graph for the natural resources domain.

[0015] A natural resource archive retrieval and utilization system combining large-scale model and knowledge graph technology includes:

[0016] The acquisition module is used to acquire a target natural resource archive set, wherein the archive set contains at least one natural resource archive to be processed;

[0017] The parsing module is used to parse the document object model topology based on the target natural resource archive set, and decompose it into corresponding sub-document sets according to the directory hierarchy rules;

[0018] The execution module is used to perform deep semantic understanding and structured information extraction on the sub-document collection through a pre-set large model of natural resources, and output structured metadata describing the core elements of the archive.

[0019] The building module is used to identify business entities and their semantic relationships in the natural resources domain based on structured metadata, and to construct RDF triples with attribute constraints; wherein, the identified business entities include land parcel entities with vector boundary coordinate point sequences;

[0020] The noise reduction module is used to perform geometric feature-preserving coordinate simplification on the sequence of boundary coordinate points of land parcel entity vectors using the Douglas-Puk compression algorithm, generating a set of noise-reduced key coordinate points.

[0021] The alignment module is used to perform vector geometric superposition operations on the polygonal boundaries of two land parcel entities based on the denoised key coordinate point set, calculate the area ratio of the overlapping area according to the intersection-union ratio, and perform entity spatial alignment when the intersection-union ratio exceeds a preset threshold to obtain a set of spatially aligned triples.

[0022] The fusion module is used to fuse spatially aligned sets of triples to construct a knowledge graph in the field of natural resources.

[0023] The above-described solution of the present invention has at least the following beneficial effects:

[0024] By parsing the document object model topology and decomposing sub-documents according to directory hierarchy rules, the limitations of complex formats and coarse structure parsing in traditional archive processing are overcome, and the fine-grained decomposition of archives is achieved. This set of sub-documents layered according to business logic effectively reduces the interference of redundant information, makes deep semantic understanding more targeted, and greatly improves the automation and accuracy of archive preprocessing.

[0025] By employing a pre-defined large-scale model of natural resources to perform deep semantic understanding, compared with traditional rule matching or general models, it can accurately identify domain-specific terms (such as "boundary point coordinates" and "land use code") and implicit business logic (such as ownership change association). The output structured metadata is more in line with the core elements of the archives. This process solves the problem of insufficient understanding of professional semantics in traditional methods and improves the integrity of metadata.

[0026] Constructing RDF triples with attribute constraints not only records the semantic relationships between entities, but also enhances the accuracy of these relationships through attribute constraints (such as the area and coordinate range of a land parcel). In particular, incorporating the "vector boundary coordinate sequence" of land parcel entities into the entity definition achieves a fusion of spatial and semantic features.

[0027] The Douglas-Puk algorithm is used to perform geometric feature-preserving simplification on the vector boundary coordinate point sequence. While removing noise points (such as redundant coordinates caused by measurement errors), the core geometric outline of the plot is fully preserved. This process reduces the amount of coordinate data and avoids boundary matching deviations caused by coordinate noise.

[0028] Entity spatial alignment is achieved through vector geometric superposition based on intersection-union ratio (IU). Compared with traditional alignment methods that rely on text similarity, this method solves the problem of identifying "same place, different names" and "same name, different places" from the perspective of essential spatial features. The quantitative alignment logic when the IU exceeds a preset threshold improves the accuracy of cross-archive matching of land parcel entities and realizes knowledge association in the spatial dimension. Attached Figure Description

[0029] Figure 1This is a flowchart illustrating a method for retrieving and utilizing natural resource archives that combines large-scale modeling and knowledge graph technology, as provided in an embodiment of the present invention.

[0030] Figure 2 This is a schematic diagram of a natural resource archive retrieval and utilization system that combines large model and knowledge graph technology, provided by an embodiment of the present invention. Detailed Implementation

[0031] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0032] like Figure 1 As shown, an embodiment of the present invention proposes a method for retrieving and utilizing natural resource archives that combines large-scale models and knowledge graph technology. The method includes the following steps:

[0033] Step 1: Obtain the target natural resource archive set, which contains at least one natural resource archive to be processed;

[0034] Step 2: Based on the target natural resource archive set obtained in Step 1, parse the document object model topology and generate corresponding sub-document sets according to the directory hierarchy rules;

[0035] Step 3: Using a pre-defined large model of natural resources, perform deep semantic understanding and structured information extraction on the sub-document set generated in Step 2, and output structured metadata describing the core elements of the archive.

[0036] Step 4: Based on the structured metadata output in Step 3, identify business entities in the natural resources domain and their semantic relationships, and construct RDF triples with attribute constraints; wherein, the identified business entities include land parcel entities with vector boundary coordinate point sequences;

[0037] Step 5: The sequence of boundary coordinate points of the land parcel entity vector extracted in Step 4 is simplified by geometric feature preservation using the Douglas-Puk compression algorithm to generate a set of denoised key coordinate points.

[0038] Step 6: Based on the key coordinate point set after noise reduction in Step 5, perform vector geometric superposition operation on the polygonal boundaries of the two land parcel entities, calculate the area ratio of the overlapping area according to the intersection-union ratio, and when the intersection-union ratio exceeds the preset threshold, perform entity spatial alignment to obtain a set of spatially aligned triplets.

[0039] Step 7: Merge the spatially aligned triples from Step 6 to construct a knowledge graph for the natural resources domain.

[0040] In this embodiment of the invention, by integrating a set containing at least one file to be processed, omissions of file information are avoided; it is compatible with datasets containing single or multiple natural resource files, adapting to file processing needs of different scales; by parsing the Document Object Model (DOM) topology, it breaks through the limitations of the traditional coarse-grained processing of "whole files" in file processing, and can accurately identify the hierarchical relationships within the document (such as chapters, clauses, attachments, etc.); it generates a set of sub-documents according to the directory hierarchy rules, so that each sub-document focuses on a specific business theme (such as "approval process" and "ownership certificate"), reduces redundant information interference, provides more precise granularity for deep semantic understanding, and improves the efficiency and accuracy of subsequent analysis; it adopts a preset large model of natural resource domain, which, compared with general models or rule matching methods, can accurately identify domain-specific terms (such as "boundary points" and "land use pre-approval") and implicit business logic (such as "the association between approval process and policy basis"); the output structured metadata directly corresponds to the core elements of the file (such as land coordinates, approval date, and owner), solving the problem of fragmented information extraction in traditional methods.

[0041] Based on structured metadata, business entities (such as land parcels, departments, and policies) and their semantic relationships are identified. Attribute-constrained RDF triples (e.g., "Land Parcel A - Approving Unit - Natural Resources Bureau (Approval Date: 2023-01-01)") are constructed, which more accurately reflect the actual business situation compared to triples without attribute constraints. For the first time, "vector boundary coordinate point sequence" is used as the core attribute of land parcel entities, breaking through the limitations of traditional entities relying solely on text descriptions and achieving a fusion of spatial and semantic features. While maintaining the geometric contours of the land parcels (such as boundary shape and inflection point positions), measurement errors or redundant coordinate points are removed, reducing the data volume (typically compressible by 40%-60%) and lowering subsequent computational complexity. Through "geometric feature-preserving" denoising, interference from noisy coordinates on boundary matching is avoided, ensuring that the key coordinate point set truly reflects the spatial morphology of the land parcels, providing high-quality data for accurate spatial alignment.

[0042] By employing vector geometric overlay operations and intersection-union ratio (IUU) calculations, the correlation between land parcels is determined based on essential spatial characteristics, resolving the misjudgment issues of "same location, different names" and "same name, different locations" caused by traditional reliance on text similarity (such as name matching). Using an IUU threshold (such as 80%) as the alignment criterion, quantitative decision-making for spatial alignment is achieved, increasing the accuracy of cross-archive land parcel entity matching to over 90%, effectively breaking down spatial information silos. By integrating spatially aligned triples, the knowledge graph not only includes semantic relationships between entities (such as "land parcel - policy basis" and "land parcel - ownership change"), but also integrates accurate spatial relationships (such as "adjacent land parcels" and "inclusion relationship"), breaking through the limitation of traditional knowledge graphs that can only express semantic relationships. This knowledge graph can support location-based retrieval (such as "query all industrial land archives in XX area"), time-dimensional tracing (such as "historical approval records of land parcel A"), and cross-entity correlation analysis (such as "the impact of policy changes on land use").

[0043] In a preferred embodiment of the present invention, step 2 includes:

[0044] Step 21: Based on the natural resource archive file obtained in Step 1, parse and generate a Document Object Model (DOM) tree structure; recursively calculate the depth value of each node using the DOM tree as the processing object to generate a DOM tree with depth annotation, wherein the root node depth is initialized to 0, and the child node depth value = parent node depth value + 1.

[0045] Step 22, using the depth-annotated DOM tree output from Step 21 as input:

[0046] Filter the set of nodes with a depth value ≥ 2, and perform the following for each filtered node:

[0047] Extract the text blocks of all leaf nodes in the subtree of the given node, arrange the text blocks in the original character offset order to form an ordered sequence, and calculate the spacing between adjacent text blocks in the sequence, where the spacing value = the starting offset of the subsequent text block - the ending offset of the previous text block;

[0048] Record the maximum spacing value of the current node. When the maximum spacing value is less than the preset threshold, merge all text blocks in the sequence to generate a sub-document, and output the sub-document set and the corresponding DOM node path and maximum spacing value.

[0049] Step 23: Using the set of sub-documents output in Step 22 as the processing object, assign a globally unique identifier to each sub-document, associate it with the DOM node path and maximum spacing value output in Step 22, and bind the source file number to generate structured traceability metadata.

[0050] In this embodiment of the invention, the specific implementation process of step 21 is as follows:

[0051] First, for each natural resource archive file (such as PDF, Word, scanned document to text, etc.) obtained in Step 1, its internal structure is parsed using a document parsing tool (such as PDFBox for PDF, POI for Word, etc.) to generate a corresponding Document Object Model (DOM) tree structure. The nodes of this DOM tree correspond to different content units in the archive. For example, the root node represents the entire archive file, first-level child nodes correspond to chapter titles, second-level child nodes correspond to paragraphs under chapters, third-level child nodes correspond to sentences or phrases within paragraphs, and leaf nodes correspond to the smallest text fragments.

[0052] Next, using the generated DOM tree as the processing object, the depth value of each node is calculated recursively: the depth of the root node of the DOM tree is initialized to 0; starting from the root node, all its direct child nodes are traversed in turn, and a depth value is assigned to each child node, the depth value of the child node is equal to the depth value of its parent node plus 1; then, using these child nodes as parent nodes, their child nodes are processed recursively, and the depth is calculated according to the rule of "child node depth value = parent node depth value + 1", until all nodes in the DOM tree have been traversed; finally, a DOM tree with a clear depth label is generated for each node, for example, the depth of the root node (the entire document) is 0, the depth of the "Chapter 1" node is 1, the depth of the "Section 1.1" node is 2, the depth of the "Paragraph 1.1.1" node is 3, and so on.

[0053] The specific implementation process of step 22 above is as follows:

[0054] Using the depth-annotated DOM tree output in step 21 as input, first filter out all nodes with a depth value greater than or equal to 2 to form a set of nodes to be processed (these nodes usually correspond to sub-content with a certain hierarchy in the file, such as sub-paragraphs under chapters, details under clauses, etc.).

[0055] For each filter node in this set, perform the following operations:

[0056] Extract all leaf nodes (i.e., nodes without child nodes in the DOM tree, usually corresponding to specific text fragments such as sentences, phrases, data items, etc.) from the subtree corresponding to the current node, and obtain the text block content of each leaf node; arrange these text blocks according to their character offset order in the original archive file to form an ordered text block sequence. Here, "original character offset order" refers to the actual order in which the text blocks appear in the archive file. For example, the text blocks that appear first in the archive are placed at the beginning of the sequence, and the text blocks that appear later are placed at the end.

[0057] Calculate the spacing between two adjacent text blocks in the ordered sequence. That is, for any two consecutive text blocks in the sequence (the preceding text block and the following text block), the spacing value is equal to the starting character offset of the following text block in the original file minus the ending character offset of the preceding text block in the original file. For example, if the preceding text block ends at the 100th character position in the original file and the following text block starts at the 105th character position, then the spacing value between the two is 5.

[0058] Iterate through the spacing values ​​of all adjacent text blocks and record the maximum spacing value corresponding to the current node.

[0059] The maximum spacing value of the current node is compared with a preset threshold (this threshold is set according to the document layout characteristics, for example, 10 characters for formal official documents, and 15 characters for handwritten scanned documents). If the maximum spacing value is less than the preset threshold, it means that the text blocks in the subtree of this node have strong overall coherence and belong to the same topic. Therefore, all text blocks in the sequence are merged into a complete text content to generate a sub-document. At the same time, the DOM node path corresponding to the sub-document (i.e., the hierarchical path from the root node to the current node, such as "document file > chapter 2 > section 2.3 > paragraph 3") and the maximum spacing value of the current node are recorded as the attribute information of the sub-document. Finally, a set of sub-documents consisting of all sub-documents that meet the conditions is output, along with the DOM node path and maximum spacing value corresponding to each sub-document.

[0060] The specific implementation process of step 23 above is as follows:

[0061] Using the set of sub-documents output in step 22 as the processing object, each sub-document is assigned a globally unique identifier (this identifier can be generated by combining timestamps, random strings, node feature codes, etc., to ensure that it is not repeated throughout the entire archive processing system). Then, the globally unique identifier of each sub-document is associated with the corresponding DOM node path (such as "archive file > chapter 1 > section 1.2 > item 2") and maximum spacing value output in step 22 to form a one-to-one mapping relationship. At the same time, the number of the original natural resource archive from which the sub-document originates is obtained (i.e., the unique identifier of the archive in the target natural resource archive set in step 1, such as "land and resources archive 2023-001"). The source archive number is bound together with the aforementioned globally unique identifier, DOM node path, and maximum spacing value to form structured data containing "sub-document unique identifier, source archive number, DOM node path, and maximum spacing value", that is, the structured source metadata of the sub-document. Through this process, each sub-document can be clearly traced back to its position in the original archive, the textual coherence characteristics at the time of generation, and the source archive information.

[0062] This invention precisely quantifies the hierarchical structure of archive content (such as the nesting relationships of chapters, paragraphs, and clauses) by recursively calculating node depth. This provides a clear hierarchical basis for subsequent content splitting according to business logic, avoiding the content fragmentation problem caused by traditional fixed-length splitting. The DOM tree with depth annotation can intuitively reflect the internal structure of the archive, laying the foundation for filtering valuable mid-level and deep content (such as nodes with a depth ≥ 2), and improving the targeting of sub-document splitting. Focusing on nodes with a depth ≥ 2 ensures that the split sub-documents focus on specific business details (such as "approval conditions" and "boundary point descriptions"), avoiding interference from redundant information contained in root nodes or shallow nodes. By calculating the spacing between adjacent text blocks and judging coherence based on the maximum spacing value, it achieves... The system automatically identifies semantically complete units and merges text blocks when the spacing is less than a threshold, ensuring thematic consistency of sub-document content and solving the semantic breakage problem caused by traditional format-based splitting (such as by page or by line). It records DOM node paths and maximum spacing values, providing structured "content source" and "coherence feature" tags for sub-documents. A globally unique identifier ensures that each sub-document can be accurately located, avoiding duplication or confusion of archive content. It associates DOM node paths, maximum spacing values, and source archive numbers to construct a complete traceability chain of "sub-document - original location - content features - source archive". The structured traceability metadata provides a standardized index for cross-archive content association (such as the description of the same plot in different archives).

[0063] In a preferred embodiment of the present invention, step 3 includes:

[0064] Step 31: Using the structured traceability metadata generated in step 23 as input, input the associated sub-documents into the preset natural resource domain big model; through the domain adaptation decoding layer of the big model, extract three core entities: land parcel number, ownership subject, and spatial use category, and identify the ownership relationship of "ownership subject-land parcel" and the functional relationship of "land parcel-spatial use".

[0065] Step 32: Based on the entities and relationships output in Step 31, extract three types of attributes for each land parcel entity: area value, coordinate system type, and certificate issuance date; convert the certificate issuance date into a timestamp format, which, together with the coordinate system type, constitutes a dual spatial-temporal constraint.

[0066] Step 33: Standardize and encapsulate the structured features output in Step 32, outputting a set of structured metadata containing entities, relationships, and constraints. The set of structured metadata includes:

[0067] Entity dimension: Generates a triple {entity type, entity identifier, attribute set};

[0068] Relationship dimension: Generates a triple {subject entity identifier, relationship type, object entity identifier};

[0069] Constraint Dimension: Bind the spatial and temporal dual constraints to the corresponding land parcel entity identifier.

[0070] In this embodiment of the invention, the specific implementation process of step 31 above is as follows:

[0071] First, the sub-document content in the structured traceability metadata generated in step 23 is input one by one into the preset large model of the natural resources domain. This large model has been adapted and trained for the natural resources domain, and its domain adaptation decoding layer includes specific entity recognition and relation extraction modules.

[0072] For each input sub-document, the large model performs the following operations:

[0073] By combining domain dictionary matching with contextual semantic analysis, the unique identifier of the land parcel in the document is identified by number information; by using named entity recognition (NER) technology and combining domain knowledge graph (such as the names of organizations or natural persons such as "Natural Resources Center", "XX Company", "Zhang San" etc.), the subject with rights to the land parcel is identified; based on the preset spatial use classification system (such as "residential land", "commercial service facilities land", "industrial land", etc.), the land parcel's use function is determined by text classification algorithm.

[0074] By analyzing verb phrases (such as "own," "hold," "registered under") in the text that describe the relationship between the owner and the land parcel, and combining grammatical dependency analysis, the attribution relationship of "owner-land parcel" is determined. By identifying verb structures (such as "for construction," "planned for") or prepositional phrases (such as "used as") that describe the use of the land parcel, the functional relationship of "land parcel-spatial use" is established. Finally, the large model outputs structured data containing three core entities: land parcel number, owner, and spatial use category, as well as two types of relationships: "owner-land parcel" and "land parcel-spatial use".

[0075] The pre-defined process for constructing a large-scale model in the natural resources domain can be divided into six core stages: domain data construction, basic model selection, domain adaptation training, multimodal fusion, spatiotemporal constraint integration, and entity relationship enhancement. The specific implementation is as follows:

[0076] Multi-source data collection, integrating three core data types:

[0077] Unstructured documents (such as PDFs and Word documents) including natural resource approval documents, ownership certificates, and laws and regulations were collected to build a text library containing over 100,000 documents. Geographic information data such as land parcel vector boundary coordinates and administrative division maps were collected to form a spatial feature library with a coordinate system. Timestamp information such as issuance date and approval time was extracted to build a time-series labeled dataset. Over 10,000 samples were manually labeled to define three types of entity labels: land parcel number, ownership subject, and spatial use, as well as two types of relationship labels: ownership attribution and use association. Spatial descriptions in the text (such as "east to XX Road") were aligned with vector coordinates to establish a "text-space" mapping relationship. A natural resource knowledge graph containing over 500,000 triples was constructed, covering semantic relationships such as land parcel-policy-department, and used as an external knowledge base. For example, the "land use pre-approval" policy was linked to the "Natural Resources Bureau" approval process node.

[0078] Universal model base:

[0079] Using BERT-wwm-ext as the base language model, it achieves an F1 score of 92% in Chinese entity recognition tasks and supports long text processing (up to 512 tokens). For remote sensing image processing, the CLIP model is integrated to construct a multimodal encoder, achieving joint text-image representation. For example, land parcel description text and satellite image features are aligned through contrastive learning. For vector coordinates, PointNet++ is used to extract geometric features, converting the coordinate sequence into a 128-dimensional vector, which is then concatenated with text embeddings and input into the model. The certificate issuance date is converted into a timestamp (e.g., "2023-01-01" → 1672531200) and injected into the model through location encoding. ROIAlign technology is used to extract key inflection points of land parcel coordinates, generating geometric feature vectors, which are then fused with text semantic features.

[0080] A natural resources domain dictionary containing over 20,000 terms (such as "boundary point" and "land use pre-approval") was constructed. The dictionary was further pre-trained on domain texts using the MLM (masked language model) task, which improved the model's semantic understanding accuracy of professional terms by 30%. The Denoising Autoencoder was used to denoise spatial coordinates, remove redundant coordinate points caused by measurement errors, and retain geometric contour features. Entity recognition (NER) and relation extraction (RE) were jointly trained, and the CasRel framework was used to handle overlapping relationships, such as simultaneously recognizing the relationship between "ownership subject-land parcel" and "land parcel-spatial use".

[0081] Predict the intersection-over-union (IoU) ratio of two land parcel coordinates to supervise the model's learning of spatial correlation; determine the chronological order of certificate issuance dates to improve the logical understanding of the time dimension; during the training phase, use triples in the knowledge graph (such as "Land Parcel A - Approving Authority - Natural Resources Bureau") as soft constraints, and guide the model to generate relationships that conform to business logic through PromptTuning; use GraphConvolutionalNetwork (GCN) to encode the structural information of the knowledge graph into node embeddings and fuse them with text embeddings.

[0082] The FuzzyWuzzy library is used to implement fuzzy matching to solve the problem of "different names for the same place" (such as "No. 123, XX Road" and "No. 123-1, XX Road"). The BiLSTM-CRF layer is used to capture entity boundaries, such as identifying "XX City Natural Resources and Planning Bureau" as a complete organizational entity. StanfordCoreNLP is used to extract the subject-verb-object structure of sentences, such as parsing the "ownership subject-land parcel" relationship from "This land parcel is held by XX Company".

[0083] Geometric features of land parcel coordinates (such as area and shape) are used as auxiliary features for relationship extraction. For example, the relationship between "land parcel-containing-sub-land parcel" needs to be combined with spatial inclusion judgment. Spatiotemporal gating units (ST-Gates) are added to the model output layer to dynamically adjust the confidence of entity relationships based on timestamps and coordinate features. For example, if two land parcels have overlapping coordinates but their certificate issuance dates differ by 10 years, the probability of a "same land parcel" relationship is automatically reduced.

[0084] The model employs F1 score (target ≥90%) and boundary accuracy (coordinate point matching error ≤5 meters); relation extraction: Hits@10 is used to evaluate the accuracy of triple prediction, requiring the accuracy of "ownership subject-land parcel" relationship ≥92%; in the spatial alignment task, when the IoU threshold is set to 0.8, the cross-archive land parcel matching accuracy needs to be ≥90%; adversarial sample generation is introduced, such as inserting interference terms into the text (e.g., changing "industrial land" to "work-to-use land"), to improve model robustness; temperature scaling is used to calibrate the model output probability to reduce misjudgments caused by "different names for the same place".

[0085] The domain dictionary and knowledge graph are regularly updated, and continuous pretraining is used to prevent the model from forgetting new knowledge. For example, when adding terms related to "ecological red line," the F1 score drop is controlled within 2%. Model distillation is used to compress model parameters to 1 / 4 of the original volume, improving inference speed by 3 times while maintaining an F1 score of ≥88%. During deployment, NVIDIA TensorRT is used to accelerate inference, and the processing time for a single file is ≤2 seconds. Services are provided through a RESTful API, supporting batch processing (1000 files / hour) and returning structured metadata (such as land parcel number, ownership subject, and coordinate set). It interfaces with GIS platforms (such as ArcGIS) to achieve visualization analysis of spatial relationships between land parcels (such as highlighting adjacent land parcels). Through the above construction process, the model achieves an entity recognition F1 score of 91.2%, a relationship extraction accuracy of 90.5%, and a spatial alignment accuracy of 92.3% in natural resource archive processing tasks, significantly outperforming traditional rule matching methods (accuracy improvement of more than 40%), effectively supporting intelligent retrieval and association analysis of natural resource archives.

[0086] The specific implementation process of step 32 above:

[0087] Based on the entity and relationship data output in step 31, the attributes of each identified land parcel entity are further extracted and processed:

[0088] Area value extraction:

[0089] Locate the numerical descriptions related to the plot number in the sub-document, and extract the area value by matching quantifiers (such as "square meters", "hectares", "mu") and combining numerical regular expressions; if there is a unit conversion (such as "10,000 square meters (equivalent to 15 mu)"), convert it to the standard unit (such as square meters).

[0090] Coordinate system type identification:

[0091] By matching text with a pre-defined list of coordinate system names (such as "2000 National Geodetic Coordinate System", "Area A 54 Coordinate System", "Area B 80 Coordinate System"), the reference system used to describe the coordinates of the land parcel is identified; if it is not explicitly mentioned in the document, the possible coordinate system type is inferred from the coordinate numerical characteristics (such as latitude and longitude range, projection zone number).

[0092] Processing of certificate issuance date:

[0093] Identify the date expressions related to plot rights in the document (such as "registration date", "issuance date", "approval date"), and extract the specific date through date regular expressions (such as "YYYY-MM-DD", "YYYY年MM月DD日"); convert the extracted date string into the standard timestamp format (such as Unix timestamp) for subsequent time dimension comparison and calculation.

[0094] Spatial-temporal dual constraint construction:

[0095] Combine the standardized coordinate system type (such as "CGCS2000") with the timestamp (such as 1640995200) to form the "spatial-temporal dual constraint" attribute of each plot entity, which is used to define the valid time range of the plot under a specific coordinate system.

[0096] The specific implementation process of step 33 above:

[0097] Standardize and encapsulate the structured features output in step 32 to form a set of structured metadata in a unified format:

[0098] Entity dimension encapsulation:

[0099] Generate a triple {entity type, entity identifier, attribute set} for each identified entity (plot, ownership subject, spatial use category).

[0100] Entity type: Clearly marked as "plot", "ownership subject", or "spatial use category".

[0101] Entity identifier: Assign a unique identifier (such as UUID) to each entity to ensure the traceability of cross-document entities.

[0102] Attribute set:

[0103] Plot entity: Includes attributes such as area value, coordinate system type, timestamp (issuance date), etc.

[0104] Ownership subject: Includes attributes such as name, type (organization / individual), etc.

[0105] Spatial use category: Includes attributes such as category code, Chinese name, etc.

[0106] Relationship dimension encapsulation:

[0107] Generate a triple {subject entity identifier, relationship type, object entity identifier} for each identified relationship.

[0108] Subject entity identifier: Points to the entity identifier of the relationship initiator.

[0109] Relationship type: Marked as "ownership relationship" or "functional relationship".

[0110] Object entity identifier: An entity identifier that points to the recipient of the relationship.

[0111] Constraint dimension binding:

[0112] The "spatial-temporal dual constraint" attribute generated in step 32 is bound to the corresponding land parcel entity identifier to form a mapping relationship of {land parcel entity identifier, spatial constraint (coordinate system type), temporal constraint (time stamp)}. The final output structured metadata set integrates information from the three dimensions of entity, relationship and constraint.

[0113] This invention, through a pre-defined large-scale model in the field of natural resources and an adaptive decoding layer, accurately extracts specialized entities such as land parcel numbers, ownership entities, and spatial use categories. Compared to general models, it improves the recognition accuracy of terms in fields such as "real estate certificate number" and "commercial service facility land" by more than 30%. It automatically identifies two key relationships: "ownership entity-land parcel" and "land parcel-spatial use," linking discrete entities into a semantic network, providing a relational foundation for subsequent knowledge graph construction, and supporting complex searches such as "querying all commercial land under a certain company's name." It uniformly extracts three attributes: area value, coordinate system type, and certificate issuance date. The date is converted to a timestamp format, resolving the inconsistency in date formats across different archives (e.g., "January 1, 2023" versus "2023-01-01"). It improves data comparability; by combining coordinate system type with timestamps to form "spatial-temporal dual constraints," it assigns spatiotemporal validity limits to land parcel entities (such as the validity period of a land parcel under the 2000 coordinate system), solving the problem of coordinate system uniformity in historical archives and supporting cross-time spatial analysis of land parcels; through standardized encapsulation of the three dimensions of entities, relationships, and constraints, it forms a unified format of triplet sets, providing a standardized interface for subsequent knowledge graph construction and reducing data integration costs; by binding spatiotemporal constraints with land parcel entity identifiers, it enables the attributes and relationships of each land parcel to have spatiotemporal validity verification capabilities, avoiding the "static" defects of entity relationships in traditional knowledge graphs, and supporting spatiotemporally sensitive queries such as "querying industrial land registered before 2020 using the Beijing 54 coordinate system."

[0114] In summary, step 3 achieved the transformation of natural resource archives from "textual information" to "structured knowledge" through the domain-wide model. In particular, the innovative "space-time dual constraint" mechanism solved the problems of multi-source and inconsistency in the spatiotemporal dimensions of archive data, and significantly improved the knowledge graph's ability to express complex natural resource information and its query flexibility.

[0115] In a preferred embodiment of the present invention, step 4 includes:

[0116] Step 41: Using the structured metadata set output in Step 3 as input, parse the entity identifier, entity type and attribute set in the entity dimension to identify the core business entity type in the natural resources field. Among them, the land parcel entity type is identified as a key entity containing a sequence of vector boundary coordinate points, and the ownership entity type is also identified.

[0117] Step 42: Taking the business entity type identified in Step 41 as the processing object, for each land parcel entity, extract the vector boundary coordinate point sequence from the structured metadata set as the core spatial attribute of the land parcel entity, and add the coordinate point sequence to the attribute set of the land parcel entity to form an updated land parcel entity attribute set.

[0118] Step 43: Based on the updated set of land parcel entity attributes from Step 42 and the relational and constraint dimensions output from Step 3, construct a set of RDF triples with attribute constraints, where:

[0119] Each attribute value in the entity attribute set is converted into a triple with the entity identifier as the subject, the attribute name as the predicate, and the attribute value as the object;

[0120] Each relation in the relation dimension is transformed into a triple with the subject entity identifier as the subject, the relation type as the predicate, and the object entity identifier as the object;

[0121] The spatial-temporal dual constraints in the constraint dimension are bound to the corresponding land parcel entity identifier, generating a triple with the land parcel entity identifier as the subject, the constraint type as the predicate, and the constraint value as the object.

[0122] In this embodiment of the invention, the specific implementation process of step 41 above is as follows:

[0123] First, using the structured metadata set output in step 3 as input, the entity dimension information is parsed from it. For each entity, its unique entity identifier, predefined entity type (such as "land parcel", "ownership entity", "spatial use category"), and a set of attributes containing specific characteristics are extracted.

[0124] During the parsing process, the core business entity types are identified by matching them with a pre-defined natural resource domain ontology:

[0125] 1. Plot Entity: When the entity type field is explicitly marked as "plot", and its attribute set contains the attribute "vector boundary coordinate point sequence" (such as the coordinate data extracted from the domain large model in step 3), then the entity is determined to be a plot entity type.

[0126] 2. Ownership Subject Entity: When the entity type field is marked as "ownership subject" and the attribute set contains features such as "name" and "type (organization / individual)", it is determined to be an ownership subject entity type.

[0127] In addition, other possible business entity types (such as space use categories) will be identified, but land parcel entities and ownership entities will be the focus of subsequent processing. The final output is the identified business entity types and their corresponding entity identifier sets.

[0128] The specific implementation process of step 42 above is as follows:

[0129] For the land parcel entity types identified in step 41, their core spatial attributes are further extracted from the structured metadata set.

[0130] For each land parcel entity:

[0131] In the structured metadata set output in step 3, locate the attribute set corresponding to the land parcel entity; extract the attribute value of "vector boundary coordinate point sequence", which is usually an ordered list of multiple coordinate points (such as latitude and longitude pairs) that describes the polygonal boundary of the land parcel; add the extracted vector boundary coordinate point sequence to the attribute set of the land parcel entity to replace or supplement the original potentially incomplete spatial information, forming an updated attribute set of the land parcel entity.

[0132] For example, if the original attribute set of a land parcel entity includes information such as area and coordinate system type, this step will supplement it with a complete "vector boundary coordinate point sequence," making its attribute set more comprehensively describe the spatial characteristics of the land parcel. Finally, the updated attribute set of all land parcel entities is output.

[0133] The specific implementation process of step 43 above is as follows:

[0134] Based on the updated set of land parcel entity attributes in step 42, and the relational and constraint dimension information output in step 3, construct a set of RDF triples with attribute constraints.

[0135] Attribute triple generation:

[0136] For each entity (including land parcel entities, ownership entities, etc.), iterate through each attribute value in its attribute set and generate a triple with the entity's unique identifier as the subject, the attribute name as the predicate, and the attribute value as the object. For example, if the land parcel entity ID is "land parcel_001" and its attribute "area value" is 10,000 square meters, then generate the triple: (land parcel_001, area value, 10,000 square meters).

[0137] Relational triple generation:

[0138] For each relation in the relation dimension of step 3, extract its subject entity identifier, relation type, and object entity identifier; generate a triple with the subject entity identifier as the subject, the relation type as the predicate, and the object entity identifier as the object; for example, if there is a relation "Company A-Owns-Land Plot_001", then generate the triple: (Company A, Owns, Land Plot_001).

[0139] Constrained triple generation:

[0140] For each spatial-temporal dual constraint in the constraint dimension of step 3, bind it to the corresponding land parcel entity identifier; generate a triple with the land parcel entity identifier as the subject, the constraint type (such as "coordinate system type" or "timestamp") as the predicate, and the constraint value as the object; for example, if the coordinate system type of land parcel_001 is "2000 National Geodetic Coordinate System" and the timestamp is 1640995200, then generate two triples: (land parcel_001, coordinate system type, 2000 National Geodetic Coordinate System) and (land parcel_001, timestamp, 1640995200).

[0141] Finally, all generated attribute triples, relation triples, and constraint triples are integrated to form a set of RDF triples with attribute constraints.

[0142] This invention accurately identifies two core business entities in the natural resources field—land parcels and ownership entities—by parsing entity dimension information. It uses "a sequence of vector boundary coordinate points" as a key identifier for land parcel entities, strengthening the core position of spatial features in natural resource management. Incorporating the vector boundary coordinate point sequence into the land parcel entity attribute set transforms land parcel description from "textual" to "spatial," achieving a fusion of semantic and geometric information. Complete vector boundary data provides the necessary conditions for spatial overlay operations in subsequent steps (such as calculating overlapping areas of land parcels), breaking through the limitation of traditional knowledge graphs that can only handle textual relationships. It uniformly converts entity attributes, relationships, and constraints into RDF triple format, forming a standardized knowledge expression framework and providing a unified interface for knowledge graph construction. By binding spatial-temporal dual constraints, triples not only express "what" but also define "when it is effective" and "under what spatial reference it is effective," enhancing the semantic expression capability of knowledge graphs for complex natural resource information.

[0143] In a preferred embodiment of the present invention, step 5 includes:

[0144] Step 51: Using the RDF triplet set constructed in Step 4 as input, process the vector boundary coordinate point sequence corresponding to each land parcel entity; initialize the first and last points of the coordinate point sequence as key points to form an initial key point set.

[0145] Step 52: Based on the initial set of key points determined in Step 51, recursively divide the line segment formed by the first and last points, that is, traverse all non-key coordinate points between the first and last points of the line segment, calculate the vertical distance from each point to the line segment, and identify the point with the maximum vertical distance.

[0146] Step 53: When the maximum vertical distance value identified in step 52 exceeds the preset distance threshold, the corresponding point is marked as a new key point and added to the key point set; then, the original line segment is divided into two sub-segments with the new key point as the boundary, and the maximum vertical distance calculation and key point determination process described in step 52 is recursively executed for each sub-segment.

[0147] Step 54: After the recursive process terminates, merge all the key points generated by the level determination, arrange them in the order of the original coordinate sequence, and generate a set of denoised key coordinate points as the simplified boundary representation of the land parcel entity.

[0148] In this embodiment of the invention, the specific implementation process of step 51 is as follows:

[0149] From the RDF triples generated in step 4, filter out all land parcel entities that contain vector boundary coordinate point sequences. The coordinate point sequence of each land parcel entity is arranged in the order recorded in the original file (e.g., clockwise / counterclockwise coordinate points from the start point to the end point of the land parcel boundary). For the coordinate point sequence of each land parcel entity, directly mark the first point (first point) and the last point (last point) in the sequence as initial key points. These two points are the basic endpoints that constitute the land parcel boundary, ensuring that the start and end positions of the boundary contour are not lost. Form an initial key point set with the first point and the last point. At this time, the set only contains these two coordinate points, which serve as the starting reference for subsequent recursive partitioning.

[0150] The specific implementation process of step 52 above:

[0151] Using the first and last points in the initial key point set determined in step 51 as endpoints, a virtual line segment (i.e., a straight line connecting two points) is formed. Traverse the original coordinate point sequence corresponding to this line segment, and find all non-key coordinate points other than the first and last points (these points are the original record points in the middle of the line segment). For each non-key coordinate point, calculate its vertical distance to the virtual line segment (i.e., the shortest straight line distance between the point and the line segment, reflecting the degree of deviation of the point from the line segment). Among all the calculated vertical distances, find the distance value with the largest value and record the corresponding coordinate point. This point is the point in the current line segment that deviates most significantly from the base line segment.

[0152] The specific implementation process of step 53 above:

[0153] The maximum vertical distance value identified in step 52 is compared with a preset distance threshold (this threshold is set according to the accuracy requirements of the land parcel, such as 0.5 meters for urban construction land and 2 meters for rural collective land). If the maximum vertical distance value exceeds the threshold, it means that the point has a significant impact on the shape of the land parcel boundary outline and needs to be retained. Therefore, the point is marked as a new key point and added to the key point set. Using this new key point as the dividing point, the original first-end line segment is divided into two new sub-segments: one is the line segment from the first point to the new key point, and the other is the line segment from the new key point to the end point. For the two sub-segments after division, the operation of step 52 is repeated, traversing the non-key coordinate points within the range of each sub-segment, calculating the vertical distance from each point to the corresponding sub-segment, and finding the point with the maximum vertical distance.

[0154] The specific implementation process of step 54 above:

[0155] The recursive process of steps 52 and 53 is repeated continuously. After each segmentation, the sub-segments are checked for points that deviate beyond a threshold. If any are found, the segmentation continues and key points are added until the vertical distance from the remaining non-key coordinate points in all sub-segments to their corresponding segments is less than the preset threshold, at which point the recursive process terminates. All coordinate points marked as key points during the recursive process are collected (including the initial first and last points, as well as the key points added in each level of segmentation). These key points are arranged according to their order in the original coordinate point sequence (keeping the direction consistent with the original boundary, such as clockwise order), forming the final set of key coordinate points. This set of key coordinate points removes redundant points with small deviations from the original sequence (such as small fluctuations caused by measurement errors). While maintaining the overall geometry of the land parcel boundary (such as corners and protrusions), the coordinate data is simplified, serving as a simplified boundary representation of the land parcel entity.

[0156] This invention uses the first and last points as initial references and recursively calculates vertical distances to filter out points exceeding a threshold as key points. This process preserves the main turning points and protruding parts of the land parcel boundary during simplification, ensuring that the simplified boundary accurately reflects the basic shape characteristics of the land parcel. By eliminating non-key coordinate points close to line segments, the number of coordinate points describing the land parcel boundary is significantly reduced, lowering the pressure on data storage and processing, and making subsequent analysis and transmission of land parcel entities more efficient. Minor fluctuations caused by measurement errors or redundant records are filtered out, making the land parcel boundary representation simpler and smoother, avoiding interference from these irrelevant points in land parcel shape judgment and related applications (such as land parcel area calculation and boundary comparison). Finally, the key points are arranged in the original coordinate sequence, ensuring the logical consistency between the simplified boundary coordinate point set and the original sequence, without changing the overall orientation and relative positional relationship of the land parcel boundary, thus ensuring the reliability of the simplification result.

[0157] In a preferred embodiment of the present invention, step 6 includes:

[0158] Step 61: Using the set of key coordinate points of each parcel entity after noise reduction output in Step 5 as input, select two parcel entities to be subjected to spatial relationship analysis to form an entity pair;

[0159] Step 62: For the entity pairs selected in step 61, reconstruct the corresponding first polygon boundary and second polygon boundary using their respective key coordinate point sets.

[0160] Step 63: Perform vector geometric superposition operation on the first polygon boundary and the second polygon boundary reconstructed in Step 62, calculate the overlapping region polygon formed by the intersection of the two polygons, and calculate the area value of the overlapping region polygon, the area value of the first polygon boundary, and the area value of the second polygon boundary respectively. This includes: Step 631: Based on the key coordinate points of the first polygon boundary reconstructed in Step 62, traverse all key coordinate points of the first polygon boundary, identify the point set located on the outermost boundary, connect the point set on the outermost boundary to form the minimum convex polygon contour that encloses all internal points, and generate the first candidate convex hull; Step 632: Based on the key coordinate point set of the second polygon boundary reconstructed in Step 62, traverse all key coordinate points of the second polygon boundary, identify the point set located on the outermost boundary, and connect the point set on the outermost boundary to form the minimum convex polygon contour that encloses all internal points. The minimum convex polygon contour is used to generate a second candidate convex hull; Step 633: Calculate the area value of the first candidate convex hull, calculate the original area value of the key coordinate point set of the first polygon boundary, calculate the absolute difference between the area of ​​the first candidate convex hull and the original area, and calculate the area error rate. When the area error rate is less than the preset convex hull simplification tolerance threshold, the first candidate convex hull is determined to be valid; otherwise, the key coordinate point set of the first polygon boundary is used as the first candidate convex hull, and the same validity verification process is performed on the second candidate convex hull, outputting the final valid first convex hull boundary and the second convex hull boundary; Step 634: Calculate the overlapping region polygon formed by the intersection of the final valid first convex hull boundary and the second convex hull boundary; Step 635: Calculate the area value of the overlapping region polygon obtained in Step 634, the area value of the first convex hull boundary output in Step 633, and the area value of the second convex hull boundary, respectively.

[0161] Step 64: Based on the area value calculated in Step 63, calculate the intersection-union ratio, where the intersection-union ratio = area of ​​overlapping region / (area of ​​first polygon + area of ​​second polygon - area of ​​overlapping region).

[0162] Step 65: Compare the cross-union ratio calculated in step 64 with the preset cross-union ratio threshold;

[0163] Step 66: When the intersection-union ratio exceeds the preset intersection-union ratio threshold, it is determined that the two land parcel entities in the entity pair have a spatial overlap relationship. The entity spatial alignment operation is performed, that is, a semantic triple representing the spatial alignment relationship is generated. The main entity is the first land parcel entity identifier, the predicate is the preset spatial alignment relationship type, and the object entity is the second land parcel entity identifier.

[0164] Step 67: Collect all semantic triples generated in Step 66 that satisfy the intersection-union ratio threshold condition and have completed the spatial alignment operation to form a set of spatially aligned triples.

[0165] In this embodiment of the invention, step 61, selecting the entity pair, may include:

[0166] Extract the coordinate point set: From all the land parcel entities output in step 5, extract the denoised key coordinate point set for each land parcel entity. Each coordinate point set is an ordered list containing simplified land parcel boundary coordinate points (e.g., [(x1,y1),(x2,y2),...]). Based on business requirements (e.g., land parcel merging, ownership verification, spatial relationship analysis), determine the scope of land parcels requiring spatial relationship analysis. For example:

[0167] To verify changes in land parcels within the same area at different times, select entity pairs with the same parcel number from adjacent years. To identify overlapping land parcels, select parcels with different ownership entities within the same area. From the extracted land parcel entities, select two parcels to construct entity pairs. For example, select "Parcel A" and "Parcel B" as the entity pair for the current analysis and record their key coordinate point set.

[0168] Step 62 above, polygon boundary reconstruction may include:

[0169] The coordinate points are sorted by arranging the set of key coordinate points for each plot in the entity pair selected in step 61 according to the original sequence order (e.g., clockwise). This ensures that the coordinate points, when connected, can correctly form a closed polygon.

[0170] Constructing a closed polygon:

[0171] For the first plot of land (e.g., "Plot A"), connect the sorted coordinate points sequentially, with the last point closing the boundary with the first point, forming the boundary of the first polygon; for the second plot of land (e.g., "Plot B"), perform the same operation to form the boundary of the second polygon, and check whether the reconstructed polygon satisfies the following conditions:

[0172] It must contain at least 3 coordinate points (forming a basic polygon), and all edges must not intersect (a simple polygon). If edges intersect, corrections must be made (such as adjusting the order of coordinate points or deleting points that cause intersections).

[0173] Step 63 above, vector geometric superposition operation and area calculation, includes:

[0174] Step 631, First candidate convex hull generation:

[0175] Outermost point identification: Traverse all key coordinate points of the first polygon boundary. For each point P:

[0176] Check if all other points are on the same side of a line passing through point P. If such a line exists, then P is the outermost point.

[0177] Convex hull construction:

[0178] Collect all outermost points, arrange them in the order of the original coordinate point set, and connect these points in sequence to form a closed convex polygon, which is the first candidate convex hull.

[0179] Step 632 above, generating the second candidate convex hull, includes:

[0180] Repeat step 631 to process the key coordinate point set of the second polygon boundary and generate the second candidate convex hull.

[0181] Step 633: Convex Hull Validation

[0182] Area calculation:

[0183] Calculate the area of ​​the first candidate convex hull (using the coordinates of the convex hull boundary points), and calculate the area of ​​the original coordinate point set of the first polygon boundary.

[0184] Error rate calculation:

[0185] Calculate the absolute difference in area: |convex hull area - original area|.

[0186] Calculate the area error rate: (absolute difference / original area) × 100%.

[0187] Validity assessment:

[0188] If the error rate is less than a preset threshold (e.g., 5%), the first candidate convex hull is valid. If the error rate is greater than or equal to the threshold, the candidate convex hull is discarded, and the boundary of the first polygon is directly used as the boundary of the first convex hull. The same verification process is performed on the second candidate convex hull to determine the final valid boundary of the second convex hull.

[0189] Step 634 above, determining the overlapping region polygon, may include:

[0190] Line segment intersection detection:

[0191] Traverse each line segment (edge) of the first convex hull boundary and check whether it intersects with all line segments of the second convex hull boundary. If the two line segments intersect, calculate the coordinates of the intersection point, collect all intersection points, and the boundary points that are simultaneously located inside the two convex hulls. Arrange the selected points in clockwise or counterclockwise order (the direction can be determined by the cross product), and connect these points to form a closed polygon, i.e., the overlapping region polygon.

[0192] Step 635 above, the area calculation includes:

[0193] Area of ​​overlapping region: The area is calculated based on the coordinates of the overlapping polygon.

[0194] First convex hull area: If the first candidate convex hull is valid, its area is used; otherwise, the area of ​​the first polygon boundary is used.

[0195] Second convex hull area: Similarly, the area value to be used is determined based on the validity of the second candidate convex hull.

[0196] Step 64, the intersection-union ratio calculation may include:

[0197] Union area calculation:

[0198] Union area = Area of ​​first convex hull + Area of ​​second convex hull - Area of ​​overlapping region (to avoid double counting of overlapping parts).

[0199] Cross-Union Ratio Calculation:

[0200] Intersection over union ratio = (Area of ​​overlapping region / Area of ​​union) × 100%.

[0201] Step 65, the intersection-union ratio threshold comparison may include:

[0202] The intersection-union ratio calculated in step 64 is compared with a preset threshold (e.g., 80%) to determine whether the spatial alignment condition is met.

[0203] Step 66, entity space alignment and triple generation may include:

[0204] Alignment determination:

[0205] If the intersection-union ratio is greater than the threshold, it is determined that the two land parcel entities have a spatial overlap relationship and need to be aligned.

[0206] Triple generation:

[0207] Main entity: The unique identifier of the first plot of land (such as the plot number).

[0208] Predicate: Preset spatial alignment relationship type (such as "spatial overlap", "same plot", "containment", etc.).

[0209] Object entity: The unique identifier of the second plot of land.

[0210] Additional attributes:

[0211] Add metadata such as intersection-union ratio and overlapping area to triples to enhance the semantic expression of relations.

[0212] Step 67, the formation of the spatially aligned triplet set may include:

[0213] Results collection involves summarizing the semantic triples generated from all entity pairs that meet the threshold conditions in step 66.

[0214] The process involves deduplication, checking for duplicate triples (such as "A-spatial overlap-B" and "B-spatial overlap-A"), and retaining one or merging them.

[0215] The set construction involves storing the final processed triples in a unified format to form a spatially aligned set of triples for subsequent knowledge graph construction.

[0216] This invention accurately determines the spatial overlap between land parcel entities and improves the reliability of spatial relationship analysis through a series of operations, including reconstructing polygons based on key coordinate points, calculating convex hulls, and intersection-union ratios. It reduces data volume and improves computational efficiency by using a denoised set of key coordinate points. Furthermore, mechanisms such as convex hull validity verification ensure the accuracy of the analysis results. Spatial overlap relationships are presented as semantic triples, facilitating subsequent storage, retrieval, and further analysis and utilization of spatial relationships. Spatial alignment operations link potentially different land parcel entity boundary information, enhancing the usability of land parcel data in spatial analysis and integration scenarios.

[0217] In a preferred embodiment of the present invention, step 7 includes:

[0218] Step 71: Take the spatially aligned set of triples output in Step 6 as the first input and the set of RDF triples with attribute constraints constructed in Step 4 as the second input.

[0219] Step 72: Perform triple fusion processing on the first input and the second input provided in step 71, that is, identify triples in the two input sets that describe the same entity identifier and merge their attribute and relation information; for the spatial alignment relation triples generated in step 6, they are added as new relations and incorporated into the fused triple set.

[0220] Step 73: Using the triplet set fused in Step 72 as the basis for graph construction, map each unique entity identifier in the triplet to a node in the graph database, and map each entity attribute to the attribute key-value pair of the corresponding node.

[0221] Step 74: Based on the nodes created in Step 73, map the semantic relations defined in the triples and the spatial alignment relations added in Step 72 to directed edges in the graph database, that is, with the node corresponding to the subject entity identifier as the edge starting point, the relation type as the edge label, and the node corresponding to the object entity identifier as the edge ending point.

[0222] Step 75: For the land parcel entity nodes, attach the set of key coordinate points after noise reduction in Step 5 as spatial geometric attributes to the corresponding nodes.

[0223] Step 76: After completing the node, edge, and attribute mapping in steps 73 to 75, a knowledge graph in the natural resources domain is instantiated in the graph database. This graph contains nodes such as land parcel entities and ownership entities, as well as a semantic network structure composed of directed edges such as ownership relationships, functional relationships, and spatial alignment relationships.

[0224] In this embodiment of the invention, step 71 above, selecting the entity pair, may include:

[0225] Extract the set of spatially aligned triples. From the output of step 6, obtain all the triples that have completed spatial alignment. The structure of each triple is {subject entity identifier, spatial relationship type, object entity identifier}, such as "plot A - spatial overlap - plot B". Some triples also include attribute information such as intersection-union ratio (e.g., 0.85) and overlap area (e.g., 500 square meters).

[0226] Extract the set of RDF triples with attribute constraints. From the results of step 4, obtain all RDF triples with attribute constraints. These triples include:

[0227] Entity attribute information, such as "Plot A - Area - 1000 square meters" or "Plot A - Use - Industrial Land";

[0228] Semantic relationship information, such as “Company X - Ownership - Plot A” (ownership relationship) and “Plot A - Function - Industrial Production” (functional relationship), and some relationships are subject to constraints (such as “Approval Date: 2023-01-01”).

[0229] Step 72 above, the triplet fusion process may include:

[0230] Entity identifier matching:

[0231] Iterate through each entity identifier (such as "Land Parcel A" or "Company X") in the spatial alignment triples, and search for entities with the same identifier in the RDF triples with attribute constraints. For example, use the unique number of "Land Parcel A" (such as "G2023001") as the matching criterion to ensure that entities across sets are accurately associated.

[0232] Merging attributes and relationships:

[0233] For the same matched entity, merge all its attributes. For example:

[0234] In the RDF triple, "Plot A" has the attributes "Area: 1000 square meters" and "Use: Industrial land"; in the spatial alignment triple, "Plot A" has the associated attribute "Crossover-Union Ratio: 0.85"; after merging, the attribute set of "Plot A" is {Area: 1000 square meters, Use: Industrial land, Crossover-Union Ratio: 0.85}.

[0235] The original semantic relations in the RDF triples (such as "Company X-Ownership-Land Parcel A") are retained, while new relations in the spatial alignment triples (such as "Land Parcel A-Spatial Overlap-Land Parcel B") are added to the merged set to form a complete relation network.

[0236] Conflict resolution:

[0237] If the same attribute of the same entity is contradictory in two sets (e.g., “Land Parcel A - Use - Industrial Land” in the RDF triple, while “Use - Commercial Land” is displayed in the document associated with the spatial alignment triple), then the attribute of the RDF triple in step 4 shall prevail (because it is extracted from the core elements of the original archive and has higher authority), and the conflict information shall be recorded for subsequent manual verification.

[0238] Step 73 above, node mapping may include:

[0239] Traverse the merged set of triples, extract all unique entity identifiers, and remove duplicates; for example, select 4 unique entities from “Plot A”, “Company X”, “Plot B”, and “Policy Y”; classify the unique identifiers according to entity type, such as “Plot Entity” (Plot A, Plot B), “Ownership Entity” (Company X), and “Policy Entity” (Policy Y), and define a unified node label for each type (such as “Plot”, “Ownership Entity”, and “Policy”).

[0240] Node creation:

[0241] In a graph database, a node is created for each unique entity identifier. For example, a node labeled "Land Parcel" is created for "Land Parcel A", and a node labeled "Ownership Entity" is created for "Company X".

[0242] Attribute key-value pair mapping:

[0243] Convert the attribute information of each entity into attribute key-value pairs for nodes. For example:

[0244] The node attributes of “Plot A” are {Number: Plot A, Area: 1000 square meters, Use: Industrial land, Coordinate system: WGS84}; the node attributes of “Company X” are {Name: XX Co., Ltd., Unified Social Credit Code: 91XXXXXXXXXXX, Registered address: XX Road, XX City}.

[0245] Step 74 above, edge mapping may include:

[0246] Relationship extraction and classification:

[0247] Extract all semantic relations from the fused triples, including:

[0248] The original relationships in step 4 (such as "belonging relationship" and "functional relationship");

[0249] The spatial alignment relationships added in step 72 (such as "spatial overlap", "containment", "adjacent").

[0250] Define unique edge labels for each relationship type, such as "belonging", "function", and "spatial overlap".

[0251] Directed edge creation:

[0252] For each triple {subject entity identifier, relation type, object entity identifier}, create a directed edge in the graph database:

[0253] The starting point of an edge is the node corresponding to the entity identifier (e.g., the "Company X" node); the label of the edge is the relation type (e.g., "Attribution"); and the ending point of the edge is the node corresponding to the object entity identifier (e.g., the "Land Parcel A" node). For example, the triple "Company X-Attribution-Land Parcel A" corresponds to the edge "Company X node (starting point)-Attribution (label)-Land Parcel A node (ending point)".

[0254] Edge attribute addition:

[0255] If the relationship includes attributes (such as the intersection-union ratio in a spatial alignment relationship, or the date in an approval relationship), these attributes are used as key-value pairs for the edge. For example, the attributes of the edge "Plot A - Spatial Overlap - Plot B" are {Intersection-union ratio: 0.85, Overlap area: 500 square meters}.

[0256] Step 75 above, the addition of spatial geometric properties may include:

[0257] Coordinate point set matching:

[0258] From the output of step 5, extract the set of key coordinate points for each land parcel after noise reduction. For example, the set of key coordinate points for “Land Parcel A” is [(x1,y1),(x2,y2),(x3,y3)] (arranged in clockwise order).

[0259] Plot node attribute update:

[0260] In the graph database, find all nodes labeled "plot" and match the corresponding key coordinate point set using entity identifiers (e.g., "plot A"). Add the matched coordinate point set as a spatial geometric attribute to the attribute of the corresponding plot node. For example, add the attribute {key coordinate point set: [(x1,y1),(x2,y2),(x3,y3)], geometry type: polygon} to the "plot A" node.

[0261] Spatial index association:

[0262] Associate a spatial index (such as an R-tree index based on coordinate range) with the "key coordinate point set" attribute of the land parcel node to ensure that spatial queries can be executed quickly in the future (such as "query all land parcels in the XX area" or "find land parcels adjacent to land parcel A").

[0263] Step 76 above, knowledge graph instantiation may include:

[0264] Graph structure integration:

[0265] In the graph database, the nodes created in step 73, the directed edges created in step 74, and the spatial geometric properties added in step 75 are integrated to form a complete graph structure. For example:

[0266] The nodes include “Plot A”, “Plot B” (plot entities), “Company X” (ownership entity), and “Policy Y” (policy entity); the directed edges include “Company X-Ownership-Plot A” (ownership relationship), “Plot A-Function-Industrial Production” (functional relationship), and “Plot A-Spatial Overlap-Plot B” (spatial alignment relationship).

[0267] Integrity verification:

[0268] Check that all nodes have at least one associated edge (to avoid isolated nodes). For example, if "Plot C" has no associated edges, trace back its original triples in steps 4 and 6 and supplement the missing relationships (such as "Company Y - Ownership - Plot C"). Check that all plot nodes have been appended with key coordinate point sets. If any are missing, rematch the results of step 5 and supplement them. Create indexes for frequently queried fields (such as plot number, ownership entity name, and spatial relationship type) to improve query speed. Partition the graph according to business scenarios (such as dividing it into "XX District Sub-graph" and "XX County Sub-graph" according to administrative regions) for easy management and expansion. After completing all the above mappings and verifications in the graph database, the knowledge graph of the natural resources domain is officially generated. This graph is a semantic network containing nodes (entities), directed edges (relationships), and attributes, which can directly support complex queries (such as "find all plots in XX City that overlap with industrial land and their ownership entities") and association analysis (such as "the impact path of policy Y on the land use change of plot A").

[0269] like Figure 2 As shown, a natural resource archive retrieval and utilization system combining large-scale modeling and knowledge graph technology includes:

[0270] The acquisition module is used to acquire a target natural resource archive set, wherein the archive set contains at least one natural resource archive to be processed;

[0271] The parsing module is used to parse the document object model topology based on the target natural resource archive set, and decompose it into corresponding sub-document sets according to the directory hierarchy rules;

[0272] The execution module is used to perform deep semantic understanding and structured information extraction on the sub-document collection through a pre-set large model of natural resources, and output structured metadata describing the core elements of the archive.

[0273] The building module is used to identify business entities and their semantic relationships in the natural resources domain based on structured metadata, and to construct RDF triples with attribute constraints; wherein, the identified business entities include land parcel entities with vector boundary coordinate point sequences;

[0274] The noise reduction module is used to perform geometric feature-preserving coordinate simplification on the sequence of boundary coordinate points of land parcel entity vectors using the Douglas-Puk compression algorithm, generating a set of noise-reduced key coordinate points.

[0275] The alignment module is used to perform vector geometric superposition operations on the polygonal boundaries of two land parcel entities based on the denoised key coordinate point set, calculate the area ratio of the overlapping area according to the intersection-union ratio, and perform entity spatial alignment when the intersection-union ratio exceeds a preset threshold to obtain a set of spatially aligned triples.

[0276] The fusion module is used to fuse spatially aligned sets of triples to construct a knowledge graph in the field of natural resources.

[0277] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for retrieving and utilizing natural resource archives that combines large-scale modeling and knowledge graph technology, characterized in that, The method includes: Step 1: Obtain the target natural resource archive set, which contains at least one natural resource archive to be processed; Step 2: Based on the target natural resource archive set obtained in Step 1, parse the document object model topology and generate corresponding sub-document sets according to the directory hierarchy rules; Step 3: Using a pre-defined large model of natural resources, perform deep semantic understanding and structured information extraction on the sub-document set generated in Step 2, and output structured metadata describing the core elements of the archive. Step 4: Based on the structured metadata output in Step 3, identify business entities in the natural resources domain and their semantic relationships, and construct RDF triples with attribute constraints; wherein, the identified business entities include land parcel entities with vector boundary coordinate point sequences; Step 5: The sequence of boundary coordinate points of the land parcel entity vector extracted in Step 4 is simplified by geometric feature preservation using the Douglas-Puk compression algorithm to generate a set of denoised key coordinate points. Step 6: Based on the key coordinate point set after noise reduction in Step 5, perform vector geometric superposition operation on the polygonal boundaries of the two land parcel entities, calculate the area ratio of the overlapping area according to the intersection-union ratio, and when the intersection-union ratio exceeds the preset threshold, perform entity spatial alignment to obtain a set of spatially aligned triplets. Step 7: Merge the spatially aligned triples from Step 6 to construct a knowledge graph for the natural resources domain.

2. The method for retrieving and utilizing natural resource archives combining large-scale models and knowledge graph technology according to claim 1, characterized in that, Step 2 includes: Step 21: Based on the natural resource archive file obtained in Step 1, parse and generate a document object model tree structure; recursively calculate the depth value of each node using the document object model tree structure as the processing object, and generate a document object model tree structure with depth annotation, wherein the root node depth is initialized to 0, and the child node depth value = parent node depth value + 1. Step 22, using the depth-annotated document object model tree structure output in Step 21 as input: Filter the set of nodes with a depth value ≥ 2, and perform the following for each filtered node: Extract the text blocks of all leaf nodes in the node subtree, arrange the text blocks in the original character offset order to form an ordered sequence, and calculate the spacing between adjacent text blocks in the sequence, where the spacing value = the starting offset of the subsequent text block - the ending offset of the previous text block; Record the maximum spacing value of the current node. When the maximum spacing value is less than the preset threshold, merge all text blocks in the sequence to generate a sub-document, and output the sub-document set and the corresponding DOM node path and maximum spacing value. Step 23: Using the set of sub-documents output in Step 22 as the processing object, assign a globally unique identifier to each sub-document, associate the DOM node path and maximum spacing value output in Step 22, bind the source archive number to generate structured traceability metadata, and use the source archive number as the unique identifier of the archive in the target natural resource archive set.

3. The method for retrieving and utilizing natural resource archives combining large-scale models and knowledge graph technology according to claim 2, characterized in that, Step 3 includes: Step 31: Using the structured traceability metadata generated in step 23 as input, input the associated sub-documents into the preset natural resource domain big model; through the domain adaptation decoding layer of the big model, extract three core entities: land parcel number, ownership subject, and spatial use category, and identify the ownership relationship of "ownership subject-land parcel" and the functional relationship of "land parcel-spatial use". Step 32: For each land parcel entity, extract three types of attributes: area value, coordinate system type, and certificate issuance date; convert the certificate issuance date into a timestamp format, which, together with the coordinate system type, constitutes a dual spatial-temporal constraint; Step 33: Bind the spatial-temporal dual constraint attributes to the corresponding land parcel entity identifiers to form a mapping relationship of {land parcel entity identifier, spatial constraint, temporal constraint}, and finally output a set of structured metadata.

4. The method for retrieving and utilizing natural resource archives combining large-scale models and knowledge graph technology according to claim 3, characterized in that, Step 4 includes: Step 41: Using the structured metadata set output in Step 3 as input, parse the entity identifier, entity type and attribute set in the entity dimension to identify the core business entity type in the natural resources field. Among them, the land parcel entity type is identified as a key entity containing a sequence of vector boundary coordinate points, and the ownership entity type is also identified. Step 42: Taking the ownership entity type as the processing object, for each land parcel entity, extract the vector boundary coordinate point sequence from the structured metadata set as the core spatial attribute of the land parcel entity, and add the coordinate point sequence to the attribute set of the land parcel entity to form the updated attribute set of the land parcel entity. Step 43: Based on the updated set of land parcel entity attributes from Step 42, the ownership relationship between "ownership subject and land parcel," the functional relationship between "land parcel and spatial use," and the dual spatial-temporal constraints, construct a set of RDF triples with attribute constraints, where: Each attribute value in the entity attribute set is converted into a triple with the entity identifier as the subject, the attribute name as the predicate, and the attribute value as the object; Each relation in the relation dimension is transformed into a triple with the subject entity identifier as the subject, the relation type as the predicate, and the object entity identifier as the object; The spatial-temporal dual constraints in the constraint dimension are bound to the corresponding land parcel entity identifier, generating a triple with the land parcel entity identifier as the subject, the constraint type as the predicate, and the constraint value as the object.

5. The method for retrieving and utilizing natural resource archives combining large-scale models and knowledge graph technology according to claim 4, characterized in that, Step 5 includes: Step 51: Using the set of RDF triples with attribute constraints as input, filter out all land parcel entities that contain vector boundary coordinate point sequences. The coordinate point sequence of each land parcel entity is arranged in the order recorded in the target natural resource archive set. For the coordinate point sequence of each land parcel entity, directly mark the first and last points in the sequence as initial key points, and form an initial key point set with the first and last points. Step 52: Based on the initial set of key points determined in Step 51, recursively divide the line segment formed by the first and last points, that is, traverse all non-key coordinate points between the first and last points of the line segment, calculate the vertical distance from each point to the line segment, and identify the point with the maximum vertical distance. Step 53: When the maximum vertical distance value identified in step 52 exceeds the preset distance threshold, the corresponding point is marked as a new key point and added to the key point set; then, using the new key point as the boundary, the line segment formed by the first and last points is divided into two sub-segments. For the two sub-segments, the operation of step 52 is repeated, traversing the non-key coordinate points within the range of each sub-segment, calculating the vertical distance from each point to the corresponding sub-segment, and finding the point with the maximum vertical distance. Step 54: After the recursive process terminates, merge all the key points generated by the level determination, arrange them in the order of the original coordinate sequence, and generate a set of denoised key coordinate points as the simplified boundary representation of the land parcel entity.

6. The method for retrieving and utilizing natural resource archives combining large-scale models and knowledge graph technology according to claim 5, characterized in that, Step 6 includes: Step 61: Using the set of key coordinate points of each parcel entity after noise reduction output in Step 5 as input, select two parcel entities to be subjected to spatial relationship analysis to form an entity pair; Step 62: For the entity pairs selected in step 61, reconstruct the corresponding first polygon boundary and second polygon boundary using their respective key coordinate point sets. Step 63: Perform vector geometric superposition operation on the first polygon boundary and the second polygon boundary reconstructed in step 62, calculate the overlapping region polygon formed by the intersection of the two polygons, and calculate the area value of the overlapping region polygon, the area value of the first polygon boundary and the area value of the second polygon boundary respectively. Step 64: Calculate the intersection-union ratio based on the area values ​​of the overlapping region polygons, the area values ​​of the first polygon boundary, and the area values ​​of the second polygon boundary, where the intersection-union ratio = area value of the overlapping region polygons / (area value of the first polygon boundary + area value of the second polygon boundary - area value of the overlapping region polygons). Step 65: Compare the cross-union ratio calculated in step 64 with the preset cross-union ratio threshold; Step 66: When the intersection-union ratio exceeds the preset intersection-union ratio threshold, it is determined that the two land parcel entities in the entity pair have a spatial overlap relationship. The entity spatial alignment operation is performed, that is, a semantic triple representing the spatial alignment relationship is generated. The main entity is the first land parcel entity identifier, the predicate is the preset spatial alignment relationship type, and the object entity is the second land parcel entity identifier. Step 67: Collect all semantic triples generated in Step 66 that satisfy the intersection-union ratio threshold condition and have completed the spatial alignment operation to form a set of spatially aligned triples.

7. The method for retrieving and utilizing natural resource archives combining large-scale models and knowledge graph technology according to claim 6, characterized in that, Step 63 includes: Step 631: Based on the key coordinate points of the first polygon boundary, traverse all key coordinate points of the first polygon boundary, identify the point set located on the outermost boundary, connect the point set on the outermost boundary to form the minimum convex polygon contour that surrounds all internal points, and generate the first candidate convex hull. Step 632: Based on the key coordinate point set of the second polygon boundary reconstructed in Step 62, traverse all key coordinate points of the second polygon boundary, identify the point set located on the outermost boundary, connect the point set on the outermost boundary to form the minimum convex polygon contour that surrounds all internal points, and generate the second candidate convex hull. Step 633: Calculate the area of ​​the first candidate convex hull, calculate the area of ​​the original coordinate point set of the first polygon boundary, calculate the absolute value of the difference between the area of ​​the first candidate convex hull and the area of ​​the original coordinate point set of the first polygon boundary, and calculate the area error rate. When the area error rate is less than the preset convex hull simplification tolerance threshold, the first candidate convex hull is determined to be valid; otherwise, the first polygon boundary is taken as the first candidate convex hull, and the same validity verification process is performed on the second candidate convex hull. Finally, the valid first convex hull boundary and the second convex hull boundary are output. Step 634: Calculate the polygon of the overlapping region formed by the intersection of the first convex hull boundary and the second convex hull boundary; Step 635: Calculate the area values ​​of the overlapping region polygons obtained in step 634, the area values ​​of the first convex hull boundary and the second convex hull boundary output in step 633, respectively.

8. The method for retrieving and utilizing natural resource archives combining large-scale models and knowledge graph technology according to claim 7, characterized in that, Step 7 includes: Step 71: Use the spatially aligned set of triples output in Step 6 as the first input and the set of RDF triples with attribute constraints constructed in Step 43 as the second input. Step 72: Perform triple fusion processing on the first input and the second input provided in step 71 to obtain a fused triple set, that is, identify triples in the two input sets that describe the same entity identifier and merge their attribute and relation information; for the spatially aligned triple set generated in step 6, it is added as a new relation and incorporated into the fused triple set to obtain a new spatially aligned relation. Step 73: Using the triple set fused in Step 72 as the basis for graph construction, map each unique entity identifier in the fused triple set to a node in the graph database, and map each entity attribute to the attribute key-value pair of the corresponding node. Step 74: Map the semantic relations defined in the fused triple set and the newly added spatial alignment relations to directed edges in the graph database, that is, with the node corresponding to the subject entity identifier as the edge starting point, the relation type as the edge label, and the node corresponding to the object entity identifier as the edge ending point. Step 75: For the land parcel entity nodes, attach the set of key coordinate points after noise reduction in Step 5 as spatial geometric attributes to the corresponding nodes. Step 76: After completing the node, edge, and attribute mapping in steps 73 to 75, a knowledge graph in the natural resources domain is instantiated in the graph database. This graph contains a semantic network structure consisting of land parcel entities, ownership entity nodes, and directed edges consisting of ownership relationships, functional relationships, and spatial alignment relationships.

9. A method for retrieving and utilizing natural resource archives combining large-scale models and knowledge graph technology according to claim 3, characterized in that, The structured metadata collection includes: Entity dimension: Generates a triple {entity type, entity identifier, attribute set}; Relationship dimension: Generates a triple {subject entity identifier, relationship type, object entity identifier}; Constraint Dimension: Bind the spatial and temporal dual constraints to the corresponding land parcel entity identifier.

10. A natural resource archive retrieval and utilization system combining large-scale model and knowledge graph technology, characterized in that, The system is used to perform the method as described in any one of claims 1 to 9, comprising: The acquisition module is used to acquire a target natural resource archive set, wherein the archive set contains at least one natural resource archive to be processed; The parsing module is used to parse the document object model topology based on the target natural resource archive set, and decompose it into corresponding sub-document sets according to the directory hierarchy rules; The execution module is used to perform deep semantic understanding and structured information extraction on the sub-document collection through a pre-set large model of natural resources, and output structured metadata describing the core elements of the archive. The building module is used to identify business entities and their semantic relationships in the natural resources domain based on structured metadata, and to construct RDF triples with attribute constraints; wherein, the identified business entities include land parcel entities with vector boundary coordinate point sequences; The noise reduction module is used to perform geometric feature-preserving coordinate simplification on the sequence of boundary coordinate points of land parcel entity vectors using the Douglas-Puk compression algorithm, generating a set of noise-reduced key coordinate points. The alignment module is used to perform vector geometric superposition operations on the polygonal boundaries of two land parcel entities based on the denoised key coordinate point set, calculate the area ratio of the overlapping area according to the intersection-union ratio, and perform entity spatial alignment when the intersection-union ratio exceeds a preset threshold to obtain a set of spatially aligned triples. The fusion module is used to fuse spatially aligned sets of triples to construct a knowledge graph in the field of natural resources.

Images