A method, system, terminal and storage medium for recommending an urban renewal case

Abstract

The application discloses a kind of urban renewal case recommendation method, system, terminal and storage medium, the method includes: obtaining multi-source spatial data, and pre-processing, obtain pre-processing text data and pre-processing vector data;The pre-processing text data is carried out semantic block processing and similarity calculation and conversion processing, and the case layer semantic representation is obtained;According to pre-processing vector data, construct block layer multi-modal spatial hetero graph, and according to case layer semantic representation and block layer multi-modal spatial hetero graph, construct urban renewal case knowledge graph;Determine hetero graph neural network, and the model training of hetero graph neural network is carried out, and the urban renewal case recommendation model is obtained;Target block is obtained, and the urban renewal case recommendation model is used to recommend update case to target block, and the urban renewal case recommendation result is obtained.The present application can realize the accurate recommendation of urban renewal recommended case, and be favorable to speed up urban renewal process.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method, system, terminal, and computer-readable storage medium for recommending urban renewal cases. Background Technology

[0002] With urban development, urban renewal has become a key means to promote high-quality development and spatial governance transformation. In the implementation of urban renewal projects, drawing on typical cases (i.e., urban renewal case studies) helps improve the scientific nature of decision-making and the feasibility of practice. However, due to significant differences in spatial distribution, location conditions, and scale among urban renewal projects, the comparability and experience transferability between cases are limited.

[0003] To address these issues, traditional methods often rely on manual screening and subjective judgment to obtain recommended urban renewal cases. However, they lack a systematic similarity assessment mechanism, resulting in inaccurate recommended cases generated for urban renewal and making it difficult to meet the demand for accurate recommendations in complex urban scenarios.

[0004] Therefore, existing technologies still need to be improved and developed. Summary of the Invention

[0005] The main objective of this invention is to provide a method, system, terminal, and computer-readable storage medium for recommending urban renewal cases. This invention aims to address the problem that existing technologies rely heavily on manual screening and subjective judgment to obtain recommended urban renewal cases, but lack a systematic similarity evaluation mechanism. As a result, the recommended cases generated for urban renewal are inaccurate and cannot meet the needs for accurate recommendations in complex urban scenarios.

[0006] To achieve the above objectives, the present invention provides a method for recommending urban renewal cases, the method comprising the following steps:

[0007] Acquire multi-source spatial data and preprocess the multi-source spatial data to obtain preprocessed text data and preprocessed vector data;

[0008] The preprocessed text data is semantically segmented and similarity is calculated to obtain a candidate segment set. The candidate segment set is then transformed to obtain a case-level semantic representation.

[0009] A buffer is constructed based on the preprocessed vector data. Node attribute encoding and edge relationship construction are performed on the buffer to obtain a multimodal spatial heterogeneous graph of the land parcel layer. An urban renewal case knowledge graph is then constructed based on the semantic representation of the case layer and the multimodal spatial heterogeneous graph of the land parcel layer.

[0010] A heterogeneous graph neural network is determined, and the model is trained on the heterogeneous graph neural network based on the knowledge graph of urban renewal cases to obtain an urban renewal case recommendation model.

[0011] The target plot is obtained, and urban renewal case recommendation is performed on the target plot using the urban renewal case recommendation model to obtain the urban renewal case recommendation result.

[0012] Optionally, in the method for recommending urban renewal cases, the multi-source spatial data includes text data and vector data;

[0013] The process of acquiring multi-source spatial data and preprocessing the multi-source spatial data to obtain preprocessed text data and preprocessed vector data specifically includes:

[0014] The text data is obtained from multi-source spatial data, wherein the text data includes texts of typical urban renewal cases;

[0015] The text data is cleaned and standardized to obtain preprocessed text data;

[0016] The vector data is obtained from multi-source spatial data, wherein the vector data includes urban land use type maps, population raster data, open-source map road network data, and point of interest data;

[0017] The vector data is subjected to coordinate projection processing and region clipping processing to obtain preprocessed vector data.

[0018] Optionally, the method for recommending urban renewal cases, wherein performing semantic segmentation and similarity calculation on the preprocessed text data to obtain a candidate fragment set, and transforming the candidate fragment set to obtain a case-level semantic representation, specifically includes:

[0019] The preprocessed text data is semantically segmented to obtain multiple semantic text fragments;

[0020] Obtain preset case table information, and perform header standardization and content flattening processing on the preset case table information to obtain a unified semantic text table;

[0021] Multiple target fields are obtained from the unified semantic text table, and a hybrid retrieval strategy is used to calculate the similarity between the multiple target fields and the multiple semantic text fragments to obtain a candidate fragment set;

[0022] A preset prompt word template is determined, and the candidate fragment set is converted into structured tags using the preset prompt word template. Then, a case-level semantic representation is obtained based on the structured tags.

[0023] Optionally, the method for recommending urban renewal cases includes a hybrid retrieval strategy comprising the BestMatching 25 algorithm and the Sentence-BERT algorithm.

[0024] The method of employing a hybrid retrieval strategy to calculate the similarity between multiple target fields and multiple semantic text fragments to obtain a candidate fragment set specifically includes:

[0025] The Best Matching 25 algorithm is used to calculate the first similarity between multiple target fields and multiple semantic text fragments, resulting in multiple first similarity results;

[0026] The Sentence-BERT algorithm is used to calculate the second similarity between multiple target fields and multiple semantic text fragments, resulting in multiple second similarity results;

[0027] Multiple first similarity results are added to multiple second similarity results to obtain multiple total similarity results. A preset number of semantic text segments are selected from the multiple semantic text segments in descending order of the total similarity results to obtain a candidate segment set.

[0028] Optionally, the method for recommending urban renewal cases, wherein the step of constructing a buffer based on the preprocessed vector data, performing node attribute encoding and edge relationship construction on the buffer to obtain a multimodal spatial heterogeneous graph at the land parcel level, and constructing an urban renewal case knowledge graph based on the case layer semantic representation and the multimodal spatial heterogeneous graph at the land parcel level, specifically includes:

[0029] Obtain the address field from the semantic representation of the case layer, use geocoding technology to calculate the latitude and longitude coordinates based on the address field and the preprocessed vector data, and construct a buffer based on the latitude and longitude coordinates;

[0030] Multiple spatial features in the buffer are acquired, and node attribute encoding processing is performed on the multiple spatial features to obtain node attribute encoding results, wherein the node attribute encoding results include plot nodes, points of interest nodes, and road nodes.

[0031] Determine predefined semantic relationships, and perform edge relationship construction processing on multiple spatial elements based on the predefined semantic relationships to obtain edge relationship construction results;

[0032] Based on the node attribute encoding results and the edge relationship construction results, a multimodal spatial heterogeneous graph of the land parcel layer is constructed, and an urban renewal case knowledge graph is constructed based on the multimodal spatial heterogeneous graph of the land parcel layer and the semantic representation of the case layer.

[0033] Optionally, the method for recommending urban renewal cases, wherein determining a heterogeneous graph neural network and training the heterogeneous graph neural network based on the knowledge graph of urban renewal cases to obtain an urban renewal case recommendation model specifically includes:

[0034] A heterogeneous graph neural network is determined, and the urban renewal case knowledge graph is input into the heterogeneous graph neural network. The urban renewal case knowledge graph is encoded by the heterogeneous graph neural network to obtain a high-dimensional representation vector.

[0035] The high-dimensional representation vector is subjected to graph-level pooling to obtain a graph-level representation vector;

[0036] Calculate the positive-negative contrast loss and semantic consistency loss of the graph-level representation vector, and perform a weighted summation of the positive-negative contrast loss and the semantic consistency loss to obtain the overall loss;

[0037] The heterogeneous graph neural network is optimized based on the overall loss to obtain a city renewal case recommendation model.

[0038] Optionally, the method for recommending urban renewal cases, wherein obtaining a target plot and recommending renewal cases for the target plot using the urban renewal case recommendation model to obtain urban renewal case recommendation results specifically includes:

[0039] Obtain the target plot and input the target plot into the urban renewal case recommendation model. Calculate the target similarity results between the target plot and historical case nodes in the urban renewal case knowledge graph using the urban renewal case recommendation model.

[0040] A preset number of updated cases are selected from the historical case nodes according to the target similarity results in descending order, and the preset number of updated cases are used as the urban renewal case recommendation results.

[0041] Furthermore, to achieve the above objectives, the present invention also provides a recommendation system for urban renewal cases, wherein the recommendation system for urban renewal cases includes:

[0042] The data preprocessing module is used to acquire multi-source spatial data and preprocess the multi-source spatial data to obtain preprocessed text data and preprocessed vector data.

[0043] The case layer construction module is used to perform semantic segmentation and similarity calculation on the preprocessed text data to obtain a candidate segment set, and to transform the candidate segment set to obtain a case layer semantic representation.

[0044] The knowledge graph generation module is used to construct a buffer based on the preprocessed vector data, perform node attribute encoding and edge relationship construction on the buffer to obtain a multimodal spatial heterogeneous graph of the land parcel layer, and construct an urban renewal case knowledge graph based on the semantic representation of the case layer and the multimodal spatial heterogeneous graph of the land parcel layer.

[0045] The case recommendation model construction module is used to determine the heterogeneous graph neural network and train the heterogeneous graph neural network based on the urban renewal case knowledge graph to obtain the urban renewal case recommendation model.

[0046] The case recommendation result output module is used to obtain the target plot and recommend urban renewal cases for the target plot through the urban renewal case recommendation model to obtain the urban renewal case recommendation result.

[0047] In this invention, multi-source spatial data is acquired and preprocessed to obtain preprocessed text data and preprocessed vector data. Semantic segmentation and similarity calculation are performed on the preprocessed text data to obtain a candidate fragment set. This candidate fragment set is then transformed to obtain a case-layer semantic representation. A buffer is constructed based on the preprocessed vector data, and node attribute encoding and edge relationship construction are performed on the buffer to obtain a land parcel-layer multimodal spatial heterogeneous graph. An urban renewal case knowledge graph is constructed based on the case-layer semantic representation and the land parcel-layer multimodal spatial heterogeneous graph. A heterogeneous graph neural network is determined, and the model is trained on the urban renewal case knowledge graph to obtain an urban renewal case recommendation model. A target land parcel is acquired, and urban renewal case recommendations are made for the target land parcel using the urban renewal case recommendation model to obtain urban renewal case recommendation results. This invention constructs a case-level semantic representation and a plot-level multimodal spatial heterogeneous graph by acquiring multi-source spatial data, thereby building an urban renewal case knowledge graph. Based on the urban renewal case knowledge graph, an urban renewal case recommendation model is trained, which can achieve accurate recommendation of cases corresponding to urban renewal, thus helping to accelerate the urban renewal process. Attached Figure Description

[0048] Figure 1 This is a flowchart of a preferred embodiment of the recommended method for urban renewal cases of the present invention;

[0049] Figure 2 This is a schematic diagram of the case layer extraction process of a preferred embodiment of the urban renewal case recommendation method of the present invention;

[0050] Figure 3 This is a schematic diagram of the land parcel layer extraction process, which is a preferred embodiment of the recommended method for urban renewal cases of the present invention.

[0051] Figure 4 This is a schematic diagram of source data for a preferred embodiment of the urban renewal case recommendation method of the present invention;

[0052] Figure 5 This is a schematic diagram of the selected land parcels, representing a preferred embodiment of the urban renewal case recommendation method of the present invention.

[0053] Figure 6 This is a schematic diagram illustrating the selection of POI information points in a preferred embodiment of the urban renewal case recommendation method of the present invention.

[0054] Figure 7 This is a schematic diagram of the selected roads in a preferred embodiment of the urban renewal case recommendation method of the present invention;

[0055] Figure 8 This is a schematic diagram of the filtered data overview of a preferred embodiment of the urban renewal case recommendation method of the present invention;

[0056] Figure 9 This is a schematic diagram illustrating the definition of inter-plot relationships in a preferred embodiment of the recommended method for urban renewal cases of the present invention.

[0057] Figure 10 This is a schematic diagram illustrating the relationship between land parcels and points of interest (POIs) in a preferred embodiment of the recommended method for urban renewal cases of the present invention.

[0058] Figure 11 This is a schematic diagram illustrating the relationship between land parcels and roads in a preferred embodiment of the recommended method for urban renewal cases of the present invention.

[0059] Figure 12 This is a schematic diagram illustrating the definition of road relationships in a preferred embodiment of the recommended method for urban renewal cases of the present invention.

[0060] Figure 13 This is a schematic diagram illustrating the road-POI relationship definition in a preferred embodiment of the recommended method for urban renewal cases of the present invention.

[0061] Figure 14 This is a schematic diagram of the urban renewal case recommendation technology process, which is a preferred embodiment of the urban renewal case recommendation method of the present invention.

[0062] Figure 15 This is a structural diagram of a preferred embodiment of the recommendation system for urban renewal cases of the present invention;

[0063] Figure 16 This is a structural diagram of a preferred embodiment of the terminal of the present invention. Detailed Implementation

[0064] To make the objectives, technical solutions, and advantages of this invention clearer and more explicit, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0065] With urban development, urban renewal has become a key means to promote high-quality development and spatial governance transformation.

[0066] In implementing urban renewal projects, drawing on exemplary cases can help improve the scientific nature of decision-making and the feasibility of practice. However, due to significant differences in spatial distribution, location conditions, and scale among urban renewal projects, the comparability and experience transferability between cases are limited. Traditional methods often rely on manual screening and subjective judgment, lacking a systematic similarity assessment mechanism, making it difficult to meet the demand for accurate recommendations in complex urban scenarios. Furthermore, the legal systems, development stages, and policy environments vary considerably across regions, and directly reusing cases from other areas often faces problems such as insufficient adaptability and implementation obstacles.

[0067] Currently, urban renewal case recommendation mechanisms are still in the initial exploratory stage, and the relevant research system is not yet complete. Although some studies have attempted to introduce big data and artificial intelligence technologies to achieve case matching and decision support, the level and depth of intelligence are still insufficient in practical applications. Existing methods can be roughly divided into two categories:

[0068] One type is the label-driven method based on the CBR system. This type of method adopts a "case-label-matching" paradigm. In the case library construction, a label system is generated manually or semi-automatically to describe the core features of the cases (such as project type, land use, and renewal mode). Case matching is then achieved based on label similarity. This method has a clear theory and a well-defined implementation path, but it is highly dependent on the accuracy and completeness of structured labels. Considering the high heterogeneity of urban renewal projects and the unstructured nature of information sources, the construction cost of large-scale, high-quality labels is extremely high, making it difficult to support sustainable model training. In addition, this type of method is usually based on a static ontology structure or a single semantic level, lacking the ability to model the spatial structure of land parcels, and is difficult to handle complex recommendation tasks under multi-scale and multi-contextual conditions.

[0069] Another category is spatial similarity calculation methods based on image morphology and graphic topology. This type of research attempts to bypass the high cost of label construction by extracting geometric features from data such as remote sensing imagery and land parcel graphics, and then using deep learning or topological analysis methods to encode urban morphology, thereby calculating the spatial similarity between cases. While this method has advantages in morphological recognition and automated processing, capturing indicators such as floor area ratio and road structure, it primarily focuses on physical morphology, neglecting semantic information such as land use changes and the configuration of public service facilities. Due to the lack of contextual support, the recommendation results often lack interpretability and fail to meet the multidimensional needs of urban planning for rationality analysis and social impact assessment. Morphological similarity does not equate to logical similarity; morphological matching lacking semantic reasoning support often struggles to achieve effective transfer.

[0070] In summary, current methods for recommending urban renewal cases still have many shortcomings. There is an urgent need to construct a multi-source data-driven method that integrates semantic and spatial features, and has low labor costs and strong application feasibility.

[0071] To address the aforementioned problems, this invention proposes a multi-source heterogeneous map construction and case recommendation method for urban renewal. This method uses typical urban renewal case texts as a foundation, combined with multi-source spatial data, to construct a two-layer representation model consisting of a case layer and a plot layer. First, through large-scale modeling and text analysis techniques, semantic tags including basic information, location attributes, and renewal methods are extracted from the renewal planning text to construct a semantic representation of the case layer. Second, by combining data such as land use type, population distribution, road network structure, and POIs, a plot-level spatial map is established, encoding the spatial topological relationships of the renewal area and forming a computable heterogeneous map structure. Based on this, semantic vectors and spatial graph neural network encoding are integrated to establish a semantic-spatial joint recommendation model, comprehensively evaluating the similarity of candidate cases in both semantic and spatial dimensions. Finally, a case recommendation system with semantic transfer and spatial generalization capabilities is constructed, effectively improving the accuracy and practical feasibility of the recommendation results.

[0072] The method for recommending urban renewal cases according to a preferred embodiment of the present invention, such as... Figure 1 As shown, the recommendation method for the urban renewal case includes the following steps:

[0073] Step S10: Acquire multi-source spatial data and preprocess the multi-source spatial data to obtain preprocessed text data and preprocessed vector data. The multi-source spatial data includes text data and vector data.

[0074] The multi-source spatial data used in this invention includes textual examples of urban renewal, urban land use type maps (EULUC-China), population raster data, open-source map road network data (OSM), and point of interest (POI) data. Among them, the textual examples of urban renewal are textual data, while the urban land use type maps, population raster data, open-source map road network data, and point of interest data are all vector data.

[0075] Specifically, the text data is acquired from the multi-source spatial data, wherein the text data includes texts of typical urban renewal cases; the text data is cleaned and standardized to obtain preprocessed text data; the vector data is acquired from the multi-source spatial data, wherein the vector data includes urban land use type maps, population raster data, open-source map road network data, and point of interest data; the vector data is subjected to coordinate projection processing and regional clipping processing to obtain preprocessed vector data.

[0076] The specific contents of the multi-source spatial data are shown in Table 1 below:

[0077] Table 1: Data Sources and Descriptions

[0078]

[0079] As shown in Table 1, the urban renewal text data mainly comes from descriptions of historically updated cases. The text data includes project names, update modes, update content, and plot addresses, possessing high semantic density and suitable for structured information extraction tasks. Urban land use type data is divided into 12 typical land use function types, including residential, commercial, industrial, and green space. The data format is vector polygon, projected uniformly using the WGS84 coordinate system. Population data originates from WorldPOP raster data and is used to characterize spatial population density distribution features. Road network data is based on OSM (OpenStreetMap) open-source data, extracting elements such as arterial roads, secondary roads, branch roads, and expressways, serving as the structural basis for modeling spatial connectivity between plots. POI data covers 11 categories of points of interest, including catering, education, finance, and transportation facilities, effectively supplementing detailed information on urban functional structure.

[0080] The specific process of data preprocessing is as follows:

[0081] In the data preprocessing stage, all vector data are first subjected to unified coordinate projection and study area clipping to ensure spatial boundary consistency. Text data undergoes HTML tag cleaning and redundant information removal. All data undergoes missing value handling, field screening, and format standardization to ensure consistency across different data sources in semantic attributes, spatial scope, and temporal dimensions.

[0082] Step S20: Perform semantic segmentation and similarity calculation on the preprocessed text data to obtain a candidate segment set, and perform transformation processing on the candidate segment set to obtain a case layer semantic representation.

[0083] The construction of the urban renewal case knowledge graph model includes a case layer and a plot layer, forming a two-layer representation model. The case layer mainly contains basic project information, current status information, redevelopment intention information, and case classification information. Specific extracted information is shown in Table 2 below.

[0084] Table 2: Field Definitions of Case Layer in Urban Renewal Case Knowledge Graph

[0085]

[0086]

[0087] In Table 2, items 1-4 contain basic case information, which describes metadata such as case name, city affiliation, project address, and project scale to ensure basic indexing capabilities. Items 5-6 contain current case information, focusing on textual and structured problem tags related to issues before urban renewal. Items 7-10 contain case renovation information, reflecting the direction, methods, and land use types before and after renewal. Items 11-12 contain case classification information, including multi-level matching results of case categories and case type tags, forming "knowledge coordinates" in the semantic space of urban renewal.

[0088] Specifically, the preprocessed text data is semantically segmented to obtain multiple semantic text fragments; preset case table information is obtained, and the preset case table information is subjected to header standardization and content flattening to obtain a unified semantic text table.

[0089] The process of extracting case layer information is as follows:

[0090] like Figure 2 As shown, this invention introduces a knowledge-enhanced retrieval and generation model for urban renewal to construct a structured information extraction process. The specific extraction process is as follows: 1. Semantic segmentation: The text data set with HTML tags removed is used to extract the text data set (the original data is the crawled HTML webpage, and the data here is the text part with HTML tags removed). The process involves segmenting sentences into minimal semantic units, specifically using natural paragraphs or sentence-ending punctuation (such as periods, question marks, and exclamation marks) as segmentation boundaries. After segmentation, all sentences are combined into fragments with a total length of less than 800 characters. 2. These segmented fragments are denoted as... Each s i3. Standardize the table information by formatting the header and flattening the content into HTML format. Specifically, this involves structuring the header to use a unified name that aligns with predefined structure fields, expanding merged cells, and ensuring that each row corresponds to a single key-value pair.

[0091] The process for processing the table information is shown in Table 3 below:

[0092] Table 3: Example of processing table information

[0093] Project Name Land use type Area (square meters) XX plot Residential land 35000

[0094] The conversion result is:

[0095]

[0096] Multiple target fields are obtained from the unified semantic text table, and the Best Matching 25 algorithm is used to calculate the first similarity between the multiple target fields and the multiple semantic text segments to obtain multiple first similarity results; the Sentence-BERT algorithm is used to calculate the second similarity between the multiple target fields and the multiple semantic text segments to obtain multiple second similarity results; the multiple first similarity results and the multiple second similarity results are added together to obtain multiple total similarity results, and a preset number of semantic text segments are selected from the multiple semantic text segments in descending order of the total similarity results to obtain a candidate segment set.

[0097] Furthermore, such as Figure 2 As shown, this invention sets up semantic recall: based on each target field The case fields defined in Table 2 are descriptive terms. Then, the target field f and... The similarity between them is determined by a hybrid retrieval strategy using BM25 (Best Matching25) and Sentence-BERT algorithms. Segment relevance scoring is performed (i.e., the similarity scores between two expected segments can be obtained using the BM25 and Sentence-BERT algorithms, respectively, and are denoted as scores), forming a candidate segment set. satisfy:

[0098]

[0099] Here, λ represents the retrieval score fusion weight. The BM25 algorithm is a term-weighted retrieval model that considers term frequency and inverse document frequency factors to form a similarity score between two segments. The Sentence-BERT (SBERT) algorithm performs semantic vector representation and matching on the case text pairs. That is, it represents the entire text as a vector and then calculates the cosine similarity.

[0100] A preset prompt word template is determined, and the candidate fragment set is converted into structured tags using the preset prompt word template. Then, a case-level semantic representation is obtained based on the structured tags.

[0101] Furthermore, structured generation is performed: by using preset prompt word templates and combining the contextual understanding and structure generation capabilities of the large language model, candidate fragments for each field are generated. Transform into standardized structured tags Right now:

[0102]

[0103] Among them, Prompt is a field extraction prompt template designed for the domain, C f As a contextual input, f provides semantic basis, and f is one of the target field values.

[0104] The complete case representation after construction for:

[0105]

[0106] Among them, the output results ValueSpace(f) represents the candidate value space for field f, which can be an enumeration class or free text (i.e., the “Label Level / Value Range” description in Table 2).

[0107] Step S30: Construct a buffer based on the preprocessed vector data, perform node attribute encoding and edge relationship construction on the buffer to obtain a multimodal spatial heterogeneous graph of the land parcel layer, and construct an urban renewal case knowledge graph based on the semantic representation of the case layer and the multimodal spatial heterogeneous graph of the land parcel layer.

[0108] like Figure 3 As shown, after the case layer is constructed, the land parcel layer information is constructed. This invention uses urban land parcels as the basic unit, integrates urban land parcel vector surface data, road segment data and point of interest (POI) data to represent the cases, and establishes an urban renewal case knowledge graph.

[0109] Specifically, the address field in the semantic representation of the case layer is obtained, and the latitude and longitude coordinates are calculated based on the address field and the preprocessed vector data using geocoding technology. A buffer is then constructed based on the latitude and longitude coordinates.

[0110] First, using the "Project Address" field in the case layer, its latitude and longitude coordinates (x0, y0) are obtained through geocoding technology. Then, a buffer zone B(x0, y0) with a radius of r = 500m is constructed using this point as the center to define the geographical features representing the urban renewal project. For example... Figure 4 As shown, Figure 4 This is a schematic diagram of the source data, i.e., the data sample before filtering.

[0111] Multiple spatial features in the buffer are acquired, and node attribute encoding is performed on the multiple spatial features to obtain node attribute encoding results, wherein the node attribute encoding results include land parcel nodes, points of interest nodes, and road nodes.

[0112] In buffer B, three types of spatial elements—urban land use, POI points, and road lines—are selected as the original data sources for constructing nodes in the heterogeneous map. The selection logic is as follows (e.g., Figure 5 , Figure 6 and Figure 7 As shown):

[0113] like Figure 5 As shown, the set of land parcels G = {g i}, select plots from the entire set of plots that meet any of the following conditions: plot centroid c i ∈B, or the plot polygon g i There is a spatial intersection relationship with B:

[0114]

[0115] like Figure 6 As shown, the POI set P = {p j Select all POIs that fall within the buffer zone:

[0116]

[0117] like Figure 7 As shown, the road set R = {r k Select all road segments that intersect with the buffer zone:

[0118]

[0119] All data obtained after final filtering are as follows Figure 8 As shown, Figure 8It contains the above three data sources to characterize the surrounding environmental land parcel information of an urban renewal case.

[0120] After obtaining the three types of spatial objects within the buffer, it is necessary to encode and extract the node attributes and node relationships. The node attribute encoding method is as follows:

[0121] 1. Land Parcel Node Type Encoding Method: A mapping dictionary for 11 land use categories (including residential land and commercial service land, etc.) is established, and the text category is converted into a unique index value. The area feature is directly extracted from the original area value; the land parcel node type comes from the urban land use type map (EULUC-China) in the original data, which has the following 11 categories, as shown in Table 4 below. The encoding is the index code column in the table, and normalization is performed after dictionary mapping.

[0122] Table 4: Land Use Types

[0123]

[0124] 2. Road type coding: Establish a one-hot mapping system for 27 road grades (such as motorway, residential, etc.); Boolean attribute conversion: Convert T / F tags such as "oneway", "bridge", and "tunnel" (description of road function, whether it is a one-way road, bridge, or tunnel) into binary values ​​(0 / 1); Numerical feature standardization: Pad missing speed limit values ​​for "maxspeed" (road feature description, maxspeed is the speed limit value), and retain the original integer values ​​for "layer" (layer is the level in the original data, used to distinguish the number of road layers); Concatenate the above 6 types of features (1-dimensional road type + 3-dimensional Boolean attribute + 2-dimensional numerical feature) into a 6-dimensional feature tensor.

[0125] Among them, the 27 road grades are as follows, which are the road types in the open source road data in Table 1, including residential, tertiary, unclassified, motorway, service, secondary, primary, footway, motorway_link, trunk, track, primary_link, path, trunk_link, secondary_link, steps, unknown, living_street, pedestrian, tertiary_link, cycleway, track_grade2, track_grade1, track_grade3, track_grade4, track_grade5, and bridleway.

[0126] 3. POI Node Encoding: A hierarchical classification encoding system is designed. The major category encoding uses 14 main POI categories (e.g., healthcare, food and beverage) as the first-level classification index; the intermediate category encoding uses 125 subcategories (e.g., general hospitals, coffee shops) as the second-level classification index. The major category index (1-dimensional) and the intermediate category index (1-dimensional) are combined into a 2-dimensional classification feature vector.

[0127] Determine predefined semantic relationships, and perform edge relationship construction processing on multiple spatial elements based on the predefined semantic relationships to obtain edge relationship construction results.

[0128] This invention establishes edge E based on five predefined semantic relations, specifically:

[0129] 1. Establishment of plot-to-plot relationships (e.g.) Figure 9 As shown), the Delaunay triangulation method is used, with the geometric centroid of all plots as the basis. For the input point set, a triangulation graph is constructed (using the Delaunay triangulation method. The specific steps of Delaunay triangulation are: first, construct an initial triangle containing all points from the point set; then, insert each point sequentially and update the triangle structure; after inserting a new point, if it falls within a triangle, split that triangle into three new triangles, and perform an edge flipping operation on edges that violate the Delaunay condition (i.e., a point falls within the circumcircle of a triangle)). Delaunay(G) is used. If the centroids of two plots are connected by an edge, the two plots are considered "adjacent," and an adjacent edge is established. The expression is:

[0130]

[0131] Among them, g i ,g j For plot of land, c i ,c j For the centroid of the corresponding plot, Edges(Delaunay(G)) represents the set of edges in the triangulation graph. Its edge index is encoded by generating a long type tensor of [2, num_edges]. The first line is the source plot node index, and the second line is the target plot node index.

[0132] 2. Plot-POI relationship (e.g.) Figure 10 As shown), construct the inclusion relationship, that is,

[0133]

[0134] Its edge index encoding method is to generate a long type tensor of [2, num_edges], with the first line being the parcel node index number and the second line being the POI node index number.

[0135] 3. Relationship between land parcels and roads (e.g.) Figure 11 As shown), for each target plot g i boundary Iterate through the set of possible intersecting roads in its vicinity {r j}, utilize spatial indexing to quickly locate candidate roads; for each pair of plots and one road combination (g i ,r j ), calculate its minimum boundary distance d i,j The definition is as follows:

[0136]

[0137] Where x is the set of edges of the target plot. An edge in the path r, where y is the set of roads. j An edge in the diagram, where ||·|2 represents the Euclidean distance, is given. A spatial adjacency threshold δ = 30m is set if and only if d i,j When <δ, it is considered that plot g i With road r j If valid spatial adjacency relationships exist, then an edge relationship of type "plot-road" is constructed in the urban spatial knowledge graph, represented as follows:

[0138] Edge(g i ,r j =("adjacent",d i,j );

[0139] Among them, the edge attribute records the corresponding minimum distance value d. i,j Its edge index encoding method is to generate a long type tensor of [2, num_edges], the first line is the land node index, the second line is the road node index, and the edge attribute encoding method is the original distance value / 30m, which is normalized and stored as a float type tensor of [num_edges, 1].

[0140] 4. Road-road relationships (e.g.) Figure 12 As shown), determine whether two road segments have endpoint contact or spatial intersection. If there is an intersection, it is defined as a "connection" relationship, represented as:

[0141]

[0142] The edge index is encoded as a long type tensor of [2, num_edges], with the first line being the source road node index and the second line being the target road node index.

[0143] 5. POI - Road Relationships (e.g.) Figure 13 As shown), calculate the shortest distance from the POI to the surrounding roads, record the relationship as "reachable", and the distance d.p,r This refers to the edge attribute.

[0144]

[0145] Where, δ p To determine the accessibility threshold, we take δ. p =500m, meaning the edge is created when the distance is less than 500m. The edge index is encoded as a long type tensor of [2, num_edges], where the first line is the POI node index and the second line is the road node index. The edge attribute is encoded as the original distance value / 500 meters, normalized, and stored as a float type tensor of [num_edges, 1].

[0146] By constructing the above five types of edge relationships, a multimodal spatial heterogeneous map of the land parcel layer can be generated:

[0147]

[0148] in Representing three types of nodes: land parcels, points of interest (POIs), and roads; ε is the set of edges; T v T e These are the node type and the edge type, respectively.

[0149] Based on the node attribute encoding results and the edge relationship construction results, a multimodal spatial heterogeneous graph of the land parcel layer is constructed, and an urban renewal case knowledge graph is constructed based on the multimodal spatial heterogeneous graph of the land parcel layer and the semantic representation of the case layer.

[0150] Step S40: Determine the heterogeneous graph neural network, and train the model of the heterogeneous graph neural network according to the knowledge graph of urban renewal cases to obtain the urban renewal case recommendation model.

[0151] like Figure 14 As shown, the training and inference of the urban renewal case recommendation model: Based on the above-mentioned urban renewal case knowledge graph, that is, each case is represented by the above-mentioned case layer and multimodal space heterogeneous graph, the present invention proposes an unsupervised knowledge graph heterogeneous graph learning framework for training the urban renewal case recommendation model.

[0152] Specifically, a heterogeneous graph neural network is determined, and the urban renewal case knowledge graph is input into the heterogeneous graph neural network. The urban renewal case knowledge graph is encoded by the heterogeneous graph neural network to obtain a high-dimensional representation vector. The high-dimensional representation vector is then subjected to graph-level pooling to obtain a graph-level representation vector.

[0153] This invention first loads heterogeneous graph data to implement multiple types of nodes / edges according to the construction process described above. The encoded feature matrices (block_x, road_x, poi_x) are stored hierarchically according to type. Edge indices are constructed, generating COO-formatted edge index tensors for each relation type. Normalized weight attributes are added to distance-related relation types (block_road, poi_road). Secondly, a heterogeneous graph neural network is built using HGNN to encode the structural semantic information between nodes. Each node representation aggregates information from its neighboring nodes through multiple graph convolutions, capturing a high-dimensional representation vector of spatial structure and semantic relationships. A DiffPool graph-level pooling operation is performed: all plot representations of each case graph are aggregated to generate a fixed-dimensional graph-level representation vector.

[0154] Calculate the positive-negative contrast loss and semantic consistency loss of the graph-level representation vector, and perform a weighted summation of the positive-negative contrast loss and the semantic consistency loss to obtain the overall loss; optimize the heterogeneous graph neural network based on the overall loss to obtain the urban renewal case recommendation model.

[0155] To achieve representation learning without labeled data, this invention designs two unsupervised loss functions to jointly drive model training, as shown in the following expressions:

[0156]

[0157] in, For the overall loss, For positive and negative contrast loss, λ1 and λ2 are the weighting coefficients of the two losses mentioned above, representing the semantic consistency loss.

[0158] The positive and negative contrast loss is calculated as follows: Using the original adjacency relationships of the land parcel layer map as positive samples and distant or negative sampling nodes as negative samples, a structural contrast loss function is introduced to maintain the original spatial adjacency relationships in the embedding space. Specifically, the positive and negative contrast loss function is defined as follows:

[0159]

[0160] in, Indicates positive sample pairs. H represents a negative sample pair. i h j h k Let i, j, and k represent the embeddings of nodes i, j, and k, respectively, where i and j are... The two values, i and k, are The two values ​​in the equation are sim(·) representing cosine similarity and σ(·) representing the sigmoid function.

[0161] For nodes with similar structural attributes but unknown labels, semantic similarity is calculated using their attribute vectors. The attribute space is then aligned with the embedding space, subject to the following constraints:

[0162]

[0163] Where, x i ,x j These are the semantic vectors for the two cases, sim attr (·) represents the similarity of attribute vectors, sim emb (·) indicates the similarity in the embedding space.

[0164] Step S50: Obtain the target plot and recommend urban renewal cases for the target plot through the urban renewal case recommendation model to obtain the urban renewal case recommendation result.

[0165] Specifically, a target plot is acquired and input into the urban renewal case recommendation model. The urban renewal case recommendation model calculates multiple target similarity results between the target plot and historical case nodes in the urban renewal case knowledge graph. A preset number of renewal cases are selected from the historical case nodes in descending order of the target similarity results, and the preset number of renewal cases are used as the urban renewal case recommendation results.

[0166] After completing the embedding training, for the target plot v q By calculating the cosine similarity between the updated case and the historical case node in the embedding space, the top k most similar updated cases are obtained as the recommendation results, expressed as:

[0167]

[0168] Among them, c j As a historical case node, These are the embedding vectors for the target plot and the historical case node, respectively.

[0169] This invention constructs a multi-dimensional evaluation index system to measure the rationality of unsupervised training results, covering the following three core indicators:

[0170] 1. Structure Accuracy: Based on known case category labels, calculate the consistency rate of the structure type between the recommended cases and the target plots, i.e. whether the plot type and update mode match.

[0171] 2. Semantic Similarity Score: Based on the structured fields of the case (such as subject type, update path, and function transformation, etc.), the semantic consistency of the recommended cases is calculated by using TF-IDF vector and cosine similarity.

[0172] 3. Recommendation Accuracy (Top-K Accuracy): This is calculated by determining whether the recommended results on the test set contain the true optimal case, and then calculating the Top-1 and Top-K hit rates, defined as follows:

[0173]

[0174] in, Let y represent the top K recommendation results for the i-th sample. i It is the true best matching case.

[0175] Furthermore, such as Figure 15 As shown, based on the above-mentioned method for recommending urban renewal cases, the present invention also provides a recommendation system for urban renewal cases, wherein the recommendation system for urban renewal cases includes:

[0176] The data preprocessing module 51 is used to acquire multi-source spatial data and preprocess the multi-source spatial data to obtain preprocessed text data and preprocessed vector data.

[0177] The case layer construction module 52 is used to perform semantic segmentation and similarity calculation on the preprocessed text data to obtain a candidate segment set, and to perform transformation processing on the candidate segment set to obtain a case layer semantic representation;

[0178] The knowledge graph generation module 53 is used to construct a buffer based on the preprocessed vector data, perform node attribute encoding and edge relationship construction on the buffer to obtain a multimodal spatial heterogeneous graph of the land parcel layer, and construct an urban renewal case knowledge graph based on the semantic representation of the case layer and the multimodal spatial heterogeneous graph of the land parcel layer.

[0179] The case recommendation model construction module 54 is used to determine the heterogeneous graph neural network and train the heterogeneous graph neural network according to the urban renewal case knowledge graph to obtain the urban renewal case recommendation model.

[0180] The case recommendation result output module 55 is used to obtain the target plot and recommend urban renewal cases for the target plot through the urban renewal case recommendation model to obtain the urban renewal case recommendation result.

[0181] Furthermore, such as Figure 16 As shown, based on the recommended methods and systems of the above-mentioned urban renewal cases, the present invention also provides a terminal, which includes a processor 10, a memory 20 and a display 30. Figure 16 Only some of the terminal components are shown; however, it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0182] In some embodiments, the memory 20 may be an internal storage unit of the terminal, such as a hard disk or memory. In other embodiments, the memory 20 may be an external storage device of the terminal, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc. Further, the memory 20 may include both internal and external storage devices. The memory 20 is used to store application software and various types of data installed on the terminal, such as the program code installed on the terminal. The memory 20 can also be used to temporarily store data that has been output or will be output. In one embodiment, the memory 20 stores a recommendation program 40 for urban renewal cases, which can be executed by the processor 10 to implement the urban renewal case recommendation method of this application.

[0183] In some embodiments, the processor 10 may be a central processing unit (CPU), a microprocessor, or other data processing chip, used to run program code stored in the memory 20 or process data, such as executing the recommended method for the urban renewal case.

[0184] In some embodiments, the display 30 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. The display 30 is used to display information on the terminal and to display a visual user interface.

[0185] In one embodiment, the steps of the urban renewal case recommendation method are implemented when the processor 10 executes the urban renewal case recommendation program 40 in the memory 20.

[0186] In summary, this invention provides a method, system, and terminal for recommending urban renewal cases. The method includes: acquiring multi-source spatial data and preprocessing the multi-source spatial data to obtain preprocessed text data and preprocessed vector data; performing semantic segmentation and similarity calculation on the preprocessed text data to obtain a candidate fragment set, and performing transformation processing on the candidate fragment set to obtain a case-layer semantic representation; constructing a buffer based on the preprocessed vector data, performing node attribute encoding and edge relationship construction processing on the buffer to obtain a plot-layer multimodal spatial heterogeneous graph, and constructing an urban renewal case knowledge graph based on the case-layer semantic representation and the plot-layer multimodal spatial heterogeneous graph; determining a heterogeneous graph neural network, and training the heterogeneous graph neural network based on the urban renewal case knowledge graph to obtain an urban renewal case recommendation model; acquiring a target plot, and recommending renewal cases for the target plot through the urban renewal case recommendation model to obtain an urban renewal case recommendation result. This invention constructs a case-level semantic representation and a plot-level multimodal spatial heterogeneous graph by acquiring multi-source spatial data, thereby building an urban renewal case knowledge graph. Based on the urban renewal case knowledge graph, an urban renewal case recommendation model is trained, which can achieve accurate recommendation of cases corresponding to urban renewal, thus helping to accelerate the urban renewal process.

[0187] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal that includes that element.

[0188] Of course, those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware (such as a processor, controller, etc.). The program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The computer-readable storage medium can be a memory, magnetic disk, optical disk, etc.

[0189] It should be understood that the application of the present invention is not limited to the examples above. Those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A method for recommending urban renewal cases, characterized in that, The recommended methods for urban renewal cases include: Acquire multi-source spatial data and preprocess the multi-source spatial data to obtain preprocessed text data and preprocessed vector data; The preprocessed text data is semantically segmented and similarity is calculated to obtain a candidate segment set. The candidate segment set is then transformed to obtain a case-level semantic representation. A buffer is constructed based on the preprocessed vector data. Node attribute encoding and edge relationship construction are performed on the buffer to obtain a multimodal spatial heterogeneous graph of the land parcel layer. An urban renewal case knowledge graph is then constructed based on the semantic representation of the case layer and the multimodal spatial heterogeneous graph of the land parcel layer. The process involves constructing a buffer based on the preprocessed vector data, performing node attribute encoding and edge relationship construction on the buffer to obtain a multimodal spatial heterogeneous graph at the land parcel level, and constructing an urban renewal case knowledge graph based on the semantic representation of the case layer and the multimodal spatial heterogeneous graph at the land parcel level. Specifically, this includes: Obtain the address field from the semantic representation of the case layer, use geocoding technology to calculate the latitude and longitude coordinates based on the address field and the preprocessed vector data, and construct a buffer based on the latitude and longitude coordinates; Multiple spatial features in the buffer are acquired, and node attribute encoding processing is performed on the multiple spatial features to obtain node attribute encoding results, wherein the node attribute encoding results include plot nodes, points of interest nodes, and road nodes. Determine predefined semantic relationships, and perform edge relationship construction processing on multiple spatial elements based on the predefined semantic relationships to obtain edge relationship construction results; A multimodal spatial heterogeneous graph of land parcels is constructed based on the node attribute encoding results and the edge relationship construction results, and an urban renewal case knowledge graph is constructed based on the multimodal spatial heterogeneous graph of land parcels and the semantic representation of the case layer. A heterogeneous graph neural network is determined, and the model is trained on the heterogeneous graph neural network based on the knowledge graph of urban renewal cases to obtain an urban renewal case recommendation model. The target plot is obtained, and urban renewal case recommendation is performed on the target plot using the urban renewal case recommendation model to obtain the urban renewal case recommendation result.

2. The method for recommending urban renewal cases according to claim 1, characterized in that, The multi-source spatial data includes text data and vector data; The process of acquiring multi-source spatial data and preprocessing the multi-source spatial data to obtain preprocessed text data and preprocessed vector data specifically includes: The text data is obtained from multi-source spatial data, wherein the text data includes texts of typical urban renewal cases; The text data is cleaned and standardized to obtain preprocessed text data; The vector data is obtained from multi-source spatial data, wherein the vector data includes urban land use type maps, population raster data, open-source map road network data, and point of interest data; The vector data is subjected to coordinate projection processing and region clipping processing to obtain preprocessed vector data.

3. The method for recommending urban renewal cases according to claim 1, characterized in that, The process of semantic segmentation and similarity calculation of the preprocessed text data to obtain a candidate segment set, and then transforming the candidate segment set to obtain a case-level semantic representation, specifically includes: The preprocessed text data is semantically segmented to obtain multiple semantic text fragments; Obtain preset case table information, and perform header standardization and content flattening processing on the preset case table information to obtain a unified semantic text table; Multiple target fields are obtained from the unified semantic text table, and a hybrid retrieval strategy is used to calculate the similarity between the multiple target fields and the multiple semantic text fragments to obtain a candidate fragment set; A preset prompt word template is determined, and the candidate fragment set is converted into structured tags using the preset prompt word template. Then, a case-level semantic representation is obtained based on the structured tags.

4. The method for recommending urban renewal cases according to claim 3, characterized in that, The hybrid retrieval strategy includes the Best Matching 25 algorithm and the Sentence-BERT algorithm; The method of employing a hybrid retrieval strategy to calculate the similarity between multiple target fields and multiple semantic text fragments to obtain a candidate fragment set specifically includes: The Best Matching 25 algorithm is used to calculate the first similarity between multiple target fields and multiple semantic text fragments, resulting in multiple first similarity results; The Sentence-BERT algorithm is used to calculate the second similarity between multiple target fields and multiple semantic text fragments, resulting in multiple second similarity results; Multiple first similarity results are added to multiple second similarity results to obtain multiple total similarity results. A preset number of semantic text segments are selected from the multiple semantic text segments in descending order of the total similarity results to obtain a candidate segment set.

5. The method for recommending urban renewal cases according to claim 1, characterized in that, The process of determining a heterogeneous graph neural network and training the network based on the urban renewal case knowledge graph to obtain an urban renewal case recommendation model specifically includes: A heterogeneous graph neural network is determined, and the urban renewal case knowledge graph is input into the heterogeneous graph neural network. The urban renewal case knowledge graph is encoded by the heterogeneous graph neural network to obtain a high-dimensional representation vector. The high-dimensional representation vector is subjected to graph-level pooling to obtain a graph-level representation vector; Calculate the positive-negative contrast loss and semantic consistency loss of the graph-level representation vector, and perform a weighted summation of the positive-negative contrast loss and the semantic consistency loss to obtain the overall loss; The heterogeneous graph neural network is optimized based on the overall loss to obtain a city renewal case recommendation model.

6. The method for recommending urban renewal cases according to claim 1, characterized in that, The process of acquiring the target land parcel and recommending urban renewal cases for the target land parcel using the urban renewal case recommendation model to obtain urban renewal case recommendation results specifically includes: Obtain the target plot and input the target plot into the urban renewal case recommendation model. Calculate the target similarity results between the target plot and historical case nodes in the urban renewal case knowledge graph using the urban renewal case recommendation model. A preset number of updated cases are selected from the historical case nodes according to the target similarity results in descending order, and the preset number of updated cases are used as the urban renewal case recommendation results.

7. A recommendation system for urban renewal cases, characterized in that, The urban renewal case recommendation system is used to implement the urban renewal case recommendation method as described in any one of claims 1-6, wherein the urban renewal case recommendation system includes: The data preprocessing module is used to acquire multi-source spatial data and preprocess the multi-source spatial data to obtain preprocessed text data and preprocessed vector data. The case layer construction module is used to perform semantic segmentation and similarity calculation on the preprocessed text data to obtain a candidate segment set, and to transform the candidate segment set to obtain a case layer semantic representation. The knowledge graph generation module is used to construct a buffer based on the preprocessed vector data, perform node attribute encoding and edge relationship construction on the buffer to obtain a multimodal spatial heterogeneous graph of the land parcel layer, and construct an urban renewal case knowledge graph based on the semantic representation of the case layer and the multimodal spatial heterogeneous graph of the land parcel layer. The case recommendation model construction module is used to determine the heterogeneous graph neural network and train the heterogeneous graph neural network based on the urban renewal case knowledge graph to obtain the urban renewal case recommendation model. The case recommendation result output module is used to obtain the target plot and recommend urban renewal cases for the target plot through the urban renewal case recommendation model to obtain the urban renewal case recommendation result.

8. A terminal, characterized in that, The terminal includes: a memory, a processor, and a recommendation program for urban renewal cases stored in the memory and executable on the processor. When the recommendation program for urban renewal cases is executed by the processor, it implements the steps of the recommendation method for urban renewal cases as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a recommendation program for urban renewal cases, which, when executed by a processor, implements the steps of the urban renewal case recommendation method as described in any one of claims 1-6.

Images