A method and system for urban renewal area and time sequence demarcation based on network evaluation data

Abstract

The application discloses a kind of urban renewal area and time sequence demarcation method and system based on network evaluation data, comprising: collecting network evaluation data and urban renewal text;Filter out negative feedback evaluation, identify city problem;Analysis building surface damage or aging;The city problem point set is divided into point type facility renewal area, line type street style renewal area, face type environment renovation area, and the update time sequence of all regions is calculated, and the update time sequence stage is divided;By Word2Vec feature vectorization algorithm, the core features of urban renewal area and urban renewal text are respectively output as vector set;Cosine similarity calculation formula is calculated, the similarity value of each pair of vector set is obtained, and the maximum value is obtained The highest similarity matching result;Collect and compare the number of negative feedback evaluations before and after updating, if the number of negative feedback evaluations increases, determine that the region update fails, return to city problem identification step automatically;The application effectively realizes the accurate planning and efficient and orderly implementation of urban renewal.

Description

TECHNICAL FIELD

[0001] The present application relates to a city renewal area and timing division method and system, in particular to a city renewal area and timing division method and system based on network evaluation data. BACKGROUND

[0002] In recent years, with the development of cities and changes in residents' needs, traditional urban renewal models face many challenges. Especially in the process of city renewal area and timing division, traditional methods are based on data collection and experience judgment, relying on field reconnaissance and field mapping, which increases the time and labor cost. At the same time, the traditional questionnaire survey of public participation research method has small sample size and difficulty in collection, which reduces the accuracy and speed of city problem and renewal area identification, and is difficult to effectively solve practical problems, and may even cause resource waste or new contradictions. With the wide application of network evaluation data, especially on social media and review platforms, residents' feedback has gradually become an important reference information for urban renewal. However, the current use of these data is still preliminary, mainly relying on simple statistics and analysis, lacking systematic and intelligent processing means. Traditional renewal area division is usually based on administrative division or historical problems, and cannot effectively reflect the real-time demand changes of residents. With the help of artificial intelligence and big data technology, especially the application of natural language processing and deep learning models, it is more feasible to divide the city renewal area and timing based on network evaluation data. SUMMARY

[0003] The purpose of the present application is to provide a city renewal area and timing division method and system based on network evaluation data, which can improve the accuracy of city problem identification and the accuracy of city renewal scheme matching.

[0004] Technical scheme: The city renewal area and timing division method based on network evaluation data provided by the present application comprises:

[0005] (1) Collecting network evaluation data and city renewal text; the network evaluation data includes evaluation text, evaluation photo, number of likes and time information, and the network evaluation data carries location information and time information;

[0006] (2) Based on the collected network evaluation data, city associated semantic extraction and sentiment semantic analysis are carried out, negative feedback evaluation is screened out, and city problems are identified;

[0007] (3) The collected evaluation photos are preprocessed by using image preprocessing technology, and the buildings with abnormalities are identified by using residual network model, and the damage or aging of the surface of the buildings is analyzed;

[0008] (4) merging the urban problems and the abnormal buildings to generate a set of urban problem points, dividing the set of urban problem points into a point-type facility renewal area, a line-type street style renewal area, and a surface-type environment renovation area, and calculating a renewal time sequence order of all areas and dividing the renewal time sequence stage;

[0009] (5) using a Word2Vec feature vectorization algorithm, outputting core features and urban renewal texts of the urban renewal area as a vector set, respectively; the core features include function, location, core problem, and facility type; calculating a cosine similarity calculation formula to obtain a similarity value of each pair of vector sets, and taking the maximum value to obtain the highest similarity matching result;

[0010] (6) collecting and comparing the number of negative feedback evaluations before and after the update, if the number of negative feedback evaluations increases, determining that the update of the area is invalid, and returning to the step of identifying urban problems.

[0011] Further, the step (1) comprises:

[0012] Through social media, network evaluation data carrying location information and time information are obtained, including evaluation texts, evaluation photos, and the number of likes, the obtained data is processed in real time through an edge computing device, and the processed data is stored;

[0013] The urban renewal texts obtained from the planning institutes in various places are converted into a text format and cleaned, the content of urban renewal is identified from the cleaned texts, and the texts of the content are converted into structured data and saved.

[0014] Further, the step (2) comprises:

[0015] The collected evaluation texts are text cleaned and segmented to obtain fixed-dimension vectors and location encodings; the processed vectors and location encodings are input into a Transformer model to output a probability distribution representing a sentiment polarity, and texts with a negative sentiment probability greater than a preset threshold are regarded as negative feedback evaluations; the semantic relationship between words is captured through the self-attention mechanism of the Transformer model, the text corresponding to the negative feedback evaluation and the context are associated, and the key semantic information related to the urban problem targeted by the negative feedback evaluation is identified;

[0016] The key semantic information, time information, and location information corresponding to the screened negative feedback evaluations are input into a Latent Dirichlet Allocation generative topic model LDA, the topics of the texts located at the same location at the same time node are identified, the topics with an occurrence frequency higher than a preset number of times are regarded as persistent urban problems of the area, and the topics with an occurrence frequency lower than a preset number of times are regarded as occasional urban problems;

[0017] Combine the name, date, spatial location and negative feedback evaluation quantity of the occasional urban problem to construct a case library of urban occasional problems and network evaluation association over the years, and train the Prophet trend prediction model through the data in the case library to predict the occurrence probability and region of the occasional urban problem in the next year, and patrol the region with a probability higher than the preset proportion to determine the spatial location of the occasional urban problem.

[0018] The case library contains the name, date, spatial location and negative feedback evaluation quantity of the occasional urban problem.

[0019] Further, the step (3) comprises:

[0020] The evaluation photos collected are preprocessed by using image preprocessing technology, and the preprocessing includes: using a median filter algorithm to remove random noise in the image, applying a histogram equalization technique to improve the contrast and brightness of the image and enhance the details of the building, and performing geometric and color correction;

[0021] A preset proportion is divided from the preprocessed evaluation photos, and the abnormalities of the buildings therein are labeled to obtain an abnormal image data set, and the abnormalities are damage or aging; a pre-trained ResNet residual network model is called through a deep learning framework, and supervised learning is performed on the abnormal image data set of the building to fine-tune the model parameters; the fine-tuned model is used to detect the abnormal part of the building in all preprocessed evaluation photos, identify the area with damage or aging, and locate it in combination with the spatial coordinates;

[0022] Three-dimensional point cloud data of the building with abnormalities is generated by field measurement.

[0023] Further, the step (4) comprises:

[0024] A minimum spanning tree is generated from the urban problem point set by Kruskal algorithm, and linear structure that meets the conditions is marked as a linear street landscape renewal area by setting a minimum length threshold and a maximum density threshold;

[0025] The remaining point set in the urban problem point set is analyzed by DBSCAN density clustering algorithm, and the minimum point density area where the number of points in the neighborhood reaches is marked as a face-type environment improvement area, and the remaining isolated point set is marked as a point-type facility renewal area;

[0026] The urgency weighted index of each region is calculated by combining the number of negative feedback evaluations, the number of visitors and the population density data to obtain the renewal time sequence sorting of all regions, and then the regions are divided into several renewal time sequence stages by the natural breakpoint method.

[0027] Further, the Kruskal algorithm is used to generate a minimum spanning tree from the city problem point set, comprising:

[0028] The identified abnormal point is taken as a node v i , and the coordinates are represented as (x i , y i ). Let V = {v1, v2,..., vn} be the node set, n be the total number of nodes, i = 1, 2,..., n.

[0029] The weight of all edges is calculated, and E = {e1, e2,..., em} is the edge set, m is the total number of edges, each edge e m connects two nodes and is represented as e k = (v k , v i ), j = 1, 2,..., n; k = 1, 2,..., m; and the Manhattan distance between nodes is taken as the weight w(e k ) of each edge:

[0030] w(e k ) = |x i -x j | + |y i -y j |.

[0031] The edge set E is sorted, and the sorted edge set is sorted in ascending order of weight:

[0032] E ′ = {e′1, e′2,..., e′ m}

[0033] w(e′1) ≤ w(e′2) ≤... ≤ w(e′ m )

[0034] Initialize the spanning tree, and let the empty edge set be T. Add each edge that meets the condition to the set T, and finally the set T contains n-1 edges connecting all nodes, and the total weight is the minimum;

[0035] Add edges step by step, perform loop check from the edge with the smallest weight, and use the Union-Find data structure to check whether the two nodes v i and v j connected by the sorted edge e′ k belong to different sets; if v i and v j do not belong to the same set, add this edge to the set T and merge the sets containing v i and v j ; if v i and v j belong to the same set, discard this edge.i and v j If the same set has been visited, then skip this edge to ensure no loop is formed;

[0036] The linear structure meeting the condition is marked as a linear street landscape renewal area by setting a minimum length threshold and a maximum density threshold, comprising:

[0037] Set the minimum length threshold as L min , retain all edges with length l≥L min in the minimum spanning tree MST, and screen out the linear structure formed by the edges;

[0038] Set the maximum density threshold as D max , control the density of the linear structure, calculate the total length L, node density D nodes , edge density D edges and comprehensive density D com of the linear structure, assume that the linear structure is composed of m edges e1, e2,…, e m , the length of each edge is l(e1), l(e2),…, l(e m ), then the total length The node density The edge density The comprehensive density D com =α×D nodes +β×D edges , wherein α represents the weight of the node density D nodes in the comprehensive density D com , β represents the weight of the edge density D edges in the comprehensive density D com , retain the linear structure with D com <D max , and mark it as a linear street landscape renewal area.

[0039] Further, the remaining point set is analyzed by the DBSCAN density clustering algorithm, the minimum point density area with the number of points in the neighborhood reaching the minimum point density is marked as a face-type environmental improvement area, and the remaining isolated point set is marked as a point-type facility renewal area, comprising:

[0040] Set all point sets of the remaining points p as D, that is, D={p1, p2,…, p n}; set the neighborhood radius as ∈, and all points q with a distance from p not more than the neighborhood radius ∈ form a neighborhood N ∈ (p), that is:

[0041] N ∈ (p)={q∈D|distance(p,q)≤∈}

[0042] Let the threshold for the number of points in the neighborhood be MinPts. If the number of points q, m, satisfies the predetermined threshold, i.e., m≥MinPts, then p is considered a core point.

[0043] p and its neighboring points expand to form a dense region C, where |C| is the number of points in C, and S min Let |C| be the threshold of C; if |C|≥S min Then it is marked as a surface-type environmental remediation area; if |C| min This area is then marked as a point-type facility renewal area.

[0044] Furthermore, the urgency-weighted index for each region is calculated by combining the number of negative feedback evaluations, visitor numbers, and population density data to obtain the update time-series ranking of all regions, including:

[0045] The weights for the number of negative feedbacks, the number of visitors, and the population density data are defined as w1, w2, and w3, respectively.

[0046] Standardize the data on the number of negative feedbacks, the number of visitors, and the population density.

[0047] Calculate the urgency-weighted index I for each update region using standardized data:

[0048] I = w1 × Norm (Number of negative feedback) + w2 × Norm (Number of visitors) + w2 × Norm (Population density)

[0049] Norm(·) represents the standardized value;

[0050] The urgency-weighted index I of the calculated update regions is sorted by size to form an update time series table. The higher the index, the more urgent the update need of the region.

[0051] Furthermore, the Word2Vec feature vectorization algorithm is used to output the core features of the urban renewal area and the urban renewal text as vector sets, including:

[0052] Define the characteristics of the urban renewal area as T region The urban renewal text includes functions, location, core issues, and facility types. text :

[0053] T region ={Function, Location, Core Issue, Facility Type}

[0054] T text ={Urban Renewal Text Content}

[0055] Let each word x in the core features a The vector representation is v(x) a ), by weighted average T​region Transforming into V region , T text Transforming into V text , that is:

[0056] v(x a ) = Word2Vec(x a )

[0057]

[0058] wherein, lambda a is the weight of the word x a , mu b is the weight of the word x b ;

[0059] The calculation formula of the cosine similarity is used to obtain the similarity value of each pair of vector sets, and the maximum value is taken to obtain the matching result with the highest similarity, comprising: using the feature vector V region and the text vector V text Calculate the similarity through the cosine similarity:

[0060]

[0061] The calculated similarity value represents the matching degree between the characteristics of the region and the urban renewal text strategy; the higher the similarity, the more suitable the urban renewal text strategy is for the update of the region.

[0062] Based on the same inventive concept, the application also provides a city renewal area and time sequence determination system based on network evaluation data, comprising:

[0063] A collection module is used to collect network evaluation data and urban renewal texts; the network evaluation data includes evaluation texts, evaluation photos and the number of likes, and the network evaluation data carries location information and time information;

[0064] A city problem identification module is used to perform city correlation semantic extraction and sentiment semantic analysis based on the collected network evaluation data, filter out negative feedback evaluations, and identify city problems;

[0065] An anomaly identification module is used to preprocess the collected evaluation photos by using image preprocessing technology, and identify buildings with abnormalities through a residual network model to analyze the damage or aging of the building surface;

[0066] A division module is used to combine city problems and buildings with abnormalities to generate a city problem point set, divide the city problem point set into point-type facility renewal areas, line-type street style renewal areas and surface-type environment improvement areas, calculate the update time sequence of all areas, and divide the update time sequence stage;

[0067] a calculation module for outputting core features and urban renewal texts of the urban renewal area as vector sets respectively through a Word2Vec feature vectorization algorithm, wherein the core features include function, location, core problem and facility type; and a calculation formula of cosine similarity is used to obtain a similarity value of each pair of vector sets, and the maximum value is taken to obtain a matching result with the highest similarity;

[0068] a verification module for collecting and comparing the number of negative feedback evaluations before and after the update, and determining that the area update is invalid if the number of negative feedback evaluations increases, and returning to the urban problem identification module.

[0069] Advantages: Compared with the prior art, the present application reduces the time cost of area analysis and division by about 30% through the integration of network evaluation data, problem point set analysis and intelligent algorithm, and reduces resource consumption by about 25% through automated data processing and real-time calculation. The present application shortens the urban renewal area analysis work which traditionally takes 4-6 weeks to 1 week with the help of automated analysis technology, which significantly shortens the planning and implementation cycle, optimizes the allocation and utilization efficiency of resources, and reduces the time cost and resource consumption of urban renewal area and timing division. Compared with the traditional method, the present application further improves the accuracy of urban problem identification by about 50% through the application of natural language processing algorithm based on Transformer and LDA generative probability model, and improves the accuracy of matching by about 45% through the cosine similarity calculation strategy. Based on the deep semantic analysis of negative feedback data and the DBSCAN density clustering algorithm, the present application can accurately identify potential persistent and occasional problem areas in the city, and through intelligent strategy matching, ensure that each update area can get the most suitable solution, which improves the overall effect of urban renewal and the scientificity of planning, improves the accuracy of urban problem identification and the accuracy of urban renewal scheme matching. Compared with the traditional method, the present application expands the applicable range of urban renewal scheme in different area types by about 35% and improves the design depth by about 25% through image preprocessing technology and Word2Vec feature vectorization algorithm for the scientific division of face-type environment improvement areas and point-type facility renewal areas, which improves the flexibility of urban renewal planning to meet the diversified urban development needs, and expands the applicable range and design depth of urban renewal scheme in different area types. BRIEF DESCRIPTION OF DRAWINGS

[0070] Figure 1 is a method flowchart of the present application;

[0071] Figure 2 is a building anomaly identification schematic diagram of an embodiment of the present application;

[0072] Figure 3A schematic diagram of the urban renewal area and time sequence demarcation of the embodiment of the present application. DETAILED DESCRIPTION

[0073] The technical solutions in the embodiments of the present application will be described clearly and completely below in combination with the drawings and embodiments. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0074] Embodiment 1: As shown in the accompanying drawings, the urban renewal area and time sequence demarcation method based on network evaluation data of the present embodiment comprises: Figure 1

[0075] (1) Collecting network evaluation data and urban renewal texts; the network evaluation data comprises evaluation texts, evaluation photos and the number of likes, and the network evaluation data carries location information and time information;

[0076] (2) Performing urban correlation semantic extraction and sentiment semantic analysis based on the collected network evaluation data, screening out negative feedback evaluations, and identifying urban problems;

[0077] (3) Preprocessing the collected evaluation photos by using image preprocessing technology, and identifying the existing abnormal buildings by using a residual network model, and analyzing the damage or aging of the surface of the buildings;

[0078] (4) Merging the urban problems and the existing abnormal buildings to generate a set of urban problem points, dividing the set of urban problem points into point-type facility renewal areas, line-type street landscape renewal areas and surface-type environment renovation areas, and calculating the renewal time sequence order of all the areas and dividing the renewal time sequence stages;

[0079] (5) Outputting the core features of the urban renewal areas and the urban renewal texts into vector sets respectively by using a Word2Vec feature vectorization algorithm; the core features comprise functions, locations, core problems and facility types; calculating a cosine similarity calculation formula to obtain the similarity value of each pair of vector sets, and taking the maximum value to obtain the matching result with the highest similarity;

[0080] (6) Collecting and comparing the number of negative feedback evaluations before and after the renewal, and if the number of negative feedback evaluations increases, it is determined that the renewal of the area is invalid, and the step of identifying urban problems is returned.

[0081] Specifically, step (1) network evaluation data and urban renewal text collection

[0082] ​Through the data acquisition module, connect to the social media API interface, obtain network evaluation data carrying location information including evaluation text, evaluation photos, like number, time information, and through the NVIDIA Jetson Nano edge computing device for real-time processing and data transmission. Through the ScanSnap iX1600 scanner, 500 sets of urban renewal texts are collected and stored in a big data management platform supporting structured retrieval.

[0083] Step (2) Identification of Urban Problems Based on Negative Feedback Evaluation

[0084] Based on the evaluation text, negative feedback evaluation is screened, and persistent and occasional urban problems are identified. Through the Transformer model based on self-attention mechanism (Self-Attention), urban-related semantic extraction and sentiment semantic analysis of evaluation text data are performed, and negative feedback evaluation is screened. Further persistent urban problem identification, the screened key semantics, time information and location information are used as input parameters, and the LDA generative topic model is used to identify the frequently occurring urban problems in the region. For occasional urban problems, a historical urban occasional problem-network evaluation correlation case library is constructed, a Prophet trend prediction model is trained, negative feedback evaluation is used as the input of the case library, and the occurrence probability and risk area of occasional urban problems are predicted. A UAV equipped with a LiDAR laser radar sensor is used to patrol the risk area to determine the problem point.

[0085] Step (3) Architectural Style Diagnosis Based on Residual Network Model

[0086] Through image preprocessing technology, the evaluation pictures are denoised, enhanced and corrected. Through the PyTorch deep learning framework, the pre-trained ResNet residual network model is called, and the model is fine-tuned to detect the abnormal parts of the building. A patrol vehicle equipped with a light projector and a high-resolution camera is used to obtain 3D point cloud data of abnormal location buildings through structured light technology, and the deformation and damage of the building surface are analyzed.

[0087] Step (4) Urban Renewal Area and Time Sequence Delimitation

[0088] The urban problem point set is divided into point-type facility renewal area, linear street style renewal area, and surface-type environment improvement area. The Kruskal algorithm is used to generate a minimum spanning tree from the urban problem point set. By setting a minimum length threshold and a maximum density threshold, linear structures that meet the conditions are marked as linear street style renewal areas. The DBSCAN density clustering algorithm is used to analyze the remaining point set. The minimum point density area is marked as a surface-type environment improvement area if the number of points in the neighborhood reaches the minimum point density. The remaining isolated point set is marked as a point-type facility renewal area. The urgency weighted index of each renewal area is calculated by combining the number of negative feedback, the number of visitors, and the population density data. The update timing sequence of all areas is obtained, and the natural breakpoint method is used to divide the update timing sequence into three stages (5-year update, 10-year update, and 15-year update), as shown in Figure 3 .

[0089] Step (5) City renewal strategy matching

[0090] The core features (function, location, core problem, and facility type) of the city renewal area and the city renewal text are output as vector sets through the Word2Vec feature vectorization algorithm. The cosine similarity calculation formula is used to calculate the similarity value of each pair of vector sets, and the maximum value is taken to obtain the highest similarity matching result.

[0091] Step (6) City renewal effect verification and iteration

[0092] The number of negative feedback evaluations before and after the update is collected and compared. If the number of negative feedback evaluations increases, it is considered that the area has updated effectively, and the process returns to step (2) to identify and inspect the problems again.

[0093] The embodiment can further include step (7): digital sand table display of city renewal plan

[0094] A city renewal strategy development sand table is constructed. Through augmented reality (AR) and virtual reality (VR) devices, the overall layout and strategic planning of city renewal are displayed. Through the Tableau data visualization platform, the city problems based on network evaluation data analysis are displayed in the form of heat maps and trend charts. The MATSim time sequence simulation tool is used to demonstrate the city renewal timing animation.

[0095] Further, in step (1), the DGX computer workstation with a built-in data collection module connects to the social media API interface to obtain network evaluation data of the target city carrying location information and time information, including evaluation text, URL of evaluation photos, and number of likes. The NVIDIA Jetson Nano edge computing device is used to process the obtained data in real time, and the processed data is stored in the workstation.

[0096] 500 sets of urban renewal texts obtained from planning institutes across the country are scanned at high resolution by the ScanSnap iX1600 scanner, and the scanned results are converted into searchable text format using OCR technology. The text data is cleaned and formatted, and NLP techniques and keyword matching are used to identify urban renewal-related content from the cleaned text. Specific information is extracted, and the text is converted into structured data and uploaded and stored in the Google BigQuery big data management platform that supports structured retrieval.

[0097] In step (2), the evaluation text obtained in step one is cleaned and segmented using a DGX computer workstation, and the GloVe word embedding model is used to convert words into fixed-dimensional vectors. The processed vector sequence and position encoding are input into the Transformer model, and the probability distribution representing the sentiment polarity is output. Texts with negative sentiment probability greater than 0.7 are selected as negative feedback evaluations. The self-attention mechanism of the Transformer model captures the semantic relationship between words, and the text related to negative feedback evaluations is associated with its context to identify the key semantic information related to specific urban problems targeted by negative feedback evaluations.

[0098] Further input the selected key semantic information, time information and location information into the LDA generative topic model to identify text topics at the same time node and in the same location. Topics with an occurrence frequency higher than 10 times are considered as persistent urban problems in the region, and topics with an occurrence frequency lower than 5 times are considered as occasional urban problems.

[0099] The names of occasional urban problems, dates, spatial locations, and the number of negative feedback evaluations are combined to construct a case library of urban occasional problems and network evaluation correlations over the years. The data is fitted into a model to train the Prophet trend prediction model, and the occurrence probability and risk area of occasional urban problems in the next year are predicted. The unmanned aerial vehicle equipped with LiDAR laser radar sensor is used to conduct field inspection of the risk area with a probability higher than 50%, and the spatial location of the occasional urban problem is determined.

[0100] In step (3), the evaluation photos obtained in step one are preprocessed using a DGX computer workstation, including removing random noise in the image using a median filter algorithm; applying histogram equalization technology to improve the contrast and brightness of the image and enhance the details of the building; and performing geometric and color correction to ensure the authenticity and accuracy of the image.

[0101] Select 10% of the processed photos for manual annotation of building abnormalities. Call the pre-trained ResNet residual network model through the PyTorch deep learning framework, and use the annotated building abnormal image dataset for supervised learning to fine-tune the model parameters. Use the fine-tuned model to detect abnormalities in all evaluation photos, identify areas with damage, aging or other problems, and locate them using spatial coordinates, as shown in Figure 2 .

[0102] Use a patrol vehicle equipped with a light projector and a high-resolution camera. Combine structured light technology to project known patterns of light, and calculate the deformation of the light on the abnormal building surface. Generate high-precision three-dimensional point cloud data for abnormal buildings by field measurement, and analyze the deformation and damage of the building surface.

[0103] In step (4), the Kruskal algorithm is used to generate a minimum spanning tree from the city problem point set, which means:

[0104] (411) Take the identified city problem point set (buildings, facilities and other areas that need to be updated in the city) as the node v i , and express the coordinates as (x i ,y i ). Let V = {v1, v2, …, v n} be the node set.

[0105] (412) Calculate the weight of all edges. Let E = {e1, e2, …, e m} be the edge set, each edge e k connects two nodes and is expressed as e k = (v i ,v j ), and the Manhattan distance between nodes is used as the weight w(e k ) of each edge:

[0106] w(e k ) = |x i -x j | + |y i -y j |.

[0107] (413) Sort the edge set E. The sorted edge set is sorted in ascending order of weight:

[0108] E ′ = {e′1, e′2, …, e′ m}

[0109] w(e′1) ≤ w(e′2) ≤ … ≤ w(e′ m )

[0110] (414) Initialize the spanning tree. Set empty edge set T Add each edge that meets the condition to set T, and the final set T contains n-1 edges connecting all nodes with the minimum total weight.

[0111] (415) Add edges step by step, perform loop check from the edge with the smallest weight, i.e. use Union-Find data structure to check the sorted edge e' k Two connected nodes v i and v j whether belong to different sets. If v i and v j are not in the same set, add this edge to set T and merge the sets where v i and v j are located; if v i and v j are in the same set, skip this edge to ensure that no loop is formed.

[0112] By setting the minimum length threshold and the maximum density threshold, linear structure that meets the condition is marked as linear street landscape renewal area, which refers to:

[0113] (421) Set the minimum length threshold L min , keep all edges with length l≥L min in the minimum spanning tree (MST), and select the linear structure formed by these edges.

[0114] (422) Set the maximum density threshold D max to control the density of linear structure, calculate the total length L, node density D nodes , edge density D edges and comprehensive density D com of linear structure, assuming that the linear structure is composed of m edges e1, e2, …, e m , the length of each edge is l(e1), l(e2), …, l(e m ), then the total length Node density Edge density Comprehensive density D com = α × D nodes + β × D edges (α + β = 1), where α represents the weight of node density D nodes in the comprehensive density D com , and β represents the weight of edge density D edges in the comprehensive density D com , keep the linear structure with D com <D max and mark it as a linear street landscape renewal area.

[0115] The remaining point set is analyzed by the DBSCAN density clustering algorithm, and the point number in the neighborhood is marked as a surface type environment improvement area, and the remaining isolated point set is marked as a point type facility update area, which means:

[0116] (431) Let all the point sets of the remaining point p be D, that is, D={p1, p2, …, p n}. Let the neighborhood radius be ∈, and all points q with a distance from p not exceeding the neighborhood radius ∈ form a neighborhood N ∈ (p), that is:

[0117] N ∈ (p)={q∈D|distance(p,q)≤∈}

[0118] (432) Let the point number threshold in the neighborhood be MinPts, and if the point number m satisfies the preset threshold, that is, m≥MinPts, then p is considered to be a core point.

[0119] (433) p and other points in its neighborhood are expanded to form a dense region C, |C| is the point number in C, and S min is the threshold of C. If |C|≥S min , it is marked as a surface type environment improvement area; if |C|<S min , it is marked as a point type facility update area.

[0120] The point number represents the number of points in a certain area, which is used to represent the degree of density; and the determination method of the minimum point number dense region is:

[0121] Define the neighborhood, draw a circle with a radius of ∈ for each point by DBSCAN;

[0122] Define the minimum point number MinPts, which is the minimum number of points that must be contained in the specified field; when there are at least MinPts points in the neighborhood, it is considered to be a core point, which constitutes a dense region;

[0123] Mark the points that satisfy the condition that the point number in the neighborhood is greater than or equal to MinPts as core points of the dense region;

[0124] Classify the points within ∈ range adjacent to the core point into the same cluster to form a cluster;

[0125] Identify and cluster the dense region to form a minimum point number dense region.

[0126] The urgency weighted index of each update area is calculated by combining the negative feedback number in the network evaluation, the visitor number, and the population density data, and the update time sequence of all areas is sorted, which means:

[0127] (441)Define the weights of the negative feedback quantity, the visitor quantity, and the population density data as w1, w2, and w3, respectively.

[0128] (442)Standardize the negative feedback quantity, the visitor quantity, and the population density data (Norm(·) represents the standardized value) so that the above data can be compared on the same scale:

[0129]

[0130] (443)Calculate the urgency weighted index I of each update area using the standardized data:

[0131] I = w1 x Norm(negative feedback quantity) + w2 x Norm(visitor quantity) + w2 x Norm(population density)

[0132] (444)Sort the urgency weighted index I of the update area according to the calculation result in descending order to form an update time sequence table, and the higher the index, the more urgent the update demand of the area.

[0133] In step (5), the core features (function, location, core problem, and facility type) of the urban renewal area and the urban renewal text are respectively output as vector sets by the Word2Vec feature vectorization algorithm, which means:

[0134] (51) Set the features of the urban renewal area as T region , including function, location, core problem, and facility type, and the urban renewal text as T text :

[0135] T region = {function, location, core problem, facility type}

[0136] T text = {urban renewal text content}

[0137] (52) Set the vector representation of each word x a in the core features as v(x a ), and convert T region into V region and T text into V text by weighted average (λ a is the weight of word x a , and μ b is the weight of word x b ), that is:

[0138] v(x a ) = Word2Vec(x a )

[0139]

[0140]

[0141] wherein, lambda a is the weight of the word x a , mu b is the weight of the word x b ;

[0142] The similarity value of each pair of vector sets is obtained by calculating the cosine similarity calculation formula, and the maximum value is taken to obtain the matching result with the highest similarity. It means: using the feature vector V region and the text vector V text The similarity is calculated by the cosine similarity:

[0143]

[0144] The calculated similarity value represents the matching degree between the characteristics of the region and a certain urban renewal text strategy. The higher the similarity, the more suitable the urban renewal text strategy is for the update of the region.

[0145] Based on the same inventive concept, the embodiment also provides a city renewal area and time sequence division system based on network evaluation data, comprising:

[0146] A collection module is configured to collect network evaluation data and urban renewal texts. The network evaluation data includes evaluation texts, evaluation photos, and the number of likes, and carries location information and time information.

[0147] A city problem identification module is configured to perform city-related semantic extraction and sentiment semantic analysis based on the collected network evaluation data, filter out negative feedback evaluations, and identify city problems.

[0148] An anomaly identification module is configured to preprocess the collected evaluation photos using image preprocessing technology, and identify buildings with abnormalities through a residual network model to analyze the damage or aging of the building surface.

[0149] A division module is configured to combine city problems and buildings with abnormalities to generate a city problem point set, divide the city problem point set into point-type facility renewal areas, line-type street landscape renewal areas, and face-type environment improvement areas, calculate the update time sequence of all areas, and divide the update time sequence stage.

[0150] A calculation module is configured to output core features and urban renewal texts of an urban renewal area into vector sets respectively by a Word2Vec feature vectorization algorithm; the core features include function, location, core problem and facility type; a cosine similarity calculation formula is calculated to obtain a similarity value of each pair of vector sets, and a maximum value is taken to obtain a matching result with the highest similarity;

[0151] A verification module is configured to collect and compare the number of negative feedback evaluations before and after the update, and if the number of negative feedback evaluations increases, it is determined that the area update is invalid, and the city problem identification module is returned.

Claims

1. A method for delineating urban renewal areas and time sequences based on network evaluation data, characterized in that, include: (1) Collect online evaluation data and urban renewal texts; The online evaluation data includes evaluation text, evaluation photos, and the number of likes. The online evaluation data also carries location and time information; including: By acquiring online evaluation data with location and time information through social media, including evaluation text, evaluation photos and the number of likes, the acquired data is processed in real time through edge computing devices and the processed data is stored. Urban renewal texts obtained from planning institutes in various regions are converted into text format and cleaned. The content of urban renewal is identified from the cleaned text, and the text of the content is converted into structured data and saved. (2) Based on the collected network evaluation data, perform urban association semantic extraction and sentiment semantic analysis to screen out negative feedback evaluations and identify urban problems; including: The collected evaluation texts are cleaned and segmented to obtain fixed-dimensional vectors and positional codes. The processed vectors and positional codes are then input into a Transformer model, which outputs a probability distribution representing sentiment polarity. Texts with negative sentiment probabilities greater than a preset threshold are used as negative feedback evaluations. The self-attention mechanism of the Transformer model is used to capture semantic relationships between words, and the text corresponding to the negative feedback evaluation is associated with the context to identify the key semantic information related to urban issues targeted by the negative feedback evaluation. The key semantic information, time information and location information corresponding to the selected negative feedback evaluations are input into the LatentDirichlet Allocation generative topic model (LDA) to identify the topics of texts located in the same location at the same time point. Topics that appear more frequently than a preset number of times are identified as persistent urban problems in the region, while topics that appear less frequently than a preset number of times are identified as occasional urban problems. By combining the names, dates, spatial locations, and negative feedback evaluations of the sporadic urban problems, a case library of historical sporadic urban problems and online evaluations is constructed. The Prophet trend prediction model is trained using data from the case library to predict the probability and region of sporadic urban problems in the coming year. Areas with a probability higher than a preset proportion are inspected to determine the spatial location of sporadic urban problems. The case library contains the name, date, spatial location, and number of negative feedback evaluations for incidental urban issues. (3) Use image preprocessing technology to preprocess the collected evaluation photos, and use the residual network model to identify buildings with abnormalities and analyze the damage or aging of the building surface. (4) Merge urban problems and buildings with abnormalities to generate a set of urban problem points, divide the urban problem point set into point-type facility renewal areas, linear street appearance renewal areas and surface-type environmental improvement areas, and calculate the renewal time sequence of all areas and divide the renewal time sequence stages. (5) Using the Word2Vec feature vectorization algorithm, the core features of the urban renewal area and the urban renewal text are output as vector sets respectively; the core features include function, location, core issues and facility type; the similarity value of each pair of vector sets is obtained by using the cosine similarity calculation formula, and the maximum value is taken to obtain the matching result with the highest similarity. (6) Collect and compare the number of negative feedback evaluations before and after the update. If the number of negative feedback evaluations increases, the update of the area is deemed to have failed, and the process returns to the step of identifying urban problems.

2. The method for urban renewal area and time series delineation based on network evaluation data according to claim 1, characterized in that, Step (3) includes: The collected evaluation photos are preprocessed using image preprocessing techniques. The preprocessing includes: removing random noise from the image using a median filtering algorithm, applying histogram equalization to improve the contrast and brightness of the image and enhance building details, and performing geometric and color correction. An abnormal image dataset is obtained by dividing the pre-processed evaluation photos into a predetermined proportion and labeling the abnormal buildings within them. The abnormalities are defined as damage or aging. A pre-trained ResNet residual network model is called through a deep learning framework, and supervised learning is performed using the abnormal building image dataset to fine-tune the model parameters. The fine-tuned model is then used to detect abnormal parts of buildings in all pre-processed evaluation photos, identify areas with damage or aging, and locate them using spatial coordinates. 3D point cloud data of buildings with anomalies are generated through on-site measurements.

3. The method for delineating urban renewal areas and time sequences based on network evaluation data according to claim 1, characterized in that, Step (4) includes: The Kruskal algorithm is used to generate a minimum spanning tree from the set of urban problem points. By setting minimum length threshold and maximum density threshold, linear structures that meet the conditions are selected and marked as linear streetscape update areas. The DBSCAN density clustering algorithm is used to perform cluster analysis on the remaining point set in the urban problem point set. The dense area with the minimum number of points in the neighborhood is marked as the area of ​​surface environmental improvement, and the isolated point set is marked as the area of ​​point facility renewal. By combining the number of negative feedback evaluations, visitor numbers, and population density data, the urgency weighted index of each region is calculated, and the update time sequence of all regions is obtained. Then, the natural breakpoint method is used to divide it into several update time sequence stages.

4. The method for delineating urban renewal areas and time sequences based on network evaluation data according to claim 3, characterized in that, The process of generating a minimum spanning tree from the set of urban problem points using Kruskal's algorithm includes: The identified abnormal points will be used as nodes. And represent the coordinates as ,set up For a set of nodes, The total number of nodes. ; Calculate the weights of all edges, let... Let be the set of edges. The total number of edges, each edge Connect two nodes and represent as , ; Use the Manhattan distance between nodes as the weight of each edge. : ； Set of opposite edges Sort the edges, and then sort the edges in ascending order of their weights: ， ， Initialize the spanning tree with an empty edge set. Add each edge that satisfies the condition to the set. In the final set Includes connections to all nodes The edge with the minimum total weight; Edges are added incrementally, and cycle checks are performed starting with the edge with the smallest weight. The sorted edges are checked using a Union-Find data structure. Two connected nodes and Do they belong to different sets? If and If the edge is not in the same set, add it to the set. and merge and The set in which it belongs; if and If the elements are already in the same set, skip this edge to ensure that a cycle is not formed. The process of selecting linear structures that meet certain conditions and marking them as linear streetscape update areas by setting minimum length thresholds and maximum density thresholds includes: Let the minimum length threshold be Preserve all lengths in the minimum spanning tree (MST). The edges are selected, and the linear structure formed by the edges is filtered out. Let the maximum density threshold be Used to control the density of linear structures and calculate the total length of linear structures. Node density Edge density and overall density Assuming the linear structure is composed of Strip edge Composed of, each side has a length of Then the total length Node density edge density Overall density ,in Represents node density In overall density The weights in Represents edge density In overall density The weights in the data are retained. The linear structure is marked as a linear streetscape renewal area.

5. The method for delineating urban renewal areas and time sequences based on network evaluation data according to claim 3, characterized in that, The remaining point set is clustered using the DBSCAN density clustering algorithm. Areas with a high density of points within their neighborhoods are marked as area-type environmental remediation areas, while isolated point sets are marked as point-type facility renewal areas. This includes: Let the remaining points be The set of all points is ,Right now Let the neighborhood radius be... Distance point No more than the neighborhood radius All points Forming a neighborhood ,Right now: ， Let the threshold for the number of points in the neighborhood be . ,like Number of points Meeting the reservation threshold, i.e. ,but It is considered the core point; It expands to form a dense region with other points in its neighborhood. , yes The number of points in the middle, for The threshold; if Then mark it as an area for surface environmental remediation; if This area is then marked as a point-type facility renewal area.

6. The method for delineating urban renewal areas and time sequences based on network evaluation data according to claim 3, characterized in that, The urgency-weighted index for each region is calculated by combining the number of negative feedback evaluations, visitor numbers, and population density data, resulting in an update time-series ranking of all regions, including: The weights for the number of negative feedbacks, the number of visitors, and the population density data are defined as follows: ; Standardize the data on the number of negative feedbacks, the number of visitors, and the population density. The urgency-weighted index for each update region was calculated using standardized data. : ， This represents the standardized value; Based on the calculation Sort by size to form an update time series table; the higher the index, the more urgent the update need for that area.

7. The method for delineating urban renewal areas and time sequences based on network evaluation data according to claim 1, characterized in that, The Word2Vec feature vectorization algorithm outputs the core features of the urban renewal area and the urban renewal text as vector sets, including: The characteristics of urban renewal areas are defined as follows: The urban renewal text includes functions, location, core issues, and facility types. : ， ， Let each word in the core features The vector representation is By weighted average Transform into ,Will Transform into ,Right now: ， ， ， in, It is a word The weight, It is a word The weights; The process involves obtaining the similarity value for each pair of vector sets using the cosine similarity calculation formula, and taking the maximum value to obtain the matching result with the highest similarity. This includes using feature vectors... and text vector Calculate similarity using cosine similarity: ， The calculated similarity value represents the degree of matching between the features of the region and a certain urban renewal text strategy; the higher the similarity, the more suitable the urban renewal text strategy is for the renewal of the region.

8. A system for delineating urban renewal areas and time sequences based on network evaluation data, characterized in that, include: The data collection module is used to collect online evaluation data and urban renewal text. The online evaluation data includes evaluation text, evaluation photos, and the number of likes. The online evaluation data also carries location and time information; including: By acquiring online evaluation data with location and time information through social media, including evaluation text, evaluation photos and the number of likes, the acquired data is processed in real time through edge computing devices and the processed data is stored. Urban renewal texts obtained from planning institutes in various regions are converted into text format and cleaned. The content of urban renewal is identified from the cleaned text, and the text of the content is converted into structured data and saved. The urban problem identification module is used to extract urban-related semantics and perform sentiment semantic analysis based on collected online evaluation data, filter out negative feedback evaluations, and identify urban problems; including: The collected evaluation texts are cleaned and segmented to obtain fixed-dimensional vectors and positional codes. The processed vectors and positional codes are then input into a Transformer model, which outputs a probability distribution representing sentiment polarity. Texts with negative sentiment probabilities greater than a preset threshold are used as negative feedback evaluations. The self-attention mechanism of the Transformer model is used to capture semantic relationships between words, and the text corresponding to the negative feedback evaluation is associated with the context to identify the key semantic information related to urban issues targeted by the negative feedback evaluation. The key semantic information, time information and location information corresponding to the selected negative feedback evaluations are input into the LatentDirichlet Allocation generative topic model (LDA) to identify the topics of texts located in the same location at the same time point. Topics that appear more frequently than a preset number of times are identified as persistent urban problems in the region, while topics that appear less frequently than a preset number of times are identified as occasional urban problems. By combining the names, dates, spatial locations, and negative feedback evaluations of the sporadic urban problems, a case library of historical sporadic urban problems and online evaluations is constructed. The Prophet trend prediction model is trained using data from the case library to predict the probability and region of sporadic urban problems in the coming year. Areas with a probability higher than a preset proportion are inspected to determine the spatial location of sporadic urban problems. The case library contains the name, date, spatial location, and number of negative feedback evaluations for incidental urban issues. The anomaly identification module is used to preprocess the collected evaluation photos using image preprocessing technology, and to identify buildings with anomalies through a residual network model, and to analyze the damage or aging of the building surface. The segmentation module is used to merge urban problems and buildings with abnormalities to generate a set of urban problem points. The set of urban problem points is divided into point-type facility renewal areas, linear street appearance renewal areas, and area-type environmental improvement areas. The module also calculates the update sequence order of all areas and divides the update sequence stages. The calculation module is used to output the core features of the urban renewal area and the urban renewal text as vector sets through the Word2Vec feature vectorization algorithm. The core features include function, location, core issues and facility types. The similarity value of each pair of vector sets is obtained through the cosine similarity calculation formula, and the maximum value is taken to obtain the matching result with the highest similarity. The verification module is used to collect and compare the number of negative feedback evaluations before and after the update. If the number of negative feedback evaluations increases, the update for that area is determined to be invalid, and the system is returned to the city problem identification module.

Images