A personalized interest area extraction method and system based on interest units

Abstract

The embodiment of this specification provides a method and system for extracting personalized interest areas based on interest units. Among them, the method includes: extracting urban road data within the research area, and dividing the research area into different interest units based on the urban road data; dynamically allocating corresponding weight parameters to the convolutional layer based on the check-in data of the interest units, extracting the time interval features and spatial distance features of the interest units, and fusing the above features to form spatio-temporal fusion features; using the spatio-temporal fusion features as input, clustering the interest units through an improved clustering algorithm, and dividing different interest areas to achieve the extraction of personalized interest areas.

Description

Technical Field

[0001] This document relates to the technical field of urban planning, and particularly to a personalized interest area extraction method and system based on interest units. Background Art

[0002] Currently, the traditional methods for obtaining a user's Region of Interest (ROI) mainly use grid or clustering methods. However, these methods have a common problem, that is, they lack sufficient consideration of the actual geographical scope. As a result, the ROI boundary may not be precise enough, it may be too large or too small, making the final obtained ROI boundary have a certain degree of ambiguity. To solve this problem, a personalized interest area extraction method based on interest units is proposed. Summary of the Invention

[0003] One or more embodiments of this specification provide a personalized interest area extraction method based on interest units, including:

[0004] Extracting urban road data within the research area, and dividing the research area into different interest units based on the urban road data;

[0005] Based on the check-in data of interest units, dynamically allocate corresponding weight parameters to the convolutional layer, extract the time interval feature and spatial distance feature of the interest unit, and fuse the above features to form a spatio-temporal fusion feature;

[0006] Using the spatio-temporal fusion feature as input, clustering the interest units through an improved clustering algorithm, dividing different interest areas, and realizing the extraction of personalized interest areas.

[0007] Further, the urban road data includes highway, arterial road, and urban arterial road data in OpenStreetMap;

[0008] Extracting urban road data within the research area, and dividing the research area into different interest units specifically includes:

[0009] For the research area, screening highway, arterial road, and urban arterial road data in OpenStreetMap as urban road data;

[0010] Preprocessing the screened urban road data to form new road data, and the specific steps include:

[0011] When there are topological errors in the new road data resulting in irregular grids, modifying the topological errors to make the new road data form closed units, and finally forming interest units based on the urban road data.

[0012] Further, preprocess the selected urban road data to form new road data, which specifically includes:

[0013] Remove duplicate, hanging, and independent roads, and extend and complete the disconnected roads;

[0014] Generate buffers with different ranges for roads of different levels;

[0015] Based on the urban road data and administrative region data, remove the noise and redundant information in the urban road data, integrate the overlapping parts of the road and administrative region data through overlay analysis, and clean the overlapping areas of the administrative region data.

[0016] Further, based on the check-in data of interest units, dynamically assign corresponding weight parameters to the convolutional layer, and extract the time interval features and spatial distance features of the interest units, which specifically includes:

[0017] Set the spatial graph G = {V, E, A}, where the nodes V = {v0, v1, v2... v n-1} are interest units, the edge set E = {e0, e1, e3,... e m-1}, each e i represents the edge connecting two nodes, and the adjacency matrix A represents the spatial relationship between nodes. A ij represents the Haversine distance between the interest units v i and v j ;

[0018] Based on the check-in data of interest units, analyze the check-in frequencies of different interest units, and calculate the activity levels of each interest unit according to the check-in frequencies; according to the activity levels of each interest unit, dynamically assign different convolutional weights to the convolutional layer, and then apply them to the multi-layer graph convolution and dilated convolution network to capture the complex spatio-temporal relationships between interest units;

[0019] Perform graph convolution operations after normalizing the adjacency matrix, and introduce the above-mentioned convolutional weights in each layer of the graph convolutional layer. Stack multiple layers of graph convolution to increase the network depth and extract the spatial distance features of interest units;

[0020] Introduce a dilated convolution network, expand the convolution receptive field by introducing a dilation factor in the time dimension; on this basis, further introduce a dynamic convolution mechanism to dynamically adjust the convolutional kernel weights to adapt to the feature changes at different time scales, and extract the time interval features of interest units.

[0021] Further, the method for fusing the spatial distance features and time interval features to form spatio-temporal fusion features is as follows:

[0022] Establish the mapping relationship between the spatial interval feature and the time interval feature in the same dimensional space, and fuse the spatial interval feature and the time interval feature in a homomorphic manner to obtain the spatio-temporal fusion feature f m 。

[0023] Furthermore, taking the spatio-temporal fusion feature as the input, cluster the interest units through an improved clustering algorithm, divide different interest regions, and the specific method for extracting the personalized interest region is as follows:

[0024] Design a particle swarm algorithm to optimize the clustering center, and adopt an unsupervised classification method to divide similar interest units into the same region;

[0025] The specific method for designing a particle swarm algorithm to optimize the clustering center is as follows:

[0026] First, randomly generate k particles, and each particle contains the position of a possible initial clustering center;

[0027] For each particle, calculate the fitness using the objective function of K-means clustering, and the objective function is to maximize the sum of the distance differences between the central cluster and its affiliated clustering center;

[0028] Calculate the particle clustering center position and the moving direction to update the particle position, and the specific calculation method is as follows:

[0029] x l (t + 1) = x l (t) + v l (t + 1);

[0030] v l (t + 1) = w·v l (t) + c1·r1·(pbest l - x l (t)) + c2·r2·(gbest - x l (t));

[0031] Where x l (t) and v l (t) respectively represent the particle clustering center position and the moving direction, pbest l is to find the optimal initial clustering center position, gbest represents the global best clustering center, c1 and c2 are acceleration coefficients, and r1 and r2 are random numbers;

[0032] Repeat the above steps of calculating the fitness and updating the particle position until it stops after meeting the maximum number of iterations; select the clustering center corresponding to the particle with the minimum fitness as the initial clustering center, and use the iterative K-means algorithm multiple times to assign the POI dataset to the nearest clustering center.

[0033] Further, the method further includes:

[0034] When the number of clusters grows close to the number of true categories, analyze the internal and external relationships of each cluster obtained by clustering by calculating the weighted variance ratio standard index and silhouette coefficient of each cluster. Select the optimal clustering result according to the analysis result, and perform regional clustering based on the best clustering result to achieve personalized region of interest extraction.

[0035] One or more embodiments of this specification provide a personalized region of interest extraction system based on interest units, including:

[0036] Interest unit division module: used to extract urban road data within the research area, and divide the research area into different interest units based on the urban road data;

[0037] Feature fusion module: Dynamically allocate corresponding weight parameters to the convolutional layer based on the check-in data of interest units, extract the time interval feature and spatial distance feature of the interest units, and fuse the above features to form a spatio-temporal fusion feature;

[0038] Region of interest extraction module: Take the spatio-temporal fusion feature as input, cluster the interest units through an improved clustering algorithm, divide different regions of interest, and achieve the extraction of personalized regions of interest.

[0039] One or more embodiments of this specification provide an electronic device, including:

[0040] A processor; and a memory arranged to store computer-executable instructions; the computer-executable instructions, when executed, cause the processor to implement the steps of the above-mentioned personalized region of interest extraction method based on interest units.

[0041] One or more embodiments of this specification provide a storage medium for storing computer-executable instructions, and the computer-executable instructions, when executed, implement the steps of the personalized region of interest extraction method based on interest units as described in any one of claims 1 to 7.

[0042] The above description is only an overview of the technical solution of the present invention. In order to be able to understand the technical means of the present invention more clearly, it can be implemented according to the content of the specification. And in order to make the above and other objects, features and advantages of the present invention more obvious and understandable, the following specifically describes the specific embodiments of the present invention. Description of the Drawings

[0043] To more clearly illustrate the technical solutions in the prior art of this specification, the following will briefly introduce the accompanying drawings required for use in the embodiments or the description of the prior art. Obviously, the accompanying drawings in the following description are only some embodiments recorded in this specification. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained based on these drawings.

[0044] Figure 1 Flowchart of a personalized interest area extraction method based on interest units provided for one or more embodiments of this specification;

[0045] Figure 2 Specific implementation flowchart of a personalized interest area extraction method based on interest units provided for one or more embodiments of this specification;

[0046] Figure 3 Schematic diagram of the composition of a personalized interest area extraction system based on interest units provided for one or more embodiments of this specification;

[0047] Figure 4 Schematic diagram of the structure of an electronic device provided for one or more embodiments of this specification. Specific implementation manner

[0048] In order to enable those skilled in the art of this technology to better understand the technical solutions in one or more embodiments of this specification, the following will clearly and completely describe the technical solutions in one or more embodiments of this specification in conjunction with the accompanying drawings in one or more embodiments of this specification. Obviously, the described embodiments are only some embodiments of this specification, rather than all embodiments. Based on one or more embodiments of this specification, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of this document.

[0049] Method embodiment

[0050] According to an embodiment of the present invention, a personalized interest area extraction method based on interest units is provided. Figure 1 Flowchart of a personalized interest area extraction method based on interest units provided for one or more embodiments of this specification. Figure 2 Specific implementation flowchart of a personalized interest area extraction method based on interest units provided for one or more embodiments of this specification, as Figure 1 and Figure 2 shown. The personalized interest area extraction method based on interest units according to the embodiment of the present invention specifically includes:

[0051] S1. Extract urban road data within the research area, and divide the research area into different interest units based on the urban road data.

[0052] Using the interest units divided by urban road data as the smallest units of the user interest area ROI can better capture the geographical relationships between interest areas and plays an important role in accurately extracting the ROI. The urban road data includes highway, arterial road, and urban arterial road data in OpenStreetMap.

[0053] The specific method for extracting urban road data within the research area and dividing the research area into different interest units based on the urban road data is as follows:

[0054] For the research area, screen highway, arterial road, and urban arterial road data in OpenStreetMap as urban road data, and preprocess the screened urban road data to form new road data. Specifically, the preprocessing of urban road data specifically includes:

[0055] Remove duplicate, hanging, and isolated roads, and extend and fill in the disconnected roads; generate buffers with different ranges for roads of different levels, such as generating 40m, 20m, and 10m buffers for first-level, second-level, and third-level roads respectively; based on urban road data and administrative region data, remove the noise and redundant information of urban road data, extract the overlapping part of the road and administrative region data through overlay analysis, and clean the overlapping area of the administrative region data as the new road data.

[0056] When there are topological errors in the new road data resulting in irregular grids, modify the topological errors. The topological errors include hanging points, intersection points, etc. After modification, make the new road data form closed units, and finally form interest units based on urban road data.

[0057] S2. Dynamically allocate corresponding weight parameters to the convolutional layer based on the check-in data of the interest units, extract the time interval feature and spatial distance feature of the interest units, and fuse the above features to form a spatio-temporal fusion feature.

[0058] In the urban area, each interest unit does not exist independently, but is interconnected with and affected by surrounding units. Therefore, ROI extraction is not only affected by the spatio-temporal characteristics of a single interest unit, but also by surrounding units. Set the spatial graph G = {V, E, A}, where the nodes V = {v0, v1, v2……v n-1} are interest units, and each node v i contains a d-dimensional feature {x1, x2,…, x d}, representing all POI (point of interest) category information in the interest unit; the edge set E = {e0, e1, e3,…em-1}, each e i represents an edge connecting two nodes. If two interest units have a common edge, the corresponding two nodes are connected by an edge; the adjacency matrix A represents the spatial relationship between nodes, and A ij represents the Haversine distance between interest units v i and v j .

[0059] In the encoding framework, based on the check-in data of interest units, analyze the check-in frequencies of different interest units, calculate the activity of each interest unit through the check-in frequency, and dynamically assign different convolution weights to the convolutional layer based on the calculation of the activity of the interest unit by the check-in frequency, and then apply it to the multi-layer graph convolution and dilated convolution network to capture the complex spatio-temporal relationship between interest units;

[0060] After normalizing the adjacency matrix, perform graph convolution operations as follows:

[0061] Normalize the adjacency matrix:

[0062]

[0063] Perform graph convolution operations:

[0064]

[0065] Among them, X j is the feature vector of node j, and W is the class feature of the spatial graph nodes;

[0066] Introduce the convolution weights dynamically assigned according to the activity of each interest unit in each layer of the graph convolutional layer;

[0067] Increase the network depth by stacking multiple graph convolutional networks to extract the spatial distance features of interest units;

[0068] Introduce a dilated convolution network, and expand the convolution receptive field by introducing a dilation factor in the time dimension. On this basis, further introduce a dynamic convolution mechanism to dynamically adjust the convolution kernel weights to adapt to the feature changes of different time scales and extract the time interval features of interest units.

[0069] Establish a mapping relationship between the spatial distance feature and the time interval feature in the same dimensional space, and fuse the spatial interval feature and the time interval feature in a homomorphic manner to obtain the spatio-temporal fusion feature f m .

[0070] In the decoding framework, first, f mInput into a decoder consisting of 4 convolutional layers; secondly, for each convolutional layer, perform a convolution operation using the transpose \(w\) of the convolutional kernel corresponding to that layer, and sum the convolution results; finally, add the corresponding bias term. That is, calculate to obtain the result after reconstructing the fused features, so as to acquire the spatio-temporal fused features.

[0071] In the loss function, in order to prevent the model from overfitting and improve its generalization ability, L2 regularization is adopted, in the form of where \(\lambda\) is the regularization coefficient, and cross-validation is introduced to determine its value, \(\|W\|\) F is the Frobenius norm of the weight matrix.

[0072] S3. Take the spatio-temporal fused features as the input, and cluster the interest units through an improved clustering algorithm to divide different interest regions, so as to realize the extraction of personalized interest regions.

[0073] After extracting the spatio-temporal fused features, design a particle swarm algorithm to optimize the clustering center, and adopt an unsupervised classification method to divide similar interest units into the same region to realize the extraction of personalized user interest regions. Specifically:

[0074] The specific method for designing a particle swarm algorithm to optimize the clustering center is as follows:

[0075] Firstly, randomly generate \(k\) particles, and each particle contains the position of a possible initial clustering center;

[0076] For each particle, calculate the fitness using the objective function of K-means clustering, and the objective function is to maximize the sum of the distance differences between the center clustering and its affiliated clustering center;

[0077] Calculate the particle clustering center position and the moving direction to update the particle position. The specific calculation method is as follows:

[0078] x l (t + 1)=x l (t)+v l (t + 1);

[0079] v l (t + 1)=w·v l (t)+c1·r1·(pbest l -x l (t))+c2·r2·(gbest - x l (t));

[0080] where x l (t) and v l (t) respectively represent the particle clustering center position and the moving direction, pbest lIt is to find the optimal initial clustering center position. gbest represents the global best clustering center, c1 and c2 are acceleration coefficients, and r1 and r2 are random numbers;

[0081] Repeat the above steps to calculate the fitness and update the particle positions until the maximum number of iterations is reached and then stop; Select the clustering center corresponding to the particle with the minimum fitness as the initial clustering center, and use the iterative K-means algorithm multiple times to assign the ROI dataset to the nearest clustering center.

[0082] As the number of clusters increases, the aggregation degree of each cluster increases significantly, so the variance ratio standard index will also increase. However, when the number of clusters grows close to the number of true categories, further increasing the number of clusters will not significantly improve the aggregation degree, resulting in the growth rate of the variance ratio standard index gradually slowing down and finally leveling off. Considering the importance differences of the result categories, the result categories are introduced as weights into the formula of the variance ratio standard to improve the accuracy and stability of the clustering results. Specifically:

[0083] Calculate the weighted variance ratio standard index (WCH) as follows:

[0084]

[0085] where n is the number of ROI samples, k is the number of clusters, w B , w k represent the between-cluster and within-cluster weights respectively;

[0086] Among them:

[0087]

[0088] B k is the weighted covariance matrix between ROI clusters, m i and w i are the center point and weight factor of the i-th cluster respectively; W is the weighted covariance matrix of the ROI within the cluster, tr is the trace of the matrix, m i and w i are the center point and weight factor of the i-th cluster respectively; Therefore, in the WCH k graph, an upward-sloping and then flattening trend is formed, and the k value corresponding to the inflection point is the optimal number of clusters;

[0089] Calculate the silhouette coefficient to evaluate the overall distribution of the clustering quality. The specific calculation method is as follows:

[0090]

[0091] SC(i) represents the silhouette coefficient of the i-th cluster, and a(i) and b(i) respectively represent the average distance between the i-th cluster and other samples within the same cluster and the average distance to the samples in the nearest other cluster. When SC(i) is close to 1, it indicates that the ROI sample clustering is reasonable; when SC(i) is close to -1, it indicates that the ROI set classification is unreasonable and should be classified into other clusters; when SC(i) is close to 0, it indicates that the ROI set is located at the boundary of two clusters. Therefore, the k value corresponding to when SC(i) is close to 1 is used as the optimal clustering quality.

[0092] Finally, the optimal clustering result is selected according to the analysis results of the optimal number of clusters and the optimal clustering quality, and regional clustering is performed based on the optimal clustering result to achieve personalized region of interest extraction.

[0093] System embodiment

[0094] According to an embodiment of the present invention, a personalized region of interest extraction system based on interest units is provided. Figure 3 The composition schematic diagram of a personalized region of interest extraction system based on interest units provided for one or more embodiments of this specification is as Figure 3 shown. The personalized region of interest extraction system based on interest units according to an embodiment of the present invention specifically includes:

[0095] Interest unit division module 30: used to extract urban road data within the research area and divide the research area into different interest units based on the urban road data;

[0096] Feature fusion module 32: used to dynamically allocate corresponding weight parameters to the convolutional layer based on the check-in data of the interest units, extract the time interval feature and spatial distance feature of the interest units, and fuse the above features to form a spatio-temporal fusion feature;

[0097] Region of interest extraction module 34: taking the spatio-temporal fusion feature as input, clustering the interest units through an improved clustering algorithm, dividing different regions of interest, and realizing the extraction of personalized regions of interest.

[0098] The embodiment of the present invention is a system embodiment corresponding to the above method embodiment. The specific operations of each module can be understood by referring to the description of the method embodiment and will not be elaborated here.

[0099] Device embodiment 1

[0100] The embodiment of the present invention provides an electronic device, as Figure 4 shown, including: a memory 40, a processor 42, and a computer program stored on the memory 40 and executable on the processor 42. When the computer program is executed by the processor 42, the following method steps are implemented:

[0101] S1. Extract the urban road data within the research area, and divide the research area into different interest units based on the urban road data;

[0102] S2. Dynamically assign corresponding weight parameters to the convolutional layer based on the check-in data of the interest units, extract the time interval features and spatial distance features of the interest units, and fuse the above features to form spatio-temporal fusion features;

[0103] S3. Use the spatio-temporal fusion features as input, cluster the interest units through an improved clustering algorithm, divide different interest areas, and achieve the extraction of personalized interest areas.

[0104] Device Embodiment Two

[0105] An embodiment of the present invention provides a computer-readable storage medium, on which an implementation program for information transmission is stored. When the program is executed by a processor 42, the following method steps are implemented:

[0106] S1. Extract the urban road data within the research area, and divide the research area into different interest units based on the urban road data;

[0107] S2. Dynamically assign corresponding weight parameters to the convolutional layer based on the check-in data of the interest units, extract the time interval features and spatial distance features of the interest units, and fuse the above features to form spatio-temporal fusion features;

[0108] S3. Use the spatio-temporal fusion features as input, cluster the interest units through an improved clustering algorithm, divide different interest areas, and achieve the extraction of personalized interest areas.

[0109] The computer-readable storage medium described in this embodiment includes but is not limited to: ROM, RAM, magnetic disk or optical disc, etc.

[0110] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A personalized interest area extraction method based on interest units, characterized in that, Including: Extract the urban road data within the research area, and divide the research area into different interest units based on the urban road data; Based on the check-in data of the interest units, dynamically assign corresponding weight parameters to the convolutional layer, extract the time interval features and spatial distance features of the interest units, and fuse the above features to form spatio-temporal fusion features; specifically including: Set up the spatial graph G = {V, E, A}, where the nodes V = {v0, v1, v2 …… v n-1} are interest units, the edge set E = {e0, e1, e3, … e m-1}, each e i represents an edge connecting two nodes, and the adjacency matrix A represents the spatial relationship between nodes. A ij represents the Haversine distance between the interest units v i and v j ; Based on the check-in data of the interest units, analyze the check-in frequencies of different interest units, and calculate the activity of each interest unit according to the check-in frequency; according to the activity of each interest unit, dynamically assign different convolutional weights to the convolutional layer, and then apply them to the multi-layer graph convolution and dilated convolution network to capture the complex spatio-temporal relationships between interest units; Perform graph convolution operation after normalizing the adjacency matrix, and introduce the convolutional weights in each layer of the graph convolutional layer, stack multiple layers of graph convolution to increase the network depth, and extract the spatial distance features of the interest units; Introduce a dilated convolution network, expand the convolutional receptive field by introducing a dilation factor in the time dimension; on this basis, further introduce a dynamic convolution mechanism to dynamically adjust the convolutional kernel weights to adapt to the feature changes of different time scales, and extract the time interval features of the interest units; The specific method for fusing the spatial distance features and time interval features to form spatio-temporal fusion features is: Establish a mapping relationship between the spatial distance feature and the time interval feature in the same dimensional space, and fuse the spatial interval feature and the time interval feature in a homomorphic manner to obtain the spatio-temporal fusion feature f m ; Take the spatio-temporal fusion features as the input, and cluster the interest units through an improved clustering algorithm to divide different interest areas, and realize the extraction of personalized interest areas; the specific method is: Design a particle swarm algorithm to optimize the clustering center, and adopt an unsupervised classification method to divide similar interest units into the same area; The specific method for designing a particle swarm algorithm to optimize the clustering center is: First, randomly generate k particles, and each particle contains the position of a possible initial clustering center; For each particle, calculate the fitness using the objective function of K-means clustering, and the objective function is to maximize the sum of the distance differences between the central cluster and its affiliated clustering center; Calculate the position of the particle clustering center and the moving direction to update the particle position, and the specific calculation method is as follows: x l (t + 1)= x l (t)+ v l (t + 1); v l (t + 1)= w·v l (t)+ c1·r1·(pbest l - x l (t))+ c2·r2·(gbest - x l (t)); where x l (t) and v l (t) represent the position of the particle clustering center and the moving direction respectively, pbest l is the optimal initial clustering center position found, gbest represents the global best clustering center, c1 and c2 are acceleration coefficients, and r1 and r2 are random numbers; Repeat the above steps to calculate the process of fitness and update the particle position until it stops after reaching the maximum number of iterations; select the clustering center corresponding to the particle with the minimum fitness as the initial clustering center, and use the iterative K-means algorithm multiple times to assign the POI dataset to the nearest clustering center.

2. The method according to claim 1, wherein The urban road data includes highway, arterial road and urban arterial road data in OpenStreetMap; The extraction of the urban road data within the research area and the division of the research area into different interest units based on the urban road data specifically includes: For the research area, screen the highway, arterial road and urban arterial road data in OpenStreetMap as urban road data; Preprocess the screened urban road data to form new road data; When there are topological errors in the new road data resulting in irregular grids, modify the topological errors to make the new road data form closed units, and finally form interest units based on the urban road data.

3. The method according to claim 2, wherein preprocessing the selected urban road data to form new road data specifically includes: removing duplicate, hanging and independent roads, and extending and filling in the disconnected roads; generating buffers with different ranges for roads of different levels; based on the urban road data and administrative region data, removing the noise and redundant information of the urban road data, integrating the overlapping parts of the road and administrative region data through overlay analysis, and cleaning the overlapping areas of the administrative region data.

4. The method according to claim 1, characterized in that, The method further includes: when the number of clusters grows close to the number of true categories, analyzing the internal and external relationships of each cluster obtained by clustering by calculating the weighted variance ratio standard index and silhouette coefficient of each cluster, selecting the optimal clustering result according to the analysis result, and performing regional clustering based on the best clustering result to achieve personalized interest area extraction.

5. A personalized interest area extraction system based on interest units, characterized in that, including: Interest unit division module: used to extract the urban road data in the research area, and divide the research area into different interest units based on the urban road data; Feature fusion module: based on the check-in data of the interest unit, dynamically assign corresponding weight parameters to the convolutional layer, extract the time interval feature and spatial distance feature of the interest unit, and fuse the above features to form a spatio-temporal fusion feature; Specifically: Set up a spatial graph G = {V, E, A}, where the nodes V = {v0, v1, v2... v n-1} are interest units, the edge set E = {e0, e1, e3,... e m-1}, each e i represents an edge connecting two nodes, and the adjacency matrix A represents the spatial relationship between nodes. A ij represents the Haversine distance between the interest units v i and v j ; Based on the check-in data of the interest unit, analyze the check-in frequency of different interest units, and calculate the activity of each interest unit according to the check-in frequency; according to the activity of each interest unit, dynamically assign different convolutional weights to the convolutional layer, and then apply them to the multi-layer graph convolution and dilated convolution network to capture the complex spatio-temporal relationship between interest units; perform graph convolution operation after normalizing the adjacency matrix, and introduce the convolutional weight in each layer of the graph convolutional layer, stack multiple layers of graph convolution to increase the network depth, and extract the spatial distance feature of the interest unit; introduce a dilated convolution network, expand the convolutional receptive field by introducing a dilation factor in the time dimension; on this basis, further introduce a dynamic convolution mechanism to dynamically adjust the convolutional kernel weight to adapt to the feature changes of different time scales, and extract the time interval feature of the interest unit; The specific method for fusing the spatial distance feature and the time interval feature to form a spatio-temporal fusion feature is: Establish a mapping relationship between the spatial distance feature and the time interval feature in the same dimensional space, fuse the spatial interval feature and the time interval feature in a homomorphic manner, and obtain the spatio-temporal fusion feature f m ; Interest area extraction module: taking the spatio-temporal fusion feature as input, clustering the interest units through an improved clustering algorithm, dividing different interest areas, and realizing the extraction of personalized interest areas; specifically, design a particle swarm optimization algorithm to optimize the clustering center, adopt an unsupervised classification method, and divide similar interest units into the same area; The specific method for designing a particle swarm optimization algorithm to optimize the clustering center is: First, randomly generate k particles, and each particle contains the position of a possible initial clustering center; For each particle, calculate the fitness using the objective function of K-means clustering, and the objective function is to maximize the sum of the distance differences between the central cluster and its affiliated clustering center; Calculate the position of the particle clustering center and the moving direction to update the particle position, and the specific calculation method is as follows: x l (t + 1) = x l (t) + v l (t + 1); v l (t + 1)= w·v l (t)+ c1·r1·(pbest l - x l (t))+ c2·r2·(gbest - x l (t)); where x l (t) and v l (t) represent the particle clustering center position and the moving direction respectively, pbest l is the found optimal initial clustering center position, gbest represents the global best clustering center, c1 and c2 are acceleration coefficients, and r1 and r2 are random numbers; Repeat the above steps of calculating fitness and updating the particle positions until it stops after meeting the maximum number of iterations; select the cluster center corresponding to the particle with the minimum fitness as the initial cluster center, and use the iterative K-means algorithm multiple times to assign the POI dataset to the nearest cluster center.

6. An electronic device, characterized in that, Including: A processor; And a memory arranged to store computer-executable instructions; The computer-executable instructions, when executed, cause the processor to implement the steps of the personalized interest area extraction method based on interest units according to any one of claims 1 to 4.

7. A storage medium, characterized in that, For storing computer-executable instructions, the computer-executable instructions, when executed, implement the steps of the personalized interest area extraction method based on interest units according to any one of claims 1 to 4.

Images