Task processing method, neural network and training method thereof, device and medium

Abstract

The present disclosure provides a task processing method, a neural network and a training method thereof, a device and a medium, and relates to the field of artificial intelligence, in particular to image processing technology, electronic map technology, and deep learning technology. The method comprises: determining a first feature vector of each of a plurality of city regions based on city data; determining the affiliation of each city region and a plurality of preset semantic clusters based on the first feature vector of each city region; determining a semantic cluster feature vector of each preset semantic cluster based on the first feature vector of at least one city region included in the preset semantic cluster; determining a second feature vector of a target city region based on the affiliation of the target city region and the plurality of preset semantic clusters and the semantic cluster feature vector of each of the plurality of preset semantic clusters; aggregating the first feature vector and the second feature vector of the target city region to obtain a third feature vector; and obtaining a task processing result based on the third feature vector of the target city region.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the field of artificial intelligence, in particular to image processing technology, electronic map technology, and deep learning technology, and more particularly to a task processing method, a neural network training method, a neural network for task processing, an electronic device, a computer readable storage medium, and a computer program product. BACKGROUND

[0002] Artificial intelligence is a discipline that studies enabling computers to simulate some thinking processes and intelligent behaviors (such as learning, reasoning, thinking, planning, etc.) of human beings, which has both hardware level technology and software level technology. Artificial intelligence hardware technology generally includes technologies such as sensors, special artificial intelligence chips, cloud computing, distributed storage, big data processing, etc.; artificial intelligence software technology mainly includes computer vision technology, speech recognition technology, natural language processing technology, and machine learning / deep learning, big data processing technology, knowledge graph technology, etc.

[0003] A region is a basic spatial unit constituting a city, different regions contain different building environments and human social and economic activities, forming the complexity and diversity of the city. Quantitatively representing the city region is a key content in city intelligent computing, which can serve a variety of downstream applications (for example, traffic flow prediction, population prediction), help understand the city development, better plan the city, and provide sound social services.

[0004] The methods described in this section can not necessarily be the methods previously conceived or employed. Unless otherwise indicated, nothing in this section should be assumed to be prior art merely because of its inclusion in this section. Similarly, unless otherwise indicated, matters discussed in this section should not be assumed to be prior art to any application described herein. SUMMARY

[0005] The present disclosure provides a task processing method, a neural network training method, a neural network for task processing, an electronic device, a computer readable storage medium, and a computer program product.

[0006] According to an aspect of the present disclosure, a task processing method is provided, comprising: determining a plurality of city areas in a target city; determining a first feature vector of each of the plurality of city areas based on city data of the city area, wherein the city data comprises at least one of image data, point of interest data and road network data of the corresponding city area; determining a membership relationship between each of the plurality of city areas and a plurality of preset semantic clusters based on the first feature vector of the city area; determining a semantic cluster feature vector of each of the plurality of preset semantic clusters based on the first feature vector of at least one city area included in the preset semantic cluster; determining a second feature vector of a target city area in the plurality of city areas based on the membership relationship between the target city area and the plurality of preset semantic clusters and the semantic cluster feature vector of the plurality of preset semantic clusters; aggregating the first feature vector and the second feature vector of the target city area to obtain a third feature vector of the city area; and obtaining a task processing result based on the third feature vector of the target city area.

[0007] According to another aspect of the present disclosure, a method for training a neural network is provided, where the neural network comprises a local feature extraction subnetwork, a dependency determination subnetwork, a semantic cluster feature extraction subnetwork, a global feature extraction subnetwork, an aggregation layer, and a prediction subnetwork. The method comprises: determining a plurality of city areas in a target city, where the plurality of city areas comprises a plurality of sample city areas; obtaining real labels of the plurality of sample city areas respectively; inputting city data of each of the plurality of city areas into the local feature extraction subnetwork respectively to obtain first feature vectors of the plurality of city areas respectively, where the city data comprises at least one of image data, point of interest data and road network data of the corresponding city area; inputting the first feature vector of each of the plurality of city areas into the dependency determination subnetwork to obtain a dependency between the city area and a plurality of preset semantic clusters; inputting the first feature vectors of at least one city area comprised by each of the plurality of preset semantic clusters into the semantic cluster feature extraction subnetwork to obtain a semantic cluster feature vector of the preset semantic cluster; inputting the dependency between the plurality of preset semantic clusters and the plurality of sample city areas and the semantic cluster feature vectors of the plurality of preset semantic clusters into the global feature extraction subnetwork to obtain a second feature vector of each of the plurality of sample city areas; inputting the first feature vector and the second feature vector of each of the plurality of sample city areas into the aggregation layer to obtain a third feature vector of each of the plurality of sample city areas; inputting the third feature vector of each of the plurality of sample city areas into the prediction subnetwork to obtain predicted labels of the plurality of sample city areas respectively; calculating a loss value based on the real labels and the predicted labels of the plurality of sample city areas respectively; and training the neural network based on the loss value.

[0008] According to another aspect of the present disclosure, a neural network for task processing includes: a local feature extraction subnetwork configured to determine a first feature vector of each of a plurality of city areas based on city data of the city area, wherein the city data includes at least one of image data, point of interest data, and road network data of the corresponding city area; a membership determination subnetwork configured to determine, for each of the plurality of city areas, a membership between the city area and a plurality of preset semantic clusters based on the first feature vector of the city area; a semantic cluster feature extraction subnetwork configured to determine, for each of the plurality of preset semantic clusters, a semantic cluster feature vector of the preset semantic cluster based on the first feature vector of at least one city area included in the preset semantic cluster; a global feature extraction subnetwork configured to determine a second feature vector of a target city area in the plurality of city areas based on the membership of the target city area and the plurality of preset semantic clusters and the semantic cluster feature vector of each of the plurality of preset semantic clusters; an aggregation layer configured to aggregate the first feature vector of the target city area and the second feature vector of the target city area to obtain a third feature vector of the target city area; and a prediction subnetwork configured to obtain a task processing result based on the third feature vector of the target city area.

[0009] According to another aspect of the present disclosure, an electronic device is provided, including: at least one processor; and a memory communicatively connected with the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the above method.

[0010] According to another aspect of the present disclosure, a non-transitory computer readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable a computer to perform the above method.

[0011] According to another aspect of the present disclosure, a computer program product is provided, including a computer program, wherein the computer program, when executed by a processor, implements the above method.

[0012] According to one or more embodiments of the present disclosure, by determining the dependency relationship between a plurality of city areas and a plurality of preset semantic clusters pre-constructed, and determining a feature vector of each preset semantic cluster based on a first feature vector of at least one city area included in the preset semantic cluster, then determining a second feature vector of a target city area according to the dependency relationship between the target city area and the plurality of preset semantic clusters and the feature vector of each preset semantic cluster, and aggregating the first feature vector and the second feature vector of the target city area to obtain a third feature vector for a downstream task, the mining of regional semantic characteristics and the modeling of the relationship between similar regions at a global and long distance are realized, the feature information of similar regions at a long distance is used to strengthen the representation of the current region, and the accuracy of the result of the downstream task processing is improved.

[0013] It should be understood that the matters described in this section are not intended to identify key or important features of the embodiments of the present disclosure, nor are they used to limit the scope of the present disclosure. Other features of the present disclosure will become apparent from the following description. BRIEF DESCRIPTION OF DRAWINGS

[0014] The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain exemplary implementations of the application. The illustrated embodiments are merely examples and do not limit the scope of the claims. In all the drawings, like reference numerals refer to like elements throughout the accompanying drawings.

[0015] Figure 1 A flowchart of a task processing method according to an exemplary embodiment of the present disclosure is shown;

[0016] Figure 2 A flowchart of a task processing method according to an exemplary embodiment of the present disclosure is shown;

[0017] Figures 3A-3B A schematic diagram of a plurality of city areas according to an exemplary embodiment of the present disclosure is shown;

[0018] Figure 4 A flowchart of a training method of a neural network according to an exemplary embodiment of the present disclosure is shown;

[0019] Figure 5 A flowchart of a training method of a neural network according to an exemplary embodiment of the present disclosure is shown;

[0020] Figure 6 A structural block diagram of a neural network for task processing according to an exemplary embodiment of the present disclosure is shown;

[0021] Figure 7 A structural block diagram of a neural network for task processing according to an exemplary embodiment of the present disclosure is shown; and

[0022] Figure 8 Fig. 1 shows a structural block diagram of an exemplary electronic device that can be used to implement embodiments of the present disclosure. DETAILED DESCRIPTION

[0023] Exemplary embodiments of the present disclosure are described below with reference to the accompanying drawings, which include various details of the embodiments of the present disclosure to assist in understanding them. These should be considered in their context only. Thus, those of ordinary skill in the art will recognize the various changes and modifications of the embodiments described herein, without departing from the scope of the present disclosure. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.

[0024] In the present disclosure, the terms "first", "second", and the like are used to describe various elements only and are not intended to limit the positional relationship, the time sequence relationship, or the importance relationship of the elements, and such terms are only used to distinguish one element from another. In some examples, the first element and the second element can refer to the same instance of the element, and in some cases, based on the context of the description, they can also refer to different instances.

[0025] The terms used in the description of various described examples in the present disclosure are only for the purpose of describing specific examples and are not intended to be limiting. Unless the number of elements is specifically limited, the element can be one or more, if not specifically limited by the context. In addition, the term "and / or" used in the present disclosure encompasses any one of the listed items and all possible combinations thereof.

[0026] In the related art, when completing a downstream task by using urban data (e.g., image data, point of interest data) of a city area, the features of the target area are usually learned based only on the relevant data of the target area, or the features of the target area are reinforced using the features of the neighboring area of the target area, but the task result based on such features in the downstream task is poor.

[0027] To solve the above problems, the present disclosure determines the subordinate relationship between a plurality of city areas and a plurality of pre-constructed preset semantic clusters, and determines the feature vector of each preset semantic cluster based on the first feature vector of each city area included in the preset semantic cluster, and then determines the second feature vector of the target city area according to the subordinate relationship of the target city area and the plurality of preset semantic clusters and the feature vector of each preset semantic cluster, and aggregates the first feature vector and the second feature vector of the city area to obtain the third feature vector for the downstream task, which realizes the mining of the semantic characteristics of the area and the modeling of the relationship between similar areas at a global and long distance, and reinforces the representation of the current area using the feature information of similar areas at a long distance, and improves the accuracy of the processing result of the downstream task.

[0028] The embodiments of this disclosure will now be described in detail with reference to the accompanying drawings.

[0029] According to one aspect of this disclosure, a task processing method is provided. For example... Figure 1 As shown, the task processing method includes: Step S101, determining multiple urban areas in the target city; Step S102, determining the first feature vector of each of the multiple urban areas based on the urban data of each urban area, wherein the urban data includes at least one of the image data, point of interest data, and road network data of the corresponding urban area; Step S103, for each urban area, determining the subordinate relationship between the urban area and multiple preset semantic clusters based on the first feature vector of the urban area; Step S104, for each preset semantic cluster, determining the semantic cluster feature vector of the preset semantic cluster based on the first feature vector of each of the at least one urban area included in the preset semantic cluster; Step S105, determining the second feature vector of the target urban area based on the subordinate relationship between the target urban area and the multiple preset semantic clusters and the semantic cluster feature vectors of the multiple preset semantic clusters; Step S106, aggregating the first feature vector and the second feature vector of the target urban area to obtain the third feature vector of the target urban area; and Step S107, obtaining the task processing result based on the third feature vector of the target urban area.

[0030] Therefore, by determining the hierarchical relationship between multiple city regions and multiple pre-constructed preset semantic clusters, and determining the feature vector of each preset semantic cluster based on the first feature vector of at least one city region included in each preset semantic cluster, and then determining the second feature vector of the city region based on the hierarchical relationship between the target city region and multiple preset semantic clusters and the feature vector of each preset semantic cluster, and aggregating the first and second feature vectors of the city region to obtain the third feature vector for downstream tasks, the system realizes the mining of regional semantic characteristics and the modeling of the relationship between global, long-distance similar regions. By using the feature information of distant similar regions to strengthen the representation of the current region, the system improves the accuracy of downstream task processing results.

[0031] According to some embodiments, before performing the above steps, the target city and its scope can be determined first. After determining the scope of the target city, step S101 can be performed to determine multiple urban areas of the target city. In some embodiments, a rasterization method can be used to divide the target city into N = H × W non-intersecting fixed-size grids, with each grid representing a urban area. V = {v1, v2, ..., v...} can be used. N} represents a set of urban regions. It can be understood that other ways can also be used to divide the target city. In some embodiments, the target city can also be divided into multiple urban regions according to the road network structure of the city, or the target city is divided in other ways, which are not limited herein.

[0032] According to some embodiments, each urban region can have a feature vector x i (i.e., the first feature vector) to describe the region, which can be based on various urban data (e.g., point of interest (POI) data, satellite image data) and can be constructed according to a specific downstream task.

[0033] After determining the multiple urban regions, each urban region can be taken as a node, and an urban region graph (URG) is constructed to describe the characteristics of the regions and the relationship between the regions. First, the urban region graph is defined as G = (V, E, A, X) formally, where V represents a set of nodes composed of urban regions, X represents a node feature matrix, E represents edges on the graph, and A represents a corresponding adjacency matrix. If A ij = 1, it means that there is an edge between the urban regions v i and v j , that is, the two urban regions have an adjacency relationship. The construction of the first feature vector of the urban region and the construction of the adjacency relationship between the urban regions will be described in detail below.

[0034] According to some embodiments, as shown in Figure 2 , the task processing method further includes: step S202, determining the adjacency relationship between the multiple urban regions. It can be understood that Figure 2 the operations of steps S201, S203, S205-S206, S208-S210 in Figure 1 are similar to steps S101-S107 in , which are not limited herein.

[0035] The present disclosure constructs the relationship between the regions from two angles: (1) spatial proximity; (2) road network connectivity. According to some embodiments, two urban regions in the multiple urban regions have an adjacency relationship if they meet at least one of the following conditions: adjacent two urban regions; and two urban regions less than a preset distance in road network data.

[0036] (1) Spatial Proximity

[0037] Generally speaking, adjacent regions in a city often have close contact and exhibit similar characteristics (may have similar representations in the latent space). Therefore, each region can be connected with its surrounding adjacent regions, and the corresponding position of the urban region graph adjacency matrix is set as Aij = 1, that is, the two regions have an adjacency relationship. It can be understood that, as shown in Figure 3A FIG. 6, when the plurality of city regions form a rectangular grid, each city region is adjacent to 8 city regions (there are 8 neighbors around the center region of the 3x3 grid), and when the plurality of city regions form other shapes, the adjacency relationship can also be determined according to the geometric relationship between the regions, which is not limited herein.

[0038] (2) Road network connectivity

[0039] In order to model more complex relationships between city regions, road network data can be introduced to establish edges between regions. As the most core part of the urban transportation system, the road network plays an important role in capturing the functional association across regions. Some functional regions can be composed of regions that are not spatially adjacent but are connected by the road network (for example, a commercial street). Therefore, road network data can be introduced to more comprehensively capture the relevance between regions. Specifically, if two regions v i and v j are connected in the road network, an edge can be added between the regions, and A ij = 1. In some embodiments, connectivity in the road network is defined as: if two regions are each other's neighbors within n-hop in the road network graph formed by the road network data, it is considered that the two regions are connected in the road network. In an exemplary embodiment, n is 5, and as shown in Figure 3B FIG. 7, the intersections I1-I5 are not more than 5-hop, so any two of the city regions v 10 , v 11 , and v 12 have an adjacency relationship. It can be understood that n-hop (or 5-hop) is only an example of a preset distance, and other preset distances can also be used as a constraint condition when determining two regions connected in the road network data, which is not limited herein.

[0040] According to some embodiments, in order to describe the characteristics of the city regions, the present disclosure constructs POI features and image features based on POI basic attribute data and satellite map data. That is, the first feature vector includes an image feature vector and a point of interest feature vector corresponding to the city region. Step S203 of determining the first feature vector of each of the plurality of city regions based on the city data of each of the plurality of city regions can include: determining the image feature vector of each of the plurality of city regions based on the image data of each of the plurality of city regions; and determining the point of interest feature vector of each of the plurality of city regions based on the point of interest data of each of the plurality of city regions.

[0041] In some embodiments, the POI basic attribute data provides basic information of the POI, such as the name, location and category of the POI, based on which the function facilities in a region can be described. The satellite image can be, for example, a 256x256 RGB image, which can depict the appearance characteristics of each region. It can be understood that the above POI basic attribute categories, the size of the satellite image and the color mode are only an example and are not intended to limit the scope of the present disclosure. The POI data can also include other categories of basic attributes, and the satellite image can also be of other sizes or contain other information (e.g., terrain information, ground building information), which are not limited herein.

[0042] According to some embodiments, the point of interest feature vector can include at least one of a first point of interest feature sub-vector, a second point of interest feature sub-vector, and a third point of interest feature sub-vector.

[0043] The first point of interest feature sub-vector indicates at least one of a point of interest category distribution histogram in the corresponding urban region and a point of interest category distribution histogram of a surrounding urban region of the corresponding urban region, the point of interest category distribution histogram indicating a proportion of the point of interest of each preset category in the first preset category set in the corresponding region. In some embodiments, the proportion of various types of POIs in each region and in the surrounding 3x3 region can be counted, and a probability distribution histogram vector can be calculated as a POI category distribution feature. In an exemplary embodiment, the POI categories (included in the first preset category set) can include bus stations, hospitals, schools, restaurants, subway stations, markets, and the like, which are not limited herein.

[0044] The second point of interest feature sub-vector indicates the nearest distance of the point of interest of the corresponding urban region and each preset category in the second preset category set, respectively. In some embodiments, in order to measure the convenience of residents in the region to obtain various basic life facilities, a series of POI radius features can be calculated. Each POI radius is defined as the nearest distance from the region to the POI.

[0045] The third point of interest feature sub-vector indicates whether each preset category of the point of interest in the third preset category set is included in a preset range around the corresponding urban region. In some embodiments, in order to more directly represent whether the infrastructure of a region is perfect, a Boolean feature, infrastructure index, can be defined. When various basic life facilities (i.e., in the third preset category set) are contained within 1 km around a region, the feature takes 1, otherwise 0.

[0046] It can be understood that the POI categories included in the first preset category set, the second preset category set and the third preset category set can be set according to requirements. The first preset category set, the second preset category set and the third preset category set can be completely or partially the same, or different, which is not limited herein.

[0047] In some embodiments, the point of interest feature vector can include or be derived based on the first point of interest feature sub-vector, the second point of interest feature sub-vector and the third point of interest feature sub-vector. In an exemplary embodiment, the three point of interest feature sub-vectors can be directly spliced or aggregated in other manners to obtain the point of interest feature vector of the corresponding urban area.

[0048] According to some embodiments, the appearance features of the urban area can be extracted based on satellite maps. In an exemplary embodiment, considering that the pixel-level image data has a high dimension, directly using it to train the model can easily cause overfitting, and therefore a pre-trained VGG16 model is used as a feature extractor to extract semantic features from the original satellite map. Specifically, the last two fully connected layers of the pre-trained VGG16 model are removed, the satellite image of each region is input, and then the 4096-dimensional vector output by the model is used as the image feature of the region. It can be understood that other existing neural network models or self-built neural network models can also be used to process the satellite image data to obtain image features, or other means can be used to extract features from the satellite image data, or image data with other auxiliary information (e.g., terrain information, ground building information, or image data collected in the urban area, etc.) is used, which is not limited herein.

[0049] In some embodiments, the image feature vector and the point of interest feature vector of the urban area can be aggregated to obtain the first feature vector of the urban area. In an exemplary embodiment, the image feature vector and the point of interest feature vector can be directly spliced or aggregated in other manners to obtain the first feature vector of the corresponding urban area.

[0050] Thus, the region representation containing the image data information and the POI data information of the corresponding urban area can be obtained through the above steps. How to strengthen the representation of each urban area by using the relationship between the urban areas will be introduced below.

[0051] According to some embodiments, when the urban region graph (URG) is constructed, the local relationship between adjacent regions can be captured to strengthen the representation of each urban area. As shown in FIG. 6, the relationship between the urban areas can be represented by a graph, and the nodes of the graph represent the urban areas, and the edges of the graph represent the relationship between the urban areas. Figure 2As shown, the task processing method further includes: step S204, for each of the multiple city regions, updating the first feature vector of that city region based on the first feature vectors of one or more city regions that are adjacent to that city region. Thus, in this way, each city region can capture information from its neighboring city regions, thereby strengthening the representation of that city region. It is understood that, for example, linear summation, concatenation, multilayer perceptron processing, graph neural networks, or various other methods and any combination thereof can be used to update the first feature vector of the city region based on the first feature vectors of one or more neighboring city regions; no limitation is made here.

[0052] According to some embodiments, graph neural networks can be used to capture local relationships between regions and initially learn region representations. Step S204, for each of multiple city regions, updating the first feature vector of that city region based on the first feature vectors of one or more city regions adjacent to it, includes: processing a city region graph including multiple city regions and their adjacency relationships using a trained graph attention network to update the first feature vector of each city region. Since the correlation between regions may be non-linear, using a graph attention network can better learn the relationships between regions.

[0053] In one exemplary embodiment, the first feature vector of each city region can be updated in the following manner. For each region v i , using x i To represent its feature vector, we can first calculate the attention score between each region and its neighbors or adjacent regions:

[0054]

[0055] Where, x j Indicates region v i The neighbor or adjacent area v j eigenvectors ( Indicates region v i In the urban area map URG (neighboring or adjacent regions), W and a represent the learnable model parameters. This represents the vector concatenation operation, where σ represents the LeakyReLU activation function. Then, the attention score can be normalized using the softmax function to obtain the attention weights for feature aggregation as follows:

[0056]

[0057] According to the attention weight, the local relationship between regions can be captured and the representation of the region (i.e., the first feature vector of the urban region) can be obtained:

[0058]

[0059] Based on the urban region graph constructed above, the present disclosure organizes the urban regions into a hierarchical structure, clusters the regions with similar semantics together, enables similar but distant regions to interact with each other, and provides distant supervision signals for the regions with sparse local labels (neighbor regions have no labels) to perform representation learning.

[0060] Suppose there are K hidden nodes representing K different semantic clusters in the city, and the urban regions in different clusters have different semantics. A hierarchical structure can be constructed and the representations of the regions and semantic clusters can be learned. First, the regions can be assigned to different semantic clusters according to the features of the regions, and the representation of the semantic cluster can be learned based on the representation of the region through a region→semantic cluster information collection process. Then, the learned semantic cluster representation is used as global interaction information shared between regions, and is fed back to similar but distant regions through a semantic cluster→region direction reverse mapping, so that the global interaction information can be used to enhance the representation of the region. The specific steps are as follows:

[0061] According to some embodiments, in step S205, for each urban region in the plurality of urban regions, the membership relationship between the urban region and a plurality of preset semantic clusters can be determined based on the first feature vector of the urban region. A distribution matrix can be defined to represent the membership relationship of the region to the K semantic clusters. The distribution matrix can be obtained by training a neural network that performs one or more steps of the task processing method. It can be understood that the number K of preset semantic clusters is a hyperparameter, which can be determined according to the demand before training the corresponding neural network, and the present disclosure does not intend to limit its specific value.

[0062] According to some embodiments, the membership relationship between each urban region in the plurality of urban regions and the plurality of preset semantic clusters includes the probability that the urban region belongs to each of the plurality of preset semantic clusters. In an exemplary embodiment, B ij represents the probability (or membership degree) that the i-th region belongs to the j-th semantic cluster, and has ∑ 1≤j≤K B ij = 1. For the local representation of each region v i A linear transformation and a row-wise softmax function can be used to calculate the distribution matrix:

[0063] ​

[0064] where W B denotes the training parameter matrix of linear transformation. This assignment matrix will serve as the information transfer channel between urban areas and semantic clusters.

[0065] According to some embodiments, after obtaining the membership relationship between the urban areas and the plurality of preset semantic clusters (i.e., the assignment matrix B), the urban areas included in each preset semantic cluster can be determined. Step S205, for each of the plurality of urban areas, determining the membership relationship between the urban area and the plurality of preset semantic clusters based on the first feature vector of the urban area can include: determining the preset semantic cluster with the highest probability in the plurality of preset semantic clusters corresponding to the urban area as the preset semantic cluster including the urban area. This step can be obtained by calculating the binary assignment matrix . Each row of the matrix is a one-hot vector, where the value of 1 is taken at the position corresponding to the maximum element of each row of the original assignment matrix B, and the values of other elements are 0 (formal representation ). That is, each area will be assigned to the most likely preset semantic cluster.

[0066] According to some embodiments, based on the representation of the area and the membership relationship of the assignment matrix, the representation of the preset semantic cluster, i.e., the semantic cluster feature vector, can be learned. Step S206, for each of the plurality of preset semantic clusters, determining the semantic cluster feature vector of the preset semantic cluster based on the first feature vector of at least one urban area included in the preset semantic cluster can include: determining the sum of the first feature vectors of the at least one urban area included in the preset semantic cluster as the semantic cluster feature vector of the preset semantic cluster. The semantic cluster feature vector can be represented as:

[0067]

[0068] where h j denotes the semantic cluster feature vector of the preset semantic cluster-j. By binarizing the assignment matrix, the membership relationship of the area can be constrained so that each area is only assigned to the most likely semantic cluster, which can avoid the representation of the semantic cluster being dominated by a large number of areas with low membership degree.

[0069] After the above area→semantic cluster mapping, the preset semantic clusters divide the urban areas into K groups, and the feature vector of each cluster summarizes the similar semantic information of the areas in the cluster. Then, considering that there may be connections between different semantic clusters, the semantic cluster feature vector of each preset semantic cluster can be updated based on the feature vectors of other preset semantic clusters.

[0070] According to some embodiments, as Figure 2As shown, the task processing method further includes: in step S207, for each target preset semantic cluster in the plurality of preset semantic clusters, updating the semantic cluster feature vector of the target preset semantic cluster based on the semantic cluster feature vector of each preset semantic cluster in the plurality of preset semantic clusters and the weight of each preset semantic cluster with respect to the target preset semantic cluster. Thus, by using the feature vector of each preset semantic cluster and the weight corresponding to the target preset semantic cluster (the preset semantic cluster whose feature vector is currently to be updated) to update the feature vector of the target preset semantic cluster, the representation of each preset semantic cluster can learn the information of other semantic clusters. It can be understood that the feature vector of the target preset semantic cluster can be updated based on the feature vector of each preset semantic cluster and the corresponding weight using linear summation, splicing, multilayer perception, graph neural network or other various manners and any combination thereof, which is not limited herein.

[0071] According to some embodiments, in step S207, for each target preset semantic cluster in the plurality of preset semantic clusters, updating the semantic cluster feature vector of the target preset semantic cluster based on the semantic cluster feature vector of each preset semantic cluster in the plurality of preset semantic clusters and the weight of each preset semantic cluster with respect to the target preset semantic cluster can include: processing a complete graph composed of the plurality of preset semantic clusters using the trained graph convolution network to update the semantic cluster feature vector of each preset semantic cluster in the plurality of preset semantic clusters. Thus, by forming a complete graph composed of the plurality of preset semantic clusters and using the graph convolution network for processing, the relationship between the preset semantic clusters can be fully learned.

[0072] In an exemplary embodiment, the updated semantic cluster feature vector of the preset semantic cluster can be represented as:

[0073] h′ i =σ(∑ 1≤j≤K e ij W h h j )

[0074] wherein h′ i represents the updated semantic cluster representation, W h represents a transformation matrix shared by all semantic clusters, e ij represents a parameterized edge weight between semantic cluster-i and semantic cluster-j, which will be trained together with W h .

[0075] According to some embodiments, after obtaining the semantic cluster feature vector of each preset semantic cluster, the global knowledge in the semantic cluster representation can be returned to each region through a semantic cluster→region knowledge sharing, so as to enhance the representation of the region. In some embodiments, at step S208, for each of the plurality of city regions, a second feature vector of the city region is determined based on the affiliation between the city region and the plurality of preset semantic clusters and the semantic cluster feature vector of each of the plurality of preset semantic clusters. Thus, according to the affiliation between the city region and the preset semantic clusters, the preset semantic clusters related to each city region can be determined, and the second feature vector of the city region can be obtained using the semantic cluster feature vectors of the preset semantic clusters, so as to realize the above-mentioned semantic cluster→region knowledge sharing.

[0076] It can be understood that the second feature vector of the city region can be determined based on the affiliation between the city region and the plurality of preset semantic clusters and the corresponding semantic cluster feature vectors in various ways, for example, the semantic cluster feature vectors of the preset semantic clusters that include the city region can be aggregated, or the semantic cluster feature vectors of the plurality of preset semantic clusters can be aggregated based on the probability that the city region belongs to each preset semantic cluster, or processed in other ways to obtain the second feature vector of the city region, which is not limited herein.

[0077] In some embodiments, the allocation matrix can be reused to return the global knowledge in the preset semantic cluster representation. At step S208, determining the second feature vector of the target city region based on the affiliation between the target city region and the plurality of preset semantic clusters and the semantic cluster feature vector of each of the plurality of preset semantic clusters can include: aggregating the semantic cluster feature vectors of the plurality of preset semantic clusters based on the probability that the target city region belongs to each of the plurality of preset semantic clusters to obtain the second feature vector of the target city region. The present disclosure uses the original B instead of the binary allocation matrix This is because some slightly weaker semantic clusters can still provide some useful auxiliary information to enrich the representation of the region. The reverse knowledge sharing can be represented as:

[0078]

[0079] wherein, represents the global relationship-aware region representation returned from the semantic cluster, i.e., the second feature vector, W r is the parameter matrix to be learned.

[0080] After obtaining the second feature vector, step S209 of aggregating the local region feature (the first feature vector / the updated first feature vector) and the global region representation (the second feature vector) can be performed to finally obtain an enhanced region representation (i.e., the third feature vector).

[0081]

[0082] wherein AGG(·) represents an aggregation function, which can be concatenation, summation, etc. In this way, the region representation contains global shared knowledge. Meanwhile, regions far away from each other can interact with each other, and regions with sparse local labels can also obtain certain supervision signals from distant labeled regions, and better perform representation learning.

[0083] After obtaining the third feature vector as the final representation of the city region, various downstream tasks can be performed using the third feature vector of the target city region to obtain the final task processing result. It can be understood that the downstream tasks may, for example, include village-in-city region identification, demolition region identification, high population density region identification, etc. The target city region can be a single city region or multiple city regions, which are not limited herein.

[0084] According to another aspect of the present disclosure, a method for training a neural network is disclosed. The neural network includes a local feature extraction subnetwork, a dependency determination subnetwork, a semantic cluster feature extraction subnetwork, a global feature extraction subnetwork, an aggregation layer, and a prediction subnetwork. As Figure 4 ​As shown, the method comprises: step S401, determining a plurality of city regions in a target city, wherein the plurality of city regions comprises a plurality of sample city regions; step S402, obtaining a true label of each of the plurality of sample city regions; step S403, inputting city data of each of the plurality of city regions into a local feature extraction sub-network respectively to obtain a first feature vector of each of the plurality of city regions, wherein the city data comprises at least one of image data, point of interest data and road network data of the corresponding city region; step S404, for each of the plurality of city regions, inputting the first feature vector of the city region into a membership determination sub-network to obtain a membership between the city region and a plurality of preset semantic clusters; step S405, for each of the plurality of preset semantic clusters, inputting the first feature vector of at least one city region included in the preset semantic cluster into a semantic cluster feature extraction sub-network to obtain a semantic cluster feature vector of the preset semantic cluster; step S406, for each of the plurality of sample city regions, inputting the membership between the sample city region and the plurality of preset semantic clusters and the semantic cluster feature vector of each of the plurality of preset semantic clusters into a global feature extraction sub-network to obtain a second feature vector of the sample city region; step S407, for each of the plurality of sample city regions, inputting the first feature vector and the second feature vector of the sample city region into an aggregation layer to obtain a third feature vector of the sample city region; step S408, inputting the third feature vector of each of the plurality of sample city regions into a prediction sub-network to obtain a predicted label of each of the plurality of sample city regions; step S409, calculating a loss value based on the true label and the predicted label of each of the plurality of sample city regions; and step S410, training the neural network based on the loss value. It can be understood that, Figure 4 The operations of step S401, step S403-step S408 in the method of Figure 1 The operations of step S101-step S107 in the method of

[0085] Therefore, by using semantic clusters, region semantics specific mining and modeling of global and long distance relationships are achieved, so that more sufficient region representation learning of city data can be obtained, especially in a label sparse scenario, supervision signals can be obtained from long distance labeled data, so that limited labeled data can be fully utilized to improve the effect of region representation learning.

[0086] According to some embodiments, the neural network can further comprise a local feature updating sub-network. As Figure 5As shown, the method for training the neural network can further include: step S502, determining the adjacency relationship between the plurality of urban areas; and step S505, for each of the plurality of urban areas, inputting the first feature vector of one or more urban areas having an adjacency relationship with the urban area into the local feature updating sub-network, to obtain an updated first feature vector of the urban area. It can be understood that, Figure 5 The operations of steps S501, S503-S504, S506-S507, S509-S513 in Figure 4 The operations of steps S401-S410 in

[0087] According to some embodiments, the local feature updating sub-network can be a graph attention network, and can be configured to process the urban area graph including the plurality of urban areas and the adjacency relationship between the plurality of urban areas. The step S511 of training the neural network based on the loss value can include training the parameters W and a in the graph attention network based on the loss value.

[0088] According to some embodiments, the membership relationship between each of the plurality of urban areas and the plurality of preset semantic clusters can include a probability that the urban area belongs to each of the plurality of preset semantic clusters, i.e., an assignment matrix B.

[0089] According to some embodiments, the membership relationship determining sub-network can be configured to, for each of the plurality of urban areas, determine a probability that the urban area belongs to each of the plurality of preset semantic clusters based on a product of the first feature vector of the urban area and the second parameter matrix. In some embodiments, the assignment matrix B can be represented as:

[0090]

[0091] wherein W B represents a trained parameter matrix of linear transformation, i.e., the second parameter matrix. In some embodiments, the step S511 of training the neural network based on the loss value can include training the second parameter matrix based on the loss value.

[0092] According to some embodiments, the neural network can further include a semantic cluster feature updating sub-network, as shown in Figure 5As shown, the method for training the neural network can further include: in step S508, inputting, for each of the plurality of preset semantic clusters, a semantic cluster feature vector of the each of the plurality of preset semantic clusters into the semantic cluster feature updating subnetwork to obtain an updated semantic cluster feature vector of the each of the plurality of preset semantic clusters. Training the neural network based on the loss value in step S511 can include training the semantic cluster feature updating subnetwork based on the loss value.

[0093] According to some embodiments, the plurality of preset semantic clusters can constitute a complete graph, and the semantic cluster feature updating subnetwork can be a graph convolution network. Training the neural network based on the loss value in step S511 can include jointly training a transformation matrix W shared by all semantic clusters in the graph convolution network based on the loss value. h and a parameterized edge weight e between any two semantic clusters-i and semantic cluster-j ij .

[0094] According to some embodiments, the global feature extraction subnetwork can be configured to, for each of the plurality of city regions, aggregate the semantic cluster feature vectors of the plurality of preset semantic clusters based on probabilities that the city region belongs to each of the plurality of preset semantic clusters to obtain a second feature vector of the city region.

[0095] According to some embodiments, the global feature extraction subnetwork can be configured to, for each of the plurality of sample city regions, determine a second feature vector of the city region based on a sum of products of a semantic cluster feature vector of each of the plurality of preset semantic clusters, a first parameter matrix, and a probability that the sample city region belongs to the preset semantic cluster. The second feature vector can be represented as:

[0096]

[0097] wherein W r is a parameter matrix to be learned, i.e., the first parameter matrix. Training the neural network based on the loss value in step S511 can include training the first parameter matrix based on the loss value.

[0098] According to another aspect of the present disclosure, a neural network for task processing is provided. As shown in FIG. 7, the neural network can include a global feature extraction subnetwork 701, a semantic cluster feature updating subnetwork 702, and a semantic cluster generation subnetwork 703. Figure 6As shown, the neural network 600 includes: a local feature extraction subnetwork 602, configured to determine a first feature vector for each of the multiple urban areas based on urban data 614 of each urban area, wherein the urban data includes at least one of image data, point of interest data, and road network data of the corresponding urban area; a subordination determination subnetwork 604, configured to determine the subordination relationship between each urban area and multiple preset semantic clusters based on the first feature vector of that urban area; and a semantic cluster feature extraction subnetwork 606, configured to extract features from each of the multiple preset semantic clusters based on the first feature vector of that urban area. Let the semantic cluster include at least one city region, each with its own first feature vector. The semantic cluster feature vector of this preset semantic cluster is determined. A global feature extraction subnetwork 608 is configured to determine a second feature vector of the target city region based on the hierarchical relationship between the target city region and multiple preset semantic clusters, as well as the semantic cluster feature vectors of each preset semantic cluster. An aggregation layer 610 is configured to aggregate the first and second feature vectors of the target city region to obtain a third feature vector. A prediction subnetwork 612 is configured to obtain the task processing result 616 based on the third feature vector of the target city region. It can be understood that the operations of subnetworks 602-612 in neural network 600 are respectively related to... Figure 1 The operations of steps S102-S107 are similar and will not be described in detail here.

[0099] In some embodiments, the arrow pointing from the subnetwork 604 to the global feature extraction subnetwork 608 can refer to the assignment matrix B, and the arrow pointing from the subnetwork 604 to the semantic cluster feature extraction subnetwork 606 can refer to the binarized assignment matrix.

[0100] According to some embodiments, such as Figure 7 As shown, the neural network 700 further includes a local feature update subnetwork 718, configured to update the first feature vector of each of the multiple urban regions based on the first feature vectors of one or more adjacent urban regions. It is understood that the operations of subnetworks 702-712, input 714, and output 716 in neural network 700 are similar to those of subnetworks 602-612, input 614, and output 616 in neural network 600, and will not be elaborated upon here.

[0101] In some embodiments, the arrow pointing from the affiliation determination subnetwork 704 to the global feature extraction subnetwork 708 can refer to an assignment matrix B, and the arrow pointing from the affiliation determination subnetwork 704 to the semantic cluster feature extraction subnetwork 706 can refer to a binarized assignment matrix

[0102] According to some embodiments, the local feature update subnetwork can be a graph attention network, and can be configured to process a city region graph comprising a plurality of city regions and adjacency relations between the plurality of city regions.

[0103] According to some embodiments, as shown in Figure 7 According to some embodiments, the neural network 700 further comprises a semantic cluster feature update subnetwork 720 configured to, for each target preset semantic cluster of the plurality of preset semantic clusters, update a semantic cluster feature vector of the target preset semantic cluster based on the semantic cluster feature vector of each preset semantic cluster of the plurality of preset semantic clusters and a weight of each preset semantic cluster with respect to the target preset semantic cluster.

[0104] According to some embodiments, the plurality of preset semantic clusters can constitute a complete graph, and the semantic cluster feature update subnetwork can be a graph convolution network.

[0105] It can be understood that part of the steps, sub-steps, or structures, subnetworks, modules, etc. in the neural network in the present disclosure (for example, aggregation function, activation function, loss function, etc.) can be added, adjusted, or replaced according to requirements, but does not affect the embodiment of the inventive concept of the present disclosure. The scheme obtained by such modification is within the protection scope of the present disclosure.

[0106] In the technical scheme of the present disclosure, the collection, storage, use, processing, transmission, provision and disclosure of user personal information involved in the technical scheme comply with the relevant legal regulations and do not violate public order and good customs.

[0107] According to the embodiments of the present disclosure, an electronic device, a readable storage medium and a computer program product are also provided.

[0108] Reference Figure 8The present invention describes a structural block diagram of an electronic device 800 that can serve as a server or client of the present disclosure, which is an example of a hardware device that can be applied to various aspects of the present disclosure. The electronic device is intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0109] like Figure 8 As shown, device 800 includes a computing unit 801, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 802 or a computer program loaded from storage unit 808 into random access memory (RAM) 803. RAM 803 may also store various programs and data required for the operation of device 800. The computing unit 801, ROM 802, and RAM 803 are interconnected via bus 804. Input / output (I / O) interface 805 is also connected to bus 804.

[0110] Multiple components in device 800 are connected to I / O interface 805, including: input unit 806, output unit 807, storage unit 808, and communication unit 809. Input unit 806 can be any type of device capable of inputting information to device 800. Input unit 806 can receive input numerical or character information and generate key signal inputs related to user settings and / or function control of the electronic device, and can include, but is not limited to, a mouse, keyboard, touchscreen, trackpad, trackball, joystick, microphone, and / or remote control. Output unit 807 can be any type of device capable of presenting information, and can include, but is not limited to, a monitor, speaker, video / audio output terminal, vibrator, and / or printer. Storage unit 808 can include, but is not limited to, a hard disk and an optical disk. Communication unit 809 allows device 800 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks, and can include, but is not limited to, modems, network cards, infrared communication devices, wireless communication transceivers, and / or chipsets, such as Bluetooth™ devices, 802.11 devices, WiFi devices, WiMax devices, cellular communication devices, and / or the like.

[0111] The computing unit 801 can be various general and / or special purpose processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various computing units running machine learning network algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 801 performs various methods and processes described above, such as the task processing method and / or the training method of the neural network. For example, in some embodiments, the task processing method and / or the training method of the neural network can be implemented as a computer software program tangibly embodied in a machine-readable medium, such as the storage unit 808. In some embodiments, part or all of the computer program can be loaded and / or installed onto the device 800 via the ROM 802 and / or the communication unit 809. When the computer program is loaded onto the RAM 803 and executed by the computing unit 801, one or more steps of the task processing method and / or the training method of the neural network described above can be performed. Alternatively, in other embodiments, the computing unit 801 can be configured to perform the task processing method and / or the training method of the neural network by any other suitable means, such as by means of firmware.

[0112] Various implementations of the systems and techniques described above can be realized in digital electronic circuitry, integrated circuitry, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a system on a chip (SOC), a programmable logic device (CPLD), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include implementation in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which can be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device.

[0113] Program code for carrying out methods of the present disclosure can be written in any combination of one or more programming languages. The program code can be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the program code, when executed by the processor or controller, produces a means for implementing the functions / acts specified in the flowcharts and / or block diagrams. The program code can be executed entirely on a machine, partially on a machine, partially on a machine and partially on a remote machine or entirely on a remote machine or server.

[0114] In the context of the present disclosure, a machine-readable medium can be a tangible medium that contains or stores a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include but is not limited to an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of the machine-readable storage medium will include one or more lines of electrical connections, portable computer disks, hard disk drives, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or Flash memory), optical fibers, portable compact disc read-only memories (CD-ROMs), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0115] To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the computer. Other kinds of devices can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form, including acoustic, speech, or tactile input.

[0116] The systems and techniques described here can be implemented in a computing system that includes a back end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front end component (e.g., a user computer having a graphical user interface or a Web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back end, middleware, or front end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), and the Internet.

[0117] The computer system can include clients and servers. This relationship can be remote, such that the servers are distributed across many clients. The relationship can also be such that the server is remote from the client. The servers can be cloud servers, also known as cloud computing servers or cloud hosts, which are a host product in the cloud computing service system to solve the defects of large management difficulty and weak business scalability in traditional physical hosts and VPS services ("Virtual Private Server", or simply "VPS"). The servers can also be servers of a distributed system, or servers combined with a blockchain.

[0118] It should be understood that the various forms of flow shown above can be used to reorder, add, or delete steps. For example, the steps described in the present disclosure can be executed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions disclosed in the present disclosure can be achieved, which is not limited herein.

[0119] Although embodiments or examples of the present disclosure have been described with reference to the accompanying drawings, it should be understood that the above-described methods, systems, and devices are merely exemplary embodiments or examples, and the scope of the present disclosure is not limited by these embodiments or examples, but is only limited by the granted claims and their equivalent scope. Various elements in the embodiments or examples can be omitted or replaced by equivalent elements. In addition, each step can be performed in an order different from that described in the present disclosure. Further, various elements in the embodiments or examples can be combined in various ways. It is important that many of the elements described herein can be replaced by equivalent elements that appear after the present disclosure as technology evolves.

Claims

1. A task processing method, comprising: Identify multiple urban areas within the target city; Based on the urban data of each of the plurality of urban areas, a first feature vector is determined for each of the plurality of urban areas, wherein the urban data includes at least one of the image data, point of interest data, and road network data of the corresponding urban area; For each of the multiple city regions, the subordinate relationship between the city region and multiple preset semantic clusters is determined based on the first feature vector of the city region. For each of the plurality of preset semantic clusters, based on the first feature vector of each of the at least one city region included in the preset semantic cluster, the semantic cluster feature vector of the preset semantic cluster is determined, wherein the subordinate relationship between each city region in the plurality of city regions and the plurality of preset semantic clusters includes the probability that the city region belongs to each of the plurality of preset semantic clusters. Based on the probability that the target city region belongs to each of the multiple preset semantic clusters in the multiple city regions, the semantic cluster feature vectors of each of the multiple preset semantic clusters are aggregated to obtain the second feature vector of the target city region. The first and second feature vectors of the target city area are aggregated to obtain the third feature vector of the city area; and The task processing result is obtained based on the third feature vector of the target city area.

2. The method according to claim 1, wherein, For each of the multiple city regions, based on the first feature vector of that city region, the affiliation relationship between that city region and multiple preset semantic clusters is determined, including: The preset semantic cluster with the highest probability of corresponding to the city area among the multiple preset semantic clusters is determined as the preset semantic cluster that includes the city area.

3. The method according to claim 1 or 2, further comprising: For each target preset semantic cluster among the plurality of preset semantic clusters, the semantic cluster feature vector of the target preset semantic cluster is updated based on the semantic cluster feature vector of each preset semantic cluster among the plurality of preset semantic clusters and the weight of each preset semantic cluster with respect to the target preset semantic cluster.

4. The method according to claim 3, wherein, For each target preset semantic cluster among the plurality of preset semantic clusters, the semantic cluster feature vector of the target preset semantic cluster is updated based on the semantic cluster feature vector of each preset semantic cluster and the weight of each preset semantic cluster with respect to the target preset semantic cluster, including: The trained graph convolutional network is used to process the complete graph composed of the multiple preset semantic clusters to update the semantic cluster feature vector of each preset semantic cluster.

5. The method according to claim 1 or 2, wherein, For each of the plurality of preset semantic clusters, based on the first feature vectors of at least one urban area included in the preset semantic cluster, the semantic cluster feature vector of the preset semantic cluster is determined to include: The sum of the first feature vectors of each of the at least one urban area included in the preset semantic cluster is determined as the semantic cluster feature vector of the preset semantic cluster.

6. The method according to claim 1 or 2, further comprising: Determine the adjacency relationships between the multiple city areas; as well as For each of the plurality of city regions, the first feature vector of that city region is updated based on the first feature vectors of one or more city regions that are adjacent to that city region.

7. The method according to claim 6, wherein, Two city areas among the multiple city areas that meet at least one of the following conditions are adjacent: Two adjacent urban areas; and Two urban areas within a preset distance in the road network data.

8. The method according to claim 6, wherein, For each of the plurality of city regions, updating the first feature vector of that city region based on the first feature vectors of one or more city regions that are adjacent to it includes: The trained graph attention network is used to process the city region map, which includes the multiple city regions and the adjacency relationships between the multiple city regions, in order to update the first feature vector of each of the multiple city regions.

9. The method according to claim 1 or 2, wherein, The first feature vector includes the image feature vector of the corresponding urban area and the interest point feature vector of the corresponding urban area. Determining the first feature vector for each of the plurality of urban areas based on the urban data of each urban area includes: Based on the image data of each of the multiple urban areas, determine the image feature vector of that urban area; and Based on the point of interest data of each of the multiple city regions, the point of interest feature vector of that city region is determined.

10. The method according to claim 9, wherein, The interest point feature vector includes at least one of the following: The first interest point feature vector indicates at least one of the interest point category distribution histogram within the corresponding urban area and the interest point category distribution histogram within the surrounding urban areas of the corresponding urban area. The interest point category distribution histogram indicates the proportion of interest points of each preset category in the first preset category set within the corresponding area. The second interest point feature vector indicates the nearest distance of the corresponding urban area to each preset category of interest points in the second preset category set. as well as The third interest point feature vector indicates whether the corresponding urban area is surrounded by interest points of each preset category in the third preset category set within a preset range.

11. A method for training a neural network, wherein, The neural network includes a local feature extraction subnetwork, a membership determination subnetwork, a semantic cluster feature extraction subnetwork, a global feature extraction subnetwork, an aggregation layer, and a prediction subnetwork. The method includes: Identify multiple urban areas within the target city, wherein the multiple urban areas include multiple sample urban areas; Obtain the true labels for each of the multiple sample city regions; The city data of each of the plurality of city regions is input into the local feature extraction subnetwork to obtain the first feature vector of each of the plurality of city regions, wherein the city data includes at least one of the image data, point of interest data and road network data of the corresponding city region; For each of the multiple city regions, the first feature vector of the city region is input into the subordinate relationship determination sub-network to obtain the subordinate relationship between the city region and multiple preset semantic clusters; For each of the plurality of preset semantic clusters, the first feature vector of each of the at least one urban area included in the preset semantic cluster is input into the semantic cluster feature extraction sub-network to obtain the semantic cluster feature vector of the preset semantic cluster. For each of the multiple sample city regions, the subordinate relationship between the sample city region and the multiple preset semantic clusters, as well as the semantic cluster feature vectors of each of the multiple preset semantic clusters, are input into the global feature extraction subnetwork to obtain the second feature vector of the sample city region. For each of the multiple sample city regions, the first feature vector and the second feature vector of the sample city region are input into the aggregation layer to obtain the third feature vector of the sample city region. The third feature vector of each of the multiple sample city regions is input into the prediction sub-network to obtain the prediction label of each of the multiple sample city regions. Based on the true and predicted labels of the multiple sample city regions, the loss value is calculated; and The neural network is trained based on the loss value.

12. The method according to claim 11, wherein, The subordinate relationship between each of the plurality of city regions and the plurality of preset semantic clusters includes the probability that the city region belongs to each of the plurality of preset semantic clusters. The global feature extraction subnetwork is configured to aggregate the semantic cluster feature vectors of each of the multiple preset semantic clusters based on the probability that the city region belongs to each of the multiple preset semantic clusters for each city region, so as to obtain the second feature vector of the city region.

13. The method according to claim 12, wherein, The global feature extraction subnetwork is configured to, for each of the plurality of sample city regions, determine the second feature vector of the sample city region based on the sum of the products of the semantic cluster feature vector of each of the plurality of preset semantic clusters, the first parameter matrix, and the probability that the sample city region belongs to the preset semantic cluster. Training the neural network based on the loss value includes: The first parameter matrix is ​​trained based on the loss value.

14. The method according to claim 12, wherein, The subnetwork for determining membership is configured to, for each of the plurality of city regions, determine the probability that the city region belongs to each of the plurality of preset semantic clusters based on the product of the first feature vector and the second parameter matrix of the city region. Training the neural network based on the loss value includes: The second parameter matrix is ​​trained based on the loss value.

15. The method according to claim 11 or 12, wherein, The neural network further includes a semantic cluster feature update subnetwork, and the method further includes: For each of the plurality of preset semantic clusters, the semantic cluster feature vector of each preset semantic cluster is input into the semantic cluster feature update sub-network to obtain the updated semantic cluster feature vector of each preset semantic cluster. Training the neural network based on the loss value includes: Based on the loss value, the semantic cluster feature update subnetwork is trained.

16. The method according to claim 15, wherein, The multiple preset semantic clusters constitute a complete graph, and the semantic cluster feature update sub-network is a graph convolutional network.

17. The method according to claim 11 or 12, wherein, The neural network further includes a local feature update subnetwork, and the method further includes: Determine the adjacency relationships between the multiple city areas; and For each of the multiple city regions, the first feature vectors of one or more adjacent city regions are input into the local feature update subnetwork to obtain the updated first feature vector of that city region. Training the neural network based on the loss value includes: Based on the loss value, the local feature update subnetwork is trained.

18. The method according to claim 17, wherein, The local feature update subnetwork is a graph attention network, configured to process a city region graph that includes the multiple city regions and the adjacency relationships between the multiple city regions.

19. A neural network for task processing, comprising: A local feature extraction subnetwork is configured to determine a first feature vector for each of the multiple urban regions based on urban data of each urban region, wherein the urban data includes at least one of image data, point of interest data, and road network data of the corresponding urban region; The subordinate relationship determination subnetwork is configured to determine the subordinate relationship between a city region and multiple preset semantic clusters based on a first feature vector of the city region for each of the multiple city regions. A semantic cluster feature extraction subnetwork is configured to determine the semantic cluster feature vector of each of the plurality of preset semantic clusters based on the first feature vector of each of the at least one city region included in the preset semantic cluster, wherein the subordinate relationship between each city region and the plurality of preset semantic clusters includes the probability that the city region belongs to each of the plurality of preset semantic clusters. The global feature extraction subnetwork is configured to aggregate the semantic cluster feature vectors of each of the multiple preset semantic clusters based on the probability that the target city region in the multiple city regions belongs to each of the multiple preset semantic clusters, so as to obtain the second feature vector of the target city region. An aggregation layer is configured to aggregate the first feature vector and the second feature vector of the target city region to obtain a third feature vector of the target city region; and The prediction subnetwork is configured to obtain the task processing result based on the third feature vector of the target city region.

20. The neural network of claim 19, further comprising: The semantic cluster feature update subnetwork is configured to update the semantic cluster feature vector of the target preset semantic cluster for each of the plurality of preset semantic clusters, based on the semantic cluster feature vector of each preset semantic cluster and the weight of each preset semantic cluster with respect to the target preset semantic cluster.

21. The neural network according to claim 20, wherein, The multiple preset semantic clusters constitute a complete graph, and the semantic cluster feature update sub-network is a graph convolutional network.

22. The neural network according to claim 19, further comprising: The local feature update subnetwork is configured to update the first feature vector of each of the plurality of city regions based on the first feature vectors of one or more city regions that are adjacent to that city region.

23. The neural network according to claim 22, wherein, The local feature update subnetwork is a graph attention network, configured to process a city region graph that includes the multiple city regions and the adjacency relationships between the multiple city regions.

24. An electronic device comprising: At least one processor; as well as A memory that is communicatively connected to the at least one processor; in The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-18.

25. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-18.

26. A computer program product comprising a computer program, wherein, When the computer program is executed by a processor, it implements the method of any one of claims 1-18.

Images