Dynamic graph reconstruction method, apparatus, device, readable storage medium and program product

Abstract

The present disclosure provides a dynamic graph reconstruction method and device, electronic equipment, readable storage medium and program product, relating to artificial intelligence technical fields such as degree learning, dynamic graph reconstruction, and variational inference. The method comprises: obtaining a target dynamic graph of a target modality, and determining a first implicit representation of each node according to snapshots of each time point constituting the target dynamic graph; obtaining an auxiliary dynamic graph of another modality different from the target modality, and determining a second implicit representation of each node according to snapshots of each time point constituting the auxiliary dynamic graph; splicing the first implicit representation and the second implicit representation of the same node to obtain a spliced feature of each node; performing nonlinear processing on the spliced feature of each node, and reconstructing a new dynamic graph based on the nonlinear feature obtained by processing. The method can effectively improve the generally existing data sparsity problem by combining the auxiliary dynamic graph of another modality on the basis of the target dynamic graph of the target modality.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of data processing, in particular to the technical field of artificial intelligence such as deep learning, dynamic graph reconstruction, variational inference, and the like, and more particularly to a dynamic graph reconstruction method and device, an electronic device, a computer readable storage medium, and a computer program product. BACKGROUND

[0002] A dynamic graph is a kind of graph structure that is ubiquitous, which means that the structure (including nodes and edges) of the graph evolves over time, such as traffic networks, social networks, financial transaction networks, and the like. Inference or prediction of the graph structure at a future time is a task of practical significance, which can be referred to as the problem of dynamic graph reconstruction.

[0003] How to solve the problem of data sparsity that is ubiquitous in the task of dynamic graph reconstruction is a focus of research for those skilled in the art. SUMMARY

[0004] Embodiments of the present disclosure provide a dynamic graph reconstruction method and device, an electronic device, a computer readable storage medium, and a computer program product.

[0005] In a first aspect, embodiments of the present disclosure provide a dynamic graph reconstruction method, including: obtaining a target dynamic graph of a target modality, and determining first implicit representations of nodes according to snapshots of each time constituting the target dynamic graph; obtaining an auxiliary dynamic graph of another modality different from the target modality, and determining second implicit representations of the nodes according to snapshots of each time constituting the auxiliary dynamic graph; splicing the first implicit representations and the second implicit representations of the same nodes to obtain spliced features of the nodes; performing nonlinear processing on the spliced features of the nodes, and reconstructing a new dynamic graph based on the nonlinear features obtained by the processing.

[0006] In a second aspect, embodiments of the present disclosure provide a dynamic graph reconstruction device, including: a first implicit representation determination unit configured to obtain a target dynamic graph of a target modality, and determine first implicit representations of nodes according to snapshots of each time constituting the target dynamic graph; a second implicit representation determination unit configured to obtain an auxiliary dynamic graph of another modality different from the target modality, and determine second implicit representations of the nodes according to snapshots of each time constituting the auxiliary dynamic graph; an implicit representation splicing unit configured to splice the first implicit representations and the second implicit representations of the same nodes to obtain spliced features of the nodes; and a new dynamic graph reconstruction unit configured to perform nonlinear processing on the spliced features of the nodes, and reconstruct a new dynamic graph based on the nonlinear features obtained by the processing.

[0007] In a third aspect, an electronic device is provided, and the electronic device includes at least one processor, and a memory connected with the at least one processor in communication; 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 implement the dynamic graph reconstruction method as described in any implementation manner of the first aspect.

[0008] In a fourth aspect, a non-transitory computer-readable storage medium storing computer instructions is provided, and the computer instructions are used to enable a computer to implement the dynamic graph reconstruction method as described in any implementation manner of the first aspect.

[0009] In a fifth aspect, a computer program product including a computer program is provided, and the computer program is executable by a processor to implement the steps of the dynamic graph reconstruction method as described in any implementation manner of the first aspect.

[0010] The dynamic graph reconstruction scheme provided by the present disclosure combines the auxiliary dynamic graph of other modalities on the basis of the target dynamic graph of the target modality, so as to assist in improving the data sparsity problem by combining the dynamic graphs of different modalities of other information sources, and obtain the related node information in the comprehensive dynamic graph of different modalities by splicing the first implicit representation and the second implicit representation of the same node, and obtain a new dynamic graph more consistent with the actual situation through nonlinear processing, thereby practically improving the data sparsity problem.

[0011] It should be understood that the content described in this part is not intended to identify key or important features of the embodiments of the present disclosure, nor is it used to limit the scope of the present disclosure. Other features of the present disclosure will become apparent through the following description. BRIEF DESCRIPTION OF DRAWINGS

[0012] Other features, objects, and advantages of the present disclosure will become more apparent through reading the following detailed description of non-limiting embodiments made with reference to the following drawings:

[0013] Figure 1 is an exemplary system architecture to which the present disclosure can be applied;

[0014] Figure 2 is a flowchart of a dynamic graph reconstruction method provided by an embodiment of the present disclosure;

[0015] Figure 3 is a flowchart of another dynamic graph reconstruction method provided by an embodiment of the present disclosure;

[0016] Figure 4 is a flowchart of a dynamic graph reconstruction method in an application scenario provided by an embodiment of the present disclosure;

[0017] Figure 5 A structural block diagram of a dynamic graph reconstruction device provided in this disclosure embodiment;

[0018] Figure 6 This is a schematic diagram of the structure of an electronic device suitable for performing a dynamic graph reconstruction method, provided as an embodiment of the present disclosure. Detailed Implementation

[0019] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding; these should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description. It should be noted that, unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other.

[0020] The collection, storage, use, processing, transmission, provision, and disclosure of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0021] Figure 1 An exemplary system architecture 100 is shown, in which embodiments of the dynamic graph reconstruction methods, apparatuses, electronic devices, and computer-readable storage media of this disclosure can be applied.

[0022] like Figure 1 As shown, system architecture 100 may include terminal devices 101, 102, and 103, a network 104, and a server 105. Network 104 serves as the medium for providing communication links between terminal devices 101, 102, and 103 and server 105. Network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0023] Users can use terminal devices 101, 102, and 103 to interact with server 105 via network 104 to receive or send messages, etc. Various applications for enabling information communication between the terminal devices 101, 102, and 103 and server 105 can be installed. These applications include directed graph analysis applications, dynamic graph reconstruction applications, and instant messaging applications.

[0024] The terminal devices 101, 102, 103 and the server 105 can be hardware or software. When the terminal devices 101, 102, 103 are hardware, they can be various electronic devices with a display screen, including but not limited to a smart phone, a tablet computer, a laptop computer, a desktop computer, and the like; when the terminal devices 101, 102, 103 are software, they can be installed in the above-mentioned electronic devices, and can be implemented as multiple software or software modules, or as a single software or software module, which is not specifically limited here. When the server 105 is hardware, it can be implemented as a distributed server cluster composed of multiple servers, or as a single server; when the server is software, it can be implemented as multiple software or software modules, or as a single software or software module, which is not specifically limited here.

[0025] The server 105 can provide various services through various built-in applications. Taking a dynamic graph reconstruction application that can solve the problem of sparse directed graph data as an example, the server 105 can achieve the following effects when running the dynamic graph reconstruction application: in addition to obtaining a target dynamic graph of a target modality, the server 105 can also combine an auxiliary dynamic graph of another modality different from the target modality to comprehensively record relevant information in the dynamic graphs of different modalities, and reconstruct a new dynamic graph that is more dense than the target dynamic graph data.

[0026] It should be noted that the target dynamic graph and the auxiliary dynamic graph of different modalities can be obtained from the terminal devices 101, 102, 103 through the network 104, or can be pre-stored in the server 105 locally through various ways. Therefore, when the server 105 detects that the local has already stored these data (for example, before starting to process the remaining to-be-processed tasks), it can choose to directly obtain these data from the local, and in this case, the example system architecture 100 can also not include the terminal devices 101, 102, 103 and the network 104.

[0027] Since dynamic graph reconstruction needs to occupy more computing resources and stronger computing capability, the dynamic graph reconstruction method provided in each of the subsequent embodiments of the present disclosure is generally executed by a server 105 with stronger computing capability and more computing resources, and accordingly, the dynamic graph reconstruction apparatus is generally also arranged in the server 105. However, it should also be pointed out that when the terminal devices 101, 102 and 103 also have computing capability and computing resources that meet the requirements, the terminal devices 101, 102 and 103 can also complete the above-mentioned operations by the dynamic graph reconstruction application installed thereon, and then output the same result as the server 105. Especially in the case where there are multiple terminal devices with different computing capabilities, but the dynamic graph reconstruction application judges that the terminal device has strong computing capability and has more remaining computing resources, the terminal device can be allowed to execute the above-mentioned operations, so as to appropriately reduce the computing pressure of the server 105, and accordingly, the dynamic graph reconstruction apparatus can also be arranged in the terminal devices 101, 102 and 103. In this case, the example system architecture 100 can also not include the server 105 and the network 104.

[0028] It should be understood that Figure 1 The number of terminal devices, networks and servers in the system architecture 100 is only illustrative. According to the implementation needs, there can be any number of terminal devices, networks and servers.

[0029] It should be understood that Figure 2 , Figure 2 A flowchart of a dynamic graph reconstruction method provided by an embodiment of the present disclosure is shown in FIG. 2, wherein the flowchart 200 includes the following steps:

[0030] Step 201: Obtain a target dynamic graph of a target modality, and determine a first implicit representation of each node according to snapshots of each time point constituting the target dynamic graph.

[0031] This step aims to obtain a target dynamic graph of a target modality by an execution subject of the dynamic graph reconstruction method (for example, the server 105 shown in FIG. 1), and determine a first implicit representation of each node according to snapshots of each time point constituting the target dynamic graph. Figure 1 This step aims to obtain a target dynamic graph of a target modality by an execution subject of the dynamic graph reconstruction method (for example, the server 105 shown in FIG. 1), and determine a first implicit representation of each node according to snapshots of each time point constituting the target dynamic graph.

[0032] Generally, a dynamic graph is composed of at least two snapshots, assuming that the target dynamic graph is composed of T snapshots from time 1 to time T, there are data sparsity problems in multiple snapshots or at least one snapshot has missing node information to be completed, these snapshots with problems can be referred to as target snapshots. For data sparsity problems, the target snapshot can be each snapshot; for partial node information missing, the target snapshot is usually other time snapshots except the first snapshot without missing information, and the last T time snapshot is more likely to be used to make full use of the information recorded in the multiple time snapshots to complete the missing partial node information in the T time snapshot.

[0033] Implicit representation is derived from implicit neural representation, which is a new method of parameterizing various signals. Traditional signal representation is usually discrete, for example, an image is a discrete grid of pixels, an audio signal is a discrete sample of amplitude, and a three-dimensional shape is usually parameterized as a voxel, a point cloud or a grid of meshes; on the contrary, implicit neural representation parameterizes signals as continuous functions that map signal domains (i.e. coordinates such as image pixel coordinates) to any position at that coordinate (for images, R, G, B colors). Of course, these functions are usually intractable, i.e. it is impossible to "write down" a function that parameterizes a natural image as a mathematical formula. Therefore, implicit neural representation approximates this "natural representation" function through a neural network, and the representation constructed based on this is as close to the actual situation as possible, which is the implicit representation.

[0034] That is, the first implicit representation is generated in this step, and it is expected that the characteristics of each node hidden in each snapshot are better presented by implicit neural representation.

[0035] According to different fields of the target dynamic graph, nodes are used to refer to different objects. Taking a dynamic graph for recording traffic flow information as an example, traffic flow information describes the flow information between various traffic objects appearing in the target area, which can be traffic lights, vehicles, bus stops, intersections, buildings, etc. That is, each traffic object will be an independent node, and the trajectory information between nodes will depict the traffic flow. Taking a dynamic graph for recording social interaction information as an example, it describes the social information between various users, so each node refers to a user. Other fields will also have other meanings adaptively, which will not be listed one by one here.

[0036] Step 202: obtaining an auxiliary dynamic graph of other modalities different from the target modalities, and determining a second implicit representation of each node according to each time snapshot constituting the auxiliary dynamic graph;

[0037] The step is to first acquire, by the execution subject, an auxiliary dynamic graph of another modality different from the target modality to which the target dynamic graph belongs, and then determine the second implicit representation of each node according to the snapshots constituting the auxiliary dynamic graph.

[0038] Different from step 201, the step is to acquire, from other information sources, an auxiliary dynamic graph of another modality that can also record relevant information in other graph forms, which can also be referred to as an auxiliary modality dynamic graph, and then determine the second implicit representation of each node according to the snapshots constituting the auxiliary dynamic graph, to distinguish from the first implicit representation corresponding to the target dynamic graph.

[0039] That is, the modalities of the target dynamic graph and the auxiliary dynamic graph are different, that is, the acquisition channels and information forms of the two dynamic graphs are different, but they record part of the same information in the same field (for example, they record the same traffic flow information and social network information), which can also be simply understood as different forms of dynamic graphs, in order to complete the missing part of the node and increase the sparse data from another angle by the information contained in another modality through different graph forms and different information acquisition channels, so as to improve the data density.

[0040] It should be noted that the number of auxiliary dynamic graphs can be multiple, that is, multiple auxiliary dynamic graphs of different modalities can be combined at the same time to improve the final reconstruction effect.

[0041] Step 203: Splicing the first implicit representation and the second implicit representation of the same node to obtain the spliced features of each node.

[0042] On the basis of steps 201 and 202, the step is to splice, by the execution subject, the first implicit representation and the second implicit representation of the same node, so as to fuse the implicit representations of the same node from different modalities in a spliced manner, and then obtain the spliced features of each node.

[0043] Step 204: Nonlinear processing of the spliced features of each node, and reconstruction of a new dynamic graph based on the nonlinear features obtained by processing.

[0044] On the basis of step 203, the step is to nonlinearly process, by the execution subject, the spliced features of each node, and reconstruct a new dynamic graph based on the nonlinear features obtained by processing.

[0045] The reason for nonlinearly processing the spliced features is that the interaction characteristics between nodes in actual situations are not suitable for fitting using a "linear" method, but are more suitable for fitting using a "nonlinear" method, so that the nonlinear features obtained by processing can be closer to the actual situation.

[0046] That is, the activities in the real world are participated by the objects with subjective consciousness, and the interaction characteristics thereof do not strictly follow the linear law, that is, the motion trend thereof cannot be described by a simple linear law, and therefore it is more appropriate to fit the interaction characteristics thereof in a nonlinear manner.

[0047] The dynamic graph reconstruction method provided by the embodiments of the present disclosure combines the auxiliary dynamic graph of other modalities on the basis of the target dynamic graph of the target modality, so as to assist in improving the data sparsity problem by combining the dynamic graphs of different modalities of other information sources, and obtain the related node information in the comprehensive different modal dynamic graphs by splicing the first implicit representation and the second implicit representation of the same node, and obtain a new dynamic graph more in line with the actual situation through nonlinear processing, thereby practically improving the data sparsity problem.

[0048] Please refer to Figure 3 , Figure 3 The flowchart of another dynamic graph reconstruction method provided by the embodiments of the present disclosure is shown in FIG. 3, wherein the flow 300 includes the following steps:

[0049] Step 301: obtaining a target dynamic graph of a target modality, and determining a first implicit representation of each node according to snapshots of each time point constituting the target dynamic graph;

[0050] Step 302: obtaining an auxiliary dynamic graph of other modalities different from the target modality, and determining a second implicit representation of each node according to snapshots of each time point constituting the auxiliary dynamic graph by a Bayesian graph representation learning method;

[0051] On the basis of step 202 of flow 200, this step aims to further determine the second implicit representation of each node according to snapshots of each time point constituting the auxiliary dynamic graph by using the Bayesian graph representation learning method.

[0052] The Bayesian graph representation learning method is a specific learning method for graph-form data under the Bayesian learning method, and aims to apply the idea of Bayesian learning to graph-form data. Of course, in addition to using the Bayesian graph representation learning method, other learning methods that can achieve similar effects can also be used, which are not listed one by one here.

[0053] Step 303: splicing the first attraction flow implicit representation and the second attraction flow implicit representation of the same node as an attraction object to obtain a spliced attraction flow feature;

[0054] Step 304: splicing the first emission flow implicit representation and the second emission flow implicit representation of the same node as an emission object to obtain a spliced emission flow feature;

[0055] For step 203 in process 200, the embodiment provides a specific implementation by steps 303-304, that is, the same type of splicing is performed on the implicit representation of the traffic of each node as an attractor and an emitter, respectively, to further refine the splicing accuracy of the implicit representation.

[0056] The implicit representation of the node attracting traffic is the implicit representation of the traffic presented by the node as an attractor attracting other nodes; on the contrary, the implicit representation of the node emitting traffic is the implicit representation of the traffic presented by the node as an emitter to other nodes. In short, the attracting traffic and the emitting traffic correspond to the traffic presented by the two actions of "arrival" and "departure", respectively.

[0057] Specifically, if combined with a specific traffic flow scenario, the attracting traffic and the emitting traffic correspond to the traffic presented by the two actions of "entry" and "exit", respectively. The attracting traffic and the emitting traffic in other scenarios will also be adjusted slightly in meaning according to the actual situation in the scene.

[0058] Step 305: processing the spliced attracting traffic features by using a first nonlinear layer to obtain first nonlinear features;

[0059] Step 306: processing the spliced emitting traffic features by using a second nonlinear layer to obtain second nonlinear features;

[0060] Corresponding to steps 303-304, steps 305-306 aim to process the spliced attracting traffic features and emitting traffic features by different nonlinear processing layers, respectively, to obtain corresponding nonlinear features, so as to avoid affecting the purity of the features due to the mutual mixing of different types of traffic features during nonlinear processing.

[0061] Step 307: processing the first nonlinear features and the second nonlinear features by using a preset decoder to obtain a new dynamic graph with a target modality as a processing result.

[0062] On the basis of steps 305 and 306, this step aims to process the first nonlinear features and the second nonlinear features by using a preset decoder by the above-mentioned execution subject to obtain a new dynamic graph with a target modality as a processing result.

[0063] That is, the decoder corresponds to the "encoder" for generating the implicit representation in steps 301 and 302, and aims to restore the nonlinear features of each node to the form of the dynamic graph of the target modality, and present the new dynamic graph obtained after reconstruction.

[0064] On the basis of any of the above embodiments, for the part of how to splice the spliced features in step 203, the embodiment further specifically provides another implementation:

[0065] The evidence lower bound manner under the variational inference principle is taken as the objective function;

[0066] The first implicit representation and the second implicit representation of the same node are spliced under the guidance of the objective function, and the spliced features of each node are obtained.

[0067] Among them, variational inference (VI) is a large class of methods in Bayesian approximate inference methods, which skillfully transforms the posterior inference problem into an optimization problem for solving. In the application of probability models, a central task is to calculate the posterior probability distribution of latent variables Z given the observed (visible) data variables X, and to calculate the expectation about this probability distribution. For many models in practical applications, it is infeasible to calculate the posterior probability distribution or the expectation about this posterior probability distribution. This may be due to the high dimensionality of the latent space, which makes it impossible to calculate directly, or due to the particularly complex form of the posterior probability distribution, which makes the expectation impossible to obtain an analytical solution.

[0068] For a general function f(x), f can be considered as a real number operator with respect to x, which maps a real number x to a real number f(x). Then by analogy with this mode, suppose there is a function operator F, which is a function operator with respect to f(x), which can map f(x) to a real number F(f(x)). For f(x), the extreme value of f(x) can be obtained by changing x, while in variational, this x is replaced by a function y(x), i.e. by changing x to change y(x), and finally make F(y(x)) obtain the extreme value.

[0069] Variation: refers to the variation of functional, which ultimately seeks extreme functions, which make the functional obtain maximum or minimum value. For example, there are countless paths from point A to point B, and each path is a function. The length of each function (path) is a number, and you choose the shortest or longest path from the countless paths. This is the problem of finding the extreme value of the functional.

[0070] On this basis, variational inference can be understood as finding a simple distribution to approximate the posterior probability density that cannot be solved in the inference problem. In mathematical language, it is to minimize, but because the posterior probability density in the KL divergence (Kullback-Leibler divergence) is also unsolvable, the problem is transformed from minimizing the KL divergence to maximizing the evidence lower bound (Evidence Lower Bound, ELBO, here the evidence refers to the probability density of data or observable variables) by means of the derivation idea of EM (Expectation-maximization, Expectation-maximization).

[0071] That is, the evidence lower bound method is used as the objective function to guide the splicing of features in this embodiment, and the characteristics of the evidence lower bound method are used to obtain better spliced features, so that the new dynamic graph reconstructed finally can eliminate the data sparsity problem or complete the missing node information as much as possible.

[0072] Further, considering that the implicit representation corresponding to the auxiliary dynamic graph can be subject to an arbitrary prior distribution, in order to reduce the difference between the cross-modal, the prior distributions of the first implicit representation and the second implicit representation can also be aligned.

[0073] Furthermore, in order to prevent the information in the auxiliary dynamic graph of the different modal from the target dynamic graph from interfering with the prior distribution of the implicit representation from the target dynamic graph, the gradient information of the second implicit representation can also be discarded in the back propagation process of training.

[0074] To deepen the understanding, the disclosure also gives a specific implementation scheme in combination with one specific application scenario, please see the flowchart as shown in Figure 4 .

[0075] 1. Given a target modal dynamic graph , where the snapshot at time t is denoted as . and ε (t) represent the node set and directed edge set of G (t) . In addition, the node attribute of G (t) is denoted as , and the adjacency matrix is denoted as . At the same time, the auxiliary modal dynamic graph is denoted as , and the adjacency matrix of G is denoted as .

[0076] 2. Process each snapshot constituting the dynamic graph using an encoder such as VGAE (Variational Graph Auto-Encoder, VGAE) to obtain the implicit space representation of the directed graph {G (1) , G (2) , …, G (T)} of the target modal (denoted as {Z (1) , …, Z (T)}) and learn through variational inference. The prior distribution and posterior distribution of Z (t) are denoted as p(Z (t) |A (＜t) , X (＜t) ) and q(Z (t) |A (t) , X (t) , h (t-1) , respectively.), where h (t-1 represents the time-dependent hidden state variable at time t-1;

[0077] 3、Similarly, through Bayesian table representation learning, the auxiliary modal graph hidden representation

[0078] 4、Assume conditional dependence on and subject to Gaussian distribution, then the posterior distribution of is expressed as:

[0079]

[0080] 5、Since the prior distribution of can be any distribution, in order to reduce the difference between cross-modal, the prior distribution of is aligned with the prior distribution of Z (t) :

[0081] 6、Through the principle of variational inference, the ELBO is calculated as the objective function for optimization, so that the information of the auxiliary modal and the information of the target modal are fused.

[0082] 7、In order to prevent the information of the auxiliary modal from interfering with the prior distribution of the target modal, in the back propagation process in the training process, the gradient information of is discarded.

[0083] 8、Finally, the hidden representations of the auxiliary modal and the target modal are spliced and input into two nonlinear layers: and and further input into the decoder to realize the reconstruction of the target modal graph.

[0084] The scheme provided by the embodiment reduces the transfer cost between different modalities through modal adaptation, and uses the knowledge of the supplementary modal to make up for the lack of dynamic graph data of the target modal or solve the problem of data sparsity, thereby enhancing the reconstruction performance of the dynamic graph. On the other hand, the scheme also provides an end-to-end deep learning framework, reducing the dependence on data preprocessing.

[0085] Further referring to Figure 5 , as an implementation of the method shown in the above figures, the disclosure provides an embodiment of a dynamic graph reconstruction device, which corresponds to the method embodiment shown in Figure 2 . The device can be applied to various electronic devices.

[0086] As Figure 5As shown, the dynamic graph reconstruction apparatus 500 in this embodiment can include a first implicit representation determining unit 501, a second implicit representation determining unit 502, an implicit representation splicing unit 503, and a new dynamic graph reconstruction unit 504. The first implicit representation determining unit 501 is configured to obtain a target dynamic graph of a target modality, and determine first implicit representations of nodes according to snapshots at different time points constituting the target dynamic graph. The second implicit representation determining unit 502 is configured to obtain an auxiliary dynamic graph of another modality different from the target modality, and determine second implicit representations of nodes according to snapshots at different time points constituting the auxiliary dynamic graph. The implicit representation splicing unit 503 is configured to splice the first implicit representations and the second implicit representations of the same nodes to obtain spliced features of the nodes. The new dynamic graph reconstruction unit 504 is configured to perform nonlinear processing on the spliced features of the nodes, and reconstruct a new dynamic graph based on nonlinear features obtained by the processing.

[0087] In this embodiment, the first implicit representation determining unit 501, the second implicit representation determining unit 502, the implicit representation splicing unit 503, and the new dynamic graph reconstruction unit 504 in the dynamic graph reconstruction apparatus 500 can perform specific processing and bring about technical effects, which can be respectively referred to Figure 2 The related descriptions of steps 201-204 in the corresponding embodiment will not be repeated here.

[0088] In some optional implementation manners of this embodiment, the second implicit representation determining unit 502 can include a second implicit representation determining subunit configured to determine second implicit representations of nodes according to snapshots at different time points constituting the auxiliary dynamic graph. The second implicit representation determining subunit can be further configured to:

[0089] determine the second implicit representations of the nodes by representing the snapshots at different time points constituting the auxiliary dynamic graph by a Bayesian graph.

[0090] In some optional implementation manners of this embodiment, the implicit representation splicing unit 503 can be further configured to:

[0091] splice the first attraction flow implicit representations and the second attraction flow implicit representations of the same nodes as attractors to obtain spliced attraction flow features;

[0092] splice the first emission flow implicit representations and the second emission flow implicit representations of the same nodes as emitters to obtain spliced emission flow features.

[0093] In some optional implementation manners of this embodiment, the new dynamic graph reconstruction unit 504 can be further configured to:

[0094] process the spliced attraction flow features by a first nonlinear layer to obtain first nonlinear features;

[0095] The spliced-out traffic features are processed by using a second nonlinear layer to obtain second nonlinear features;

[0096] The first nonlinear features and the second nonlinear features are processed by using a preset decoder to obtain a new dynamic graph of the target modality as a processing result.

[0097] In some optional implementations of the embodiment, the implicit representation splicing unit 503 can be further configured to:

[0098] The evidence lower bound mode under the variational inference principle is used as a target function;

[0099] The first implicit representation and the second implicit representation of the same node are guided to be spliced by using the target function to obtain spliced features of each node.

[0100] In some optional implementations of the embodiment, the dynamic graph reconstruction apparatus 500 can further include:

[0101] The prior distribution alignment unit is configured to align prior distributions of the first implicit representation and the second implicit representation before splicing the first implicit representation and the second implicit representation of the same node to obtain spliced features of each node.

[0102] In some optional implementations of the embodiment, the dynamic graph reconstruction apparatus 500 can further include:

[0103] The gradient information discarding unit is configured to discard gradient information of the second implicit representation in the back propagation process.

[0104] The embodiment as the device embodiment corresponding to the above-mentioned method embodiment exists, and the dynamic graph reconstruction apparatus provided by the embodiment combines the auxiliary dynamic graph of other modalities on the basis of the target dynamic graph of the target modality, so as to assist in improving the data sparsity problem by combining different modal dynamic graphs of other information sources, and the related node information in the comprehensive different modal dynamic graphs is obtained by splicing the first implicit representation and the second implicit representation of the same node, and a new dynamic graph more consistent with the actual situation can be obtained after nonlinear processing, thereby realizing the practical improvement of the data sparsity problem.

[0105] According to the embodiments of the present disclosure, the present disclosure further provides an electronic device, which comprises at least one processor and a memory connected with the at least one processor in communication; 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 implement the dynamic graph reconstruction method described in any of the above embodiments.

[0106] According to embodiments of this disclosure, this disclosure also provides a readable storage medium storing computer instructions that enable a computer to implement the dynamic graph reconstruction method described in any of the above embodiments when executed.

[0107] According to embodiments of this disclosure, this disclosure also provides a computer program product that, when executed by a processor, can implement the dynamic graph reconstruction method described in any of the above embodiments.

[0108] Figure 6 A schematic block diagram of an example electronic device 600 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may 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 6 As shown, device 600 includes a computing unit 601, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 602 or a computer program loaded from storage unit 608 into random access memory (RAM) 603. RAM 603 may also store various programs and data required for the operation of device 600. The computing unit 601, ROM 602, and RAM 603 are interconnected via bus 604. Input / output (I / O) interface 605 is also connected to bus 604.

[0110] Multiple components in device 600 are connected to I / O interface 605, including: input unit 606, such as keyboard, mouse, etc.; output unit 607, such as various types of monitors, speakers, etc.; storage unit 608, such as disk, optical disk, etc.; and communication unit 609, such as network card, modem, wireless transceiver, etc. Communication unit 609 allows device 600 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0111] The computing unit 601 can be various general and / or special purpose processing components with processing and computing capabilities. Some examples of the computing unit 601 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 model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The computing unit 601 performs various methods and processes described above, such as the dynamic graph reconstruction method. For example, in some embodiments, the dynamic graph reconstruction method can be implemented as a computer software program tangibly embodied in a machine-readable medium, such as the storage unit 608. In some embodiments, part or all of the computer program can be loaded and / or installed onto the device 600 via the ROM 602 and / or the communication unit 609. When the computer program is loaded onto the RAM 603 and executed by the computing unit 601, one or more steps of the dynamic graph reconstruction method described above can be performed. Alternatively, in other embodiments, the computing unit 601 can be configured to perform the dynamic graph reconstruction method by any other appropriate 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 (PLD), a 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 the functions / operations specified in the flowcharts and / or the block diagrams. The program code can be executed entirely on a machine, partially on a machine, partially on a machine as a stand-alone software package, partially on a machine and partially on a remote machine or entirely on a remote machine or server.

[0114] In the context of this 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 implemented by a computer program running on the respective computers and having a client-server relationship with one another. 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 virtual private server (VPS) services.

[0118] According to the technical scheme of the embodiment of the present disclosure, the auxiliary dynamic graph of other modalities is combined on the basis of the target dynamic graph of the target modality, so as to assist in improving the data sparsity problem by combining the dynamic graphs of different modalities of other information sources, and the related node information in the comprehensive dynamic graph of different modalities is obtained by splicing the first implicit representation and the second implicit representation of the same node, and a new dynamic graph more in line with the actual situation can be obtained through nonlinear processing, thereby realizing the practical improvement of the data sparsity problem.

[0119] It should be understood that the various forms of the flow shown above can be reordered, added, or deleted steps. For example, each step 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 scheme of the present disclosure can be achieved, and the present disclosure is not limited herein.

[0120] The above detailed description does not constitute a limitation on the protection scope of the present disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent replacements and improvements within the spirit and principles of the present disclosure should be included in the protection scope of the present disclosure.

Claims

1. A dynamic graph reconstruction method, comprising: obtaining a target dynamic graph of a target area traffic flow, and determining first implicit representations of nodes according to snapshots constituting the target dynamic graph; obtaining an auxiliary dynamic graph of an area traffic flow different from the target area traffic flow, and determining second implicit representations of nodes according to snapshots constituting the auxiliary dynamic graph; splicing the first implicit representations and the second implicit representations of the same nodes to obtain spliced features of the nodes, including splicing first and second implicit representations of attracting flow of the same nodes as attractors to obtain spliced attracting flow features, the attracting flow corresponding to flow presented by vehicles entering; and splicing first and second implicit representations of emitting flow of the same nodes as emitters to obtain spliced emitting flow features, the emitting flow corresponding to flow presented by vehicles leaving; performing nonlinear processing on the spliced features of the nodes by using a preset decoder, and reconstructing a new dynamic graph of the target area traffic flow based on nonlinear features obtained by the processing.

2. The method of claim 1, wherein, The determining of the second implicit representations of the nodes according to the snapshots constituting the auxiliary dynamic graph comprises: determining the second implicit representations of the nodes by using a Bayesian graph representation learning method on the snapshots constituting the auxiliary dynamic graph.

3. The method of claim 1, wherein, The performing of the nonlinear processing on the spliced features of the nodes by using the preset decoder, and the reconstructing of the new dynamic graph of the target area traffic flow based on the nonlinear features obtained by the processing comprises: processing the spliced attracting flow features by using a first nonlinear layer to obtain first nonlinear features; processing the spliced emitting flow features by using a second nonlinear layer to obtain second nonlinear features; processing the first nonlinear features and the second nonlinear features by using the preset decoder to obtain, as a processing result, the new dynamic graph of the area traffic flow as the target area traffic flow.

4. The method of claim 1, wherein, The splicing of the first implicit representations and the second implicit representations of the same nodes to obtain the spliced features of the nodes comprises: using a lower bound of evidence under a variational inference principle as a target function; using the target function to guide the splicing of the first implicit representations and the second implicit representations of the same nodes to obtain the spliced features of the nodes.

5. The method according to any one of claims 1 to 4, wherein, Before the splicing of the first implicit representations and the second implicit representations of the same nodes to obtain the spliced features of the nodes, the method further comprises: aligning prior distributions of the first implicit representations and the second implicit representations.

6. The method of claim 5, further comprising: discarding gradient information of the second implicit representations in a back propagation process.

7. A dynamic graph reconstruction apparatus, comprising: a first implicit representation determining unit configured to obtain a target dynamic graph of a target area traffic flow, and determine first implicit representations of nodes according to snapshots constituting the target dynamic graph; a second implicit representation determining unit configured to obtain an auxiliary dynamic graph of an area traffic flow different from the target area traffic flow, and determine second implicit representations of nodes according to snapshots constituting the auxiliary dynamic graph; and The implicit representation splicing unit is configured to splice the first and second implicit representations of the same node to obtain spliced features of each node, including: splicing the first and second implicit representations of the same node as an attractor to obtain spliced attractor flow features, the attractor flow corresponding to the flow presented by vehicles entering; and splicing the first and second implicit representations of the same node as an emitter to obtain spliced emitter flow features, the emitter flow corresponding to the flow presented by vehicles exiting. The new dynamic map reconstruction unit is configured to perform nonlinear processing on the spliced features of each node by using a preset decoder, and reconstruct a new dynamic map of the target area traffic flow based on the nonlinear features obtained by processing.

8. The apparatus of claim 7, wherein, The second implicit representation determination unit includes a second implicit representation determination subunit configured to determine the second implicit representation of each node according to the snapshots of each time point constituting the auxiliary dynamic map, and the second implicit representation determination subunit is further configured to: determine the second implicit representation of each node by representing the snapshots of each time point constituting the auxiliary dynamic map by a Bayesian graph representation learning method.

9. The apparatus of claim 7, wherein, The new dynamic map reconstruction unit is further configured to: process the spliced attractor flow features by using a first nonlinear layer to obtain first nonlinear features; process the spliced emitter flow features by using a second nonlinear layer to obtain second nonlinear features; process the first and second nonlinear features by using a preset decoder to obtain a new dynamic map of the target area traffic flow as a processing result.

10. The apparatus of claim 7, wherein, The implicit representation splicing unit is further configured to: use the evidence lower bound method under the variational inference principle as a target function; use the target function to guide the splicing of the first and second implicit representations of the same node to obtain the spliced features of each node.

11. The apparatus of any one of claims 7-10, further comprising: a prior distribution alignment unit configured to align prior distributions of the first and second implicit representations of the same node before splicing the first and second implicit representations of the same node to obtain the spliced features of each node.

12. The apparatus of claim 11, further comprising: a gradient information discarding unit configured to discard gradient information of the second implicit representation in a backpropagation process.

13. An electronic device, comprising: at least one processor; and a memory connected to the at least one processor in communication; 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 dynamic map reconstruction method of any one of claims 1-6.

14. A non-transitory computer-readable storage medium storing computer instructions for causing a computer to perform the dynamic map reconstruction method of any one of claims 1-6. ​ 15. A computer program product comprising a computer program which, when executed by a processor, implements the steps of the dynamic graph reconstruction method according to any one of claims 1-6.

Images