Next point of interest recommendation method based on adaptive spatiotemporal graph representation learning

Abstract

The embodiment of this specification provides a method for recommending the next point of interest based on adaptive spatiotemporal graph representation learning, including: collecting user identity information and user visited location information; using the user identity information and the user visited location information as input, based on a preset geographic map mapping model embedded with door control, obtaining geographic map semantic information; using the user identity information and the user visited location information as input, based on a preset sequence graph mapping model embedded with door control, obtaining sequence graph semantic information; based on semantic joint framework learning, according to the user visited location information, the geographic map semantic information and the sequence graph semantic information, achieving geographic map and sequence graph consistency distribution. The technical solution provided by this application is used to solve the problem of low accuracy in recommending the next point of interest.

Description

Technical Field

[0001] This document relates to the technical field of next point of interest recommendation, and in particular to a next point of interest recommendation method based on adaptive spatiotemporal graph representation learning. Background Art

[0002] With the rapid development of the Internet and the advent of the big data era, the explosive growth of check-in and rating data for points of interest has provided a good background and data foundation for the development of the next point of interest recommendation technology.

[0003] Existing technologies regard places visited by users as nodes, and based on graph convolutional neural networks, realize "data sharing" between nodes, and finally recommend the next point of interest based on the "shared" data.

[0004] However, the hierarchical structure of graph convolutional neural networks limits the weight variation range of each node information stored in each node, which reduces the information propagation effect and thus reduces the accuracy of the next point of interest recommendation. Summary of the invention

[0005] In view of the above analysis, the present application aims to propose a next point of interest recommendation method and device based on adaptive spatiotemporal graph representation learning to improve the accuracy of the next point of interest recommendation.

[0006] In a first aspect, one or more embodiments of this specification provide a next point of interest recommendation method based on adaptive spatiotemporal graph representation learning, including:

[0007] Collect user identity information and user visit location information;

[0008] Taking the user identity information and the user visit location information as input, based on a preset geographic map surveying and mapping model embedded with door control, obtain geographic map semantic information;

[0009] Taking the user identity information and the user visit location information as input, based on a preset sequence diagram mapping model embedded in door control, obtaining sequence diagram semantic information;

[0010] Based on semantic joint framework learning, the geographical map and the sequence map semantic information are realized according to the user's access location information, the geographical map semantic information and the sequence map semantic information. Figure 1 Consistent distribution.

[0011] Furthermore, the user identity information and the user visit location information are used as inputs, and based on a preset geographic map surveying and mapping model embedded in door control, the geographic map semantic information is obtained, including:

[0012] Determine the master node information and the neighbor node information according to the user identity information and the user access location information;

[0013] Inputting the master node information and the neighbor node information into the adaptive gate of the geographic map surveying and mapping model to obtain gated data that considers the importance of the master node information and the neighbor node information;

[0014] Determine the master node weight and neighbor node weight according to the gating data;

[0015] The geographic map semantic information is determined according to the master node weight, the neighbor node weight, the master node information and the neighbor node information.

[0016] Furthermore, the user identity information and the user access location information are used as inputs, and based on a preset sequence diagram mapping model embedded in door control, sequence diagram semantic information is obtained, including:

[0017] Obtaining a sequence diagram according to the user identity information and the user visit location information;

[0018] According to the sequence diagram, determining an initial user preference value;

[0019] Forming a common feature by combining the initial user preference value and the user's visited location information;

[0020] Inputting the common feature into the adaptive gate of the sequential graph mapping model to obtain a gate control signal;

[0021] The initial user preference value is adjusted according to the gate control signal to obtain the sequence diagram semantic information.

[0022] Furthermore, the semantic joint framework learning is based on the user's access location information, the geographic map semantic information and the sequence map semantic information to achieve geographic map and sequence map semantic information. Figure 1 Consistent distribution, including:

[0023] Determining a behavior anchor point sequence according to the user's visited location information;

[0024] Determining a geographic map anchor point sequence according to the geographic map semantic information;

[0025] Determining a sequence diagram anchor point sequence according to the sequence diagram semantic information;

[0026] Determining a first similarity between the behavior anchor point sequence and the geographic map anchor point sequence;

[0027] Determining a second similarity between the behavior anchor point sequence and the sequence graph anchor point sequence;

[0028] According to the first similarity and the second similarity, the geographic map and the sequence are realized. Figure 1 Consistent distribution.

[0029] Furthermore, the geographic map and the sequence are realized according to the first similarity and the second similarity. Figure 1 Consistent distribution, including:

[0030] Substituting the first similarity and the second similarity into a preset soft label cross entropy loss function, and calculating the consistency parameter of the geographic map and the sequence map;

[0031] When it is determined that the consistency parameter has not reached the preset value, the above process is repeated until the consistency parameter reaches the preset value.

[0032] In a second aspect, one or more embodiments of this specification provide a device for recommending a next point of interest based on an adaptive spatiotemporal graph, including:

[0033] Acquisition module, geographic map module, sequence map module and semantic association module;

[0034] The collection module is used to collect user identity information and user visit location information;

[0035] The geographic map module is used to obtain geographic map semantic information based on a preset geographic map surveying and mapping model embedded in door control by taking the user identity information and the user visit location information as input;

[0036] The sequence diagram module is used to obtain sequence diagram semantic information based on the preset sequence diagram mapping model embedded in the door control by taking the user identity information and the user visit location information as input;

[0037] The semantic association module is used to learn based on the semantic association framework, and realize the geographic map and sequence map semantic information according to the user's access location information, the geographic map semantic information and the sequence map semantic information. Figure 1 Consistent distribution.

[0038] Furthermore, the geographic map module is used to determine the main node information and the neighbor node information based on the user identity information and the user visit location information; input the main node information and the neighbor node information into the adaptive gate of the geographic map surveying and mapping model to obtain gating data that considers the importance of the main node information and the neighbor node information; determine the main node weight and the neighbor node weight based on the gating data; determine the geographic map semantic information based on the main node weight, the neighbor node weight, the main node information and the neighbor node information.

[0039] Furthermore, the sequence diagram module is used to obtain a sequence diagram based on the user identity information and the user visit location information; determine an initial user preference value based on the sequence diagram; concatenate the initial user preference value and the user visit location information to obtain a common feature; input the common feature into the control gate of the sequence diagram mapping model to obtain a gate control signal; adjust the initial user preference value according to the gate control signal to obtain the sequence diagram semantic information.

[0040] Furthermore, the semantic union module is used to determine a behavior anchor sequence according to the user's visited location information; determine a geographic map anchor sequence according to the geographic map semantic information; determine a sequence map anchor sequence according to the sequence map semantic information; determine a first similarity between the behavior anchor sequence and the geographic map anchor sequence; determine a second similarity between the behavior anchor sequence and the sequence map anchor sequence; and implement the geographic map and the sequence map anchor sequence according to the first similarity and the second similarity. Figure 1 Consistent distribution.

[0041] In a third aspect, one or more embodiments of this specification provide a storage medium, including:

[0042] Used to store computer executable instructions, which implement the method described in any one of the first aspects when executed.

[0043] Compared with the prior art, this application can at least achieve the following technical effects:

[0044] By introducing a gate control mechanism into the geographic map mapping model, the information weight of each node can be adjusted according to different scenarios. By introducing a gate control mechanism into the sequential map mapping model, the influence of user preferences on the calculation results is strengthened. Finally, the results of the geographic map mapping model and the sequential map mapping model are fused to further improve the calculation efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate one or more embodiments of this specification or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0046] Figure 1 A flowchart of a method for recommending a next point of interest based on an adaptive spatiotemporal graph is provided for one or more embodiments of this specification. DETAILED DESCRIPTION

[0047] In order to enable those skilled in the art to better understand the technical solutions in one or more embodiments of this specification, the following will be combined with the drawings in one or more embodiments of this specification to clearly and completely describe the technical solutions in one or more embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on one or more embodiments of this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this document.

[0048] A point of interest is a coordinate in a geographic location system, which can refer to restaurants, stadiums, clothing stores and other stores. There are many types of points of interest, and each point of interest contains multiple attributes such as name, category, geographic location, brand, rating, number of check-ins, etc. The point of interest recommendation technology aims to combine the user's historical visit records to points of interest with the rich features of the points of interest to recommend the places that users are most likely to be interested in.

[0049] The next POI recommendation needs to be based on the information contained in the existing POIs, so the weight distribution problem between POIs needs to be considered. For example, if the user's existing POI A is the neighborhood where he lives, then when the user chooses the next POI B, he will most likely choose restaurants (B1), barber shops (B2), and parks (B3) around the neighborhood. Therefore, in the specific calculation process, POI A is the main node, and POIs B1, B2, and B3 are neighboring nodes. The weight of the information contained in POI A in POIs B1, B2, and B3 is increased to increase the probability that POIs B1, B2, and B3 are selected as the next POI.

[0050] For another example, when predicting the next point of interest based on the tourist spots that the user has visited, since there is no strong correlation between the points of interest as in the previous example, it is necessary to collect as many points of interest as possible to facilitate the determination of the next point of interest. It is necessary to increase the weight of the information of each neighbor node in the main node to achieve the determination of the next point of interest based on more neighbor nodes.

[0051] However, the existing graph convolutional neural network has only one weight distribution rule when calculating, and cannot change the weight distribution rule according to different scenarios. That is, the hierarchical structure of the graph convolutional neural network will limit the weight change range of each node information stored in each node. In addition, this weight distribution rule cannot well reflect user preferences.

[0052] In view of the above technical problems, the present application embodiment provides a next point of interest recommendation method based on adaptive spatiotemporal graph representation learning, such as Figure 1 As shown, the following steps are included:

[0053] Step 1: Collect user identity information and user visit location information.

[0054] In the embodiment of the present application, the user identity information includes but is not limited to name, gender and ID number. The user visit location information includes but is not limited to the identifier, coordinates and visit order of the visit location.

[0055] Step 2: Taking the user identity information and the user's visited location information as input, based on the preset geographic map surveying and mapping model embedded with door control, obtain the geographic map semantic information.

[0056] In the embodiment of the present application, in order to reasonably determine the weight of each node information for different scenarios, the present application embeds an adaptive gate on the basis of the graph neural network model. That is, the geographic map surveying and mapping model is based on the graph neural network model. After embedding the adaptive gate, the weight of each node information is adjusted based on the gate control.

[0057] Specifically, the main node information and neighbor node information are determined according to the user identity information and the user access location information; the main node information and neighbor node information are input into the adaptive gate of the geographic map surveying and mapping model to obtain the gating data that considers the importance of the main node information and the neighbor node information; the main node weight and the neighbor node weight are determined according to the gating data; the geographic map semantic information is determined according to the main node weight, the neighbor node weight, the main node information and the neighbor node information.

[0058] Among them, the specific calculation process of the adaptive gate is:

[0059] R Geo =σ(W1x1+W2I i→j )

[0060] O i→j =R Geo ·x1+(1-R Geo )·I i→j

[0061] Where x1 and I i→j They represent the main node information and neighbor node information of the graph neural network respectively, and W1 and W2 represent the information corresponding to the input x1 and I i→j The linear transformation weight matrix, σ represents the sigmoid activation function, which is used to convert the result of the linear transformation into a value between 0 and 1 as a gate control signal. Geo Represents the gate control signal, which determines x1 and I i→j The contribution ratio in the final output. i→j is the output after gate adjustment, which is x1 and I i→j From the above formula, we can see that the geographic map adaptive gate allows the model to dynamically adjust x1 and I i→jThe relative importance in the final output helps the GCN in the geographic graph model to dynamically adjust the main node information and neighbor node information, so as to learn to retain more important node information to improve the recommendation accuracy of the geographic graph.

[0062] Step 3: Taking the user identity information and the user access location information as input, based on the preset sequence diagram mapping model embedded in the door control, the sequence diagram semantic information is obtained.

[0063] In the embodiment of the present application, the semantic information of the sequence graph is used to describe the user's preference, that is, the probability of the user appearing at the corresponding node. In order to make the calculation result better reflect the user's preference, the present application embeds an adaptive gate in the graph neural network and optimizes the user's preference value based on the adaptive gate.

[0064] Specifically, a sequence diagram is obtained based on user identity information and user visit location information; an initial user preference value is determined based on the sequence diagram; common features are formed by splicing the initial user preference value and user visit location information; the common features are input into an adaptive gate of a sequence diagram mapping model to obtain a gate control signal; the initial user preference value is adjusted based on the gate control signal to obtain sequence diagram semantic information.

[0065] Among them, the formula related to the adaptive gate is:

[0066] c f =concat(M s,u ,t e )

[0067] R seq =σ(W·c f +b)

[0068] e s,u =R seq ×M s,u

[0069] Among them, M s,u is the initial user access preference learned by the random walk neural network, t e represents the attribute information of the target visited location, W and b represent the weight and bias of the linear layer respectively, and σ is the Sigmoid function, which ensures that the value of the gate control signal is between 0 and 1. f Finally, e s,u It represents the user preference feature after gate adjustment, which dynamically adjusts the weight of the user access preference feature through the gate control signal. The sequential graph gate is mainly used to dynamically adjust the importance of the user access preference learned by the random walk network according to the attribute characteristics of the target access location.

[0070] Step 4: Based on the semantic joint framework learning, realize the geographic map and sequence map semantic information according to the user's access location information, geographic map semantic information and sequence map semantic information. Figure 1 Consistent distribution.

[0071] In the embodiment of the present application, step 4 is specifically as follows:

[0072] Determine the behavior anchor sequence according to the user's visited location information; determine the geographic map anchor sequence according to the geographic map semantic information; determine the sequence map anchor sequence according to the sequence map semantic information; determine the first similarity between the behavior anchor sequence and the geographic map anchor sequence; determine the second similarity between the behavior anchor sequence and the sequence map anchor sequence; implement the geographic map and sequence map anchor sequence based on the first similarity and the second similarity Figure 1 Consistent distribution.

[0073] Since the weights in more scenarios are taken into account, overfitting is prone to occur if hard labels are used in the calculation process. Therefore, this application introduces a soft label cross entropy loss function when constructing the loss function. Figure 1 The specific distribution of consistency is:

[0074] The first similarity and the second similarity are brought into the preset soft label cross entropy loss function to calculate the consistency parameter of the geographic map and the sequence map; when it is determined that the consistency parameter does not reach the preset value, the above process is repeated until the consistency parameter reaches the preset value.

[0075] Specifically, the expression of the soft label cross entropy loss function is:

[0076]

[0077] Where T is the temperature parameter and KL represents the KL divergence. It is worth noting that even without temperature adjustment (T=1), the soft label cross entropy loss function is still quite different from the KL divergence, because the KL divergence itself does not involve additional scaling of the gradient. In short, the soft label cross entropy loss function is regarded as the KL divergence after considering the adjustment and gradient scaling.

[0078] The calculation formula for the first similarity is:

[0079]

[0080] The calculation formula for the second similarity is:

[0081]

[0082] The calculation formula of consistency parameter is:

[0083]

[0084] The soft label cross entropy loss of H(·,·) calculates the difference between the two distributions. The formula of Φ(·,·) represents the knowledge distillation process. The temperature t is set to 0.5, and e g,u and They represent the user behavior anchor sequence and the geographic map anchor sequence of the geographic map, respectively. s,u and They represent the user behavior anchor sequence of the sequence graph and the sequence graph anchor sequence respectively, and u represents the number of users. Finally, the geographic map loss L is calculated respectively. geo , the loss L of the sequential graph seq , and combined with the loss function of the semantic joint framework, the output forms the loss of the final model.

[0085] In the embodiment of the present application, the entire method uses two models, namely a geographic map mapping model and a sequential map mapping model.

[0086] Among them, the work content of the geographic map surveying and mapping model is:

[0087] P1, build geographic map

[0088] An undirected geographic graph is constructed to describe user preferences. The representation of the geographic graph is G(E g ,A g ), where E g =(P i , P j ) is P in the geographic map i and P j The distance of undirected edges within a specific threshold Δd, A g =(i, j) is poiP i and P j The edge weight matrix of the geographical distances between them.

[0089] P2, a message passing mechanism for geographic graphs

[0090] In geographic graphs, message passing neural networks (MPNNs) are mainly used to aggregate and update geographic graph nodes. Message passing neural networks are a general network framework for graph supervised learning. They contain the commonalities of several advanced models in graph structured data to more clearly and intuitively understand the relationship between nodes and update changes. The forward transmission of message passing neural networks is mainly divided into two stages: the message passing stage and the readout stage.

[0091] In the message passing phase, the message function represents the i To P j Information transmission process:

[0092]

[0093] in is the distance kernel, that is, when d represents the distance between POIs ij As the distance kernel k(d ij ) is reduced to indicate that the farther away the POI is, the less influence it has on the user’s decision. (l) represents the trainable weight matrix, represents the POI embedding information at layer l-1 in GNN, Indicates initialization of POI embedding information. i , N j They represent the nodes P in the message transmission process. i and P j Neighbor node information. i→j represents the result of simple fusion of the main node information and the neighbor node information. In order to process the node information more flexibly and effectively, this paper introduces the geographic map adaptive gating mechanism on this basis, aiming to more effectively fuse the node information. The adjusted node information is represented by O i→j The node update function represents the node update process. We use the activation function LeakyRelu to update the node. The random gradient of the LeakyRelu function that is less than 0 introduces randomness, which can avoid falling into the local optimal solution during the message passing process.

[0094]

[0095] in Represents the aggregated messages adjusted by the gating mechanism. Represents node V j neighbor nodes.

[0096] The message passing mechanism is mainly propagated in the L layer. Finally, we integrate the update information into a list, which represents the delivery update results of all user check-in points in the L layer.

[0097]

[0098] In the readout phase, a multi-head attention mechanism is used to aggregate node-level representations to obtain graph-level representations. The multi-head attention mechanism can solve the problem that the model over-focuses on its own position when encoding information at the current position. Therefore, the multi-head attention mechanism can aggregate high-order geographic representations in user check-in behavior, ultimately achieving the goal of semantically capturing topological geographic influences.

[0099]

[0100]

[0101]

[0102] Among them, Q r , K r 、V r is the transformation matrix of query, key and value, is the importance of the i-th node information, R is the number of heads in the multi-head attention mechanism, SU represents the user's historical check-in sequence, and L i (1) Represents the geographic coding of the POI sequence of layer L. Finally, the aggregated updated geographic semantic representation e is obtained g,u .

[0103] The work content of the sequence diagram mapping model is:

[0104] P1. Construction sequence diagram

[0105] The representation method of the sequence diagram is G(V s,u ,E s,u ), represents the sequential graph formed by user u and its historical check-in sequence su, where edge E s,u =(v i , v j ) indicates that the user is from the point of interest v i To the point of interest j The check-in process, the sequence diagram mainly describes the relationship between the check-in sequence of users to the points of interest.

[0106] P2, Graph Kernel Neural Network

[0107] Firstly, the kernel matrix of the random walk kernel is constructed, and the kernel matrix is ​​used to perform random walk on the direct product graph in the sequence graph. Finally, a graph filter is defined to optimize the extraction of structural features in the sequence graph.

[0108] Definition of random walk kernel: The random walk kernel with step size p on the direct product graph is defined as follows, where e represents the unit vector, A x Represents G x The adjacency matrix of .

[0109]

[0110] Define direct product graph: Define two graphs G a (V a ,E a ) and G b (V b ,E b ), then the direct product graph is G x (V x ,E x ), where V a and V b Represents Figure G a and Gb The node E a and E b Represent the edges in the graph respectively. . In the direct product graph, V x ={(v a ,v b ):v a ∈V a ∧v b ∈V b}, E x ={{(v a ,v b ),(u a ,u b )}:{v a ,u a}∈E∧{v b ,u b}∈E b}.

[0111] Definition of graph filter: Graph filter is an operation to enhance or attenuate the intensity of each frequency component in the graph signal spectrum. Graph filter is mainly used to extract the structural features in the sequence graph extracted by the graph kernel, so as to capture the high-order sequence substructure of user check-in. The graph filter G of each layer bi With bn nodes, the adjacency matrix Parameters can be assigned to the graph filter. Since the graph kernel can implicitly represent the data in the feature space and use the kernel function to calculate the inner product in the space, the similarity between a pair of graphs can be quantified. Therefore, the present invention chooses to use its characteristics to quantify the similarity between each sequential graph and the graph filter of the user.

[0112] Given different random walk step sizes P = {0, ..., p}, and a series of graph filters G mentioned above bi = {G b1 , ..., G bn}, and for each sequence graph G s,u Construct the matrix M∈R N×(p+1) .

[0113] M o,q =k (q-1) (G s,u ,G bi )

[0114] k (q-1) represents a random walk kernel with a step size of q-1. Here, o and q represent the indices of each element of the matrix.

[0115] Finally, the matrix M is expanded into a one-dimensional vector and put into the fully connected layer of the graph convolutional neural network to generate the initial sequential semantic representation M s，u .

[0116]

[0117] User Preferences s,u It reflects the high-order sequence substructure captured by the graph kernel neural network in the sequence graph, and then the user preference M learned from the sequence graph is s,u Combined with the adaptive gating mechanism, the importance of relevant information in user preferences is adjusted according to the attribute information of the target embedding, and finally a sequential graph semantic representation is generated. s,u .

[0118] The embodiment of the present application provides a next point of interest recommendation device based on adaptive spatiotemporal graph representation learning, including: a collection module, a geographic graph module, a sequence graph module and a semantic association module;

[0119] The collection module is used to collect user identity information and user visit location information;

[0120] The geographic map module is used to obtain geographic map semantic information based on a preset geographic map surveying and mapping model embedded in door control by taking the user identity information and the user visit location information as input;

[0121] The sequence diagram module is used to obtain sequence diagram semantic information based on a preset sequence diagram mapping model embedded in door control by taking the user identity information and the user visit location information as input;

[0122] The semantic association module is used to learn based on the semantic association framework, and realize the geographic map and sequence map semantic information according to the user's access location information, the geographic map semantic information and the sequence map semantic information. Figure 1 Consistent distribution.

[0123] In an embodiment of the present application, the geographic map module is used to determine the main node information and the neighbor node information based on the user identity information and the user visit location information; input the main node information and the neighbor node information into the adaptive gate of the geographic map surveying and mapping model to obtain gating data that considers the importance of the main node information and the neighbor node information; determine the main node weight and the neighbor node weight based on the gating data; determine the geographic map semantic information based on the main node weight, the neighbor node weight, the main node information and the neighbor node information.

[0124] In an embodiment of the present application, the sequence diagram module is used to obtain a sequence diagram based on the user identity information and the user visit location information; determine an initial user preference value based on the sequence diagram; concatenate the initial user preference value and the user visit location information to obtain a common feature; input the common feature into the control gate of the sequence diagram mapping model to obtain a gate control signal; adjust the initial user preference value according to the gate control signal to obtain the sequence diagram semantic information.

[0125] In an embodiment of the present application, the semantic union module is used to determine a behavior anchor sequence according to the user's access location information; determine a geographic map anchor sequence according to the geographic map semantic information; determine a sequence map anchor sequence according to the sequence map semantic information; determine a first similarity between the behavior anchor sequence and the geographic map anchor sequence; determine a second similarity between the behavior anchor sequence and the sequence map anchor sequence; and implement the geographic map and the sequence map anchor sequence according to the first similarity and the second similarity. Figure 1 Consistent distribution.

[0126] The present application provides a storage medium, including:

[0127] Used to store computer executable instructions, which implement the method described in any of the above embodiments when executed.

[0128] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0129] In the 1930s, improvements to a technology could be clearly distinguished as hardware improvements (for example, improvements to the circuit structure of diodes, transistors, switches, etc.) or software improvements (improvements to the method flow). However, with the development of technology, many improvements to the method flow today can be regarded as direct improvements to the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved method flow into the hardware circuit. Therefore, it cannot be said that an improvement in a method flow cannot be implemented using a hardware entity module. For example, a programmable logic device (PLD) (such as a field programmable gate array (FPGA)) is such an integrated circuit whose logical function is determined by the user's programming of the device. Designers can "integrate" a digital method on a PLD by programming it themselves, without having to ask a chip manufacturer to design and produce a dedicated integrated circuit chip. Moreover, nowadays, instead of manually making integrated circuit chips, this kind of programming is mostly implemented by "logic compiler" software, which is similar to the software compiler used when developing and writing programs, and the original code before compilation must also be written in a specific programming language, which is called hardware description language (HDL). There is not only one HDL, but many kinds, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. The most commonly used ones are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also know that it is only necessary to program the method flow slightly in the above-mentioned hardware description languages ​​and program it into the integrated circuit, and then it is easy to obtain the hardware circuit that implements the logic method flow.

[0130] The controller can be implemented in any appropriate manner, for example, the controller can take the form of a microprocessor or processor and a computer-readable medium storing a computer-readable program code (such as software or firmware) that can be executed by the (micro)processor, a logic gate, a switch, an application-specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320. The memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art also know that in addition to implementing the controller in a purely computer-readable program code manner, the controller can be implemented in the form of a logic gate, a switch, an application-specific integrated circuit, a programmable logic controller, and an embedded microcontroller by logically programming the method steps. Therefore, this controller can be considered as a hardware component, and the devices included therein for implementing various functions can also be regarded as structures within the hardware component. Or even, the devices for implementing various functions can be regarded as both software modules for implementing the method and structures within the hardware component.

[0131] The methods, devices, modules or units described in the above embodiments may be implemented by a computer chip or entity, or by a product having a certain function. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0132] For the convenience of description, the above devices are described in terms of functions and are divided into various units. Of course, when implementing the embodiments of this specification, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0133] It will be appreciated by those skilled in the art that one or more embodiments of this specification may be provided as methods, methods or computer program products. Therefore, one or more embodiments of this specification may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0134] This specification is described with reference to the flowcharts and / or block diagrams of the methods, devices (methods), and computer program products according to the embodiments of this specification. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0135] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0136] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0137] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0138] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0139] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0140] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0141] One or more embodiments of the present specification may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. One or more embodiments of the present specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0142] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the method embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0143] The above description is only an embodiment of this document and is not intended to limit this document. For those skilled in the art, this document may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this document should be included in the scope of the claims of this document.

Claims

1. A next point of interest recommendation method based on adaptive spatiotemporal graph representation learning, characterized in that: include: Collect user identity information and user visit location information; Taking the user identity information and the user visit location information as input, based on a preset geographic map surveying and mapping model embedded with door control, obtain geographic map semantic information; Taking the user identity information and the user visit location information as input, based on a preset sequence diagram mapping model embedded in door control, obtaining sequence diagram semantic information; Based on the consistency framework learning, according to the user's access location information, the geographic map semantic information and the sequence map semantic information, the geographic map and the sequence map consistent distribution is achieved; The method of obtaining geographic map semantic information based on a preset geographic map surveying and mapping model embedded with door control by taking the user identity information and the user visit location information as input includes: Determine the master node information and the neighbor node information according to the user identity information and the user access location information; Inputting the master node information and the neighbor node information into the adaptive gate of the geographic map surveying and mapping model to obtain gated data that considers the importance of the master node information and the neighbor node information; Determine the master node weight and neighbor node weight according to the gating data; Determining the geographic map semantic information according to the master node weight, the neighbor node weight, the master node information and the neighbor node information; The consistent framework-based learning realizes consistent distribution of geographic maps and sequential maps according to the user's access location information, the geographic map semantic information, and the sequential map semantic information, including: Determining a behavior anchor point sequence according to the user's visited location information; Determining a map anchor point sequence according to the geographic map semantic information; Determining a sequence diagram anchor point sequence according to the sequence diagram semantic information; Determining a first similarity between the behavior anchor point sequence and the map anchor point sequence; Determining a second similarity between the behavior anchor point sequence and the sequence graph anchor point sequence; Achieving consistent distribution of the geographic graph and the sequence graph according to the first similarity and the second similarity; The step of achieving consistent distribution of the geographic graph and the sequence graph according to the first similarity and the second similarity includes: Substituting the first similarity and the second similarity into a preset soft label cross entropy loss function, and calculating the consistency parameter of the geographic map and the sequence map; When it is determined that the consistency parameter has not reached the preset value, repeating the above process until the consistency parameter reaches the preset value; Where H represents the KL divergence between the first similarity and the second similarity output, in a temperature-adjusted form; T is the temperature parameter, and KL represents the KL divergence; The calculation formula for the first similarity is: The calculation formula for the second similarity is: The calculation formula of the consistency parameter is: where H(·‖·) represents the difference between the two distributions calculated by the soft label cross entropy loss, the Φ(·,·) formula represents the knowledge distillation process, the temperature t is set to 0.5, and e g,u and They represent the user behavior anchor sequence and map anchor sequence of the geographic map, respectively, and e s,u and They represent the user behavior anchor point sequence of the sequence graph and the anchor point sequence of the sequence graph respectively. u always represents a specific user, and N is used to represent the total number of users.

2. The method according to claim 1, characterized in that The method uses the user identity information and the user's access target location information as input, and obtains sequence diagram semantic information based on a preset sequence diagram mapping model embedded in door control, including: Obtaining a sequence diagram according to the user identity information and the user visit location information; According to the sequence diagram, determining an initial user preference value; Forming a common feature based on the initial user preference value and the user's visit target location information; Inputting the common feature into the adaptive gate of the sequential graph mapping model to obtain a gate control signal; The initial user access preference value is adjusted according to the gate control information to obtain the sequence diagram semantic information.

3. An adaptive point of interest recommendation device, characterized in that: include: Acquisition module, geographic map module, sequence map module and consistency module; The collection module is used to collect user identity information and user visit location information; The geographic map module is used to obtain geographic map semantic information based on a preset geographic map surveying and mapping model embedded in door control by taking the user identity information and the user visit location information as input; The sequence diagram module is used to obtain sequence diagram semantic information based on the preset sequence diagram mapping model embedded in the door control by taking the user identity information and the user visit location information as input; The consistency module is used to achieve consistency classification of geographic map and sequence map based on consistency framework learning according to the user's access location information, the geographic map semantic information and the sequence map semantic information; The geographic map module is used to determine the main node information and the neighbor node information according to the user identity information and the user visit location information; input the main node information and the neighbor node information into the adaptive gate of the geographic map surveying and mapping model to obtain the gating data considering the importance of the main node information and the neighbor node information; determine the main node weight and the neighbor node weight according to the gating data; determine the geographic map semantic information according to the main node weight, the neighbor node weight, the main node information and the neighbor node information; The consistency module is used to determine a behavior anchor sequence according to the user's visited location information; determine a map anchor sequence according to the geographic map semantic information; determine a sequence map anchor sequence according to the sequence map semantic information; determine a first similarity between the behavior anchor sequence and the map anchor sequence; determine a second similarity between the behavior anchor sequence and the sequence map anchor sequence; Achieving consistent distribution of the geographic graph and the sequence graph according to the first similarity and the second similarity; Where H represents the KL divergence between the first similarity and the second similarity output, in a temperature-adjusted form; T is the temperature parameter, and KL represents the KL divergence; The calculation formula for the first similarity is: The calculation formula for the second similarity is: The calculation formula of the consistency parameter is: where H(·‖·) represents the difference between the two distributions calculated by the soft label cross entropy loss, the Φ(·,·) formula represents the knowledge distillation process, the temperature t is set to 0.5, and e g,u and They represent the user behavior anchor sequence and map anchor sequence of the geographic map, respectively, and e s,u and They represent the user behavior anchor point sequence of the sequence graph and the anchor point sequence of the sequence graph respectively. u always represents a specific user, and N is used to represent the total number of users.

4. The device according to claim 3, characterized in that The sequence diagram module is used to obtain a sequence diagram according to the user identity information and the user's target location information; determine the initial user preference value according to the sequence diagram; and compare the initial user preference value and the user's target location information. e Splicing to get common features; The common features are input into the control gate of the sequence diagram mapping model to obtain a gate control signal; and the sequence diagram semantic information is obtained according to the gate control information and the initial user preference value.

5. A storage medium, characterized in that: include: Used to store computer executable instructions, which implement the method of claim 1 or 2 when executed.

Images