CN113610265A - Hypergraph convolutional network-based time-space behavior prediction method and system - Google Patents
Hypergraph convolutional network-based time-space behavior prediction method and system Download PDFInfo
- Publication number
- CN113610265A CN113610265A CN202110706207.0A CN202110706207A CN113610265A CN 113610265 A CN113610265 A CN 113610265A CN 202110706207 A CN202110706207 A CN 202110706207A CN 113610265 A CN113610265 A CN 113610265A
- Authority
- CN
- China
- Prior art keywords
- user
- hypergraph
- behavior
- time
- channel
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 65
- 230000003993 interaction Effects 0.000 claims abstract description 93
- 238000012549 training Methods 0.000 claims abstract description 30
- 238000013528 artificial neural network Methods 0.000 claims abstract description 18
- 230000006399 behavior Effects 0.000 claims description 250
- 239000013598 vector Substances 0.000 claims description 53
- 239000011159 matrix material Substances 0.000 claims description 50
- 238000012545 processing Methods 0.000 claims description 18
- 230000006870 function Effects 0.000 claims description 13
- 238000004590 computer program Methods 0.000 claims description 11
- 238000010276 construction Methods 0.000 claims description 10
- 230000007246 mechanism Effects 0.000 claims description 10
- 238000000926 separation method Methods 0.000 claims description 10
- 239000000126 substance Substances 0.000 claims description 9
- 238000003860 storage Methods 0.000 claims description 8
- 230000004931 aggregating effect Effects 0.000 claims description 7
- 230000001902 propagating effect Effects 0.000 claims description 7
- 238000013527 convolutional neural network Methods 0.000 claims description 6
- 230000002776 aggregation Effects 0.000 claims description 5
- 238000004220 aggregation Methods 0.000 claims description 5
- 230000009471 action Effects 0.000 description 13
- 238000010586 diagram Methods 0.000 description 13
- 230000002452 interceptive effect Effects 0.000 description 8
- 230000008569 process Effects 0.000 description 8
- 230000005540 biological transmission Effects 0.000 description 6
- 238000004891 communication Methods 0.000 description 5
- 235000019580 granularity Nutrition 0.000 description 5
- 238000012935 Averaging Methods 0.000 description 4
- 238000005070 sampling Methods 0.000 description 4
- 238000012163 sequencing technique Methods 0.000 description 4
- 238000005520 cutting process Methods 0.000 description 3
- 238000007781 pre-processing Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 230000003542 behavioural effect Effects 0.000 description 2
- 238000009826 distribution Methods 0.000 description 2
- 238000007726 management method Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000013439 planning Methods 0.000 description 2
- 230000000644 propagated effect Effects 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000006806 disease prevention Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000007499 fusion processing Methods 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 238000010438 heat treatment Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000013468 resource allocation Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9537—Spatial or temporal dependent retrieval, e.g. spatiotemporal queries
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2216/00—Indexing scheme relating to additional aspects of information retrieval not explicitly covered by G06F16/00 and subgroups
- G06F2216/03—Data mining
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Business, Economics & Management (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Strategic Management (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Economics (AREA)
- Evolutionary Computation (AREA)
- Human Resources & Organizations (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Development Economics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a hypergraph convolutional network-based space-time behavior prediction method and a hypergraph convolutional network-based space-time behavior prediction system, wherein the method comprises the following steps: determining a user ID of a time-space behavior to be predicted and a corresponding time and place; inputting the user ID of the spatio-temporal behavior to be predicted and the corresponding time and place into a spatio-temporal behavior prediction model to obtain a behavior result corresponding to the user ID output by the spatio-temporal behavior prediction model; the spatiotemporal behavior prediction model is obtained by training a hypergraph graph convolution neural network constructed based on historical spatiotemporal behavior data of a user. The invention solves the problem of insufficient modeling of the interaction relation among the user, the time, the place and the behavior under the user time-space behavior scene at present, thereby realizing the accurate prediction of the user time-space behavior.
Description
Technical Field
The invention relates to the technical field of space-time behavior prediction, in particular to a space-time behavior prediction method and system based on a hypergraph convolutional network.
Background
The Spatiotemporal behavior Prediction (spatiomporal Activity Prediction) aims at predicting the behavior of a user under a given Spatiotemporal scene (time, place). Because the dependence of the daily life of the user on the mobile internet is enhanced, the method has very important significance and value for modeling and predicting the time-space behavior rule of the user. The rules of the user behaviors can be found through the research on the user time-space behaviors, the behavior prediction can be carried out on the observed user individuals under a specific time-space scene, the possibility, the scale and the potential development trend of certain collective behaviors can be found in the user time-space behavior rules, and a decision maker can be helped to make a more reasonable decision and carry out overall planning. In a word, the space-time behavior prediction has extremely high research value and can be widely applied to scenes such as disease prevention, smart cities, network optimization, mobile recommendation and the like.
At present, the research of user space-time behavior prediction is mainly divided into two aspects: tensor-based methods and graph-based methods. Tensor-based methods model users, time, place, and behavior separately with a quaternary tensor, and then use the idea of tensor decomposition and Collaborative Filtering (CF) to obtain eigenvectors, such as SCP, UCLAF. The graph model-based method explicitly models a graph (graph) of relationships among users, time, places and behaviors, and learning algorithms for embedding characteristics into nodes in the graph (graph) are divided into two types: one is based on the traditional graph embedding (graph embedding) method, such as CrossMap, ACTOR; one type is a convolutional neural network (GCN) based approach, such as SA-GCN.
The prior art has the following limitations: (1) neglecting the modeling of user similarity depiction and user relationship in a space-time scene, obviously, the method is very important in a complex user space-time behavior scene, and can effectively solve the problem of data sparsity; (2) the modeling of the interaction relation among the user, the time, the place and the behavior under the user space-time behavior scene lacks deeper thinking and exploration, the tensor-based method does not display the modeling interaction relation, and the graph model-based method is insufficient for modeling the relation, is mostly limited to a specific scene and lacks universality.
Disclosure of Invention
The embodiment of the invention provides a hypergraph convolutional network-based spatiotemporal behavior prediction method and a hypergraph convolutional network-based spatiotemporal behavior prediction system, which are used for solving the problems of part or all of the problems in the conventional user spatiotemporal behavior prediction technology.
In a first aspect, an embodiment of the present invention provides a hypergraph convolutional network-based spatiotemporal behavior prediction method, including:
determining a user ID of a time-space behavior to be predicted and a corresponding time and place;
inputting the user ID of the spatio-temporal behavior to be predicted and the corresponding time and place into a spatio-temporal behavior prediction model to obtain a behavior result corresponding to the user ID output by the spatio-temporal behavior prediction model;
the spatiotemporal behavior prediction model is obtained by training a hypergraph graph convolution neural network constructed based on historical spatiotemporal behavior data of a user.
Further, the spatiotemporal behavior prediction model is obtained by training a hypergraph graph convolution neural network constructed based on historical spatiotemporal behavior data of a user, and comprises the following steps:
constructing a multi-channel user relationship hypergraph and a heterogeneous interaction hypergraph based on historical spatiotemporal behavior data of a user;
carrying out iteration on the multi-channel user relationship hypergraph and the heterogeneous interaction hypergraph for L times of hypergraph convolution operation and propagation to obtain L groups of different user, time, place and behavior embedded vectors;
obtaining an embedding vector for prediction based on the L groups of different user, time, place and behavior embedding vectors, and obtaining a corresponding prediction score based on the embedding vector for prediction; wherein, L is the number of hypergraph convolution layers;
and training the graph convolution neural network to obtain the space-time behavior prediction model based on a supervised learning method for constructing an objective function based on the prediction score.
Further, the method for constructing the multi-channel user relationship hypergraph and the heterogeneous interaction hypergraph based on the historical spatiotemporal behavior data of the user comprises the following steps:
after discretizing historical spatio-temporal behavior data of a user into quadruple data, acquiring an interaction matrix containing user preference information, which specifically comprises the following steps: respectively carrying out discretization processing of user ID, discretization processing of time division, discretization processing of region division or clustering of longitude and latitude coordinates of a place and discretization processing of keyword or theme clustering of behaviors on the historical time-space behavior data of the user; the interaction matrix containing the user preference information comprises an interaction matrix of a user and a place, an interaction matrix of the user and time and an interaction matrix of the user and a behavior;
and constructing a multi-channel user relationship hypergraph comprising a separation channel, a local channel and a global channel and a heterogeneous interaction hypergraph taking a user as a center based on the quadruple data and the interaction matrix containing the user preference information.
Further, the iterating the multi-channel user relationship hypergraph and the heterogeneous interaction hypergraph to perform L times of hypergraph convolution operation and propagation to obtain L groups of different user, time, place and behavior embedding vectors, and the method comprises the following steps:
iterating the multi-channel user relationship hypergraph in the separation channel, the local channel and the global channel for L times of hypergraph convolution operation and propagation to obtain user embedded features of each channel, and aggregating the user embedded features of each channel by adopting an attention mechanism to obtain L groups of user, time, place and behavior embedded vectors fused with the user embedded features;
and carrying out L times of hypergraph convolution operation and propagation on the heterogeneous interaction hypergraph iteration to obtain L groups of user, time, place and behavior embedded vectors including node characteristics and hyperedge characteristics.
Further, the formula for propagating the multi-channel user relationship hypergraph is represented as:
wherein the content of the first and second substances,
is a user-embedded feature of each channel, DcAnd ΔcDegree matrices, H, of hypergraph nodes and hyperedges, respectivelycIs a relationship matrix of the hypergraph,
the characteristics corresponding to the three channels are obtained by respectively carrying out hypergraph convolution operation on each channel.
Further, the formula for propagating the heterogeneous interaction hypergraph is expressed as:
wherein AGG is an aggregation function,
is a user-embedded feature obtained in a multi-channel user relationship hypergraph,
is the corresponding super edge, epsilon, of the userl,εt,εaRespectively, the connection location l, the time t and the set of all the super edges of the corresponding nodes of the behavior a.
Further, the embedding vector for prediction is obtained based on the L groups of different user, time, place and behavior embedding vectors, and the formula is as follows:
wherein the content of the first and second substances,
respectively being a super-map convolution layer
Embedding characteristics corresponding to the user u, the place l, the time t and the behavior a of the layer;
w, b are the learnable weight matrix and bias vector, respectively, and q is the attention vector.
In a second aspect, an embodiment of the present invention provides a hypergraph convolutional network-based spatiotemporal behavior prediction system, including:
the user information determining unit is used for determining a user ID of the time-space behavior to be predicted and the corresponding time and place;
the space-time behavior prediction unit is used for inputting the user ID of the space-time behavior to be predicted and the corresponding time and place into a space-time behavior prediction model to obtain a behavior result corresponding to the user ID output by the space-time behavior prediction model;
the spatiotemporal behavior prediction model is obtained by training a hypergraph graph convolution neural network constructed based on historical spatiotemporal behavior data of a user.
In a third aspect, an embodiment of the present invention provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor executes the program to implement the steps of any one of the hypergraph convolutional network-based spatiotemporal behavior prediction methods provided in the first aspect.
In a fourth aspect, an embodiment of the present invention provides a non-transitory computer-readable storage medium, on which a computer program is stored, where the computer program, when executed by a processor, implements the steps of the hypergraph convolutional network-based spatiotemporal behavior prediction method as described in any one of the aspects provided above.
According to the hypergraph convolutional network-based spatiotemporal behavior prediction method and system, the user ID and the corresponding time and place are input into the spatiotemporal behavior prediction model, the behavior result corresponding to the user ID output by the spatiotemporal behavior prediction model is obtained, and the spatiotemporal behavior prediction model is obtained through hypergraph convolutional neural network training constructed based on historical spatiotemporal behavior data of the user. The invention solves the problem of insufficient modeling of the interaction relation among the user, the time, the place and the behavior under the user time-space behavior scene at present, thereby realizing the accurate prediction of the user time-space behavior.
Drawings
In order to more clearly illustrate the technical solutions of the present invention or the prior art, the drawings needed for the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and those skilled in the art can also obtain other drawings according to the drawings without creative efforts.
FIG. 1 is a schematic flow chart of a spatiotemporal behavior prediction method based on a hypergraph convolutional network provided by the invention;
FIG. 2 is a schematic diagram of a training process of a spatiotemporal behavior prediction model provided by the present invention;
FIG. 3 is a schematic diagram of a model prediction process provided by the present invention;
FIG. 4 is a schematic diagram of a user similarity pattern provided by the present invention;
FIGS. 5a and 5b are user spatiotemporal behavioral data and user centric interaction patterns provided by the present invention;
FIG. 6 is a schematic diagram of a heterogeneous interactive hypergraph construction provided by the present invention;
FIG. 7 is a schematic diagram of the construction and information dissemination of a multi-channel user relationship hypergraph provided by the present invention;
FIG. 8 is a schematic diagram of the information dissemination process in the heterogeneous interaction hypergraph provided by the present invention;
FIG. 9 is a schematic structural diagram of a spatiotemporal behavior prediction system based on a hypergraph convolutional network provided by the invention;
FIG. 10 is a schematic diagram of user spatio-temporal behavior prediction for mobile phone APP usage provided by the present invention;
FIGS. 11a and 11b are schematic diagrams of prediction of spatiotemporal behavior of a user for twitter keywords provided by the present invention;
FIG. 12 is a schematic diagram of user spatiotemporal behavior prediction for purchasing behavior in spatiotemporal scenarios provided by the present invention;
fig. 13 is a schematic structural diagram of an electronic device provided by the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings, and it is obvious that the described embodiments are some, but not all embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The technical idea of the invention is as follows: two hypergraphs are constructed to solve the limitation problem existing in the prior art: 1) modeling a user high-order relation under different time-space behavior modes into a multi-channel hypergraph (user relation hypergraph), and fusing user similarity information of different channels through hypergraph convolution propagation and Attention mechanism (Attention), thereby obtaining better user embedding characteristics; 2) the complex interaction relation in modeling the user time-space behavior is displayed by modeling the user, the time, the place and the behavior to another heterogeneous hypergraph (interaction hypergraph), and the complex interaction information is fully captured by the two-step hypergraph convolution operation of node-hypergraph edge-node.
The following describes a spatiotemporal behavior prediction method and system based on a hypergraph convolutional network provided by the present invention with reference to fig. 1 to 13.
The embodiment of the invention provides a hypergraph convolutional network-based space-time behavior prediction method. Fig. 1 is a schematic flowchart of a spatio-temporal behavior predictor method based on a hypergraph convolutional network according to an embodiment of the present invention, as shown in fig. 1, the method includes:
the spatiotemporal behavior prediction model is obtained by training a hypergraph graph convolution neural network constructed based on historical spatiotemporal behavior data of a user.
According to the prediction method provided by the embodiment of the invention, the user ID and the corresponding time and place are input into the space-time behavior prediction model to obtain the behavior result corresponding to the user ID output by the space-time behavior prediction model, and the space-time behavior prediction model is obtained by hypergraph convolutional neural network training constructed on the basis of historical space-time behavior data of the user.
Based on any of the above embodiments, as shown in fig. 2, the spatiotemporal behavior prediction model is obtained by training a hypergraph graph convolutional neural network constructed based on historical spatiotemporal behavior data of a user, and includes: step 210, constructing a multi-channel user relationship hypergraph and a heterogeneous interaction hypergraph based on historical spatiotemporal behavior data of a user;
and 240, training the graph convolution neural network to obtain the space-time behavior prediction model by a supervised learning method for constructing an objective function based on the prediction score.
Specifically, the prediction process is as shown in fig. 3, and the final prediction is made by embedding features in the user, location, time, and behavior, such as simple multiplication and summation of corresponding elements, for example as follows:
in order to optimize the model, a paired learning mode is adopted for model training, and the mode is widely applied to recommendation and prediction problems. Thus, the objective function is defined as follows:
wherein the content of the first and second substances,
representing pairs of training data with negative samples, λ is the weight of the L2 regularization term, and Θ is the trainable model parameter.
In the model training process, model hyper-parameters also need to be set, including a negative sample number sample _ number, a batch size mini _ batch _ size, an embedding size embedding _ size, a learning rate learning _ rate, an L2 regular item L2_ normalization, and a convolution layer number layer _ number. In the process of training the network, the weights and bias values of each layer of the network can be updated by a Stochastic Gradient Descent method (Stochastic Gradient decision) in the process of back propagation.
Based on any one of the embodiments, the method for constructing the multi-channel user relationship hypergraph and the heterogeneous interaction hypergraph based on the historical spatiotemporal behavior data of the user comprises the following steps:
after discretizing historical spatio-temporal behavior data of a user into quadruple data, acquiring an interaction matrix containing user preference information, which specifically comprises the following steps: respectively carrying out discretization processing of user ID, discretization processing of time division, discretization processing of region division or clustering of longitude and latitude coordinates of a place and discretization processing of keyword or theme clustering of behaviors on the historical time-space behavior data of the user; the interaction matrix containing the user preference information comprises an interaction matrix of a user and a place, an interaction matrix of the user and time and an interaction matrix of the user and a behavior;
specifically, firstly, based on the user space-time behavior data of the platform mobile phone
The interaction information of the user with the place, the time and the behavior (including different preferences of the user on the place and the behavior of the time) can be extracted. In order to more accurately depict the preference information of the user, the three kinds of interaction information need to be modeled and used
To represent a set of users, places, time slices, and behaviors (N)U,NL,NT，NARespectively representing the number of users, the number of places, the number of times and the number of behaviors), and modeling the interaction information of the users with the places, the times and the behaviors into three interaction matrixes, namely a user-place interaction matrix
User-time interaction matrix
User-behavior interaction matrix
To represent different preferences of the user for location, time and behavior, respectively. Wherein each element of the matrix is a binary value to indicate whether a corresponding interaction exists. The interaction matrix is constructed as follows: if (u, l) appears in the same user record, then Rul(u, l) ═ 1, otherwise 0.
And constructing a multi-channel user relationship hypergraph comprising a separation channel, a local channel and a global channel and a heterogeneous interaction hypergraph taking a user as a center based on the quadruple data and the interaction matrix containing the user preference information.
In particular, user spatiotemporal behavior data based on quadruplets
And the extracted interaction matrix R containing the user preference informationul，Rut，RuaTwo different hypergraphs, a multi-channel user relationship hypergraph and a heterogeneous interaction hypergraph can be constructed to respectively model a high-order relationship of a user and a complex interaction relationship under a space-time scene. In the two types of hypergraphs, the invention adopts hypergraph convolution operation to carry out information propagation and obtain the characteristics of nodes and hyperedges. And finally, based on the embedding characteristics obtained by the hypergraph convolutional network, the user time-space behavior is finally predicted.
In the space-time scene, the user similarity is multi-type and multi-granularity. For example, the spatiotemporal behavioral data of a user may reveal different types of preferences for time, place, and behavior, so the user's depiction of similarity is multi-typed. While users who tend to do the same action in the same place and users who merely like to go to similar places or have similar action preferences are representations of user similarity at different granularities. For complex user similarity, the invention proposes 7 different user relationship modes for accurate depiction, and divides them into three groups (separation mode, local mode and global mode), as shown in fig. 4. According to the user mode grouping, user relation hypergraphs of three channels (a separation channel s, a local channel l and a global channel g) are respectively constructed to depict user similarity under different granularities, and therefore the multi-channel user relation hypergraph is constructed
Wherein, the nodes are users, and the super-edge connection is the user relationship in different similarity modes.
In the spatiotemporal scenario, for a specific user, a specific time, place and behavior can determine a specific user spatiotemporal behavior, as shown in fig. 5a, that is, the user needs to interact with the place, time and behavior at the same time, and is defined as a "user-centered" interaction mode, as shown in fig. 5 b. Obviously, the relation is a one-to-many relation, the ordinary graph model cannot model the relation, and the interaction relation of 'user is taken as the center' can be modeled more naturally because one edge of the hypergraph can connect more than two nodes.
To explicitly model a "user centric" interaction relationship, another heteromorphic graph is constructed
Where the place, time, and action are taken as nodes and the user as a super edge. The specific way of constructing the heterogeneous interactive hypergraph is shown in fig. 6.
Based on any of the above embodiments, the iterating the multi-channel user relationship hypergraph and the heterogeneous interaction hypergraph to perform L times of hypergraph convolution operation and propagation to obtain L groups of different user, time, place and behavior embedding vectors, including:
iterating the multi-channel user relationship hypergraph in the separation channel, the local channel and the global channel for L times of hypergraph convolution operation and propagation to obtain user embedded features of each channel, and aggregating the user embedded features of each channel by adopting an attention mechanism to obtain L groups of user, time, place and behavior embedded vectors fused with the user embedded features;
specifically, as shown in fig. 7, in order to obtain the user relationship under the user similarity with different granularities, embedding propagation is performed on each channel through hypergraph convolution, so as to obtain the user features under the user similarity with different granularities.
Then, an attention mechanism is adopted to fuse the embedded features of the three channels, user relationship information under different granularity user similarity modes is obtained, and for a user u, a fusion process can be expressed as the following formula:
where W, b are the learnable weight matrix and bias vector, respectively. q is the attention vector. Through the attention mechanism, multi-channel user relationship information is fused, the complex user relationship in a space-time scene is captured, and better user embedding characteristics can be obtained.
And carrying out L times of hypergraph convolution operation and propagation on the heterogeneous interaction hypergraph iteration to obtain L groups of user, time, place and behavior embedded vectors including node characteristics and hyperedge characteristics.
Specifically, fig. 8 shows an information propagation process in the heterogeneous interactive hypergraph, which is mainly performed in two steps, and adopts two-step hypergraph convolution operations: firstly, the node (place, time and behavior) features are propagated to the super edge to generate super edge (user) features, and the super edge (user) features are combined with the user information learned in the user relationship graph and are added to serve as the final user features; the super-edge features are then propagated to the nodes, generating node (place, time, and behavior) features.
Based on any of the above embodiments, the formula for propagating the multi-channel user relationship hypergraph is expressed as:
wherein the content of the first and second substances,
is a user-embedded feature of each channel, DcAnd ΔcDegree matrices, H, of hypergraph nodes and hyperedges, respectivelycIs a relationship matrix of the hypergraph,
the characteristics corresponding to the three channels are obtained by respectively carrying out hypergraph convolution operation on each channel.
Specifically, the propagation update of the user-embedded features in the multi-channel hypergraph can be expressed as:
wherein
Is a user-embedded feature of a certain channel, DcAnd ΔcDegree matrices, H, of hypergraph nodes and hyperedges, respectivelycIs a relational matrix of the hypergraph. By respectively carrying out hypergraph convolution calculation on each channel, the characteristics corresponding to the three channels can be obtained
Based on any of the above embodiments, the formula for propagating the heterogeneous interaction hypergraph is expressed
Wherein AGG is an aggregation function,
is a user-embedded feature obtained in a multi-channel user relationship hypergraph,
is the corresponding super edge, epsilon, of the userl,εt,εaRespectively, the connection location l, the time t and the set of all the super edges of the corresponding nodes of the behavior a.
Specifically, the embedded feature propagation rule in the heterogeneous interaction hypergraph can be expressed as:
wherein AGG is an aggregation function, which may be a function such as a simple mean function, a mean function with sampling coefficients, maximum pooling, etc.
Is a user embedded feature obtained from a multi-channel user relationship hypergraph.
Is the corresponding super edge, epsilon, of the userl,εt,εaAre respectively provided withIs the set of all the super edges connecting the nodes corresponding to the location l, the time t and the action a. Through two-step information propagation, the complex interaction of the user and the time, the place and the action can be fully captured, and the embedded characteristics of the user, the time, the place and the action are better.
Based on any of the above embodiments, the embedding vector for prediction is obtained based on the L groups of different user, time, place, and behavior embedding vectors, and the formula is as follows:
wherein the content of the first and second substances,
respectively being a super-map convolution layer
Embedding characteristics corresponding to the user u, the place l, the time t and the behavior a of the layer;
w, b are the learnable weight matrix and bias vector, respectively, and q is the attention vector.
It should be noted that, by iteratively performing L times of hypergraph convolution propagation in the multi-channel user hypergraph and the heterogeneous interaction hypergraph, L groups of different user, location, time, and behavior embedding vectors can be obtained, and embedding of all layers is combined, such as in a connection or averaging manner, to merge information received from neighbors at different depths for prediction, and the embedding vector finally used for prediction can be expressed as formula (8).
For feature outputs in hypergraph convolutional layers, define
The embedded feature matrix of all users, places, times and behaviors of the layer is
The embedded characteristics corresponding to the user u, the location l, the time t and the action a are respectively
The invention provides a hypergraph convolutional network-based spatiotemporal behavior prediction system, and the hypergraph convolutional network-based spatiotemporal behavior prediction method described below and described above can be referred to correspondingly.
Fig. 9 is a schematic structural diagram of a spatiotemporal behavior prediction system based on a hypergraph convolutional network according to an embodiment of the present invention, as shown in fig. 9, the system includes a user information determination unit 910 and a spatiotemporal behavior prediction unit 920;
the user information determining unit 910 is configured to determine a user ID of a temporal-spatial behavior to be predicted and a corresponding time and place;
the spatiotemporal behavior prediction unit 920 is configured to input the user ID of the spatiotemporal behavior to be predicted and the corresponding time and place into a spatiotemporal behavior prediction model, so as to obtain a behavior result corresponding to the user ID output by the spatiotemporal behavior prediction model;
the spatiotemporal behavior prediction model is obtained by training a hypergraph graph convolution neural network constructed based on historical spatiotemporal behavior data of a user.
According to the prediction system provided by the embodiment of the invention, the user ID and the corresponding time and place are input into the space-time behavior prediction model to obtain the behavior result corresponding to the user ID output by the space-time behavior prediction model, and the space-time behavior prediction model is obtained by hypergraph convolutional neural network training constructed on the basis of historical space-time behavior data of the user.
Based on any one of the embodiments, the space-time behavior prediction unit comprises a hypergraph construction module, a convolution operation and propagation module, a prediction score construction module and a prediction score training module;
the hypergraph construction module is used for constructing a multi-channel user relationship hypergraph and a heterogeneous interaction hypergraph based on historical spatiotemporal behavior data of a user;
the convolution operation and propagation module is used for iterating the multi-channel user relationship hypergraph and the heterogeneous interaction hypergraph to carry out L times of hypergraph convolution operation and propagation to obtain L groups of different user, time, place and behavior embedded vectors;
the prediction score construction module is used for obtaining an embedding vector for prediction based on the L groups of different user, time, place and behavior embedding vectors and obtaining a corresponding prediction score based on the embedding vector for prediction; wherein, L is the number of hypergraph convolution layers;
and the prediction score training module is used for training the graph convolution neural network to obtain the space-time behavior prediction model based on a supervised learning method for constructing an objective function based on the prediction score.
Based on any embodiment, the hypergraph construction module comprises a data discretization processing module and a data construction module;
the data discretization processing module is used for obtaining an interaction matrix containing user preference information after discretizing historical time-space behavior data of a user into quadruple data, and specifically comprises the following steps: respectively carrying out discretization processing of user ID, discretization processing of time division, discretization processing of region division or clustering of longitude and latitude coordinates of a place and discretization processing of keyword or theme clustering of behaviors on the historical time-space behavior data of the user; the interaction matrix containing the user preference information comprises an interaction matrix of a user and a place, an interaction matrix of the user and time and an interaction matrix of the user and a behavior;
the data construction module is used for constructing a multi-channel user relationship hypergraph comprising a separation channel, a local channel and a global channel and a heterogeneous interaction hypergraph taking a user as a center based on the quadruple data and the interaction matrix containing the user preference information.
Based on any one of the embodiments, the convolution operation and propagation module comprises a multi-channel user relationship hypergraph operation and propagation module and a heterogeneous interaction hypergraph operation and propagation module;
the multi-channel user relation hypergraph operation and propagation module is used for iterating the multi-channel user relation hypergraph in the separation channel, the local channel and the global channel for L times of hypergraph convolution operation and propagation to obtain user embedding characteristics of each channel, and aggregating the user embedding characteristics of each channel by adopting an attention mechanism to obtain L groups of user, time, place and behavior embedding vectors which are fused with the user embedding characteristics;
and the heterogeneous interactive hypergraph operation and propagation module is used for carrying out L times of hypergraph convolution operation and propagation on the heterogeneous interactive hypergraph iteration to obtain L groups of user, time, place and behavior embedded vectors including node characteristics and hyperedge characteristics.
Based on any of the above embodiments, the formula for propagating the multi-channel user relationship hypergraph is expressed as:
wherein the content of the first and second substances,
is a user-embedded feature of each channel, DcAnd ΔcDegree matrices, H, of hypergraph nodes and hyperedges, respectivelycIs a relationship matrix of the hypergraph,
the characteristics corresponding to the three channels are obtained by respectively carrying out hypergraph convolution operation on each channel.
Based on any of the above embodiments, the formula for propagating the heterogeneous interaction hypergraph is expressed as:
wherein AGG is an aggregation function,
is a user-embedded feature obtained in a multi-channel user relationship hypergraph,
is the corresponding super edge, epsilon, of the userl，εt，εaRespectively, the connection location l, the time t and the set of all the super edges of the corresponding nodes of the behavior a.
Based on any of the above embodiments, the embedding vector for prediction is obtained based on the L groups of different user, time, place, and behavior embedding vectors, and the formula is as follows:
wherein the content of the first and second substances,
respectively being a super-map convolution layer
Embedding characteristics corresponding to the user u, the place l, the time t and the behavior a of the layer;
w, b are the learnable weight matrix and bias vector, respectively, and q is the attention vector.
The method and the system can be suitable for all space-time behavior prediction scenes (such as mobile APP use prediction, user commodity purchase prediction, microblog theme prediction and the like under the space-time scene). The concrete description is as follows:
specific examples of the application include: the mobile network operator wants to use the user spatiotemporal behavior prediction system to predict the usage of the mobile phone APP of the user, and then reasonably allocate network resources, as shown in fig. 10. In this embodiment, the data acquisition of the mobile network operator mainly originates from a base station accessed by the intelligent mobile terminal, including longitude and latitude coordinates of the base station, and a user ID received by the base station, that is, APP historical usage data (user ID, location information, time, behavior), and performs data mining from the APP historical usage data, and what needs to be predicted is an APP that may be used in a given user, location (base station coordinates), and time situation, so as to optimize network resource allocation.
Firstly, preprocessing user data, discretizing time (selecting a discretization mode for a specific application scene), for example, cutting a day into 24 time slices according to 24 hours a day to obtain discretized four-tuple data
The interaction matrix R containing the user preference information can be obtained from the obtained quadruple dataul，Rut，Rua. Then two hypergraphs and a multichannel user relation hypergraph are constructed according to the user data and the interaction matrix
And heterogeneous interaction hypergraphs
For user, place, time and action nodes in the hypergraph, the independent heating is adoptedThe encoding encodes the input and then compresses it into dense real-valued vectors, the first
The embedded feature matrix of all users, places, times and behaviors of the layer is
The embedded characteristics corresponding to the user u, the location l, the time t and the action a are respectively
Representing input features of users, places, time and behaviors as layer 0 features of a hypergraph neural network, performing hypergraph convolution propagation on a graph structure to capture hypergraph structure information, and updating entity features from the aspect of representation learning. For multi-channel user relationship hypergraph
Firstly, carrying out hypergraph convolution propagation on each channel, and aggregating to obtain user embedded features of three channels
The user embedding characteristics of each channel are then adaptively aggregated using an Attention mechanism (Attention) to obtain an output user embedding
For heterogeneous interaction hypergraphs
The information transmission in the method can carry out two-step transmission of 'node-super edge-node', thereby obtaining the characteristics of the node and the super edge and fully capturing the complex interaction information of the user and the time-space behavior. The hypergraph convolution operation in the interactive hypergraph can be fused with the output of the multi-channel user relationship hypergraph
The two-step hypergraph convolution can obtain the embedded characteristic output of a specific layer
After L times of hypergraph convolution propagation are carried out iteratively, L groups of different user, place, time and behavior embedding vectors can be obtained. And combining the embedding of all layers, e.g. concatenating or averaging, to combine the information received from the neighbors at different depths for prediction, resulting in a final representation p for predictionu,ql，rt,sa. Finally, multiplying the embedded representation in the corresponding dimension and summing to obtain the corresponding prediction fraction
The model training adopts a paired learning mode, the influence of positive and negative samples on the model performance is comprehensively considered through negative sampling, and the performance and the universality of the model are improved.
After a network operator obtains the prediction scores of all APPs of a user at a specific base station position and at a specific time, the system can perform sequencing, and a plurality of APPs with the highest sequencing are used as the APPs possibly interested by the user; and then, user distribution statistics is carried out on different APP types at specific time and base station positions, and possible network flow consumption of the corresponding base station at the moment is calculated, so that dynamic network resource management and allocation are carried out, and more reasonable base station site selection and layout can be carried out. In conclusion, the general framework for predicting the time-space behavior provided by the invention can be well applied to the network optimization problem.
Specific application examples are two: the prediction of the user time-space behavior provided by the invention can also be used for predicting key words (contents) of text data such as blogs and the like. Taking the keyword prediction of the twitter text as an example, as shown in fig. 11a, the keyword sent by the user on the premise of a given time and place is predicted, which can also be regarded as a kind of spatiotemporal behavior prediction. In this embodiment, tweets (or blog text data) and corresponding user anonymous ID time information, latitude and longitude coordinates, may be obtained, as shown in fig. 11 b. What needs to be predicted is that given time and place, a user may issue a tweet of declared keywords, which is essentially a prediction of the text content, although the problem can be extended to text topic prediction.
Firstly, preprocessing user data, and discretizing time (selecting a discretization mode aiming at a specific application scene), for example, dividing one day into 24 time slices by cutting the time slices at 24 hours in one day; for longitude and latitude coordinates of a place, mapping into a discretized ID representation by adopting a method of dividing a grid area or clustering; for text information, a text processing method in NLP (e.g. clustering to obtain keywords, and LDA to obtain text topics) can be used to map to discretized behaviors (e.g. keywords or topics). Thus, discretized quadruple data can be obtained
The interaction matrix R containing the user preference information can be obtained from the obtained quadruple dataul，Rut，Rua. Then two hypergraphs and a multichannel user relation hypergraph are constructed according to the user data and the interaction matrix
And heterogeneous interaction hypergraphs
For user, place, time and action nodes in the hypergraph, the input is encoded using one-hot encoding and then compressed into dense real-valued vectors, the first
The embedded feature matrix of all users, places, times and behaviors of the layer is
The embedded characteristics corresponding to the user u, the location l, the time t and the action a are respectively
Representing input features of users, places, time and behaviors as layer 0 features of a hypergraph neural network, performing hypergraph convolution propagation on a graph structure to capture hypergraph structure information, and updating entity features from the aspect of representation learning. For multi-channel user relationship hypergraph
Firstly, carrying out hypergraph convolution propagation on each channel, and aggregating to obtain user embedded features of three channels
The user embedding characteristics of each channel are then adaptively aggregated using an Attention mechanism (Attention) to obtain an output user embedding
For heterogeneous interaction hypergraphs
The information transmission in the method can carry out two-step transmission of 'node-super edge-node', thereby obtaining the characteristics of the node and the super edge and fully capturing the complex interaction information of the user and the time-space behavior. The hypergraph convolution operation in the interactive hypergraph can be fused with the output of the multi-channel user relationship hypergraph
The two-step hypergraph convolution can obtain the embedded characteristic output of a specific layer
After L times of hypergraph convolution propagation are carried out iteratively, L groups of different user, place, time and behavior embedding vectors can be obtained. And combining the embedding of all layers, e.g. concatenating or averaging, to combine information received from neighbors of different depthsLine prediction, resulting in a final representation p for predictionu，ql,rt,sa. Finally, multiplying the embedded representation in the corresponding dimension and summing to obtain the corresponding prediction fraction
The model training adopts a paired learning mode, the influence of positive and negative samples on the model performance is comprehensively considered through negative sampling, and the performance and the universality of the model are improved.
After the prediction scores of the possible text keywords of the user at a specific time and position are obtained, the system can perform sequencing, and a plurality of keywords with the highest sequencing are used for describing the contents of the text possibly published by the user, so that the text contents of the user can be predicted, and the platform can recommend the related contents to the user according to the contents. In conclusion, the general frame for predicting the space-time behavior provided by the invention can be well applied to content prediction of text information such as blogs and the like.
Specific application examples are three: the user space-time behavior prediction provided by the invention can also be used for recommending problems (such as takeaway demand prediction) in space-time scenes. As shown in fig. 12, the user spatio-temporal behavior prediction framework proposed by the present invention introduces spatio-temporal information to more reasonably model the purchasing behavior of the user, inputs the obtained user characteristics into the spatio-temporal behavior prediction network, and performs the region characteristic interaction with the spatio-temporal behavior prediction network after completing the similarity modeling according to the obtained region characteristics, the region similarity map, and the surrounding region information. In this embodiment, historical order information of the user with spatio-temporal information may be obtained, taking a take-out scene as an example, the platform may obtain information such as the ID order time, the place of placing an order, and the type of take-out, of the mobile phone user, and what needs to be predicted is the type of a possible take-out order (purchase behavior) of the user at a given time and place.
Firstly, preprocessing user data, and discretizing time (selecting a discretization mode aiming at a specific application scene), for example, dividing one day into 24 time slices by cutting the time slices at 24 hours in one day; for the longitude and latitude coordinates of the place, a method of dividing grid areas or clustering can be adopted to map the coordinates into a discretized ID tableShown in the figure. Thus, discretized quadruple data can be obtained
The interaction matrix R containing the user preference information can be obtained from the obtained quadruple dataul，Rut，Rua. Then two hypergraphs and a multichannel user relation hypergraph are constructed according to the user data and the interaction matrix
And heterogeneous interaction hypergraphs
For user, place, time and action nodes in the hypergraph, the input is encoded using one-hot encoding and then compressed into dense real-valued vectors, the first
The embedded feature matrix of all users, places, times and behaviors of the layer is
The embedded characteristics corresponding to the user u, the location l, the time t and the action a are respectively
Representing input features of users, places, time and behaviors as layer 0 features of a hypergraph neural network, performing hypergraph convolution propagation on a graph structure to capture hypergraph structure information, and updating entity features from the aspect of representation learning. For multi-channel user relationship hypergraph
Firstly, carrying out hypergraph convolution propagation on each channel, and aggregating to obtain users of three channelsEmbedded features
The user embedding characteristics of each channel are then adaptively aggregated using an Attention mechanism (Attention) to obtain an output user embedding
For heterogeneous interaction hypergraphs
The information transmission in the method can carry out two-step transmission of 'node-super edge-node', thereby obtaining the characteristics of the node and the super edge and fully capturing the complex interaction information of the user and the time-space behavior. The hypergraph convolution operation in the interactive hypergraph can be fused with the output of the multi-channel user relationship hypergraph
The two-step hypergraph convolution can obtain the embedded characteristic output of a specific layer
After L times of hypergraph convolution propagation are carried out iteratively, L groups of different user, place, time and behavior embedding vectors can be obtained. And combining the embedding of all layers, e.g. concatenating or averaging, to combine the information received from the neighbors at different depths for prediction, resulting in a final representation p for predictionu,ql,rt,sa. Finally, multiplying the embedded representation in the corresponding dimension and summing to obtain the corresponding prediction fraction
The model training adopts a paired learning mode, the influence of positive and negative samples on the model performance is comprehensively considered through negative sampling, and the performance and the universality of the model are improved.
After the predicted scores of the takeaway varieties which are possibly ordered by the user at a specific time and position are obtained, the system can perform sorting, and a plurality of takeaway varieties with the highest sorting are recommended to the user. From the perspective of the platform, the most popular takeout types under a specific time-space scene can be obtained according to statistics of the distribution of all the takeout types at a specific time and place, so that the efficiency of management services such as takeout merchant site selection or rider route planning is improved. In conclusion, the general space-time behavior prediction framework provided by the invention can be well applied to the recommendation problem in the air scene such as takeaway order prediction.
Fig. 13 is a schematic structural diagram of an electronic device according to an embodiment of the present invention, and as shown in fig. 13, the electronic device may include: a processor (processor)1310, a communication Interface (Communications Interface)1320, a memory (memory)1330 and a communication bus 1340, wherein the processor 1310, the communication Interface 1320 and the memory 1330 communicate with each other via a communication bus 840. The processor 1310 may invoke logic instructions in the memory 1330 to perform a hypergraph convolutional network-based spatiotemporal behavior prediction method comprising: determining a user ID of a time-space behavior to be predicted and a corresponding time and place; inputting the user ID of the spatio-temporal behavior to be predicted and the corresponding time and place into a spatio-temporal behavior prediction model to obtain a behavior result corresponding to the user ID output by the spatio-temporal behavior prediction model; the spatiotemporal behavior prediction model is obtained by training a hypergraph graph convolution neural network constructed based on historical spatiotemporal behavior data of a user.
In addition, the logic instructions in the memory 1330 may be implemented in software functional units and stored in a computer readable storage medium when sold or used as a stand-alone product. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.
In another aspect, an embodiment of the present invention further provides a computer program product, where the computer program product includes a computer program stored on a non-transitory computer-readable storage medium, the computer program includes program instructions, and when the program instructions are executed by a computer, the computer can execute the hypergraph convolution network-based spatiotemporal behavior prediction method provided by the above methods, and the method includes: determining a user ID of a time-space behavior to be predicted and a corresponding time and place; inputting the user ID of the spatio-temporal behavior to be predicted and the corresponding time and place into a spatio-temporal behavior prediction model to obtain a behavior result corresponding to the user ID output by the spatio-temporal behavior prediction model; the spatiotemporal behavior prediction model is obtained by training a hypergraph graph convolution neural network constructed based on historical spatiotemporal behavior data of a user.
In yet another aspect, an embodiment of the present invention further provides a non-transitory computer-readable storage medium, on which a computer program is stored, the computer program being implemented by a processor to perform the above-mentioned methods for predicting spatiotemporal behavior based on a hypergraph convolutional network, the method including: determining a user ID of a time-space behavior to be predicted and a corresponding time and place; inputting the user ID of the spatio-temporal behavior to be predicted and the corresponding time and place into a spatio-temporal behavior prediction model to obtain a behavior result corresponding to the user ID output by the spatio-temporal behavior prediction model; the spatiotemporal behavior prediction model is obtained by training a hypergraph graph convolution neural network constructed based on historical spatiotemporal behavior data of a user.
The above-described embodiments of the apparatus are merely illustrative, and the units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment. One of ordinary skill in the art can understand and implement it without inventive effort.
Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware. With this understanding in mind, the above-described technical solutions may be embodied in the form of a software product, which can be stored in a computer-readable storage medium such as ROM/RAM, magnetic disk, optical disk, etc., and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in the embodiments or some parts of the embodiments.
Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (10)
1. A spatial-temporal behavior prediction method based on a hypergraph convolutional network is characterized by comprising the following steps:
determining a user ID of a time-space behavior to be predicted and a corresponding time and place;
inputting the user ID of the spatio-temporal behavior to be predicted and the corresponding time and place into a spatio-temporal behavior prediction model to obtain a behavior result corresponding to the user ID output by the spatio-temporal behavior prediction model;
the spatiotemporal behavior prediction model is obtained by training a hypergraph graph convolution neural network constructed based on historical spatiotemporal behavior data of a user.
2. The hypergraph convolutional network-based spatiotemporal behavior prediction method of claim 1, wherein the spatiotemporal behavior prediction model is obtained by training a hypergraph convolutional neural network constructed based on historical spatiotemporal behavior data of a user, and comprises the following steps:
constructing a multi-channel user relationship hypergraph and a heterogeneous interaction hypergraph based on historical spatiotemporal behavior data of a user;
carrying out iteration on the multi-channel user relationship hypergraph and the heterogeneous interaction hypergraph for L times of hypergraph convolution operation and propagation to obtain L groups of different user, time, place and behavior embedded vectors;
obtaining an embedding vector for prediction based on the L groups of different user, time, place and behavior embedding vectors, and obtaining a corresponding prediction score based on the embedding vector for prediction; wherein, L is the number of hypergraph convolution layers;
and training the graph convolution neural network to obtain the space-time behavior prediction model based on a supervised learning method for constructing an objective function based on the prediction score.
3. The hypergraph convolutional network-based spatiotemporal behavior prediction method of claim 2, wherein the construction of the multi-channel user relationship hypergraph and the heterogeneous interaction hypergraph based on historical spatiotemporal behavior data of the user comprises:
after discretizing historical spatio-temporal behavior data of a user into quadruple data, acquiring an interaction matrix containing user preference information, which specifically comprises the following steps: respectively carrying out discretization processing of user ID, discretization processing of time division, discretization processing of region division or clustering of longitude and latitude coordinates of a place and discretization processing of keyword or theme clustering of behaviors on the historical time-space behavior data of the user; the interaction matrix containing the user preference information comprises an interaction matrix of a user and a place, an interaction matrix of the user and time and an interaction matrix of the user and a behavior;
and constructing a multi-channel user relationship hypergraph comprising a separation channel, a local channel and a global channel and a heterogeneous interaction hypergraph taking a user as a center based on the quadruple data and the interaction matrix containing the user preference information.
4. The hypergraph convolutional network-based spatiotemporal behavior prediction method of claim 3, wherein the hypergraph convolution operation and propagation are performed on the multi-channel user relationship hypergraph and heterogeneous interaction hypergraph iteration for L times to obtain L groups of different user, time, place and behavior embedding vectors, and the method comprises the following steps:
iterating the multi-channel user relationship hypergraph in the separation channel, the local channel and the global channel for L times of hypergraph convolution operation and propagation to obtain user embedded features of each channel, and aggregating the user embedded features of each channel by adopting an attention mechanism to obtain L groups of user, time, place and behavior embedded vectors fused with the user embedded features;
and carrying out L times of hypergraph convolution operation and propagation on the heterogeneous interaction hypergraph iteration to obtain L groups of user, time, place and behavior embedded vectors including node characteristics and hyperedge characteristics.
5. The hypergraph convolutional network-based spatiotemporal behavior prediction method of claim 4, wherein the formula for propagation of the multi-channel user relationship hypergraph is represented as:
wherein the content of the first and second substances,
is a user-embedded feature of each channel, DcAnd ΔcDegree matrices, H, of hypergraph nodes and hyperedges, respectivelycIs a relationship matrix of the hypergraph,
the characteristics corresponding to the three channels are obtained by respectively carrying out hypergraph convolution operation on each channel.
6. The hypergraph convolutional network-based spatiotemporal behavior prediction method of claim 4, characterized in that the formula for propagating the heterogeneous interaction hypergraph is expressed as:
wherein AGG is an aggregation function,
is a user-embedded feature obtained in a multi-channel user relationship hypergraph,
is the corresponding super edge, epsilon, of the userl,εt,εaRespectively, the connection location l, the time t and the set of all the super edges of the corresponding nodes of the behavior a.
7. The hypergraph convolutional network-based spatiotemporal behavior prediction method of claim 2, wherein the embedding vectors for prediction are obtained based on the L groups of different user, time, place and behavior embedding vectors, and the formula is as follows:
wherein the content of the first and second substances,
respectively being a super-map convolution layer
Embedding characteristics corresponding to the user u, the place l, the time t and the behavior a of the layer;
w, b are the learnable weight matrix and bias vector, respectively, and q is the attention vector.
8. A hypergraph convolutional network-based spatiotemporal behavior prediction system, comprising:
the user information determining unit is used for determining a user ID of the time-space behavior to be predicted and the corresponding time and place;
the space-time behavior prediction unit is used for inputting the user ID of the space-time behavior to be predicted and the corresponding time and place into a space-time behavior prediction model to obtain a behavior result corresponding to the user ID output by the space-time behavior prediction model;
the spatiotemporal behavior prediction model is obtained by training a hypergraph graph convolution neural network constructed based on historical spatiotemporal behavior data of a user.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor when executing the program implements the steps of the hypergraph convolutional network-based spatiotemporal behavior prediction method of any of claims 1 to 7.
10. A non-transitory computer readable storage medium having stored thereon a computer program, which when executed by a processor, performs the steps of the hypergraph convolutional network-based spatiotemporal behavior prediction method of any of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110706207.0A CN113610265A (en) | 2021-06-24 | 2021-06-24 | Hypergraph convolutional network-based time-space behavior prediction method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110706207.0A CN113610265A (en) | 2021-06-24 | 2021-06-24 | Hypergraph convolutional network-based time-space behavior prediction method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113610265A true CN113610265A (en) | 2021-11-05 |
Family
ID=78303677
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110706207.0A Pending CN113610265A (en) | 2021-06-24 | 2021-06-24 | Hypergraph convolutional network-based time-space behavior prediction method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113610265A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113762425A (en) * | 2021-11-09 | 2021-12-07 | 天津创启投资有限公司 | Data screen display sliding door control system based on 5G intelligent space |
| CN114036400A (en) * | 2021-12-07 | 2022-02-11 | 中国计量大学 | Hypergraph-based collaborative session recommendation method |
| CN114944053A (en) * | 2022-03-16 | 2022-08-26 | 浙江工业大学 | Traffic flow prediction method based on spatio-temporal hypergraph neural network |
| CN115333957A (en) * | 2022-08-05 | 2022-11-11 | 国家电网有限公司信息通信分公司 | Service flow prediction method and system based on user behaviors and enterprise service characteristics |
| CN116192662A (en) * | 2023-05-04 | 2023-05-30 | 中国电信股份有限公司四川分公司 | Service behavior prediction and deterministic network association model based and recommendation method |
| CN116775988A (en) * | 2023-05-17 | 2023-09-19 | 广州大学 | POI recommendation method and device based on hypergraph embedding |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111860951A (en) * | 2020-06-12 | 2020-10-30 | 北京工业大学 | Rail transit passenger flow prediction method based on dynamic hypergraph convolutional network |
| CN112417219A (en) * | 2020-11-16 | 2021-02-26 | 吉林大学 | Hyper-graph convolution-based hyper-edge link prediction method |
-
2021
- 2021-06-24 CN CN202110706207.0A patent/CN113610265A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111860951A (en) * | 2020-06-12 | 2020-10-30 | 北京工业大学 | Rail transit passenger flow prediction method based on dynamic hypergraph convolutional network |
| CN112417219A (en) * | 2020-11-16 | 2021-02-26 | 吉林大学 | Hyper-graph convolution-based hyper-edge link prediction method |
Non-Patent Citations (2)
| Title |
|---|
| JINGCHENG WANG等: "Metro Passenger Flow Prediction via Dynamic Hypergraph Convolution Networks", 《IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS》, 21 April 2021 (2021-04-21) * |
| YINFENG LI等: "Spatiotemporal-aware Session-based Recommendation with Graph Neural Networks", 《ACM》, 17 October 2022 (2022-10-17) * |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113762425A (en) * | 2021-11-09 | 2021-12-07 | 天津创启投资有限公司 | Data screen display sliding door control system based on 5G intelligent space |
| CN113762425B (en) * | 2021-11-09 | 2022-04-05 | 天津创启投资有限公司 | Data screen display sliding door control system based on 5G intelligent space |
| CN114036400A (en) * | 2021-12-07 | 2022-02-11 | 中国计量大学 | Hypergraph-based collaborative session recommendation method |
| CN114036400B (en) * | 2021-12-07 | 2022-04-08 | 中国计量大学 | Hypergraph-based collaborative session recommendation method |
| CN114944053A (en) * | 2022-03-16 | 2022-08-26 | 浙江工业大学 | Traffic flow prediction method based on spatio-temporal hypergraph neural network |
| CN115333957A (en) * | 2022-08-05 | 2022-11-11 | 国家电网有限公司信息通信分公司 | Service flow prediction method and system based on user behaviors and enterprise service characteristics |
| CN115333957B (en) * | 2022-08-05 | 2023-09-05 | 国家电网有限公司信息通信分公司 | Service flow prediction method and system based on user behavior and enterprise service characteristics |
| CN116192662A (en) * | 2023-05-04 | 2023-05-30 | 中国电信股份有限公司四川分公司 | Service behavior prediction and deterministic network association model based and recommendation method |
| CN116192662B (en) * | 2023-05-04 | 2023-06-23 | 中国电信股份有限公司四川分公司 | Service behavior prediction and deterministic network association model based and recommendation method |
| CN116775988A (en) * | 2023-05-17 | 2023-09-19 | 广州大学 | POI recommendation method and device based on hypergraph embedding |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113610265A (en) | Hypergraph convolutional network-based time-space behavior prediction method and system | |
| US11238065B1 (en) | Systems and methods for generating and implementing knowledge graphs for knowledge representation and analysis | |
| WO2021203819A1 (en) | Content recommendation method and apparatus, electronic device, and storage medium | |
| CN109992710A (en) | Clicking rate predictor method, system, medium and calculating equipment | |
| WO2021196639A1 (en) | Message pushing method and apparatus, and computer device and storage medium | |
| CN106776928B (en) | Position recommendation method based on memory computing framework and fusing social contact and space-time data | |
| Lu et al. | GLR: A graph-based latent representation model for successive POI recommendation | |
| CN113722611A (en) | Method, device and equipment for recommending government affair service and computer readable storage medium | |
| Nannan et al. | Adaptive online time series prediction based on a novel dynamic fuzzy cognitive map | |
| Sandeep Kumar et al. | Location identification for real estate investment using data analytics | |
| Ma et al. | Exploring multiple spatio-temporal information for point-of-interest recommendation | |
| CN114511387A (en) | Product recommendation method and device, electronic equipment and storage medium | |
| Wu et al. | A federated deep learning framework for privacy-preserving consumer electronics recommendations | |
| CN116628345B (en) | Content recommendation method and device, electronic equipment and storage medium | |
| CN116821519A (en) | Intelligent recommendation method for system filtering and noise reduction based on graph structure | |
| CN116955788A (en) | Method, device, equipment, storage medium and program product for processing content | |
| CN115618079A (en) | Session recommendation method, device, electronic equipment and storage medium | |
| CN111428741B (en) | Network community discovery method and device, electronic equipment and readable storage medium | |
| CN115482019A (en) | Activity attention prediction method and device, electronic equipment and storage medium | |
| CN115455276A (en) | Method and device for recommending object, computer equipment and storage medium | |
| CN115705384A (en) | Decoupling recommendation method, system, terminal and medium based on knowledge graph fusion | |
| CN114596108A (en) | Object recommendation method and device, electronic equipment and storage medium | |
| Doan et al. | Pacela: A neural framework for user visitation in location-based social networks | |
| Black et al. | Clustering spatio-temporal bi-partite graphs for finding crowdsourcing communities in IoMT networks | |
| Zhang | [Retracted] Design and Implementation of Sports Industry Information Service Management System Based on Data Mining |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| EE01 | Entry into force of recordation of patent licensing contract |
Application publication date: 20211105 Assignee: INSTITUTE FOR ELECTRONICS AND INFORMATION TECHNOLOGY IN TIANJIN, TSINGHUA University Assignor: TSINGHUA University Contract record no.: X2023110000009 Denomination of invention: A spatiotemporal behavior prediction method and system based on hypergraph convolution network License type: Common License Record date: 20230112 |
|
| EE01 | Entry into force of recordation of patent licensing contract |