CN111695719A - User value prediction method and system - Google Patents
User value prediction method and system Download PDFInfo
- Publication number
- CN111695719A CN111695719A CN202010313398.XA CN202010313398A CN111695719A CN 111695719 A CN111695719 A CN 111695719A CN 202010313398 A CN202010313398 A CN 202010313398A CN 111695719 A CN111695719 A CN 111695719A
- Authority
- CN
- China
- Prior art keywords
- user
- information
- graph
- network
- user value
- 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
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
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- 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
- G06F16/9536—Search customisation based on social or collaborative filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Business, Economics & Management (AREA)
- Evolutionary Computation (AREA)
- Human Resources & Organizations (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Strategic Management (AREA)
- Health & Medical Sciences (AREA)
- Economics (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Development Economics (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Game Theory and Decision Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Entrepreneurship & Innovation (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The embodiment of the invention provides a user value prediction method and a user value prediction system, wherein the method comprises the following steps: obtaining portrait information, transaction history information and social interaction information of each user, inputting the portrait information and the transaction history information into a multilayer neural network, and constructing a characterization vector of each user; constructing a social network among users according to the social interaction information and the characterization vectors, and constructing a corresponding master network according to the social network; respectively inputting the social network and the master network into a graph neural network to obtain a plurality of graph characteristic vectors, and fusing the plurality of graph characteristic vectors through an attention mechanism to obtain a graph characteristic target vector; and inputting the graph representation target vector into the trained user value prediction model to obtain a predicted value of the user value. According to the embodiment of the invention, the information such as the portrait, the transaction history and the like of the user is fully utilized, and the hidden information in the social relationship among the users is utilized, so that the user value prediction result is more accurate.
Description
Technical Field
The invention relates to the technical field of user value prediction, in particular to a user value prediction method and a user value prediction system.
Background
User value prediction (Customer value prediction) refers to estimation and prediction of the amount of money consumed by a user on a shop or a platform in a future period of time, and a typical application scenario is that a shop or a platform predicts the future purchasing ability of the user according to data such as portrait of the user, historical transactions and the like, so that different user relationship management (Customer relationship management) strategies are adopted for users with different purchasing abilities to achieve a better marketing effect. In the existing practical production application, an RFM Model (Recency Frequency Monetary Model) is often used for evaluating the user value, and the Model provides three important indexes for measuring the user value: last consumption, frequency of consumption, and amount of consumption. However, the weights among the three indexes need to be adjusted by experts according to different industries, and meanwhile, fine-grained prediction of user value cannot be performed only by using the three indexes.
Currently, besides the RFM model in traditional management, there is a method of using machine learning algorithm to predict the user value. However, the existing machine learning algorithms are based on that experts manually extract a large number of relevant features, and then the features are input into machine learning models such as random forests and the like to carry out regression on user values. The existing method focuses more on the behavior characteristics of the user, and mutual influence between the user and the user is not sufficiently considered, so that the existing user value prediction accuracy is low.
Therefore, a method and a system for predicting user value are needed to solve the above problems.
Disclosure of Invention
Aiming at the problems in the prior art, the embodiment of the invention provides a user value prediction method and a user value prediction system.
In a first aspect, an embodiment of the present invention provides a user value prediction method, including:
obtaining portrait information, transaction history information and social interaction information of each user, inputting the portrait information and the transaction history information into a multilayer neural network, and constructing a characterization vector of each user;
constructing a social network among users according to the social interaction information and the characterization vectors, and constructing a corresponding master network according to the social network;
respectively inputting the social network and the master mask network into a graph neural network to obtain a plurality of graph characteristic vectors, and fusing the plurality of graph characteristic vectors through an attention mechanism to obtain a graph characteristic target vector;
inputting the chart characteristic target vector into a trained user value prediction model to obtain a predicted value of the user value; the trained user value prediction model is constructed through a multilayer neural network and is obtained by training sample portrait information, sample transaction history information and sample social interaction information.
Further, the obtaining of portrait information, transaction history information and social interaction information of each user, inputting the portrait information and the transaction history information into a multilayer neural network, and constructing a characterization vector of each user includes:
respectively carrying out feature screening on the portrait information, the transaction history information and the social interaction information of each user to obtain a user portrait information dimension, a transaction history information dimension and a social interaction information dimension;
coding the user portrait information dimension and the transaction history information dimension respectively, and splicing the coded user portrait information dimension and the transaction history information dimension to obtain a spliced information dimension;
and coding the spliced information dimensions, and inputting the information dimensions into a multilayer neural network to obtain the characterization vector of each user.
Further, the constructing a social network between users according to the social interaction information and the characterization vector, and constructing a corresponding master network according to the social network includes:
and according to the social interaction information and the characterization vectors, constructing a social network among the users, and according to a plurality of preset mother boards, extracting a mother board network corresponding to each preset mother board from the social network.
Further, characterized in that the graph characterizes target vectors as:
wherein the content of the first and second substances,
representing a graph feature vector output by the ith node in the nth graph from the neural network of the (l + 1) th layer graph, ai,nIndicating the attention parameter of the ith node to the nth graph,
the graph representing the ith node characterizes the target vector.
Further, after predicting the user value of each user based on the trained user value prediction model and the graph characteristic target vector to obtain a predicted value of the user value, the method further includes:
and obtaining a real value of the user value, comparing the real value with the predicted value of the user value, and iteratively updating the weight in the user value prediction model through a back propagation algorithm to obtain an updated user value prediction model.
Further, a Dropout layer is arranged in the multilayer neural network, and a regular term is arranged in a loss function of the user value prediction model.
Further, before the inputting the graph characteristic target vector into the trained user value prediction model to obtain a predicted value of the user value, the method further includes:
and carrying out model evaluation on the user value prediction model through the average absolute error, the root mean square error, the normalized average absolute error and the normalized mean square error.
In a second aspect, an embodiment of the present invention provides a user value prediction system, including:
the acquisition module is used for acquiring portrait information, transaction history information and social interaction information of each user, inputting the portrait information and the transaction history information into a multilayer neural network, and constructing a characterization vector of each user;
the social network construction module is used for constructing a social network among users according to the social interaction information and the characterization vectors, and constructing a corresponding master network according to the social network;
the graph representation learning module is used for respectively inputting the social network and the master network into a graph neural network to obtain a plurality of graph representation vectors, and fusing the plurality of graph representation vectors through an attention mechanism to obtain a graph representation target vector;
the user value prediction module is used for inputting the chart characteristic target vector into a trained user value prediction model to obtain a predicted value of the user value; the trained user value prediction model is constructed through a multilayer neural network and is obtained by training sample portrait information, sample transaction history information and sample social interaction information.
In a third aspect, an embodiment of the present invention provides an electronic device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the steps of the method provided in the first aspect when executing the program.
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, which when executed by a processor, implements the steps of the method as provided in the first aspect.
The user value prediction method and the user value prediction system provided by the embodiment of the invention fully utilize the self portrait, transaction history and other information of the users, and better utilize the hidden information in the social relationship among the users, so that the user value prediction result is more accurate, and better prediction performance is achieved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in 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 user value prediction method according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of user characterization learning according to an embodiment of the present invention;
fig. 3 is a schematic diagram of a directional master formed by three nodes according to an embodiment of the present invention;
fig. 4 is a schematic diagram of a non-directional master composed of four nodes according to an embodiment of the present invention;
fig. 5 is an extraction schematic diagram of a master network according to an embodiment of the present invention;
FIG. 6 is a schematic diagram illustrating the calculation of graph characterization information according to an embodiment of the present invention;
FIG. 7 is a schematic structural diagram of a user value prediction system according to an embodiment of the present invention;
fig. 8 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, 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 existing user value prediction method only considers the information of the personal dimension of the user from the aspect of feature extraction, does not well utilize the social interaction information between the user and the user, and the performance of a model depends on the manually extracted features seriously. Meanwhile, the existing method uses machine learning models with relatively basic functions such as random forests and the like, so that the learning capability is very limited.
Fig. 1 is a schematic flow chart of a user value prediction method provided in an embodiment of the present invention, and as shown in fig. 1, the embodiment of the present invention provides a user value prediction method, including:
In the embodiment of the invention, the data of portraits, transactions, social network relations and the like of the user in the previous quarter are obtained, so that the user value of the user in the next quarter is predicted. The existing user value is roughly defined as the following two: the first is the total amount the user spends at the store or platform; the second is the net profit earned by the store or platform on the user. According to the embodiment of the invention, under the E-commerce platform, the commodity cost price is greatly influenced by factors such as manufacturers, logistics and seasons, and the total amount spent by the user on the shop or the platform is predicted by mining the consumption potential of the user. Further, a characterization vector is learned for each user based on the portrait information and transaction history information for each user.
Specifically, in an embodiment of the present invention, the portrait information of the user includes: registration time, user type, gender, age, constellation, age group, mobile phone model, frequent region, frequent city level, number of common shipping addresses, and the like. The user type, gender, age, constellation, age group, mobile phone model, frequently-located region and frequently-located city level are all category variables, and the category variables need to be subjected to one-hot coding. For the missing value, 0-value padding, mean padding, special-value padding, and the like may be selected according to the characteristics, for example, 2 (male: 0, female: 1) for the missing value in gender; for an abnormal value, it is refilled, for example, if the age exceeds 100 years, the item is regarded as a missing value, and the filling is performed according to the processing method of the missing value.
Further, the user historical transaction information includes: the user's historical total spending amount, last spending amount, average spending amount, total amount of orders, number of days from last spending, average spending interval number of days, historical total return amount, last return amount, number of days from last return, spending amount of last year/quarter/month and spending number of last year/quarter/month. In general, the user's historical transaction data includes both consumption and return portions, as well as coarse-grained (year) and fine-grained (month) related features over time. The feature of this dimension delicately characterizes the indicators in the RFM model from various aspects.
Further, the social interaction information among the users includes: and the user can edit the order, cut price and like information. In the embodiment of the invention, the social network among the users is reproduced by using the social interaction information among the users, so that different interaction types are not distinguished, and as long as certain social interaction exists between the two users, the two users are considered to have social connection. The social interaction behavior is specifically represented in that in each social interaction behavior, an initiator of the social interaction behavior and a follower of the social interaction behavior exist, for example, a user initiates a praise assistance request, namely the initiator; the user likes the assistance for others, namely the follower. Thus, the social relationships in embodiments of the invention are directed, i.e., the edges are directed by the initiator of the behavior to the follower of the behavior.
Preferably, in the embodiment of the present invention, in order to improve the accuracy of subsequent prediction, data with more dimensions may be added, so as to further increase feature information, including behavior data of a user in an application, information of a commodity purchased by the user, and the like.
And 102, constructing a social network among users according to the social interaction information and the characterization vectors, and constructing a corresponding master network according to the social network.
In the embodiment of the invention, the social network among the users is constructed based on the social relationship among the users, in the social network, the users are nodes, the characterization vectors corresponding to the users are information of the nodes, the social relationship among the users is edges among the nodes, and if the social relationship exists among the users, the edges exist among the nodes of the users. Further, based on a preset master, a master network corresponding to the preset master is extracted from the user social network.
Step 103, inputting the social network and the master network into a graph neural network respectively to obtain a plurality of graph feature vectors, and fusing the plurality of graph feature vectors through an attention mechanism to obtain a graph feature target vector.
In the embodiment of the invention, the social network and the master network obtained in the embodiment are respectively subjected to graph convolution operation through the graph neural network to obtain user graph representation information of a plurality of graphs, namely graph representation vectors, and finally, the plurality of graph representation information are fused through an attention mechanism to obtain the graph representation target vectors.
In an embodiment of the present invention, the feature of the user's current quarter is represented as (x)1,x2,x3,…,xn) And the user value predicted in the following quarter is y, and the relationship between the two is satisfied:
y＝F(x1,x2,x3,…,xn)；
further, what the user value prediction model needs to learn is a mapping function F (-) and in the model training process, model hyper-parameters (including embedding dimensions and regularization coefficient λ) need to be setθLearning rate β, etc.) in the process of training the network, the weight and bias value of each layer of the network can be updated by a Stochastic Gradient Descent method (Stochastic Gradient decision) in the process of back propagation.
The user value prediction method provided by the embodiment of the invention fully utilizes the self portrait, transaction history and other information of the user, and better utilizes the hidden information in the social relationship among the users, so that the user value prediction result is more accurate, and better prediction performance is achieved.
On the basis of the above embodiment, the obtaining portrait information, transaction history information, and social interaction information of each user, and inputting the portrait information and the transaction history information into a multilayer neural network, and constructing a characterization vector of each user includes:
respectively carrying out feature screening on the portrait information, the transaction history information and the social interaction information of each user to obtain a user portrait information dimension, a transaction history information dimension and a social interaction information dimension;
coding the user portrait information dimension and the transaction history information dimension respectively, and splicing the coded user portrait information dimension and the transaction history information dimension to obtain a spliced information dimension;
and coding the spliced information dimensions, and inputting the information dimensions into a multilayer neural network to obtain the characterization vector of each user.
In the embodiment of the invention, required information extracted from the user portrait, the transaction history information and the social interaction information in the last quarter stored in the e-commerce platform is supplemented to missing values, error values are corrected and screened out, data related to user privacy are anonymized, and in the embodiment of the invention, fields related to the user privacy are subjected to hash encryption. Next, the above information data is divided into a training set, a verification set, and a test set according to natural quarters. For each data set, the characteristics come from the portrait of the quarterly user, transaction history information, social interaction information and the like; the label, i.e., the forecast value, represents the total amount spent by the user for the next quarter.
Further, the category values in the user's portrait information and transaction history information are encoded as unique heat vectors. In the embodiment of the invention, the characteristics from the user portrait information and the transaction historical information are respectively subjected to characterization learning, then the characterization information of the user portrait dimension and the characterization information of the user historical transaction information dimension are spliced together for further characterization learning, and thus the final characterization vector of the user is obtained.
Fig. 2 is a schematic diagram of user characterization learning provided by an embodiment of the present invention, which can be referred to in fig. 2, in the embodiment of the present invention, the user characterization learning is to learn a characterization vector for each user through portrait information and historical purchase information (user transaction information) of the user. The user representation learning can be specifically divided into two steps, firstly, the user portrait information dimension and the user historical transaction information are respectively coded, then, the output of the two dimensions is spliced together for further coding, and the purpose is to learn the cross information of the two dimension characteristics. In the embodiment of the invention, when the user characterization learning is performed, the adjustable parameters include the output dimensions of three embedding layers, and the sizes of the output dimensions of the first two embedding layers need to consider both the size of the original input feature and the ratio of the sizes of the two output dimensions. The input dimension of the last embedding layer is equal to the sum of the output dimensions of the first two embedding layers, the output dimension is an adjustable parameter, and the output information is the information of the user node. It should be noted that, in the embodiment of the present invention, information of other dimensions may be added for supplementation, so as to further improve the user node information.
On the basis of the above embodiment, the constructing a social network between users according to the social interaction information and the characterization vector, and constructing a corresponding master network according to the social network includes:
and according to the social interaction information and the characterization vectors, constructing a social network among the users, and according to a plurality of preset mother boards, extracting a mother board network corresponding to each preset mother board from the social network.
In the embodiment of the invention, a social network among users is constructed based on social interaction information among the users, namely, the users are nodes, the characterization vectors corresponding to the users are information of the nodes, the social relations among the users are edges among the nodes, and if the social relations exist among the users, the edges exist among the nodes of the users. Further, based on multiple different preset masters, a master network corresponding to each preset master is extracted from the user social network, so that a series of networks with the same node information and different node connection relations are obtained.
Specifically, in the embodiment of the present invention, a directed social network (original social network) between users may be constructed according to the directed social relationship between users obtained in the foregoing embodiment. In the social network, the node is a user, the node information is the user node information obtained in the above embodiment, and the edge is constructed according to the directed social relationship among the users. If there is a social relationship between users, then there is a corresponding edge between the user nodes. For the weight of an edge, there are three construction methods: the first is that all edges are weighted by 1; the second is that the weight of the edge is the sum of the social relationship number among the users; and thirdly, learning a weight for each side according to the interaction information among the users. The embodiment of the invention adopts a first method, namely, the adjacent matrix corresponding to the user social network only has 0 and 1, and if the social relationship exists between the users, the corresponding element in the adjacent matrix is 1, otherwise, the corresponding element is 0. In the embodiment of the invention, the expansion can be performed by combining with the actual situation, specifically, the weight of the social side is set according to the strength of the social interaction among the users; and learning the weight of the social side according to the historical interactive behaviors among the users.
Further, in the embodiment of the present invention, the master network is a network structure between the nodes and the complex network. Fig. 3 is a schematic diagram of a directional master formed by three nodes according to an embodiment of the present invention, and as shown in fig. 3, the directional master formed by three nodes includes 13 types, and fig. 4 is a schematic diagram of a non-directional master formed by four nodes according to an embodiment of the present invention, and as shown in fig. 4, the non-directional master formed by four nodes includes 6 types. In the embodiment of the invention, the network based on the master processing is more beneficial to the subsequent image convolution neural network to capture some high-order information. Specifically, in the embodiment of the present invention, the method for constructing the master network includes: if the connection relationship between any three nodes in the network conforms to the structure of the preset master, the three nodes are considered to be connected in the master network corresponding to the preset master, fig. 5 is a schematic diagram of extraction of the master network provided in the embodiment of the present invention, and the extraction process can refer to fig. 5. In the embodiment of the present invention, a directional master formed by three nodes shown in fig. 3 is used for description, and preferably, in the embodiment of the present invention, after an original network (i.e., a social network) is processed by a plurality of different preset masters, a plurality of networks can be obtained. These networks have the same nodes and node information, but the edges are not identical. Preferably, in the embodiment of the present invention, a non-directional master composed of four nodes in fig. 4 may be added in combination with the actual situation, so as to further obtain a higher-order network structure.
On the basis of the above embodiment, the graph characterizes the target vector as:
wherein the content of the first and second substances,
representing a graph feature vector output by the ith node in the nth graph from the neural network of the (l + 1) th layer graph, ai,nIndicating the attention parameter of the ith node to the nth graph,
the graph representing the ith node characterizes the target vector.
In the embodiment of the invention, graph characteristic learning refers to that a node aggregates meaningful information from the neighbors of the node according to a network structure to assist the node in predicting. Specifically, assume that the input graph is the nth graph, and the characteristic of the node i is represented by Xi,nAnd the neighbor node set of the node i is Ni,nThen the graph characterization learning is expressed as:
wherein l represents the number of layers of the neural network of the graph, and F (·) represents an aggregation function; σ (-) denotes a non-linear function, such as a ReLU function; j denotes a neighbor node which is a node of the neighbor,
a graph feature vector representing the ith node of the l +1 th layer output. In an embodiment of the invention, the debuggable parameters include the dimension of the output layer of the graph neural network, the number of layers of the graph neural network, an aggregation function and a nonlinear activation function. It should be noted that the neural network operator used in the embodiment of the present invention is:
the neural network operator can be selected according to actual needs, and the embodiment of the present invention is not limited to this.
Fig. 6 is a schematic diagram of calculating graph characterization information according to an embodiment of the present invention, and as shown in fig. 6, N graphs obtained by the graph characterization learning method according to the embodiment are respectively input to a graph neural network to obtain N graph characterization vectors, and the N graph characterization vectors are fused together by an attention mechanism to obtain a graph characterization target vector:
wherein, ai,nThe attention parameter of the ith node to the nth graph is represented by the following calculation formula:
w denotes a randomly initialized adjustable parameter.
On the basis of the above embodiment, after predicting the user value of each user based on the trained user value prediction model and the graph characteristic target vector to obtain a predicted value of the user value, the method further includes:
and obtaining a real value of the user value, comparing the real value with the predicted value of the user value, and iteratively updating the weight in the user value prediction model through a back propagation algorithm to obtain an updated user value prediction model.
In the embodiment of the invention, a predicted value of the user value is obtained, and the formula of the predicted value is as follows:
in the embodiment of the invention, in order to learn the mapping function G (-), a multi-layer neural network is used for fitting the function, and the adjustable hyper-parameters comprise the depth of the neural network and the number of neurons of each layer of the neural network.
Further, as for the loss function of the model, a mean absolute error, a mean square error, or the like may be selected, wherein which loss function is selected may depend on the convergence of the model, the evaluation criterion, and the distribution of the prediction values. The embodiment of the invention adopts the average absolute error, namely:
wherein the content of the first and second substances,
indicates the predicted value of node i, yiRepresenting the true value of node i and N representing the number of samples.
On the basis of the embodiment, a Dropout layer is arranged in the multilayer neural network, and a regular term is arranged in a loss function of the user value prediction model, so that overfitting of the model is suppressed.
In the embodiment of the present invention, the loss function of the user value prediction model may select a Mean Absolute Error (MAE) or a Mean Square Error (MSE):
wherein the content of the first and second substances,
indicates the predicted value of node i, yiRepresenting the true value of node i and N representing the total number of samples. In the embodiment of the present invention, the average absolute error is used, and in practical applications, the loss function may be selected according to actual conditions of data and training, which is not limited in the embodiment of the present invention. In addition, during the model training process, the hyper-parameters to be adjusted include a learning rate, a batch size, a regularization term and a Dropout rate.
On the basis of the above embodiment, before the inputting the graph characteristic target vector into the trained user value prediction model to obtain the predicted value of the user value, the method further includes:
and carrying out model evaluation on the user value prediction model through the average absolute error, the root mean square error, the normalized average absolute error and the normalized mean square error.
Further, in the embodiment of the present invention, for the evaluation of the model, the Mean Absolute Error (MAE), the Root Mean Square Error (RMSE), the Normalized Mean Absolute Error (NMAE), and the normalized mean square error (NRMSE) may be selected to evaluate the model, and the calculation formula is as follows:
fig. 7 is a schematic structural diagram of a user value prediction system according to an embodiment of the present invention, and as shown in fig. 7, the user value prediction system according to an embodiment of the present invention includes an obtaining module 701, a social network constructing module 702, a graph characterization learning module 703 and a user value prediction module 704, where the obtaining module 701 is configured to obtain portrait information, transaction history information and social interaction information of each user, and input the portrait information and the transaction history information into a multilayer neural network to construct a characterization vector of each user; the social network construction module 702 is configured to construct a social network between users according to the social interaction information and the characterization vector, and construct a corresponding master network according to the social network; the graph representation learning module 703 is configured to input the social network and the master network into a graph neural network, respectively, to obtain a plurality of graph representation vectors, and fuse the plurality of graph representation vectors by an attention mechanism, to obtain a graph representation target vector; the user value prediction module 704 is used for inputting the chart characteristic target vector into a trained user value prediction model to obtain a predicted value of the user value; the trained user value prediction model is constructed through a multilayer neural network and is obtained by training sample portrait information, sample transaction history information and sample social interaction information.
The user value prediction system provided by the embodiment of the invention fully utilizes the self portrait, transaction history and other information of the user, and better utilizes the hidden information in the social relationship among the users, so that the user value prediction result is more accurate, and better prediction performance is achieved.
The system provided by the embodiment of the present invention is used for executing the above method embodiments, and for details of the process and the details, reference is made to the above embodiments, which are not described herein again.
Fig. 8 is a schematic structural diagram of an electronic device according to an embodiment of the present invention, and referring to fig. 8, the electronic device may include: a Processor (Processor)801, a communication Interface (Communications Interface)802, a Memory (Memory)803 and a communication bus 804, wherein the Processor 801, the communication Interface 802 and the Memory 803 complete communication with each other through the communication bus 804. The processor 801 may call logic instructions in the memory 803 to perform the following method: obtaining portrait information, transaction history information and social interaction information of each user, inputting the portrait information and the transaction history information into a multilayer neural network, and constructing a characterization vector of each user; constructing a social network among users according to the social interaction information and the characterization vectors, and constructing a corresponding master network according to the social network; respectively inputting the social network and the master mask network into a graph neural network to obtain a plurality of graph characteristic vectors, and fusing the plurality of graph characteristic vectors through an attention mechanism to obtain a graph characteristic target vector; inputting the chart characteristic target vector into a trained user value prediction model to obtain a predicted value of the user value; the trained user value prediction model is constructed through a multilayer neural network and is obtained by training sample portrait information, sample transaction history information and sample social interaction information.
In addition, the logic instructions in the memory 803 may be implemented in the form of software functional units and stored in a computer readable storage medium when the logic instructions are sold or used as independent products. 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 non-transitory computer-readable storage medium, on which a computer program is stored, where the computer program is implemented to, when executed by a processor, perform the user value prediction method provided in the foregoing embodiments, for example, including: obtaining portrait information, transaction history information and social interaction information of each user, inputting the portrait information and the transaction history information into a multilayer neural network, and constructing a characterization vector of each user; constructing a social network among users according to the social interaction information and the characterization vectors, and constructing a corresponding master network according to the social network; respectively inputting the social network and the master mask network into a graph neural network to obtain a plurality of graph characteristic vectors, and fusing the plurality of graph characteristic vectors through an attention mechanism to obtain a graph characteristic target vector; inputting the chart characteristic target vector into a trained user value prediction model to obtain a predicted value of the user value; the trained user value prediction model is constructed through a multilayer neural network and is obtained by training sample portrait information, sample transaction history information and sample social interaction information.
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 user value prediction method, comprising:
obtaining portrait information, transaction history information and social interaction information of each user, inputting the portrait information and the transaction history information into a multilayer neural network, and constructing a characterization vector of each user;
constructing a social network among users according to the social interaction information and the characterization vectors, and constructing a corresponding master network according to the social network;
respectively inputting the social network and the master mask network into a graph neural network to obtain a plurality of graph characteristic vectors, and fusing the plurality of graph characteristic vectors through an attention mechanism to obtain a graph characteristic target vector;
inputting the chart characteristic target vector into a trained user value prediction model to obtain a predicted value of the user value; the trained user value prediction model is constructed through a multilayer neural network and is obtained by training sample portrait information, sample transaction history information and sample social interaction information.
2. The method for predicting user value according to claim 1, wherein the obtaining of the portrait information, the transaction history information and the social interaction information of each user and the inputting of the portrait information and the transaction history information into the multilayer neural network, and the constructing of the characterization vector of each user comprises:
respectively carrying out feature screening on the portrait information, the transaction history information and the social interaction information of each user to obtain a user portrait information dimension, a transaction history information dimension and a social interaction information dimension;
coding the user portrait information dimension and the transaction history information dimension respectively, and splicing the coded user portrait information dimension and the transaction history information dimension to obtain a spliced information dimension;
and coding the spliced information dimensions, and inputting the information dimensions into a multilayer neural network to obtain the characterization vector of each user.
3. The method for predicting user value according to claim 1, wherein the constructing a social network between users according to the social interaction information and the characterization vector, and constructing a corresponding master network according to the social network comprises:
and according to the social interaction information and the characterization vectors, constructing a social network among the users, and according to a plurality of preset mother boards, extracting a mother board network corresponding to each preset mother board from the social network.
4. The user value prediction method of claim 1, wherein the graph characterizing target vectors is represented as:
wherein the content of the first and second substances,
representing a graph feature vector output by the ith node in the nth graph from the neural network of the (l + 1) th layer graph, ai,nIndicating the attention parameter of the ith node to the nth graph,
the graph representing the ith node characterizes the target vector.
5. The method of claim 1, wherein after predicting the user value of each user based on the trained user value prediction model and the graph characteristic target vector to obtain a predicted value of the user value, the method further comprises:
and obtaining a real value of the user value, comparing the real value with the predicted value of the user value, and iteratively updating the weight in the user value prediction model through a back propagation algorithm to obtain an updated user value prediction model.
6. The user value prediction method according to claim 1, wherein a Dropout layer is provided in the multi-layer neural network, and a regularization term is provided in a loss function of the user value prediction model.
7. The method of claim 1, wherein before inputting the graph characterization target vector into a trained user value prediction model to obtain a predicted value of user value, the method further comprises:
and carrying out model evaluation on the user value prediction model through the average absolute error, the root mean square error, the normalized average absolute error and the normalized mean square error.
8. A user value prediction system, comprising:
the acquisition module is used for acquiring portrait information, transaction history information and social interaction information of each user, inputting the portrait information and the transaction history information into a multilayer neural network, and constructing a characterization vector of each user;
the social network construction module is used for constructing a social network among users according to the social interaction information and the characterization vectors, and constructing a corresponding master network according to the social network;
the graph representation learning module is used for respectively inputting the social network and the master network into a graph neural network to obtain a plurality of graph representation vectors, and fusing the plurality of graph representation vectors through an attention mechanism to obtain a graph representation target vector;
the user value prediction module is used for inputting the chart characteristic target vector into a trained user value prediction model to obtain a predicted value of the user value; the trained user value prediction model is constructed through a multilayer neural network and is obtained by training sample portrait information, sample transaction history information and sample social interaction information.
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 performs the steps of the user value prediction method according to any of claims 1 to 7.
10. A non-transitory computer readable storage medium, having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the user value prediction method according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010313398.XA CN111695719A (en) | 2020-04-20 | 2020-04-20 | User value prediction method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010313398.XA CN111695719A (en) | 2020-04-20 | 2020-04-20 | User value prediction method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111695719A true CN111695719A (en) | 2020-09-22 |
Family
ID=72476367
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010313398.XA Pending CN111695719A (en) | 2020-04-20 | 2020-04-20 | User value prediction method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111695719A (en) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112035683A (en) * | 2020-09-30 | 2020-12-04 | 北京百度网讯科技有限公司 | User interaction information processing model generation method and user interaction information processing method |
| CN112184046A (en) * | 2020-10-12 | 2021-01-05 | 上海移卓网络科技有限公司 | Advertisement service user value evaluation method, device, equipment and storage medium |
| CN112364222A (en) * | 2021-01-13 | 2021-02-12 | 北京云真信科技有限公司 | Regional portrait method of user age, computer equipment and storage medium |
| CN112488355A (en) * | 2020-10-28 | 2021-03-12 | 华为技术有限公司 | Method and device for predicting user rating based on graph neural network |
| CN112861064A (en) * | 2021-01-20 | 2021-05-28 | 重庆第二师范学院 | Social credit evaluation source data processing method, system, terminal and medium |
| CN113516501A (en) * | 2021-05-13 | 2021-10-19 | 浙江鸿程计算机系统有限公司 | User communication behavior prediction method and device based on graph neural network |
| CN113538070A (en) * | 2020-10-30 | 2021-10-22 | 深圳市九九互动科技有限公司 | User life value cycle detection method and device and computer equipment |
| CN113569809A (en) * | 2021-08-27 | 2021-10-29 | 腾讯音乐娱乐科技（深圳）有限公司 | Image processing method, device and computer readable storage medium |
| CN113657945A (en) * | 2021-08-27 | 2021-11-16 | 建信基金管理有限责任公司 | User value prediction method, device, electronic equipment and computer storage medium |
| CN114513394A (en) * | 2022-01-13 | 2022-05-17 | 南京信息工程大学 | Attention machine drawing neural network-based signal modulation format identification method, system and device and storage medium |
| CN116127204A (en) * | 2023-04-17 | 2023-05-16 | 中国科学技术大学 | Multi-view user portrayal method, multi-view user portrayal system, apparatus, and medium |
| CN116453141A (en) * | 2023-06-13 | 2023-07-18 | 平安银行股份有限公司 | Identification method and device for bill latent passenger and electronic equipment |
| CN116703457A (en) * | 2022-12-16 | 2023-09-05 | 荣耀终端有限公司 | Prediction method and prediction device for life cycle value of user |
-
2020
- 2020-04-20 CN CN202010313398.XA patent/CN111695719A/en active Pending
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112035683A (en) * | 2020-09-30 | 2020-12-04 | 北京百度网讯科技有限公司 | User interaction information processing model generation method and user interaction information processing method |
| CN112184046A (en) * | 2020-10-12 | 2021-01-05 | 上海移卓网络科技有限公司 | Advertisement service user value evaluation method, device, equipment and storage medium |
| CN112488355A (en) * | 2020-10-28 | 2021-03-12 | 华为技术有限公司 | Method and device for predicting user rating based on graph neural network |
| CN113538070A (en) * | 2020-10-30 | 2021-10-22 | 深圳市九九互动科技有限公司 | User life value cycle detection method and device and computer equipment |
| CN113538070B (en) * | 2020-10-30 | 2023-12-26 | 深圳市九九互动科技有限公司 | User life value cycle detection method and device and computer equipment |
| CN112364222A (en) * | 2021-01-13 | 2021-02-12 | 北京云真信科技有限公司 | Regional portrait method of user age, computer equipment and storage medium |
| CN112364222B (en) * | 2021-01-13 | 2021-04-27 | 北京云真信科技有限公司 | Regional portrait method of user age, computer equipment and storage medium |
| CN112861064B (en) * | 2021-01-20 | 2023-02-03 | 重庆第二师范学院 | Social credit evaluation source data processing method, system, terminal and medium |
| CN112861064A (en) * | 2021-01-20 | 2021-05-28 | 重庆第二师范学院 | Social credit evaluation source data processing method, system, terminal and medium |
| CN113516501A (en) * | 2021-05-13 | 2021-10-19 | 浙江鸿程计算机系统有限公司 | User communication behavior prediction method and device based on graph neural network |
| CN113569809A (en) * | 2021-08-27 | 2021-10-29 | 腾讯音乐娱乐科技（深圳）有限公司 | Image processing method, device and computer readable storage medium |
| CN113657945A (en) * | 2021-08-27 | 2021-11-16 | 建信基金管理有限责任公司 | User value prediction method, device, electronic equipment and computer storage medium |
| CN114513394A (en) * | 2022-01-13 | 2022-05-17 | 南京信息工程大学 | Attention machine drawing neural network-based signal modulation format identification method, system and device and storage medium |
| CN114513394B (en) * | 2022-01-13 | 2023-12-05 | 南京信息工程大学 | Signal modulation format identification method, system and device based on attention mechanism diagram neural network and storage medium |
| CN116703457A (en) * | 2022-12-16 | 2023-09-05 | 荣耀终端有限公司 | Prediction method and prediction device for life cycle value of user |
| CN116127204A (en) * | 2023-04-17 | 2023-05-16 | 中国科学技术大学 | Multi-view user portrayal method, multi-view user portrayal system, apparatus, and medium |
| CN116453141A (en) * | 2023-06-13 | 2023-07-18 | 平安银行股份有限公司 | Identification method and device for bill latent passenger and electronic equipment |
| CN116453141B (en) * | 2023-06-13 | 2023-10-13 | 平安银行股份有限公司 | Identification method and device for bill latent passenger and electronic equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111695719A (en) | User value prediction method and system | |
| TWI631518B (en) | Computer server system having one or more computing devices and computer-implemented method of training and event classifier model | |
| CN111582538B (en) | Community value prediction method and system based on graph neural network | |
| CN111553759A (en) | Product information pushing method, device, equipment and storage medium | |
| CN107330741A (en) | Graded electron-like certificate uses Forecasting Methodology, device and electronic equipment | |
| CN111160968A (en) | SKU-level commodity sales prediction method and device | |
| CN112163963B (en) | Service recommendation method, device, computer equipment and storage medium | |
| Kullaya Swamy et al. | Bank transaction data modeling by optimized hybrid machine learning merged with ARIMA | |
| CN114549046A (en) | Sales prediction method, system, device and storage medium based on fusion model | |
| CN111080360A (en) | Behavior prediction method, model training method, device, server and storage medium | |
| CN110264270A (en) | A kind of behavior prediction method, apparatus, equipment and storage medium | |
| Xia et al. | Modeling consumer buying decision for recommendation based on multi-task deep learning | |
| CN114298851A (en) | Network user social behavior analysis method and device based on graph sign learning and storage medium | |
| CN111177581A (en) | Multi-platform-based social e-commerce website commodity recommendation method and device | |
| CN110009371B (en) | Abnormal behavior determination method, device, equipment and computer readable storage medium | |
| CN111461862B (en) | Method and device for determining target characteristics for service data | |
| CN115471260A (en) | Neural network-based sales prediction method, apparatus, device and medium | |
| CN114756720A (en) | Time sequence data prediction method and device | |
| CN113554099A (en) | Method and device for identifying abnormal commercial tenant | |
| CN117252665B (en) | Service recommendation method and device, electronic equipment and storage medium | |
| CN113379494B (en) | Commodity recommendation method and device based on heterogeneous social relationship and electronic equipment | |
| CN114692005B (en) | Sparse ultrashort sequence-oriented personalized recommendation method, system, medium and device | |
| CN111815396B (en) | Product screening method, system, equipment and storage medium based on metagraph | |
| CN115511582B (en) | Commodity recommendation system and method based on artificial intelligence | |
| CN117009883B (en) | Object classification model construction method, object classification method, device and equipment |
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 |