WO2021115269A1 - 用户集群的预测方法、装置、计算机设备和存储介质 - Google Patents

用户集群的预测方法、装置、计算机设备和存储介质 Download PDF

Info

Publication number
WO2021115269A1
WO2021115269A1 PCT/CN2020/134579 CN2020134579W WO2021115269A1 WO 2021115269 A1 WO2021115269 A1 WO 2021115269A1 CN 2020134579 W CN2020134579 W CN 2020134579W WO 2021115269 A1 WO2021115269 A1 WO 2021115269A1
Authority
WO
WIPO (PCT)
Prior art keywords
user
prediction model
feature
cluster
prediction
Prior art date
Application number
PCT/CN2020/134579
Other languages
English (en)
French (fr)
Inventor
周学立
朱恩东
张茜
凌海挺
Original Assignee
平安科技（深圳）有限公司
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 平安科技（深圳）有限公司 filed Critical 平安科技（深圳）有限公司
Publication of WO2021115269A1 publication Critical patent/WO2021115269A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24—Querying
    • G06F16/245—Query processing
    • G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
    • G06F16/2465—Query processing support for facilitating data mining operations in structured databases
    • 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
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • 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
    • 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
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0201—Market modelling; Market analysis; Collecting market data

Definitions

  • This application relates to the technical field of intelligent decision-making, in particular to a method, device, computer equipment, and storage medium for predicting user clusters.
  • machine learning is used in more and more fields to effectively analyze data in the corresponding field, such as user cluster prediction.
  • current prediction methods for user clusters generally obtain user information in a single dimension of a user, such as historical business operation information, and input the user information in the single dimension into a machine learning model to identify the user through the machine learning model. Whether a user is a target user; and so on, a machine learning model can be used to predict a user cluster that meets the conditions; however, the inventor realizes that judging whether a user is a target user is often affected by multiple factors, and only through a machine learning model Analyzing the user information of a user in a single dimension can easily cause the accuracy of the predicted user cluster to be low.
  • a method, apparatus, computer equipment, and storage medium for predicting a user cluster are provided.
  • a method for predicting user clusters includes:
  • the feature code includes user information feature codes of multiple user feature dimensions
  • Fusion processing is performed on the predicted user clusters output by each of the user prediction models to obtain the target user cluster corresponding to the initial user cluster.
  • a prediction device for user clusters includes:
  • the feature code acquisition module is used to obtain the feature code of each user in the initial user cluster; the feature code includes user information feature codes of multiple user feature dimensions;
  • the predictive user cluster acquisition module is configured to respectively input the user information feature codes of the multiple user feature dimensions of each user into the corresponding pre-trained user prediction model to obtain the predicted user cluster output by each of the user prediction models;
  • the target user cluster acquisition module is configured to perform fusion processing on the predicted user clusters output by each of the user prediction models to obtain the target user cluster corresponding to the initial user cluster.
  • a computer device including a memory and one or more processors, the memory stores computer readable instructions, and when the computer readable instructions are executed by the processor, the one or more processors execute The following steps:
  • the feature code includes user information feature codes of multiple user feature dimensions
  • Fusion processing is performed on the predicted user clusters output by each of the user prediction models to obtain the target user cluster corresponding to the initial user cluster.
  • One or more computer-readable storage media storing computer-readable instructions.
  • the one or more processors perform the following steps:
  • the feature code includes user information feature codes of multiple user feature dimensions
  • Fusion processing is performed on the predicted user clusters output by each of the user prediction models to obtain the target user cluster corresponding to the initial user cluster.
  • the prediction method, device, computer equipment and storage medium of the user cluster described above are obtained by obtaining the user information feature encoding of multiple user feature dimensions of each user in the initial user cluster; then, respectively, the user information of the multiple user feature dimensions of each user In the pre-trained user prediction model corresponding to the feature code input, the predicted user cluster output by each user prediction model is obtained; finally, the predicted user cluster output by each user prediction model is fused to obtain the target user cluster corresponding to the initial user cluster;
  • the user information feature coding based on multiple user feature dimensions of each user in the initial user cluster is used to obtain the purpose of the target user cluster, which comprehensively considers the user information feature coding of multiple user feature dimensions of each user, and passes multiple user information feature codes.
  • the prediction of the user prediction model is helpful to improve the accuracy of the predicted user cluster, avoiding the analysis of the user information of a user in a single dimension through a machine learning model, which is likely to cause the accuracy of the predicted user cluster to be low Defects.
  • Fig. 1 is an application environment diagram of a method for predicting user clusters in one or more embodiments
  • FIG. 2 is a schematic flowchart of a method for predicting user clusters according to one or more embodiments
  • FIG. 3 is a schematic flowchart of a method for predicting user clusters in another embodiment
  • Fig. 4 is a block diagram of an apparatus for predicting user clusters according to one or more embodiments
  • Figure 5 is a block diagram of a computer device according to one or more embodiments.
  • the method for predicting user clusters provided in this application can be applied to the application environment as shown in FIG. 1.
  • the terminal 110 communicates with the server 120 through the network.
  • the terminal 110 collects the user information of multiple user characteristic dimensions of each user in the initial user cluster, and sends the user information of the multiple user characteristic dimensions of each user to the server 120;
  • the user information is encoded to obtain the user information feature code of multiple user feature dimensions of each user;
  • the user information feature code of multiple user feature dimensions of each user is input into the corresponding pre-trained user prediction model to obtain each user
  • the predicted user cluster output by each user prediction model is fused to obtain the target user cluster corresponding to the initial user cluster.
  • the terminal 110 may be, but is not limited to, various personal computers, notebook computers, smart phones, and tablet computers.
  • the server 120 may be implemented by an independent server or a server cluster composed of multiple servers.
  • a method for predicting user clusters is provided. Taking the method applied to the server in FIG. 1 as an example for description, the method includes the following steps:
  • Step S201 Obtain the feature code of each user in the initial user cluster; the feature code includes user information feature codes of multiple user feature dimensions.
  • the initial user cluster refers to the user cluster that needs to filter out target users (such as core users), such as historical user clusters);
  • the user characteristic dimension refers to the dimensions used to describe user information, such as user basic information, user recent operation behavior, user Concerned business information, business information handled by users, etc.;
  • user information feature coding refers to low-dimensional feature vectors that are compressed and coded to represent the low-level semantics of user information, which can be learned through pre-trained feature embedding network models.
  • the server obtains user information of multiple user feature dimensions of each user in the initial user cluster, and encodes the user information of multiple user feature dimensions of each user through the pre-trained feature embedding network model to obtain the user information of each user.
  • User information feature coding of each user feature dimension in this way, it is beneficial to subsequently input the user information feature coding of multiple user feature dimensions of each user into the corresponding pre-trained user prediction model to obtain the predicted user output by each user prediction model Cluster.
  • the user selects the initial user cluster on the user prediction interface provided by the terminal, and the initial cluster includes user information of multiple user characteristic dimensions of each user; the terminal responds to the user's selection operation and obtains the information of each user in the initial cluster.
  • the terminal responds to the user's selection operation and obtains the information of each user in the initial cluster.
  • the server predicts the user cluster request Perform analysis to obtain the user information of multiple user feature dimensions of each user in the initial cluster, and encode the user information of multiple user feature dimensions of each user in the initial cluster according to the preset coding instruction, and obtain the user information of multiple user feature dimensions of each user in the initial cluster.
  • User information feature codes of multiple user feature dimensions of each user.
  • Step S202 respectively inputting user information feature codes of multiple user feature dimensions of each user into a corresponding pre-trained user prediction model to obtain a predicted user cluster output by each user prediction model.
  • the user prediction model is a neural network model used to identify whether a user is a target user (such as a key user), such as a convolutional neural network model, a deep learning network model, and so on.
  • a target user such as a key user
  • different user characteristic dimensions have different corresponding user prediction models.
  • user prediction models include user prediction models with partial static attributes, user prediction models with partial generalization attributes, and user prediction models with partial dynamic attributes; among them, user prediction models with partial static attributes can dig into more users. Potential improvement.
  • the user prediction model with partial generalization attributes fully avoids the problem of excessive feature dimensions and indirect overfitting of training samples.
  • the user prediction model with partial dynamic attributes can more intuitively show a user’s recent intention tendency. In order to find a better recommendation strategy.
  • the server obtains the pre-trained user prediction model corresponding to each user feature dimension, and then respectively inputs the user information feature codes of the multiple user feature dimensions of each user into the corresponding pre-trained user prediction model, and passes each
  • the pre-trained user prediction model performs a series of neural network processing on the user information feature encoding corresponding to the user feature dimension of each user, such as convolution processing, pooling processing, full connection processing, etc., to obtain the prediction of each user prediction model for each user Prediction results: From the prediction results of each user prediction model for each user, the clusters formed by users that meet the conditions are screened out, corresponding to the predicted user clusters output by each user prediction model. In this way, the user information feature encoding of multiple user feature dimensions of each user is comprehensively considered, and the prediction is made through multiple user prediction models, which is beneficial to improve the accuracy of the subsequent predicted user clusters.
  • the user information feature codes of multiple user feature dimensions corresponding to user A are A1, A2, and A3, respectively, and user B corresponds to The user information feature codes of multiple user feature dimensions are B1, B2, and B3, respectively.
  • the user information feature codes of multiple user feature dimensions corresponding to user C are C1, C2, and C3; Code A1, user information feature code B1 of user B1, user information feature code C1 of user C is input into the pre-trained user prediction model a, and the prediction results of user prediction model a for users A, B, and C are obtained. If the user A.
  • the predicted user cluster output by user prediction model a includes user A and user B; similarly, user A’s user information feature code A2, user B’s user information feature code B2, user C’s
  • the user information feature code C2 is input into the pre-trained user prediction model b to obtain the prediction results of user A, user B, and user C by user prediction model b.
  • the prediction output by user prediction model b The user cluster includes user B and user C; respectively input user information feature code A3 of user A, user information feature code B3 of user B, and user information feature code C3 of user C into the pre-trained user prediction model c to obtain user predictions Model c predicts the results of users A, B, and C. If users A and B meet the conditions, the predicted user clusters output by the user prediction model c include users A and B.
  • Step S203 Perform fusion processing on the predicted user clusters output by each user prediction model to obtain a target user cluster corresponding to the initial user cluster.
  • the target user cluster includes multiple target users.
  • the server performs fusion processing on the predicted user clusters output by each user prediction model to calculate the occurrence probability of each user in the predicted user cluster, and selects users whose occurrence probability is greater than the preset probability of occurrence from the predicted user cluster.
  • the target user As the target user; according to the target user, construct the target user cluster as the target user cluster corresponding to the initial user cluster. In this way, the purpose of obtaining the target user cluster corresponding to the initial user cluster according to the predicted user cluster output by each user prediction model is realized, and it is avoided that only a machine learning model is used to analyze the user information of a user in a single dimension, which is easy to cause predictions. The defect that the accuracy of the user cluster is low.
  • the method further includes: uploading the target user cluster to the blockchain.
  • the server may also store the target user cluster in a node of a blockchain to ensure the privacy and security of the target user cluster.
  • the server uses blockchain technology to generate the blocks of the target user cluster, and stores the blocks of the target user cluster in the nodes of the blockchain to store the target user cluster through the blockchain to prevent the target user cluster from being tampered with. Thereby ensuring the security of the obtained target user cluster.
  • the user information feature codes of multiple user feature dimensions of each user in the initial user cluster are obtained; then the user information feature codes of multiple user feature dimensions of each user are input into the corresponding pre-training In the user prediction model, the predicted user clusters output by each user prediction model are obtained; finally, the predicted user clusters output by each user prediction model are fused to obtain the target user cluster corresponding to the initial user cluster;
  • the user information feature coding of multiple user feature dimensions of each user is used to obtain the target user cluster.
  • the user information feature coding of multiple user feature dimensions of each user is comprehensively considered, and the prediction is made through multiple user prediction models. It is beneficial to improve the accuracy of the predicted user clusters, and avoids the defect that the accuracy of the predicted user clusters is low by analyzing the user information of a user in a single dimension through only a machine learning model.
  • the above step S201, obtaining the feature code of each user in the initial user cluster includes: obtaining user information of multiple user feature dimensions of each user in the initial user cluster; The user information of the feature dimension is encoded to obtain the user information feature code of the multiple user feature dimensions of each user; the user information feature code of the multiple user feature dimensions of each user is spliced to obtain the feature code of each user.
  • the server extracts the user information of multiple user feature dimensions of each user in the initial user cluster from a local database that stores user information of multiple user feature dimensions of the user; respectively, each user in the initial user cluster
  • the user information of multiple user feature dimensions is input into the pre-trained feature embedding network model, and the user information of multiple user feature dimensions of each user is encoded through the pre-trained feature embedding network model to obtain multiple user features of each user Dimensional user information feature codes; according to the preset sequence of splicing and combination, the user information feature codes of multiple user feature dimensions of each user are spliced and combined to obtain the spliced and combined user information feature codes, which are used as the characteristic codes of each user.
  • step S202 the user information feature codes of multiple user feature dimensions of each user are respectively input into the corresponding user prediction model to obtain the predicted user cluster output by each user prediction model, including: query preset The corresponding relationship between the user feature dimension and the user prediction model is obtained, and the user prediction model corresponding to the multiple user feature dimensions is obtained; the user information feature encoding input of the multiple user feature dimensions of each user is input with the multiple user feature dimensions.
  • a corresponding user prediction model is used to obtain the prediction results of each user prediction model for each user; according to the prediction results of each user prediction model for each user, the predicted user cluster output by each user prediction model is obtained.
  • the preset corresponding relationship between the user feature dimension and the user prediction model refers to a one-to-one correspondence between the user feature dimension and the user prediction model.
  • the server obtains the corresponding relationship between the preset user feature dimension and the user prediction model from the local database, and obtains the user corresponding to multiple user feature dimensions one-to-one according to the preset corresponding relationship between the user feature dimension and the user prediction model.
  • Prediction model respectively input the user information feature coding of multiple user feature dimensions of the user into a user prediction model corresponding to multiple user feature dimensions one-to-one to obtain the prediction results of each user prediction model for the user; and so on, you can get The prediction results of each user prediction model for each user; respectively, from the prediction results of each user prediction model for each user, select the target users that meet the conditions; according to the target users that meet the conditions, build the corresponding user cluster, which corresponds to each The predicted user cluster output by the user prediction model.
  • the user information feature encoding of multiple user feature dimensions of each user is comprehensively considered, and prediction is performed through multiple user prediction models, which is beneficial to improve the accuracy of the subsequent predicted user clusters.
  • obtaining the predicted user cluster output by each user prediction model includes: extracting each user prediction from the prediction results of each user prediction model for each user The predicted probability of each user by the model; select users whose predicted probability is greater than the preset probability from each user, corresponding to the target user output by each user prediction model; obtain the cluster of target users output by each user prediction model, Corresponds to the predicted user clusters as output of each user prediction model.
  • the predicted probability is used to measure whether the user is a target user, for example, to determine whether the user is a key user.
  • the target user output by each user prediction model is determined according to the prediction probability of each user prediction model for each user, which is beneficial to improve the accuracy of the predicted user cluster output by each user prediction model.
  • the pre-trained user prediction model is obtained by training in the following manner: obtaining a sample user training set; the sample user training set includes user information of each feature dimension of the sample user and the actual probability of the sample user; The user information of each feature dimension of the sample user is encoded to obtain the user information feature code of each feature dimension of the sample user; the user information feature code of each feature dimension of the sample user is input into each corresponding user prediction model to obtain each user prediction The prediction probability of the sample users in the model; according to the prediction probability of the sample users in each user prediction model and the actual probability of the sample users, the loss value of each user prediction model is calculated; each user is predicted according to the loss value of each user prediction model The model undergoes reverse training until each user prediction model meets the convergence condition; if each user prediction model meets the convergence condition, each user prediction model is corresponding to each pre-trained user prediction model.
  • the user prediction model satisfies the convergence condition, it means that the number of training times of the user prediction model reaches the preset number of training times, or the loss value of the user prediction model is less than the preset loss value.
  • the server calculates the loss value of each user prediction model according to the predicted probability of the sample user in each user prediction model and the actual probability of the sample user, combined with the preset loss function; compares the loss value of the user prediction model with the preset loss If the loss value of the user prediction model is greater than or equal to the preset loss value, the network parameter update gradient of the user prediction model is calculated according to the loss value of the user prediction model; the network parameter update gradient of the user prediction model is updated according to the user prediction model.
  • the network parameters of the prediction model are updated, and the user prediction model after the network parameter update is retrained until the loss value obtained according to the user prediction model is less than the preset loss value, then the user prediction model is used as a pre-trained user prediction model ; With reference to this method, each pre-trained user prediction model can be obtained.
  • the above step S203 after performing fusion processing on the predicted user clusters output by each user prediction model to obtain the target user cluster corresponding to the initial user cluster, further includes: obtaining the information of each target user in the target user cluster. Credit score; if the credit score is greater than or equal to the preset score, obtain the resource type corresponding to the credit score; push the resource corresponding to the resource type to the corresponding target user.
  • the credit score is used to measure the user's credit level. The higher the credit score, the higher the user's credit level; the lower the credit score, the lower the user's credit level.
  • Resources refer to products or businesses, specifically financial products or businesses in the financial field; resource types are used to indicate product types or business types, and different resource types correspond to different product types or business types.
  • the server queries the local database storing the credit scores of multiple users to obtain the credit scores of each target user in the target user cluster; compares the credit scores of each target user with a preset score, and if the credit score is greater than Or equal to the preset score, the resource type corresponding to the credit score is obtained from the local database, and the resource corresponding to the resource type is pushed to the corresponding target user terminal to display the resource corresponding to the resource type through the target user terminal, such as financial Products, financial services, etc.
  • the target user cluster after obtaining the credit score of each target user in the target user cluster, it further includes: if the credit score is less than the preset score, generating risk reminder information corresponding to the credit score; and pushing the risk reminder information to The corresponding target user.
  • the server compares the credit score of each target user with the preset score, and if the credit score is less than the preset score, obtains a preset risk reminder information template, and generates a risk corresponding to the credit score based on the preset risk reminder information template Reminder information, and push the risk reminder information to the corresponding target user terminal, so as to display the risk reminder information through the target user terminal, so as to promptly remind the user to pay attention to avoiding risks.
  • FIG. 3 another method for predicting user clusters is provided. Taking the method applied to the server in FIG. 1 as an example, the method includes the following steps:
  • Step S301 Obtain user information of multiple user characteristic dimensions of each user in the initial user cluster.
  • Step S302 Perform coding processing on user information of multiple user feature dimensions of each user to obtain user information feature codes of multiple user feature dimensions of each user.
  • Step S303 Perform splicing processing on the user information feature codes of multiple user feature dimensions of each user to obtain the feature code of each user.
  • Step S304 Query the correspondence between the preset user feature dimensions and the user prediction model to obtain a user prediction model corresponding to multiple user feature dimensions one-to-one.
  • step S305 the user information feature codes of the multiple user feature dimensions of each user are respectively input into the user prediction model corresponding to the multiple user feature dimensions one-to-one to obtain the prediction result of each user prediction model for each user.
  • Step S306 Extract the prediction probability of each user prediction model for each user from the prediction result of each user prediction model for each user.
  • Step S307 Filter out users whose predicted probability is greater than the preset probability from each user, corresponding to the target user output by each user prediction model.
  • Step S308 Obtain a cluster composed of target users output by each user prediction model, and correspond to the predicted user cluster output by each user prediction model.
  • Step S309 Perform fusion processing on the predicted user clusters output by each user prediction model to obtain the target user cluster corresponding to the initial user cluster, and upload the target user cluster to the blockchain.
  • the user information feature coding based on the multiple user feature dimensions of each user in the initial user cluster is realized to obtain the target user cluster, and the user information of the multiple user feature dimensions of each user is comprehensively considered.
  • Feature encoding and prediction through multiple user prediction models help to improve the accuracy of the predicted user clusters, avoiding the use of only one machine learning model to analyze the user information of a user in a single dimension, which is easy to cause predictions. The defect of low accuracy of user clusters.
  • this application also provides an application scenario that applies the above-mentioned prediction method of user clusters.
  • the application of the user cluster prediction method in the application scenario is as follows:
  • each user's prediction model is trained to ensure the comprehensiveness of the model effect.
  • each user prediction model model is trained and updated to detect the user prediction model completed by each training; through the calculation of each user prediction model, the business logic result of each data area is obtained .
  • the result of the list strategy is obtained for the global data.
  • the list strategy described above is a summary of the model results produced by each model. It can specifically integrate the list, supplement and delete business rules, and obtain the matching of label strategies or models for different lists. Strategy matching and so on.
  • each module can be reused and upgraded.
  • the addition and deletion of modules can also meet new business needs. For example, when there are new business requirements, there may be partial intersections in the data sections, and the corresponding subsequent models, lists, and tags can be reused directly, and updated and iterated with subsequent use.
  • the business logic of the model changes for example, new strategies are added, the corresponding data sections, or models, rules, etc. can be directly added to be compatible with the use of new strategies.
  • the bank's business operation capability is improved, and the customer experience is optimized.
  • more users can be fully covered; more business and maintenance strategies can be adopted, and a personalized user service experience can be achieved through algorithm calculations.
  • a complete and rigorous logical framework can be dynamically adjusted and added and deleted based on the results; dynamic adjustments ensure that the model itself can continuously meet the requirements of update iterations based on user samples and business migration; additions and deletions ensure The various logic modules in the link can be reused to make new business outputs faster.
  • a device for predicting user clusters including: a feature code acquisition module 410, a predicted user cluster acquisition module 420, and a target user cluster acquisition module 430, wherein:
  • the feature code obtaining module 410 is used to obtain feature codes of each user in the initial user cluster; the feature codes include user information feature codes of multiple user feature dimensions.
  • the predicted user cluster obtaining module 420 is configured to respectively input user information feature codes of multiple user feature dimensions of each user into the corresponding pre-trained user prediction model to obtain predicted user clusters output by each user prediction model.
  • the target user cluster acquisition module 430 is configured to perform fusion processing on the predicted user clusters output by each user prediction model to obtain the target user cluster corresponding to the initial user cluster.
  • the feature encoding acquisition module 410 is also used to acquire user information of multiple user feature dimensions of each user in the initial user cluster; to encode user information of multiple user feature dimensions of each user, Obtain user information feature codes of multiple user feature dimensions of each user; perform splicing processing on the user information feature codes of multiple user feature dimensions of each user to obtain the feature code of each user.
  • the predictive user cluster acquisition module 420 is also used to query the correspondence between preset user feature dimensions and user prediction models to obtain user prediction models corresponding to multiple user feature dimensions one-to-one; respectively The user information feature encoding of multiple user feature dimensions of the user input the user prediction model corresponding to multiple user feature dimensions one-to-one to obtain the prediction result of each user prediction model for each user; according to the prediction result of each user prediction model for each user , Get the predicted user cluster output by each user prediction model.
  • the predicted user cluster acquisition module 420 is also used to extract the predicted probabilities of each user prediction model for each user from the prediction results of each user prediction model for each user; and filter each user separately The user whose predicted probability is greater than the preset probability corresponds to the target user output by each user prediction model; the cluster formed by the target user output by each user prediction model is obtained, and the predicted user cluster is corresponding to the output of each user prediction model.
  • the user cluster prediction device further includes a model training module for obtaining a sample user training set; the sample user training set includes user information of each feature dimension of the sample user and the actual probability of the sample user; The user information of each feature dimension of the sample user is encoded to obtain the user information feature code of each feature dimension of the sample user; the user information feature code of each feature dimension of the sample user is input into each corresponding user prediction model to obtain each user prediction The predicted probability of the sample users in the model; according to the predicted probability of the sample users in each user prediction model and the actual probability of the sample users, the loss value of each user prediction model is calculated; each user is predicted according to the loss value of each user prediction model The model undergoes reverse training until each user prediction model meets the convergence condition; if each user prediction model meets the convergence condition, each user prediction model is corresponding to each pre-trained user prediction model.
  • a model training module for obtaining a sample user training set
  • the sample user training set includes user information of each feature dimension of the sample user and the actual probability of the sample user
  • the predicting device of the user cluster further includes a resource pushing module, which is used to obtain the credit score of each target user in the target user cluster; if the credit score is greater than or equal to the preset score, obtain the corresponding credit score Resource type; the resource corresponding to the resource type is pushed to the corresponding target user.
  • a resource pushing module which is used to obtain the credit score of each target user in the target user cluster; if the credit score is greater than or equal to the preset score, obtain the corresponding credit score Resource type; the resource corresponding to the resource type is pushed to the corresponding target user.
  • the prediction device of the user cluster further includes an information push module, configured to generate risk reminder information corresponding to the credit score if the credit score is less than the preset score; push the risk reminder information to the corresponding target user.
  • the device for predicting user clusters further includes an upload module for uploading the target user cluster to the blockchain.
  • Each module in the above-mentioned user cluster prediction device can be implemented in whole or in part by software, hardware, and a combination thereof.
  • the above-mentioned modules may be embedded in the form of hardware or independent of the processor in the computer equipment, or may be stored in the memory of the computer equipment in the form of software, so that the processor can call and execute the operations corresponding to the above-mentioned modules.
  • a computer device is provided.
  • the computer device may be a server, and its internal structure diagram may be as shown in FIG. 5.
  • the computer equipment includes a processor, a memory, and a network interface connected through a system bus. Among them, the processor of the computer device is used to provide calculation and control capabilities.
  • the memory of the computer device includes a non-volatile or volatile storage medium and internal memory.
  • the non-volatile or volatile storage medium stores an operating system, computer readable instructions, and a database.
  • the internal memory provides an environment for the operation of the operating system and computer-readable instructions in the non-volatile or volatile storage medium.
  • the database of the computer equipment is used to store the characteristic codes of each user, target user clusters and other data.
  • the network interface of the computer device is used to communicate with an external terminal through a network connection.
  • the computer-readable instructions are executed by the processor to realize a method for predicting user clusters.
  • FIG. 5 is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied.
  • the specific computer device may Including more or fewer parts than shown in the figure, or combining some parts, or having a different arrangement of parts.
  • a computer device includes a memory and one or more processors, and computer-readable instructions are stored in the memory.
  • the steps of the method for predicting a user cluster provided in any one of the embodiments of the present application are implemented .
  • One or more computer-readable storage media storing computer-readable instructions.
  • the computer-readable storage media may be non-volatile or volatile.
  • the computer-readable instructions are executed by one or more processors , Enabling one or more processors to implement the steps of the user cluster prediction method provided in any embodiment of the present application.
  • the blockchain referred to in this application is a new application mode of computer technology such as distributed data storage, point-to-point transmission, consensus mechanism, and encryption algorithm.
  • Blockchain essentially a decentralized database, is a series of data blocks associated with cryptographic methods. Each data block contains a batch of network transaction information for verification. The validity of the information (anti-counterfeiting) and the generation of the next block.
  • the blockchain can include the underlying platform of the blockchain, the platform product service layer, and the application service layer.
  • Non-volatile memory may include read-only memory (Read-Only Memory, ROM), magnetic tape, floppy disk, flash memory, or optical storage.
  • Volatile memory may include random access memory (RAM) or external cache memory.
  • RAM may be in various forms, such as static random access memory (Static Random Access Memory, SRAM) or dynamic random access memory (Dynamic Random Access Memory, DRAM), etc.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • Software Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Business, Economics & Management (AREA)
  • Accounting & Taxation (AREA)
  • Computing Systems (AREA)
  • Strategic Management (AREA)
  • Finance (AREA)
  • Artificial Intelligence (AREA)
  • Computational Linguistics (AREA)
  • Development Economics (AREA)
  • Evolutionary Computation (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Fuzzy Systems (AREA)
  • Probability & Statistics with Applications (AREA)
  • Medical Informatics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Game Theory and Decision Science (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • General Business, Economics & Management (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

一种用户集群的预测方法，涉及智能决策技术领域，包括：获取初始用户集群中的各个用户的特征编码；特征编码包括多个用户特征维度的用户信息特征编码（S201）；分别将各个用户的多个用户特征维度的用户信息特征编码输入对应的预先训练的用户预测模型中，得到各个用户预测模型输出的预测用户集群（S202）；对各个用户预测模型输出的预测用户集群进行融合处理，得到初始用户集群对应的目标用户集群（S203）。此外，该方法还涉及区块链技术，目标用户集群可存储于区块链节点中。

Description

用户集群的预测方法、装置、计算机设备和存储介质
本申请要求于2020年06月24日提交中国专利局，申请号为2020105864119，申请名称为“用户集群的预测方法、装置、计算机设备和存储介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及智能决策技术领域，特别是涉及一种用户集群的预测方法、装置、计算机设备和存储介质。
背景技术
随着机器学习的普及，越来越多的领域运用到了机器学习，以对对应领域的数据进行有效分析，比如用户集群预测。
然而，目前的用户集群的预测方法，一般是通过获取用户的单个维度的用户信息，比如历史业务操作信息，并将该单个维度的用户信息输入机器学习模型中，以通过机器学习模型判别该用户是否为目标用户；以此类推，可以通过机器学习模型预测出符合条件的用户集群；然而，发明人意识到，判别用户是否为目标用户，往往受多个因素的影响，仅仅通过一个机器学习模型，对用户的单个维度的用户信息进行分析，容易造成预测出的用户集群的准确性较低。
发明内容
根据本申请公开的各种实施例，提供一种用户集群的预测方法、装置、计算机设备和存储介质。
一种用户集群的预测方法包括：
获取初始用户集群中的各个用户的特征编码；所述特征编码包括多个用户特征维度的用户信息特征编码；
分别将所述各个用户的多个用户特征维度的用户信息特征编码输入对应的预先训练的用户预测模型中，得到各个所述用户预测模型输出的预测用户集群；及
对各个所述用户预测模型输出的预测用户集群进行融合处理，得到所述初始用户集群对应的目标用户集群。
一种用户集群的预测装置包括：
特征编码获取模块，用于获取初始用户集群中的各个用户的特征编码；所述特征编码包括多个用户特征维度的用户信息特征编码；
预测用户集群获取模块，用于分别将所述各个用户的多个用户特征维度的用户信息特 征编码输入对应的预先训练的用户预测模型中，得到各个所述用户预测模型输出的预测用户集群；及
目标用户集群获取模块，用于对各个所述用户预测模型输出的预测用户集群进行融合处理，得到所述初始用户集群对应的目标用户集群。
一种计算机设备，包括存储器和一个或多个处理器，所述存储器中储存有计算机可读指令，所述计算机可读指令被所述处理器执行时，使得所述一个或多个处理器执行以下步骤：
获取初始用户集群中的各个用户的特征编码；所述特征编码包括多个用户特征维度的用户信息特征编码；
分别将所述各个用户的多个用户特征维度的用户信息特征编码输入对应的预先训练的用户预测模型中，得到各个所述用户预测模型输出的预测用户集群；及
对各个所述用户预测模型输出的预测用户集群进行融合处理，得到所述初始用户集群对应的目标用户集群。
一个或多个存储有计算机可读指令的计算机可读存储介质，计算机可读指令被一个或多个处理器执行时，使得一个或多个处理器执行以下步骤：
获取初始用户集群中的各个用户的特征编码；所述特征编码包括多个用户特征维度的用户信息特征编码；
分别将所述各个用户的多个用户特征维度的用户信息特征编码输入对应的预先训练的用户预测模型中，得到各个所述用户预测模型输出的预测用户集群；及
对各个所述用户预测模型输出的预测用户集群进行融合处理，得到所述初始用户集群对应的目标用户集群。
上述用户集群的预测方法、装置、计算机设备和存储介质，通过获取初始用户集群中的各个用户的多个用户特征维度的用户信息特征编码；接着分别将各个用户的多个用户特征维度的用户信息特征编码输入对应的预先训练的用户预测模型中，得到各个用户预测模型输出的预测用户集群；最后对各个用户预测模型输出的预测用户集群进行融合处理，得到初始用户集群对应的目标用户集群；实现了根据初始用户集群中的各个用户的多个用户特征维度的用户信息特征编码，得到目标用户集群的目的，综合考虑了每个用户的多个用户特征维度的用户信息特征编码，且通过多个用户预测模型进行预测，有利于提高预测出的用户集群的准确性，避免了仅仅通过一个机器学习模型，对用户的单个维度的用户信息进行分析，容易造成预测出的用户集群的准确性较低的缺陷。
本申请的一个或多个实施例的细节在下面的附图和描述中提出。本申请的其它特征和优点将从说明书、附图以及权利要求书变得明显。
附图说明
为了更清楚地说明本申请实施例中的技术方案，下面将对实施例中所需要使用的附图 作简单地介绍，显而易见地，下面描述中的附图仅仅是本申请的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其它的附图。
图1为根据一个或多个实施例中用户集群的预测方法的应用环境图；
图2为根据一个或多个实施例中用户集群的预测方法的流程示意图；
图3为另一个实施例中用户集群的预测方法的流程示意图；
图4为根据一个或多个实施例中用户集群的预测装置的框图；
图5为根据一个或多个实施例中计算机设备的框图。
具体实施方式
为了使本申请的技术方案及优点更加清楚明白，以下结合附图及实施例，对本申请进行进一步详细说明。应当理解，此处描述的具体实施例仅仅用以解释本申请，并不用于限定本申请。
本申请提供的用户集群的预测方法，可以应用于如图1所示的应用环境中。其中，终端110通过网络与服务器120进行通信。终端110采集初始用户集群中的各个用户的多个用户特征维度的用户信息，并将各个用户的多个用户特征维度的用户信息发送至服务器120；服务器120对各个用户的多个用户特征维度的用户信息进行编码处理，得到各个用户的多个用户特征维度的用户信息特征编码；分别将各个用户的多个用户特征维度的用户信息特征编码输入对应的预先训练的用户预测模型中，得到各个用户预测模型输出的预测用户集群；对各个用户预测模型输出的预测用户集群进行融合处理，得到初始用户集群对应的目标用户集群。其中，终端110可以但不限于是各种个人计算机、笔记本电脑、智能手机和平板电脑便，服务器120可以用独立的服务器或者是多个服务器组成的服务器集群来实现。
在其中一个实施例中，如图2所示，提供了一种用户集群的预测方法，以该方法应用于图1中的服务器为例进行说明，包括以下步骤：
步骤S201，获取初始用户集群中的各个用户的特征编码；特征编码包括多个用户特征维度的用户信息特征编码。
其中，初始用户集群是指需要筛选出目标用户(比如核心用户)的用户集群，比如历史用户集群；用户特征维度是指用于描述用户信息的维度，比如用户基本信息、用户近期操作行为、用户关注的业务信息、用户办理的业务信息等；用户信息特征编码是指经过压缩编码后的用于表示用户信息的低层语义的低维度特征向量，可以通过预先训练的特征嵌入网络模型学习得到。
具体地，服务器获取初始用户集群中各个用户的多个用户特征维度的用户信息，通过预先训练的特征嵌入网络模型对各个用户的多个用户特征维度的用户信息进行编码处理，得到各个用户的多个用户特征维度的用户信息特征编码；这样，有利于后续分别将各个用户的多个用户特征维度的用户信息特征编码输入对应的预先训练的用户预测模型中，得到 各个用户预测模型输出的预测用户集群。
举例说明，用户在终端提供的用户预测界面上选择初始用户集群，该初始集群中包括各个用户的多个用户特征维度的用户信息；终端响应用户的选择操作，获取初始集群中的各个用户的多个用户特征维度的用户信息，根据初始集群中的各个用户的多个用户特征维度的用户信息生成用户集群预测请求，并将该用户集群预测请求发送至对应的服务器；服务器对该用户集群预测请求进行解析，得到初始集群中的各个用户的多个用户特征维度的用户信息，根据预设的编码指令对初始集群中的各个用户的多个用户特征维度的用户信息进行编码处理，得到初始集群中的各个用户的多个用户特征维度的用户信息特征编码。
步骤S202，分别将各个用户的多个用户特征维度的用户信息特征编码输入对应的预先训练的用户预测模型中，得到各个用户预测模型输出的预测用户集群。
其中，用户预测模型是一种用于识别用户是否为目标用户(比如关键用户)的神经网络模型，比如卷积神经网络模型、深度学习网络模型等。不同用户特征维度，对应的用户预测模型不一样。在实际场景中，用户预测模型包括偏静态属性的用户预测模型、偏泛化属性的用户预测模型、偏动态属性的用户预测模型；其中，偏静态属性的用户预测模型可以挖掘到更多用户的潜在提升空间，偏泛化属性的用户预测模型充分地避免了特征维度过多而间接过拟合训练样本的问题，偏动态属性的用户预测模型可以更直观地表现出用户近期的一个意图倾向，从而找到更好的推荐策略。
具体地，服务器获取与每个用户特征维度对应的预先训练的用户预测模型，然后分别将各个用户的多个用户特征维度的用户信息特征编码输入对应的预先训练的用户预测模型中，通过每个预先训练的用户预测模型对各个用户的对应用户特征维度的用户信息特征编码进行一系列神经网络处理，比如卷积处理、池化处理、全连接处理等，得到每个用户预测模型对各个用户的预测结果；分别从每个用户预测模型对各个用户的预测结果中，筛选出满足条件的用户所构成的集群，对应作为每个用户预测模型输出的预测用户集群。这样，综合考虑了每个用户的多个用户特征维度的用户信息特征编码，且通过多个用户预测模型进行预测，有利于提高后续预测出的用户集群的准确性。
举例说明，假设初始用户集群中有3个用户，分别为用户A、用户B和用户C，用户A对应的多个用户特征维度的用户信息特征编码分别是A1、A2和A3，用户B对应的多个用户特征维度的用户信息特征编码分别是B1、B2和B3，用户C对应的多个用户特征维度的用户信息特征编码分别是C1、C2和C3；接着，分别将用户A的用户信息特征编码A1、用户B的用户信息特征编码B1、用户C的用户信息特征编码C1输入预先训练的用户预测模型a中，得到用户预测模型a对用户A、用户B和用户C的预测结果，若用户A、用户B满足条件，则用户预测模型a输出的预测用户集群包括用户A、用户B；同理，分别将用户A的用户信息特征编码A2、用户B的用户信息特征编码B2、用户C的用户信息特征编码C2输入预先训练的用户预测模型b中，得到用户预测模型b对用户A、用户B和用户C的预测结果，若用户B、用户C满足条件，则用户预测模型b输出的预测用户集群包 括用户B、用户C；分别将用户A的用户信息特征编码A3、用户B的用户信息特征编码B3、用户C的用户信息特征编码C3输入预先训练的用户预测模型c中，得到用户预测模型c对用户A、用户B和用户C的预测结果，若用户A、用户B满足条件，则用户预测模型c输出的预测用户集群包括用户A、用户B。
步骤S203，对各个用户预测模型输出的预测用户集群进行融合处理，得到初始用户集群对应的目标用户集群。
其中，目标用户集群中包括多个目标用户。
具体地，服务器对各个用户预测模型输出的预测用户集群进行融合处理，以统计出预测用户集群中的各个用户的出现概率，并从预测用户集群中筛选出出现概率大于预设出现概率的用户，作为目标用户；根据目标用户，构建目标用户集群，作为初始用户集群对应的目标用户集群。这样，实现了根据各个用户预测模型输出的预测用户集群，得到初始用户集群对应的目标用户集群的目的，避免了仅仅通过一个机器学习模型，对用户的单个维度的用户信息进行分析，容易造成预测出的用户集群的准确性较低的缺陷。
进一步地，在对各个用户预测模型输出的预测用户集群进行融合处理，得到初始用户集群对应的目标用户集群之后，还包括：将目标用户集群上传至区块链中。具体来说，在得到用户集群之后，服务器还可以将目标用户集群存储于一区块链的节点中，以保证目标用户集群的私密和安全性。
举例说明，服务器利用区块链技术生成目标用户集群的区块，将目标用户集群的区块存储至区块链的节点中，以通过区块链存储目标用户集群，避免目标用户集群被篡改，从而保证了得到的目标用户集群的安全性。
上述用户集群的预测方法中，通过获取初始用户集群中的各个用户的多个用户特征维度的用户信息特征编码；接着分别将各个用户的多个用户特征维度的用户信息特征编码输入对应的预先训练的用户预测模型中，得到各个用户预测模型输出的预测用户集群；最后对各个用户预测模型输出的预测用户集群进行融合处理，得到初始用户集群对应的目标用户集群；实现了根据初始用户集群中的各个用户的多个用户特征维度的用户信息特征编码，得到目标用户集群的目的，综合考虑了每个用户的多个用户特征维度的用户信息特征编码，且通过多个用户预测模型进行预测，有利于提高预测出的用户集群的准确性，避免了仅仅通过一个机器学习模型，对用户的单个维度的用户信息进行分析，容易造成预测出的用户集群的准确性较低的缺陷。
在其中一个实施例中，上述步骤S201，获取初始用户集群中的各个用户的特征编码，包括：获取初始用户集群中的各个用户的多个用户特征维度的用户信息；对各个用户的多个用户特征维度的用户信息进行编码处理，得到各个用户的多个用户特征维度的用户信息特征编码；对各个用户的多个用户特征维度的用户信息特征编码进行拼接处理，得到各个用户的特征编码。
举例说明，服务器从存储有用户的多个用户特征维度的用户信息的本地数据库中，提 取出初始用户集群中的各个用户的多个用户特征维度的用户信息；分别将初始用户集群中的各个用户的多个用户特征维度的用户信息输入预先训练的特征嵌入网络模型，通过预先训练的特征嵌入网络模型对各个用户的多个用户特征维度的用户信息进行编码处理，得到各个用户的多个用户特征维度的用户信息特征编码；按照预设拼接组合顺序，将各个用户的多个用户特征维度的用户信息特征编码进行拼接组合，得到拼接组合后的用户信息特征编码，作为各个用户的特征编码。
在本实施例中，通过获取初始用户集群中的各个用户的特征编码，有利于后续分别将各个用户的多个用户特征维度的用户信息特征编码输入对应的预先训练的用户预测模型中，得到各个用户预测模型输出的预测用户集群。
在其中一个实施例中，上述步骤S202，分别将各个用户的多个用户特征维度的用户信息特征编码输入对应的用户预测模型中，得到各个用户预测模型输出的预测用户集群，包括：查询预设的用户特征维度与用户预测模型的对应关系，得到与多个用户特征维度一一对应的用户预测模型；分别将各个用户的多个用户特征维度的用户信息特征编码输入与多个用户特征维度一一对应的用户预测模型，得到各个用户预测模型对各个用户的预测结果；根据各个用户预测模型对各个用户的预测结果，得到各个用户预测模型输出的预测用户集群。
其中，预设的用户特征维度与用户预测模型的对应关系，是指用户特征维度与用户预测模型存在一一对应关系。
举例说明，服务器从本地数据库中获取预设的用户特征维度与用户预测模型的对应关系，根据预设的用户特征维度与用户预测模型的对应关系，得到与多个用户特征维度一一对应的用户预测模型；分别将用户的多个用户特征维度的用户信息特征编码输入与多个用户特征维度一一对应的用户预测模型，得到各个用户预测模型对该用户的预测结果；以此类推，可以得到各个用户预测模型对各个用户的预测结果；分别从各个用户预测模型对各个用户的预测结果中，筛选出满足条件的目标用户；根据满足条件的目标用户，构建对应的用户集群，对应作为每个用户预测模型输出的预测用户集群。
在本实施例中，综合考虑了每个用户的多个用户特征维度的用户信息特征编码，且通过多个用户预测模型进行预测，有利于提高后续预测出的用户集群的准确性。
在其中一个实施例中，根据各个用户预测模型对各个用户的预测结果，得到各个用户预测模型输出的预测用户集群，包括：从各个用户预测模型对各个用户的预测结果中，提取出各个用户预测模型对各个用户的预测概率；分别从各个用户中，筛选出预测概率大于预设概率的用户，对应作为各个用户预测模型输出的目标用户；获取各个用户预测模型输出的目标用户所构成的集群，对应作为各个用户预测模型输出的预测用户集群。
其中，预测概率用于衡量用户是否为目标用户，比如用于判别用户是否为关键用户。
在本实施例中，根据各个用户预测模型对各个用户的预测概率，确定各个用户预测模型输出的目标用户，后利于提高各个用户预测模型输出的预测用户集群的准确性。
在其中一个实施例中，预先训练的用户预测模型通过下述方式训练得到：获取样本用户训练集；样本用户训练集包括样本用户的各个特征维度的用户信息以及样本用户的实际概率；对样本用户的各个特征维度的用户信息进行编码处理，得到样本用户的各个特征维度的用户信息特征编码；分别将样本用户的各个特征维度的用户信息特征编码输入对应的各个用户预测模型中，得到各个用户预测模型中对样本用户的预测概率；根据各个用户预测模型中对样本用户的预测概率以及样本用户的实际概率，统计各个用户预测模型的损失值；根据各个用户预测模型的损失值，对各个用户预测模型进行反向训练，直至各个用户预测模型满足收敛条件；若各个用户预测模型满足收敛条件，则将各个用户预测模型，对应作为各个预先训练的用户预测模型。
其中，用户预测模型满足收敛条件，是指用户预测模型的训练次数达到预设训练次数，或者用户预测模型的损失值小于预设损失值。
举例说明，服务器根据各个用户预测模型中对样本用户的预测概率以及样本用户的实际概率，结合预设损失函数，计算得到各个用户预测模型的损失值；将用户预测模型的损失值与预设损失值进行比较，若用户预测模型的损失值大于或者等于预设损失值，则根据用户预测模型的损失值，计算用户预测模型的网络参数更新梯度；根据用户预测模型的网络参数更新梯度，对用户预测模型的网络参数进行更新，并对网络参数更新后的用户预测模型进行再次训练，直到根据用户预测模型得到的损失值小于预设损失值，则将该用户预测模型作为预先训练的用户预测模型；参照此方法，可以得到各个预先训练的用户预测模型。
在本实施例中，通过对各个用户预测模型进行反复训练，有利于提高用户预测模型输出的预测用户集群的准确性，从而提高了后续得到的目标用户集群的准确性，进一步提高了预测出的用户集群的准确性。
在其中一个实施例中，上述步骤S203，在对各个用户预测模型输出的预测用户集群进行融合处理，得到初始用户集群对应的目标用户集群之后，还包括：获取目标用户集群中的各个目标用户的信用分数；若信用分数大于或者等于预设分数，则获取与信用分数对应的资源类型；将资源类型对应的资源推送给对应的目标用户。
其中，信用分数用于衡量用户的信用程度，信用分数越高，用户的信用程度越高；信用分数越低，用户的信用程度越低。资源是指产品或者业务，具体是指金融领域中的金融产品或者金融业务；资源类型用于表示产品类型或者业务类型，不同资源类型，对应不同产品类型或者业务类型。
具体地，服务器从存储有多个用户的信用分数的本地数据库中，查询得到目标用户集群中的各个目标用户的信用分数；将各个目标用户的信用分数与预设分数进行比较，若信用分数大于或者等于预设分数，则从本地数据库中获取与信用分数对应的资源类型，并将资源类型对应的资源推送给对应的目标用户终端，以通过目标用户终端展示该资源类型对应的资源，比如金融产品、金融业务等。
在本实施例中，在目标用户的信用分数大于或者等于预设分数的情况下，将信用分数对应的资源类型的资源推送给目标用户，达到了精准推送的目的，进一步提高了资源推送的准确率。
在其中一个实施例中，在获取目标用户集群中的各个目标用户的信用分数之后，还包括：若信用分数小于预设分数，则生成与信用分数对应的风险提醒信息；将风险提醒信息推送给对应的目标用户。
具体地，服务器将各个目标用户的信用分数与预设分数进行比较，若信用分数小于预设分数，则获取预设风险提醒信息模板，根据预设风险提醒信息模板，生成与信用分数对应的风险提醒信息，并将风险提醒信息推送给对应的目标用户终端，以通过目标用户终端展示该风险提醒信息，便于及时提醒用户注意规避风险。
在其中一个实施例中，如图3所示，提供了另一种用户集群的预测方法，以该方法应用于图1中的服务器为例进行说明，包括以下步骤：
步骤S301，获取初始用户集群中的各个用户的多个用户特征维度的用户信息。
步骤S302，对各个用户的多个用户特征维度的用户信息进行编码处理，得到各个用户的多个用户特征维度的用户信息特征编码。
步骤S303，对各个用户的多个用户特征维度的用户信息特征编码进行拼接处理，得到各个用户的特征编码。
步骤S304，查询预设的用户特征维度与用户预测模型的对应关系，得到与多个用户特征维度一一对应的用户预测模型。
步骤S305，分别将各个用户的多个用户特征维度的用户信息特征编码输入与多个用户特征维度一一对应的用户预测模型，得到各个用户预测模型对各个用户的预测结果。
步骤S306，从各个用户预测模型对各个用户的预测结果中，提取出各个用户预测模型对各个用户的预测概率。
步骤S307，分别从各个用户中，筛选出预测概率大于预设概率的用户，对应作为各个用户预测模型输出的目标用户。
步骤S308，获取各个用户预测模型输出的目标用户所构成的集群，对应作为各个用户预测模型输出的预测用户集群。
步骤S309，对各个用户预测模型输出的预测用户集群进行融合处理，得到初始用户集群对应的目标用户集群，将目标用户集群上传至区块链中。
在本实施例中，实现了根据初始用户集群中的各个用户的多个用户特征维度的用户信息特征编码，得到目标用户集群的目的，综合考虑了每个用户的多个用户特征维度的用户信息特征编码，且通过多个用户预测模型进行预测，有利于提高预测出的用户集群的准确性，避免了仅仅通过一个机器学习模型，对用户的单个维度的用户信息进行分析，容易造成预测出的用户集群的准确性较低的缺陷。
在其中一个实施例中，本申请还提供一种应用场景，该应用场景应用上述的用户集群 的预测方法。具体地，该用户集群的预测方法在该应用场景的应用如下：
(1)接受现有可支持数据，并对现有的基本数据在数据设备上进行领域业务的区分；针对不同来源的数据和数据状态设定符合该数据领域的目标和标杆任务。
(2)针对不同来源的数据和数据状态，选择不同的用户预测模型，比如偏静态属性的用户预测模型、偏泛化属性的用户预测模型、偏动态属性的用户预测模型。
(3)通过训练数据和预测数据，对每个用户预测模型进行训练，从而保证模型效果的全面性。
(4)按照预设的业务逻辑模型算法和目标对各个用户预测模型模型进行训练和更新检测，得到各个训练完成的用户预测模型；通过各个用户预测模型的计算，得到各个数据片区的业务逻辑结果。
(5)根据上述的业务逻辑结果，比如运营名单，进行各个模型标签的设定；基于规则和其他模型规则的干预下，完成名单整合和补充删减任务。
(6)模拟下发策略在现有的测试集上进行回测，反复调整各个环节的参数和逻辑，达到整体目标最优化的结果。
(7)在新的数据样本进入的时候，会对数据的分布进行检测，判定是否有需要进行模型参数更新的需求。
(8)在使用过程中，按照预定设置的模型和规则策略，对全局的数据得到名单策略结果。如上所述的名单策略，为各个模型产出的模型结果汇总而成，具体可以针对名单进行的整合，针对业务规则进行的补充和删减，针对不同名单获取进行的标签策略的匹配、或者模型策略的匹配等等。
(9)根据客户运营反馈结果，返回进行建模环节、模型结果环节、名单整合环节的模式、参数、权重调整。
(10)进一步地，考虑到上述各个环节有着明显的解耦能力，各个模块可以进行复用提升。在进行模块的增加、删减也能够满足新的业务需求。举例来说，当有新的业务需求的同时，数据版块可能会有部分的交集，相应的后续模型、名单、标签就可以直接进行复用，并伴随后续使用进行更新迭代。再者，当该模型的业务逻辑出现变化，例如增加了新的策略的同时，可以直接的增加对应的数据板块、或者是模型、规则等，兼容新的策略使用。
上述实施例，通过设计一套合理简洁的算法模型与数据存储使用框架，来提高银行的业务经营能力，并优化提升客户的体验感受。在广大的名单群体当中，可以全面的覆盖更多的用户；并可以采取更多的经营维护策略，通过算法的计算下，达到个性化的用户服务体验。同时，一套完整缜密的逻辑框架，可以根据结果合理的进行动态调整和添加删减；通过动态调整保证了模型自身根据用户样本和业务的迁徙可以不断满足更新迭代的要求；添加删减保证了环节上的各个逻辑模块可以进行复用，更快的进行新的业务产出。
应该理解的是，虽然图2-3的流程图中的各个步骤按照箭头的指示依次显示，但是这 些步骤并不是必然按照箭头指示的顺序依次执行。除非本文中有明确的说明，这些步骤的执行并没有严格的顺序限制，这些步骤可以以其它的顺序执行。而且，图2-3中的至少一部分步骤可以包括多个步骤或者多个阶段，这些步骤或者阶段并不必然是在同一时刻执行完成，而是可以在不同的时刻执行，这些步骤或者阶段的执行顺序也不必然是依次进行，而是可以与其它步骤或者其它步骤中的步骤或者阶段的至少一部分轮流或者交替地执行。
在其中一个实施例中，如图4所示，提供了一种用户集群的预测装置，包括：特征编码获取模块410、预测用户集群获取模块420和目标用户集群获取模块430，其中：
特征编码获取模块410，用于获取初始用户集群中的各个用户的特征编码；特征编码包括多个用户特征维度的用户信息特征编码。
预测用户集群获取模块420，用于分别将各个用户的多个用户特征维度的用户信息特征编码输入对应的预先训练的用户预测模型中，得到各个用户预测模型输出的预测用户集群。
目标用户集群获取模块430，用于对各个用户预测模型输出的预测用户集群进行融合处理，得到初始用户集群对应的目标用户集群。
在其中一个实施例中，特征编码获取模块410，还用于获取初始用户集群中的各个用户的多个用户特征维度的用户信息；对各个用户的多个用户特征维度的用户信息进行编码处理，得到各个用户的多个用户特征维度的用户信息特征编码；对各个用户的多个用户特征维度的用户信息特征编码进行拼接处理，得到各个用户的特征编码。
在其中一个实施例中，预测用户集群获取模块420，还用于查询预设的用户特征维度与用户预测模型的对应关系，得到与多个用户特征维度一一对应的用户预测模型；分别将各个用户的多个用户特征维度的用户信息特征编码输入与多个用户特征维度一一对应的用户预测模型，得到各个用户预测模型对各个用户的预测结果；根据各个用户预测模型对各个用户的预测结果，得到各个用户预测模型输出的预测用户集群。
在其中一个实施例中，预测用户集群获取模块420，还用于从各个用户预测模型对各个用户的预测结果中，提取出各个用户预测模型对各个用户的预测概率；分别从各个用户中，筛选出预测概率大于预设概率的用户，对应作为各个用户预测模型输出的目标用户；获取各个用户预测模型输出的目标用户所构成的集群，对应作为各个用户预测模型输出的预测用户集群。
在其中一个实施例中，用户集群的预测装置还包括模型训练模块，用于获取样本用户训练集；样本用户训练集包括样本用户的各个特征维度的用户信息以及样本用户的实际概率；对样本用户的各个特征维度的用户信息进行编码处理，得到样本用户的各个特征维度的用户信息特征编码；分别将样本用户的各个特征维度的用户信息特征编码输入对应的各个用户预测模型中，得到各个用户预测模型中对样本用户的预测概率；根据各个用户预测模型中对样本用户的预测概率以及样本用户的实际概率，统计各个用户预测模型的损失值；根据各个用户预测模型的损失值，对各个用户预测模型进行反向训练，直至各个用户预测 模型满足收敛条件；若各个用户预测模型满足收敛条件，则将各个用户预测模型，对应作为各个预先训练的用户预测模型。
在其中一个实施例中，用户集群的预测装置还包括资源推送模块，用于获取目标用户集群中的各个目标用户的信用分数；若信用分数大于或者等于预设分数，则获取与信用分数对应的资源类型；将资源类型对应的资源推送给对应的目标用户。
在其中一个实施例中，用户集群的预测装置还包括信息推送模块，用于若信用分数小于预设分数，则生成与信用分数对应的风险提醒信息；将风险提醒信息推送给对应的目标用户。
在其中一个实施例中，用户集群的预测装置还包括上传模块，用于将目标用户集群上传至区块链中。
关于用户集群的预测装置的具体限定可以参见上文中对于用户集群的预测方法的限定，在此不再赘述。上述用户集群的预测装置中的各个模块可全部或部分通过软件、硬件及其组合来实现。上述各模块可以硬件形式内嵌于或独立于计算机设备中的处理器中，也可以以软件形式存储于计算机设备中的存储器中，以便于处理器调用执行以上各个模块对应的操作。
在其中一个实施例中，提供了一种计算机设备，该计算机设备可以是服务器，其内部结构图可以如图5所示。该计算机设备包括通过系统总线连接的处理器、存储器和网络接口。其中，该计算机设备的处理器用于提供计算和控制能力。该计算机设备的存储器包括非易失性或易失性存储介质、内存储器。该非易失性或易失性存储介质存储有操作系统、计算机可读指令和数据库。该内存储器为非易失性或易失性存储介质中的操作系统和计算机可读指令的运行提供环境。该计算机设备的数据库用于存储各个用户的特征编码、目标用户集群等数据。该计算机设备的网络接口用于与外部的终端通过网络连接通信。该计算机可读指令被处理器执行时以实现一种用户集群的预测方法。
本领域技术人员可以理解，图5中示出的结构，仅仅是与本申请方案相关的部分结构的框图，并不构成对本申请方案所应用于其上的计算机设备的限定，具体的计算机设备可以包括比图中所示更多或更少的部件，或者组合某些部件，或者具有不同的部件布置。
一种计算机设备，包括存储器和一个或多个处理器，存储器中储存有计算机可读指令，计算机可读指令被处理器执行时实现本申请任意一个实施例中提供的用户集群的预测方法的步骤。
一个或多个存储有计算机可读指令的计算机可读存储介质，所述计算机可读存储介质可以是非易失性，也可以是易失性，计算机可读指令被一个或多个处理器执行时，使得一个或多个处理器实现本申请任意一个实施例中提供的用户集群的预测方法的步骤。
本申请所指区块链是分布式数据存储、点对点传输、共识机制、加密算法等计算机技术的新型应用模式。区块链(Blockchain)，本质上是一个去中心化的数据库，是一串使用密码学方法相关联产生的数据块，每一个数据块中包含了一批次网络交易的信息，用于验 证其信息的有效性(防伪)和生成下一个区块。区块链可以包括区块链底层平台、平台产品服务层以及应用服务层等。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，是可以通过计算机可读指令来指令相关的硬件来完成，所述的计算机可读指令可存储于一非易失性计算机可读取存储介质中，该计算机可读指令在执行时，可包括如上述各方法的实施例的流程。其中，本申请所提供的各实施例中所使用的对存储器、存储、数据库或其它介质的任何引用，均可包括非易失性和易失性存储器中的至少一种。非易失性存储器可包括只读存储器(Read-Only Memory，ROM)、磁带、软盘、闪存或光存储器等。易失性存储器可包括随机存取存储器(Random Access Memory，RAM)或外部高速缓冲存储器。作为说明而非局限，RAM可以是多种形式，比如静态随机存取存储器(Static Random Access Memory，SRAM)或动态随机存取存储器(Dynamic Random Access Memory，DRAM)等。
以上实施例的各技术特征可以进行任意的组合，为使描述简洁，未对上述实施例中的各个技术特征所有可能的组合都进行描述，然而，只要这些技术特征的组合不存在矛盾，都应当认为是本说明书记载的范围。
以上所述实施例仅表达了本申请的几种实施方式，其描述较为具体和详细，但并不能因此而理解为对发明专利范围的限制。应当指出的是，对于本领域的普通技术人员来说，在不脱离本申请构思的前提下，还可以做出若干变形和改进，这些都属于本申请的保护范围。因此，本申请专利的保护范围应以所附权利要求为准。

Claims (20)

  1. 一种用户集群的预测方法，包括：
    获取初始用户集群中的各个用户的特征编码；所述特征编码包括多个用户特征维度的用户信息特征编码；
    分别将所述各个用户的多个用户特征维度的用户信息特征编码输入对应的预先训练的用户预测模型中，得到各个所述用户预测模型输出的预测用户集群；及
    对各个所述用户预测模型输出的预测用户集群进行融合处理，得到所述初始用户集群对应的目标用户集群。
  2. 根据权利要求1所述的方法，其中，所述获取初始用户集群中的各个用户的特征编码，包括：
    获取所述初始用户集群中的各个用户的多个用户特征维度的用户信息；
    对所述各个用户的多个用户特征维度的用户信息进行编码处理，得到所述各个用户的多个用户特征维度的用户信息特征编码；及
    对所述各个用户的多个用户特征维度的用户信息特征编码进行拼接处理，得到所述各个用户的特征编码。
  3. 根据权利要求1所述的方法，其中，所述分别将所述各个用户的多个用户特征维度的用户信息特征编码输入对应的用户预测模型中，得到各个所述用户预测模型输出的预测用户集群，包括：
    查询预设的用户特征维度与用户预测模型的对应关系，得到与所述多个用户特征维度一一对应的用户预测模型；
    分别将所述各个用户的多个用户特征维度的用户信息特征编码输入与所述多个用户特征维度一一对应的用户预测模型，得到各个所述用户预测模型对所述各个用户的预测结果；及
    根据各个所述用户预测模型对所述各个用户的预测结果，得到各个所述用户预测模型输出的预测用户集群。
  4. 根据权利要求3所述的方法，其中，所述根据各个所述用户预测模型对所述各个用户的预测结果，得到各个所述用户预测模型输出的预测用户集群，包括：
    从各个所述用户预测模型对所述各个用户的预测结果中，提取出各个所述用户预测模型对所述各个用户的预测概率；
    分别从所述各个用户中，筛选出所述预测概率大于预设概率的用户，对应作为各个所述用户预测模型输出的目标用户；及
    获取各个所述用户预测模型输出的目标用户所构成的集群，对应作为各个所述用户预测模型输出的预测用户集群。
  5. 根据权利要求1所述的方法，其中，所述预先训练的用户预测模型通过下述方式训练得到：
    获取样本用户训练集；所述样本用户训练集包括样本用户的各个特征维度的用户信息以及所述样本用户的实际概率；
    对所述样本用户的各个特征维度的用户信息进行编码处理，得到所述样本用户的各个特征维度的用户信息特征编码；
    分别将所述样本用户的各个特征维度的用户信息特征编码输入对应的各个用户预测模型中，得到所述各个用户预测模型中对所述样本用户的预测概率；
    根据所述各个用户预测模型中对所述样本用户的预测概率以及所述样本用户的实际概率，统计所述各个用户预测模型的损失值；
    根据所述各个用户预测模型的损失值，对所述各个用户预测模型进行反向训练，直至所述各个用户预测模型满足收敛条件；及
    若所述各个用户预测模型满足收敛条件，则将所述各个用户预测模型，对应作为各个预先训练的用户预测模型。
  6. 根据权利要求1至5任一项所述的方法，其中，在对各个所述用户预测模型输出的预测用户集群进行融合处理，得到所述初始用户集群对应的目标用户集群之后，所述方法还包括：
    获取所述目标用户集群中的各个目标用户的信用分数；
    若所述信用分数大于或者等于预设分数，则获取与所述信用分数对应的资源类型；
    将所述资源类型对应的资源推送给对应的目标用户；
    若所述信用分数小于所述预设分数，则生成与所述信用分数对应的风险提醒信息；及
    将所述风险提醒信息推送给对应的目标用户。
  7. 根据权利要求1至5任一项所述的方法，其中，在对各个所述用户预测模型输出的预测用户集群进行融合处理，得到所述初始用户集群对应的目标用户集群之后，所述方法还包括：
    将所述目标用户集群上传至区块链中。
  8. 一种用户集群的预测装置，包括：
    特征编码获取模块，用于获取初始用户集群中的各个用户的特征编码；所述特征编码包括多个用户特征维度的用户信息特征编码；
    预测用户集群获取模块，用于分别将所述各个用户的多个用户特征维度的用户信息特征编码输入对应的预先训练的用户预测模型中，得到各个所述用户预测模型输出的预测用户集群；及
    目标用户集群获取模块，用于对各个所述用户预测模型输出的预测用户集群进行融合处理，得到所述初始用户集群对应的目标用户集群。
  9. 一种计算机设备，包括存储器及一个或多个处理器，所述存储器中储存有计算机可读指令，所述计算机可读指令被所述一个或多个处理器执行时，使得所述一个或多个处理器执行以下步骤：
    获取初始用户集群中的各个用户的特征编码；所述特征编码包括多个用户特征维度的用户信息特征编码；
    分别将所述各个用户的多个用户特征维度的用户信息特征编码输入对应的预先训练的用户预测模型中，得到各个所述用户预测模型输出的预测用户集群；及
    对各个所述用户预测模型输出的预测用户集群进行融合处理，得到所述初始用户集群对应的目标用户集群。
  10. 根据权利要求9所述的计算机设备，其中，所述处理器执行所述计算机可读指令时还执行以下步骤：
    获取所述初始用户集群中的各个用户的多个用户特征维度的用户信息；
    对所述各个用户的多个用户特征维度的用户信息进行编码处理，得到所述各个用户的多个用户特征维度的用户信息特征编码；及
    对所述各个用户的多个用户特征维度的用户信息特征编码进行拼接处理，得到所述各个用户的特征编码。
  11. 根据权利要求9所述的计算机设备，其中，所述处理器执行所述计算机可读指令时还执行以下步骤：
    查询预设的用户特征维度与用户预测模型的对应关系，得到与所述多个用户特征维度一一对应的用户预测模型；
    分别将所述各个用户的多个用户特征维度的用户信息特征编码输入与所述多个用户特征维度一一对应的用户预测模型，得到各个所述用户预测模型对所述各个用户的预测结果；及
    根据各个所述用户预测模型对所述各个用户的预测结果，得到各个所述用户预测模型输出的预测用户集群。
  12. 根据权利要求11所述的计算机设备，其中，所述处理器执行所述计算机可读指令时还执行以下步骤：
    从各个所述用户预测模型对所述各个用户的预测结果中，提取出各个所述用户预测模型对所述各个用户的预测概率；
    分别从所述各个用户中，筛选出所述预测概率大于预设概率的用户，对应作为各个所述用户预测模型输出的目标用户；及
    获取各个所述用户预测模型输出的目标用户所构成的集群，对应作为各个所述用户预测模型输出的预测用户集群。
  13. 根据权利要求9所述的计算机设备，其中，所述处理器执行所述计算机可读指令时还执行以下步骤：
    获取样本用户训练集；所述样本用户训练集包括样本用户的各个特征维度的用户信息以及所述样本用户的实际概率；
    对所述样本用户的各个特征维度的用户信息进行编码处理，得到所述样本用户的各个 特征维度的用户信息特征编码；
    分别将所述样本用户的各个特征维度的用户信息特征编码输入对应的各个用户预测模型中，得到所述各个用户预测模型中对所述样本用户的预测概率；
    根据所述各个用户预测模型中对所述样本用户的预测概率以及所述样本用户的实际概率，统计所述各个用户预测模型的损失值；
    根据所述各个用户预测模型的损失值，对所述各个用户预测模型进行反向训练，直至所述各个用户预测模型满足收敛条件；及
    若所述各个用户预测模型满足收敛条件，则将所述各个用户预测模型，对应作为各个预先训练的用户预测模型。
  14. 根据权利要求9至13任一项所述的计算机设备，其中，所述处理器执行所述计算机可读指令时还执行以下步骤：
    获取所述目标用户集群中的各个目标用户的信用分数；
    若所述信用分数大于或者等于预设分数，则获取与所述信用分数对应的资源类型；
    将所述资源类型对应的资源推送给对应的目标用户；
    若所述信用分数小于所述预设分数，则生成与所述信用分数对应的风险提醒信息；及
    将所述风险提醒信息推送给对应的目标用户。
  15. 一个或多个存储有计算机可读指令的计算机可读存储介质，所述计算机可读指令被一个或多个处理器执行时，使得所述一个或多个处理器执行以下步骤：
    获取初始用户集群中的各个用户的特征编码；所述特征编码包括多个用户特征维度的用户信息特征编码；
    分别将所述各个用户的多个用户特征维度的用户信息特征编码输入对应的预先训练的用户预测模型中，得到各个所述用户预测模型输出的预测用户集群；及
    对各个所述用户预测模型输出的预测用户集群进行融合处理，得到所述初始用户集群对应的目标用户集群。
  16. 根据权利要求15所述的存储介质，其中，所述计算机可读指令被所述处理器执行时还执行以下步骤：
    获取所述初始用户集群中的各个用户的多个用户特征维度的用户信息；
    对所述各个用户的多个用户特征维度的用户信息进行编码处理，得到所述各个用户的多个用户特征维度的用户信息特征编码；及
    对所述各个用户的多个用户特征维度的用户信息特征编码进行拼接处理，得到所述各个用户的特征编码。
  17. 根据权利要求15所述的存储介质，其中，所述计算机可读指令被所述处理器执行时还执行以下步骤：
    查询预设的用户特征维度与用户预测模型的对应关系，得到与所述多个用户特征维度一一对应的用户预测模型；
    分别将所述各个用户的多个用户特征维度的用户信息特征编码输入与所述多个用户特征维度一一对应的用户预测模型，得到各个所述用户预测模型对所述各个用户的预测结果；及
    根据各个所述用户预测模型对所述各个用户的预测结果，得到各个所述用户预测模型输出的预测用户集群。
  18. 根据权利要求17所述的存储介质，其中，所述计算机可读指令被所述处理器执行时还执行以下步骤：
    从各个所述用户预测模型对所述各个用户的预测结果中，提取出各个所述用户预测模型对所述各个用户的预测概率；
    分别从所述各个用户中，筛选出所述预测概率大于预设概率的用户，对应作为各个所述用户预测模型输出的目标用户；及
    获取各个所述用户预测模型输出的目标用户所构成的集群，对应作为各个所述用户预测模型输出的预测用户集群。
  19. 根据权利要求15所述的存储介质，其中，所述计算机可读指令被所述处理器执行时还执行以下步骤：
    获取样本用户训练集；所述样本用户训练集包括样本用户的各个特征维度的用户信息以及所述样本用户的实际概率；
    对所述样本用户的各个特征维度的用户信息进行编码处理，得到所述样本用户的各个特征维度的用户信息特征编码；
    分别将所述样本用户的各个特征维度的用户信息特征编码输入对应的各个用户预测模型中，得到所述各个用户预测模型中对所述样本用户的预测概率；
    根据所述各个用户预测模型中对所述样本用户的预测概率以及所述样本用户的实际概率，统计所述各个用户预测模型的损失值；
    根据所述各个用户预测模型的损失值，对所述各个用户预测模型进行反向训练，直至所述各个用户预测模型满足收敛条件；及
    若所述各个用户预测模型满足收敛条件，则将所述各个用户预测模型，对应作为各个预先训练的用户预测模型。
  20. 根据权利要求15至19任一项所述的存储介质，其中，所述计算机可读指令被所述处理器执行时还执行以下步骤：
    获取所述目标用户集群中的各个目标用户的信用分数；
    若所述信用分数大于或者等于预设分数，则获取与所述信用分数对应的资源类型；
    将所述资源类型对应的资源推送给对应的目标用户；
    若所述信用分数小于所述预设分数，则生成与所述信用分数对应的风险提醒信息；及
    将所述风险提醒信息推送给对应的目标用户。
PCT/CN2020/134579 2020-06-24 2020-12-08 用户集群的预测方法、装置、计算机设备和存储介质 WO2021115269A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010586411.9A CN111737319B (zh) 2020-06-24 2020-06-24 用户集群的预测方法、装置、计算机设备和存储介质
CN202010586411.9 2020-06-24

Publications (1)

Publication Number Publication Date
WO2021115269A1 true WO2021115269A1 (zh) 2021-06-17

Family

ID=72650892

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/134579 WO2021115269A1 (zh) 2020-06-24 2020-12-08 用户集群的预测方法、装置、计算机设备和存储介质

Country Status (2)

Country Link
CN (1) CN111737319B (zh)
WO (1) WO2021115269A1 (zh)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111737319B (zh) * 2020-06-24 2024-03-01 平安科技（深圳）有限公司 用户集群的预测方法、装置、计算机设备和存储介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105868847A (zh) * 2016-03-24 2016-08-17 车智互联（北京）科技有限公司 一种购物行为的预测方法及装置
US20180158552A1 (en) * 2016-12-01 2018-06-07 University Of Southern California Interpretable deep learning framework for mining and predictive modeling of health care data
CN110348581A (zh) * 2019-06-19 2019-10-18 平安科技（深圳）有限公司 用户特征群中用户特征寻优方法、装置、介质及电子设备
US20200151396A1 (en) * 2018-01-31 2020-05-14 Jungle Disk, L.L.C. Natural language generation using pinned text and multiple discriminators
CN111737319A (zh) * 2020-06-24 2020-10-02 平安科技（深圳）有限公司 用户集群的预测方法、装置、计算机设备和存储介质

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107292528A (zh) * 2017-06-30 2017-10-24 阿里巴巴集团控股有限公司 车险风险预测方法、装置及服务器
CN109948633A (zh) * 2017-12-20 2019-06-28 广东欧珀移动通信有限公司 用户性别预测方法、装置、存储介质及电子设备
CN111259256B (zh) * 2020-02-13 2023-09-12 深圳市雅阅科技有限公司 内容处理方法、装置、计算机可读存储介质和计算机设备

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105868847A (zh) * 2016-03-24 2016-08-17 车智互联（北京）科技有限公司 一种购物行为的预测方法及装置
US20180158552A1 (en) * 2016-12-01 2018-06-07 University Of Southern California Interpretable deep learning framework for mining and predictive modeling of health care data
US20200151396A1 (en) * 2018-01-31 2020-05-14 Jungle Disk, L.L.C. Natural language generation using pinned text and multiple discriminators
CN110348581A (zh) * 2019-06-19 2019-10-18 平安科技（深圳）有限公司 用户特征群中用户特征寻优方法、装置、介质及电子设备
CN111737319A (zh) * 2020-06-24 2020-10-02 平安科技（深圳）有限公司 用户集群的预测方法、装置、计算机设备和存储介质

Also Published As

Publication number Publication date
CN111737319A (zh) 2020-10-02
CN111737319B (zh) 2024-03-01

Similar Documents

Publication Publication Date Title
EP3985578A1 (en) Method and system for automatically training machine learning model
CN116569189A (zh) 被配置为在机器学习模型中实现动态离群值偏倚减少的基于计算机的系统、计算组件和计算对象
US12086548B2 (en) Event extraction from documents with co-reference
US20220100772A1 (en) Context-sensitive linking of entities to private databases
CN112256886B (zh) 图谱中的概率计算方法、装置、计算机设备及存储介质
US11727031B2 (en) Systems and methods for formatting data using a recurrent neural network
WO2021139432A1 (zh) 基于人工智能的用户等级预测方法、装置、终端及介质
WO2021115269A1 (zh) 用户集群的预测方法、装置、计算机设备和存储介质
CN114781532A (zh) 机器学习模型的评估方法、装置、计算机设备及介质
CN117113222A (zh) 数据分析模型的生成方法、生成装置和电子设备
CN115952266A (zh) 问题生成方法、装置、计算机设备和存储介质
CN114692012A (zh) 一种基于Bert的神经协同过滤的电子政务推荐方法
CN114898184A (zh) 模型训练方法、数据处理方法、装置及电子设备
CN114154048A (zh) 构建推荐模型的方法、装置、电子设备及存储介质
CN118245638B (zh) 基于泛化模型的图数据预测方法、装置、设备及存储介质
CN116050508B (zh) 神经网络训练方法以及装置
CN115169321B (zh) 物流内容文本核对方法、装置、电子设备和计算机介质
US12147447B2 (en) Systems and methods for formatting data using a recurrent neural network
CN118656685B (zh) 一种衍生特征提取方法、装置、计算机设备及存储介质
CN117151247B (zh) 机器学习任务建模的方法、装置、计算机设备和存储介质
US20240338386A1 (en) Smart data signals for artificial intelligence based modeling
CN115630973A (zh) 用户数据处理方法、装置、计算机设备和存储介质
CN116957063A (zh) 一种样本预测方法、装置、设备及计算机可读存储介质
CN116756668A (zh) 资源转移异常检测方法、装置、计算机设备和存储介质
CN116091209A (zh) 信贷业务处理方法、装置、计算机设备和存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20899593

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20899593

Country of ref document: EP

Kind code of ref document: A1