WO2017215403A1 - 一种用户信用评估方法、装置及存储介质 - Google Patents

一种用户信用评估方法、装置及存储介质 Download PDF

Info

Publication number
WO2017215403A1
WO2017215403A1 PCT/CN2017/085049 CN2017085049W WO2017215403A1 WO 2017215403 A1 WO2017215403 A1 WO 2017215403A1 CN 2017085049 W CN2017085049 W CN 2017085049W WO 2017215403 A1 WO2017215403 A1 WO 2017215403A1
Authority
WO
WIPO (PCT)
Prior art keywords
user
real
feature information
offline
time
Prior art date
Application number
PCT/CN2017/085049
Other languages
English (en)
French (fr)
Inventor
陈培炫
陈谦
陈玲
Original Assignee
腾讯科技（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 腾讯科技（深圳）有限公司 filed Critical 腾讯科技（深圳）有限公司
Priority to JP2018543338A priority Critical patent/JP6732034B2/ja
Priority to EP17812514.2A priority patent/EP3471046A1/en
Priority to KR1020187029224A priority patent/KR102178633B1/ko
Publication of WO2017215403A1 publication Critical patent/WO2017215403A1/zh
Priority to US15/954,710 priority patent/US20180232805A1/en

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
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0637—Strategic management or analysis, e.g. setting a goal or target of an organisation; Planning actions based on goals; Analysis or evaluation of effectiveness of goals
    • 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
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/02—Knowledge representation; Symbolic representation
    • G06N5/022—Knowledge engineering; Knowledge acquisition
    • 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
    • G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
    • G06Q40/03—Credit; Loans; Processing thereof

Definitions

  • the present application relates to the field of Internet technologies, and in particular, to a user credit evaluation method, apparatus, and storage medium.
  • the credit evaluation method for users is usually to collect the user's personal information, and then predict the user default risk through statistical models or some prediction algorithms of machine learning, such as the commonly used FICO credit scoring system and the Zestfinace credit evaluation system.
  • the personal information (big data) used in the existing credit scoring mechanism is usually updated according to a preset update period, and the update period is generally one month or longer, and the user's situation can be referred to in the next update.
  • the lag in information has a very large impact on the accuracy of user credit evaluation.
  • This application example provides a user credit evaluation method, which includes:
  • the offline feature information is feature information of the user updated according to a preset update period
  • the real-time feature information is feature information of the user collected within a current preset time range, and the preset time range is smaller than the preset update period;
  • the comprehensive credit score of the target user is calculated according to the obtained offline credit score and real-time credit score of the target user combined with the preset comprehensive prediction model.
  • the application example further provides a user credit evaluation apparatus, and the apparatus includes:
  • An offline feature obtaining module configured to acquire offline feature information of the target user, where the offline feature information is feature information of the user updated according to a preset update period;
  • An offline scoring module configured to calculate an offline credit score of the target user according to the offline feature information of the target user and the preset offline prediction model
  • the real-time feature acquisition module is configured to acquire real-time feature information of the target user, where the real-time feature information is feature information of the user collected within a current preset time range, and the preset time range is smaller than the preset update period;
  • a real-time scoring module configured to calculate a real-time credit score of the target user according to the real-time feature information of the target user and the preset real-time prediction model;
  • the comprehensive scoring module is configured to calculate a comprehensive credit score of the target user according to the obtained offline credit score and real-time credit score of the target user combined with the preset comprehensive prediction model.
  • the application example also provides a computer readable storage medium, which is stored in a computer readable storage
  • the instructions may cause at least one processor to perform the method as described above.
  • FIG. 1 is a schematic flow chart of a user credit evaluation method in an example of the present application.
  • FIG. 2 is a schematic diagram of obtaining source information of real-time feature information and offline feature information of a user in an example of the present application;
  • FIG. 3 is a schematic flowchart of training an offline prediction model in an example of the present application.
  • FIG. 4 is a schematic flowchart of training a real-time prediction model in an example of the present application
  • FIG. 5 is a schematic flowchart of training the comprehensive prediction model in the example of the present application.
  • FIG. 6 is a schematic structural diagram of a user credit evaluation apparatus in an example of the present application.
  • FIG. 7 is a schematic structural diagram of a sample acquisition module in an example of the present application.
  • the user credit evaluation method and apparatus in the example of the present application can be implemented in, for example, a personal electric In computer systems such as brains, laptops, smart phones, tablets, and e-readers, many of them can be used in servers that provide user credit evaluation, such as back-end servers of data service platforms.
  • servers that provide user credit evaluation such as back-end servers of data service platforms.
  • the following is a description of the user credit evaluation apparatus as an execution subject of the example of the present application.
  • FIG. 1 is a schematic flowchart of a user credit evaluation method in an example of the present application. As shown in the figure, the user credit evaluation method process in this example may include:
  • the offline feature information is as shown in FIG. 2, and the user credit evaluation device may be obtained by collecting user data provided by a third party, or may be obtained from user data collected by the service platform.
  • the user credit evaluation apparatus may perform feature calculation on the user data obtained above, and convert the user attribute, user behavior or user attribute/behavior change in the user data into offline feature information in a unified format, such as digitized feature information.
  • the preset update period may be an update period in which the external manufacturer provides the user data, or may be an acquisition update period set by the user credit evaluation device. Since the big data involves a large user base, the offline feature information may include all the historical feature information of the user, and the amount of data is huge. Therefore, the preset update period is generally long, usually at least one week to one month.
  • the offline feature information may be more stable feature information of the user, such as gender, age, place of origin, occupation, income status, and the like, and may also include all historical contract credit records, which are generally stable for this type.
  • the user feature information only needs to be updated according to the preset update period, so the information of these feature categories is taken as offline feature information.
  • the offline feature information may be offline feature information of the filtered feature category, that is, the user data provided by the third party or the offline data of the user data collected by the service platform may include multiple feature categories.
  • User credit evaluation device can be from Filter out offline feature information for the specified feature category.
  • the specified feature category may be obtained by the user credit evaluation device according to the preset training sample data, where the training sample data includes samples of credit score results of multiple users and offline feature information samples of multiple feature categories of each user.
  • the training sample data is also referred to as user credit data.
  • the user credit evaluation device calculates the correlation between each feature category and the credit score result according to the credit score result samples of the plurality of users in the user credit sample data and the offline feature information samples of the plurality of feature categories of each user, thereby The feature category that has a correlation with the credit score result that reaches a preset threshold is determined as the specified feature category.
  • the offline prediction model may be a trained logistic regression classification model, or may be a trained integrated learning model, a deep learning model, a random forest model, and the like.
  • the user credit evaluation device substitutes the offline feature information of the target user into the preset offline prediction model, and then calculates the offline credit score of the target user.
  • the offline prediction model may be obtained by the user credit evaluation device according to the preset training sample data, and the training sample data may include a credit score result sample of the plurality of users and offline feature information of each user; the offline prediction model It may also be a trained offline prediction model obtained from the outside by the user credit evaluation device.
  • the real-time feature information is as shown in FIG. 2, and the user credit evaluation device can be obtained by using the user data collected by the service platform.
  • the user credit evaluation device can perform the feature calculation on the obtained user data, and convert the user attribute, user behavior or user attribute/behavior change in the user data into a real-time feature information in a unified format, such as a digitized feature. information.
  • the service platform may collect the latest feature information of the user, where the preset time range is smaller than the preset update period, for example, the feature information of the user collected in the last day, two days, or one week.
  • the user credit evaluation device may pre-set some feature categories as high-risk features.
  • the credit score of the user may be greatly affected. For example, if the user handles a specific platform lending business, handles a visa business abroad, changes in the geographical location, or a large amount of consumption in a specific field, the user credit evaluation device may have corresponding feature information for these high-risk features that require real-time attention.
  • the real-time feature information of the user is collected and recorded in real time, and the other feature information is updated as the offline feature information for the preset update period.
  • the real-time feature information may be real-time feature information of the filtered feature category, that is, the user data collected by the service platform may include real-time feature information of multiple feature categories, and the user credit evaluation device may filter out the specified feature.
  • the specified feature category may be a user credit evaluation device according to preset training sample data, where the training sample data includes a plurality of user credit score result samples and real-time feature information samples of multiple feature categories of each user,
  • the training sample data is also referred to as user credit sample data.
  • the user credit evaluation device calculates the correlation between each feature category and the credit score result according to the credit score result samples of the plurality of users in the user credit sample data and the real-time feature information samples of the plurality of feature categories of each user, thereby The feature category that has a correlation with the credit score result that reaches a preset threshold is determined as the specified feature category.
  • the real-time prediction model may be a trained logistic regression classification model, or a trained integrated learning model, a deep learning model, a random forest model, and the like.
  • the real-time credit score of the target user can be calculated by substituting the real-time feature information of the target user into the preset real-time prediction model by using the evaluation device.
  • the real-time prediction model may be obtained by the user credit evaluation device according to the preset training sample data, and the training sample data may include a credit score result sample of the plurality of users and offline feature information of each user; the real-time prediction model It may also be a trained real-time prediction model obtained from the outside by the user credit evaluation device.
  • the comprehensive prediction model may be a trained logistic regression classification model, or may be a trained integrated learning model, a deep learning model, a random forest model, a gradient lifting decision tree model, and the like.
  • the user credit evaluation device substitutes the offline credit score and the real-time credit score of the target user into the preset real-time prediction model, and the real-time credit score of the target user can be calculated.
  • the comprehensive prediction model may be obtained by the user credit evaluation device according to the preset training sample data, and the training sample data may include a plurality of user credit score result samples and off-line feature information and real-time feature information of each user, the user
  • the credit evaluation device obtains the offline credit score of each user according to the offline feature information of the user by using the offline prediction model, and obtains the real-time credit score of each user according to the real-time feature information of the user by using the real-time prediction model, according to the credit of the plurality of users
  • the comprehensive prediction model is trained by the scoring results and the offline credit score and real-time credit score of each user.
  • the real-time prediction model may also be a trained real-time prediction model obtained from the outside by the user credit evaluation device.
  • the real-time credit score of the target user can be calculated by using the comprehensive prediction model of the following logistic regression algorithm:
  • ⁇ , ⁇ , ⁇ are the parameters obtained by the training model
  • Score1 and Score2 are the offline credit score and real-time credit score of the target user respectively
  • the score is the comprehensive credit score of the target user.
  • the user credit evaluation apparatus may push product information, such as push financial product information, fixed asset management product information, etc., to the target user according to the comprehensive credit score of the target user calculated after the above steps of the example; or
  • the target user's comprehensive credit score monitors and manages the data service of the target user, such as risk management of the target user's lending business, and management advice on the target user's liquidity.
  • the offline credit score and the real-time credit score of the user are respectively calculated by acquiring the offline feature information and the real-time feature information of the user, thereby calculating the comprehensive credit score of the user, and realizing the long-term characteristic data of the combined user.
  • real-time feature data accurately predicts the user's credit status, and solves the problem of inaccurate credit estimation caused by user information lag in the prior art.
  • FIG. 3 is a schematic flowchart of training an offline prediction model in the example of the present application.
  • the offline prediction model training process in this example may include the following steps:
  • the credit score result samples of the plurality of users and the offline feature information samples of the plurality of feature categories of the respective users may be extracted according to the training sample data input to the user credit evaluation apparatus.
  • the plurality of users' credit score result samples may be calculated by using the default records of the plurality of users, that is, determining the credit scores of the plurality of users according to whether the plurality of users default, or the number and severity of the default events, and the like. Results sample.
  • the sample of the credit score results of the multiple users may also be obtained by manual scoring.
  • the user credit evaluation device may collect the The offline feature information samples of multiple feature categories of each user are obtained from the user data provided by the third party or the user data collected by the service platform.
  • the user's credit sample data includes samples of credit score results of a plurality of users and offline feature information samples of a plurality of feature categories of respective users.
  • the feature category may be, for example, age, location, gender, occupation, etc., the correlation between the feature category and the credit score result, and the effect on the user credit score result such as age, gender, occupation, etc., if the correlation is high , indicating that the feature category has a greater impact on the credit score result, and vice versa, has little effect on the credit score result, and may not consider the offline feature information of the feature category when establishing the offline prediction model.
  • the correlation between the respective feature categories and the credit score result may be calculated by using the following formula exemplarily:
  • x is the offline feature information of a certain feature category
  • y is the credit score result of the user.
  • the subscript i indicates that it corresponds to a different user.
  • S303 Determine, as a feature category of the offline feature information, a feature category that has a correlation with a credit score result that reaches a preset threshold, and select a corresponding identifier from offline feature information samples of the multiple feature categories of each user. Offline feature information for feature categories.
  • the offline prediction model may be a trained logistic regression classification model, or may be a trained integrated learning model, a deep learning model, a random forest model, and the like.
  • the offline prediction model may be a prediction formula in which the user credit evaluation device calculates the credit score of the user according to the offline feature information of the filtered corresponding feature category, and the credit score result sample of the plurality of users.
  • the offline feature information of the corresponding feature category of each user is trained and iterated to the model parameters in the prediction algorithm, so that the model parameters of the prediction formula closest to the credit score result sample can be obtained, thereby obtaining a trained offline prediction model.
  • the obtained offline feature information samples of multiple feature categories of each user may be used as offline feature information for real-time prediction models without screening. training.
  • the real-time prediction model training process in this example may include:
  • the sample of the credit score result of the plurality of users and the real-time feature information samples of the plurality of feature categories of each user may be extracted according to the training sample data input to the user credit evaluation apparatus.
  • the plurality of users' credit score result samples may be calculated by using the default records of the plurality of users, that is, determining the credit scores of the plurality of users according to whether the plurality of users default, or the number and severity of the default events, and the like. Results sample.
  • the sample of the credit score results of the multiple users may also be obtained by manual scoring.
  • the user credit evaluation apparatus may obtain real-time feature information samples of the plurality of feature categories of the respective users in the user data collected by the service platform.
  • S402. Calculate a correlation between each feature category and a credit score result according to the credit score result samples of the plurality of users and the real-time feature information samples of the plurality of feature categories of each user.
  • the user's credit sample data includes samples of credit score results of a plurality of users and offline feature information samples of a plurality of feature categories of respective users.
  • the feature category may be, for example, age, location, gender, occupation, etc., the correlation between the feature category and the credit score result, and the effect on the user credit score result such as age, gender, occupation, etc., if the correlation is high , indicating that the feature category has a greater impact on the credit score result, and vice versa, has little effect on the credit score result, and may not consider the real-time feature information of the feature category when establishing the real-time predictive model.
  • the correlation between the respective feature categories and the credit score result may be exemplarily calculated by using the following formula:
  • z is the real-time feature information of a feature category
  • y is the user's credit score result.
  • the subscript i indicates that it corresponds to a different user.
  • the correlation between the real-time feature information of each feature category and the credit score result may also be calculated by using an affinity algorithm such as an IV value and a chi-square value.
  • S403. Determine, as a feature category of the real-time feature information, a feature category that has a correlation with a credit score result that reaches a preset threshold, and select a corresponding one of the real-time feature information samples of the multiple feature categories of each user. Real-time feature information of feature categories.
  • S404 Establish a real-time prediction model according to the real-time feature information of the user corresponding to the filtered feature category, and train the real-time prediction model according to the credit score result samples of the multiple users and the real-time feature information of the corresponding feature categories of each user.
  • the real-time prediction model may be a trained logistic regression classification model, or a trained integrated learning model, a deep learning model, a random forest model, and the like.
  • the real-time prediction model may be a user credit evaluation device according to the selected corresponding feature category
  • the real-time feature information of the user is combined with the specific model parameter to calculate a prediction formula of the user's credit score, and the model parameters in the prediction formula are obtained by the credit score result samples of the plurality of users and the real-time feature information of the corresponding feature categories of each user.
  • the training iteration is performed so that the model parameters of the prediction formula closest to the sample of the credit score result can be obtained, thereby obtaining a trained real-time prediction model.
  • the foregoing S402 and S403 are optional steps.
  • the real-time feature information samples of the multiple feature categories of each user that are obtained without being filtered may be used as real-time feature information for real-time prediction models. training.
  • FIG. 5 is a schematic flow chart of training the comprehensive prediction model in the example of the present application.
  • the credit score result samples of the multiple users and the offline feature information and real-time feature information of each user may be extracted according to training sample data input to the user credit evaluation apparatus.
  • the plurality of users' credit score result samples may be calculated by using the default records of the plurality of users, that is, determining the credit scores of the plurality of users according to whether the plurality of users default, or the number and severity of the default events, and the like. Results sample.
  • the sample of the credit score results of the multiple users may also be obtained by manual scoring.
  • the user credit evaluation device may acquire the user data provided by the third party, or obtain the real-time of the multiple feature categories of each user by using the user data collected by the service platform. Feature information sample.
  • the comprehensive prediction model is established according to the offline credit score and the real-time credit score of the user, and the comprehensive prediction model is trained according to the credit score results of the multiple users and the offline credit score and the real-time credit score of each user.
  • the comprehensive prediction model may be a trained logistic regression classification model, or may be a trained integrated learning model, a deep learning model, a random forest model, a gradient lifting decision tree model, and the like.
  • the comprehensive prediction model may be a prediction formula for the user credit evaluation device to calculate a comprehensive credit score of the user according to the offline credit score and the real-time credit score of the user in combination with the specific model parameters, and the sample and calculation of the credit score result of the plurality of users.
  • the obtained offline credit score and real-time credit score of each user can perform training iteration on the model parameters in the prediction formula, so that the model parameters of the prediction formula closest to the credit score result sample can be obtained, thereby obtaining a trained comprehensive prediction model.
  • the real-time credit score of the target user can be calculated by using the comprehensive prediction model of the following logistic regression algorithm:
  • ⁇ , ⁇ , ⁇ are the model parameters obtained by the training model
  • Score1 and Score2 are the offline credit score and real-time credit score of the target user respectively
  • the score is the comprehensive credit score of the target user.
  • FIG. 6 is a schematic structural diagram of a user credit evaluation apparatus according to the present application.
  • the user credit evaluation apparatus in the example of the present application may include:
  • the offline feature acquisition module 610 is configured to acquire offline feature information of the target user, where the offline feature information is feature information of the user that is updated according to a preset update period.
  • the offline feature information is as shown in FIG. 2, and the offline feature obtaining module 610 may be obtained by collecting user data provided by a third party, or may be obtained by collecting user data collected by the service platform.
  • the offline feature obtaining module 610 can perform user function, user behavior or user attribute/line in the user data by performing feature calculation on the user data obtained above.
  • the changes are converted to offline feature information in a uniform format, such as digitized feature information.
  • the preset update period may be an update period of the user data provided by the external manufacturer, or may be an acquisition update period set by the offline feature acquisition module 610. Since the big data involves a large user base, the offline feature information may include all the historical feature information of the user, and the amount of data is huge.
  • the preset update period is generally long, usually at least one week to one month.
  • the offline feature information may be more stable feature information of the user, such as gender, age, place of origin, occupation, income status, and the like, and may also include all historical contract credit records, which are generally stable for this type.
  • the user feature information only needs to be updated according to the preset update period, so the information of these feature categories is taken as offline feature information.
  • the offline feature information may be offline feature information of the filtered feature category, that is, the user data provided by the third party or the offline data of the user data collected by the service platform may include multiple feature categories.
  • the offline feature acquisition module 610 can filter out offline feature information of the specified feature category.
  • the specified feature category may be obtained by the user credit evaluation device according to the preset training sample data, where the training sample data includes samples of credit score results of multiple users and offline feature information samples of multiple feature categories of each user.
  • the training sample data is also referred to as user credit sample data.
  • the user credit evaluation device calculates the correlation between each feature category and the credit score result according to the credit score result samples of the plurality of users in the user credit sample data and the offline feature information samples of the plurality of feature categories of each user, thereby The feature category that has a correlation with the credit score result that reaches a preset threshold is determined as the specified feature category.
  • the offline scoring module 620 is configured to calculate an offline credit score of the target user according to the offline feature information of the target user and the preset offline prediction model.
  • the offline prediction model may be a trained logistic regression classification model, or may be a trained integrated learning model, a deep learning model, a random forest model, and the like.
  • Offline review The sub-module 620 substitutes the offline feature information of the target user into the preset offline prediction model, and then calculates the offline credit score of the target user.
  • the offline prediction model may be obtained by the user credit evaluation device according to the preset training sample data, and the training sample data may include a credit score result sample of the plurality of users and offline feature information of each user; the offline prediction model It may also be a trained offline prediction model obtained from the outside by the user credit evaluation device.
  • the real-time feature acquisition module 630 is configured to acquire real-time feature information of the target user, where the real-time feature information is feature information of the user collected within a current preset time range, and the preset time range is smaller than the preset update period. .
  • the real-time feature information is as shown in FIG. 2, and the real-time feature obtaining module 630 can obtain the user data collected by the service platform.
  • the real-time feature acquisition module 630 can convert the user attribute, user behavior, or user attribute/behavior change in the user data into real-time feature information in a unified format, such as digitized feature information, by performing feature calculation on the obtained user data.
  • the service platform may collect the latest feature information of the user, where the preset time range is smaller than the preset update period, for example, the feature information of the user collected in the last day, two days, or one week.
  • the user credit evaluation device may pre-set some feature categories as high-risk features.
  • the real-time feature acquisition module 630 can perform corresponding features for these high-risk features that require real-time attention.
  • the information is collected and recorded as real-time feature information of the user in real time, and the other feature information is updated as the offline feature information for the preset update period.
  • the real-time feature information may be real-time feature information of the filtered feature category, that is, the user data provided by the third party or the user data collected by the service platform may be
  • the real-time feature information includes a plurality of feature categories
  • the real-time feature acquisition module 630 can filter out real-time feature information of the specified feature category.
  • the specified feature category may be a user credit evaluation device according to preset training sample data, where the training sample data includes a plurality of user credit score result samples and real-time feature information samples of multiple feature categories of each user,
  • the training sample data is also referred to as user credit sample data.
  • the user credit evaluation device calculates the correlation between each feature category and the credit score result according to the credit score result samples of the plurality of users in the user credit sample data and the real-time feature information samples of the plurality of feature categories of each user, thereby The feature category that has a correlation with the credit score result that reaches a preset threshold is determined as the specified feature category.
  • the real-time scoring module 640 is configured to calculate a real-time credit score of the target user according to the real-time feature information of the target user and the preset real-time prediction model.
  • the real-time prediction model may be a trained logistic regression classification model, or a trained integrated learning model, a deep learning model, a random forest model, and the like.
  • the user credit evaluation device substitutes the real-time feature information of the target user into the preset real-time prediction model, and then calculates the real-time credit score of the target user.
  • the real-time prediction model may be obtained by the user credit evaluation device according to the preset training sample data, and the training sample data may include a credit score result sample of the plurality of users and offline feature information of each user; the real-time prediction model It may also be a trained real-time prediction model obtained from the outside by the user credit evaluation device.
  • the comprehensive scoring module 650 is configured to calculate a comprehensive credit score of the target user according to the obtained offline credit score and real-time credit score of the target user and the preset comprehensive prediction model.
  • the comprehensive prediction model may be a trained logistic regression classification model, or may be a trained integrated learning model, a deep learning model, a random forest model, a gradient lifting decision tree model, and the like.
  • the comprehensive scoring module 650 substitutes the offline credit score and the real-time credit score of the target user into the preset real-time prediction model, and then the target user is calculated. Time credit score.
  • the comprehensive prediction model may be obtained by the user credit evaluation device according to the preset training sample data, and the training sample data may include a plurality of user credit score result samples and off-line feature information and real-time feature information of each user, the user
  • the credit evaluation device obtains the offline credit score of each user according to the offline feature information of the user by using the offline prediction model, and obtains the real-time credit score of each user according to the real-time feature information of the user by using the real-time prediction model, according to the credit of the plurality of users
  • the comprehensive prediction model is trained by the scoring results and the offline credit score and real-time credit score of each user.
  • the real-time prediction model may also be a trained real-time prediction model obtained from the outside by the user credit evaluation device.
  • the real-time credit score of the target user can be calculated by using the comprehensive prediction model of the following logistic regression algorithm:
  • ⁇ , ⁇ , ⁇ are the parameters obtained by the training model
  • Score1 and Score2 are the offline credit score and real-time credit score of the target user respectively
  • the score is the comprehensive credit score of the target user.
  • the offline credit score and the real-time credit score of the user are respectively calculated by acquiring the offline feature information and the real-time feature information of the user, thereby calculating the comprehensive credit score of the user, and realizing the long-term characteristic data of the combined user.
  • real-time feature data accurately predicts the user's credit status, and solves the problem of inaccurate credit estimation caused by user information lag in the prior art.
  • the user credit evaluation apparatus may further include:
  • the sample obtaining module 660 is configured to obtain a sample of credit score results of multiple users and offline feature information of each user.
  • the sample of the credit score result of the multiple users and the offline feature of each user can be extracted from the training sample data input to the user credit evaluation device.
  • the plurality of users' credit score result samples may be calculated by using the default records of the plurality of users, that is, determining the credit scores of the plurality of users according to whether the plurality of users default, or the number and severity of the default events, and the like. Results sample.
  • the sample of the credit score results of the multiple users may also be obtained by manual scoring.
  • the sample obtaining module 660 may obtain the offline feature information of each user by collecting the user data provided by the third party or the user data collected by the service platform.
  • the offline model training module 670 is configured to establish the offline prediction model according to the offline feature information of the user, and train the offline prediction model according to the credit score result samples of the multiple users and the offline feature information of each user.
  • the offline prediction model may be a trained logistic regression classification model, or may be a trained integrated learning model, a deep learning model, a random forest model, and the like.
  • the offline prediction model may be a prediction formula for calculating a credit score of the user according to the offline feature information of the filtered corresponding feature category, and the offline model training module 670 passes the credit score result of the multiple users.
  • the offline feature information of the sample and the corresponding feature category of each user is trained and iterated to the model parameters in the prediction algorithm, so that the model parameters of the prediction formula closest to the sample of the credit score result can be obtained, thereby obtaining a trained offline prediction model.
  • the sample obtaining module 660 may further include:
  • the offline sample obtaining unit 661 is configured to acquire the credit score result samples of the plurality of users and the offline feature information samples of the plurality of feature categories of the respective users.
  • the correlation calculation unit 663 is configured to calculate a correlation between each feature category and a credit score result according to the credit score result samples of the plurality of users and the offline feature information samples of the plurality of feature categories of the respective users.
  • the feature category may be, for example, age, location, gender, occupation, etc., the correlation between the feature category and the credit score result, and the effect on the user credit score result such as age, gender, occupation, etc., if the correlation is high , indicating that the feature category has a greater impact on the credit score result, and vice versa, has little effect on the credit score result, and may not consider the offline feature information of the feature category when establishing the offline prediction model.
  • the correlation between the respective feature categories and the credit score result may be calculated by using the following formula exemplarily:
  • x is the offline feature information of a certain feature category
  • y is the credit score result of the user.
  • the subscript i indicates that it corresponds to a different user.
  • the correlation between the various feature categories and the credit score result may also be calculated by using an affinity algorithm such as an IV value and a chi-square value.
  • a feature category screening unit 665 configured to determine, as a feature category of the offline feature information, a feature category that reaches a preset threshold between the scores of the credit scores, and offline features of the plurality of feature categories of the respective users The offline feature information of the corresponding feature category is filtered out in the information sample.
  • the sample obtaining module 660 is configured to obtain a sample of credit score results of multiple users and real-time feature information of each user.
  • the credit score result samples of the multiple users and the real-time feature information of each user may be extracted in the training sample data input to the user credit evaluation apparatus.
  • the plurality of users' credit score result samples may be calculated by using the default records of the plurality of users, that is, determining the credit scores of the plurality of users according to whether the plurality of users default, or the number and severity of the default events, and the like. Results sample.
  • the sample of the credit score results of the multiple users may also be obtained by manual scoring.
  • the user credit evaluation device can obtain the real-time feature information of each user in the user data collected by the service platform.
  • the user credit evaluation apparatus may further include: a real-time model training module 680, configured to establish the real-time prediction model according to real-time feature information of the user, and according to the credit score result samples of the plurality of users and real-time feature information of each user The real-time prediction model is trained.
  • a real-time model training module 680 configured to establish the real-time prediction model according to real-time feature information of the user, and according to the credit score result samples of the plurality of users and real-time feature information of each user The real-time prediction model is trained.
  • the real-time prediction model may be a trained logistic regression classification model, or a trained integrated learning model, a deep learning model, a random forest model, and the like.
  • the real-time prediction model may be a prediction formula for the user credit evaluation device to calculate the user's credit score according to the real-time feature information of the user of the corresponding corresponding feature category that is filtered, and the real-time model training module 680 passes the plurality of users.
  • the sample of the credit score result and the real-time feature information of the corresponding feature category of each user are trained and iterated to the model parameters in the prediction formula, so that the model parameters of the prediction formula closest to the sample of the credit score result can be obtained, thereby obtaining the trained real-time. Forecast model.
  • the sample obtaining module further includes, as shown in FIG. 7,
  • the real-time sample obtaining unit 662 is configured to acquire a sample of the credit score result of the plurality of users and a real-time feature information sample of the plurality of feature categories of each user.
  • the correlation calculation unit 663 is configured to calculate each feature category according to the credit score result sample of the plurality of users and the real-time feature information samples of the plurality of feature categories of each user. The correlation between credit score results.
  • the feature category may be, for example, age, location, gender, occupation, etc., the correlation between the feature category and the credit score result, and the effect on the user credit score result such as age, gender, occupation, etc., if the correlation is high , indicating that the feature category has a greater impact on the credit score result, and vice versa, has little effect on the credit score result, and may not consider the real-time feature information of the feature category when establishing the real-time predictive model.
  • the correlation between the respective feature categories and the credit score result may be exemplarily calculated by using the following formula:
  • z is the real-time feature information of a feature category
  • y is the user's credit score result.
  • the subscript i indicates that it corresponds to a different user.
  • the correlation between the real-time feature information of each feature category and the credit score result may also be calculated by using an affinity algorithm such as an IV value and a chi-square value.
  • the feature category screening unit 665 is configured to determine, as a feature category of the real-time feature information, a feature category that reaches a preset threshold between the scores of the credit scores, and real-time features of the plurality of feature categories of the respective users The real-time feature information of the corresponding feature category is filtered out in the information sample.
  • the feature class The screening unit 665 may compare the corresponding preset thresholds, determine the feature categories of the real-time feature information by reaching the required feature categories, and select from the real-time feature information samples of the plurality of feature categories of the respective users. The real-time feature information of the corresponding feature category is filtered out.
  • the sample obtaining module 660 is configured to obtain a sample of credit score results of multiple users and offline feature information and real-time feature information of each user.
  • the credit score result samples of the multiple users and the offline feature information and real-time feature information of each user may be extracted according to training sample data input to the user credit evaluation apparatus.
  • the plurality of users' credit score result samples may be calculated by using the default records of the plurality of users, that is, determining the credit scores of the plurality of users according to whether the plurality of users default, or the number and severity of the default events, and the like. Results sample.
  • the sample of the credit score results of the multiple users may also be obtained by manual scoring.
  • the sample obtaining module 660 may obtain the user data provided by the third party, or obtain the plurality of feature categories of each user by using the user data collected by the service platform. A sample of real-time feature information.
  • the offline scoring module 620 is further configured to calculate offline credit scores of the respective users according to offline feature information of the respective users and a preset offline prediction model.
  • the real-time scoring module 640 is further configured to calculate real-time credit scores of the respective users according to real-time feature information of the respective users and a preset real-time prediction model.
  • the user credit evaluation apparatus may further include:
  • the comprehensive model training module 690 is configured to establish the comprehensive prediction model according to the offline credit score and the real-time credit score of the user, and according to the credit score results of the multiple users and the offline credit score and real-time credit score of each user.
  • the comprehensive prediction model is trained.
  • the comprehensive prediction model may be a trained logistic regression classification model, or may be a trained integrated learning model, a deep learning model, a random forest model, a gradient lifting decision tree model, and the like.
  • the comprehensive prediction model may be a prediction formula for the user credit evaluation device to calculate a comprehensive credit score of the user according to the offline credit score and the real-time credit score of the user in combination with the specific model parameters, and the sample and calculation of the credit score result of the plurality of users.
  • the obtained offline credit score and real-time credit score of each user can perform training iteration on the model parameters in the prediction formula, so that the model parameters of the prediction formula closest to the credit score result sample can be obtained, thereby obtaining a trained comprehensive prediction model.
  • the real-time credit score of the target user can be calculated by using the comprehensive prediction model of the following logistic regression algorithm:
  • ⁇ , ⁇ , ⁇ are the model parameters obtained by the training model
  • Score1 and Score2 are the offline credit score and real-time credit score of the target user respectively
  • the score is the comprehensive credit score of the target user.
  • the user credit evaluation apparatus may further include an information pushing module 6100 or a service monitoring module 6110, where:
  • the information pushing module 6100 is configured to push the product information to the target user according to the comprehensive credit score of the target user, that is, push the product information, such as pushing, according to the comprehensive credit score of the target user calculated by the comprehensive scoring module 650 of the example of the present application.
  • the service monitoring module 6110 is configured to monitor and manage the data service of the target user according to the comprehensive credit score of the target user, for example, performing risk control management on the target user's loan business, and managing the target user's working capital.
  • the user credit evaluation apparatus in the example of the present application calculates the offline credit score and the real-time credit score of the user by acquiring the offline feature information and the real-time feature information of the user. Therefore, the user's comprehensive credit score is calculated, and the user's long-term feature data and real-time feature data are combined to accurately predict the user's credit status, and the problem of inaccurate credit estimation caused by user information lag in the prior art is solved.
  • the storage medium may be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

Landscapes

  • Business, Economics & Management (AREA)
  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Human Resources & Organizations (AREA)
  • Economics (AREA)
  • Strategic Management (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Finance (AREA)
  • Accounting & Taxation (AREA)
  • Marketing (AREA)
  • General Business, Economics & Management (AREA)
  • Development Economics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Educational Administration (AREA)
  • General Engineering & Computer Science (AREA)
  • Software Systems (AREA)
  • Technology Law (AREA)
  • Game Theory and Decision Science (AREA)
  • Tourism & Hospitality (AREA)
  • Quality & Reliability (AREA)
  • Operations Research (AREA)
  • Mathematical Physics (AREA)
  • Artificial Intelligence (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Computing Systems (AREA)
  • Computational Linguistics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Financial Or Insurance-Related Operations Such As Payment And Settlement (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

本申请公开了一种用户信用评估方法和装置，其中的所述方法包括：获取目标用户的离线特征信息，所述离线特征信息为按照预设更新周期进行更新的用户的特征信息；根据目标用户的离线特征信息以及预设的离线预测模型，计算目标用户的离线信用评分；获取目标用户的实时特征信息，所述实时特征信息为距离当前预设时间范围内采集到的用户的特征信息，所述预设时间范围小于所述预设更新周期；根据目标用户的实时特征信息以及预设的实时预测模型，计算目标用户的实时信用评分；根据得到的目标用户的离线信用评分和实时信用评分结合预设的综合预测模型，计算目标用户的综合信用评分。本申请还提出了相应的装置及存储介质。

Description

一种用户信用评估方法、装置及存储介质
本申请要求于2016年6月12日提交中国专利局、申请号为201610416661.1、申请名称为“一种用户信用评估方法和装置”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
本申请涉及互联网技术领域，尤其涉及一种用户信用评估方法、装置及存储介质。
背景技术
近年来，随着互联网技术的飞速发展，人们越来越多的通过互联网进行各种数据业务，而用户的信用评估也成为了一个互联网技术领域的焦点问题。
现有技术中对用户的信用评估方式通常是通过收集用户的个人信息，然后通过统计模型或机器学习的一些预测算法，对用户违约风险进行预测，例如常用的FICO信用评分系统以及Zestfinace信用评价系统。现有的信用评分机制中采用的个人信息(大数据)通常都是按照预设更新周期进行更新，更新周期一般为一个月或更长，用户发生的状况要在下次更新时才能被参考，从而造成信息滞后，对用户信用的评估准确性带来非常大的影响。
技术内容
实例本申请实例提供了一种用户信用评估方法，所述方法包括：
获取目标用户的离线特征信息，所述离线特征信息为按照预设更新周期进行更新的用户的特征信息；
根据目标用户的离线特征信息以及预设的离线预测模型，计算目标用户的离线信用评分；
获取目标用户的实时特征信息，所述实时特征信息为距离当前预设时间范围内采集到的用户的特征信息，所述预设时间范围小于所述预设更新周期；
根据目标用户的实时特征信息以及预设的实时预测模型，计算目标用户的实时信用评分；
根据得到的目标用户的离线信用评分和实时信用评分结合预设的综合预测模型，计算目标用户的综合信用评分。
相应地，本申请实例还提供了一种用户信用评估装置，所述装置包括：
离线特征获取模块，用于获取目标用户的离线特征信息，所述离线特征信息为按照预设更新周期进行更新的用户的特征信息；
离线评分模块，用于根据目标用户的离线特征信息以及预设的离线预测模型，计算目标用户的离线信用评分；
实时特征获取模块，用于获取目标用户的实时特征信息，所述实时特征信息为距离当前预设时间范围内采集到的用户的特征信息，所述预设时间范围小于所述预设更新周期；
实时评分模块，用于根据目标用户的实时特征信息以及预设的实时预测模型，计算目标用户的实时信用评分；
综合评分模块，用于根据得到的目标用户的离线信用评分和实时信用评分结合预设的综合预测模型，计算目标用户的综合信用评分。
本申请实例还提供了一种计算机可读存储介质，存储有计算机可读 指令，可以使至少一个处理器执行如上述所述的方法。
采用本申请提供的上述方案，能够提高用户信用评估的准确性。
附图说明
为了更清楚地说明本申请实例或现有技术中的技术方案，下面将对实例或现有技术描述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他的附图。
图1是本申请实例中的一种用户信用评估方法的流程示意图；
图2是本申请实例中获取用户的实时特征信息和离线特征信息的来源示意图；
图3是本申请实例中对离线预测模型进行训练的流程示意图；
图4是本申请实例中对实时预测模型进行训练的流程示意图；
图5是本申请实例中对综合预测模型进行训练的流程示意图；
图6是本申请实例中的一种用户信用评估装置的结构示意图；以及
图7是本申请实例中的样本获取模块的结构示意图。
具体实施方式
下面将结合本申请实例中的附图，对本申请实例中的技术方案进行清楚、完整地描述，显然，所描述的实例仅仅是本申请一部分实例，而不是全部的实例。基于本申请中的实例，本领域普通技术人员在没有作出创造性劳动前提下所获得的所有其他实例，都属于本申请保护的范围。
本申请实例中的用户信用评估方法和装置，可以实现在如个人电 脑、笔记本电脑、智能手机、平板电脑、电子阅读器等计算机系统中，较多的可以被采用在提供用户信用评估的服务器中，例如数据业务平台的后台服务器。下文均以用户信用评估装置作为本申请实例的执行主体进行介绍。
图1是本申请实例中的一种用户信用评估方法的流程示意图，如图所示本实例中的用户信用评估方法流程可以包括：
S101，获取目标用户的离线特征信息，所述离线特征信息为按照预设更新周期进行更新的用户的特征信息。
所述离线特征信息如图2所示，用户信用评估装置可以通过采集来自第三方提供的用户数据得到，也可以在业务平台采集得到的用户数据中获取得到。用户信用评估装置可以通过对上述得到的用户数据进行特征计算，将用户数据中的用户属性、用户行为或用户属性/行为的变化转换为统一格式的离线特征信息，例如数字化的特征信息。所述预设的更新周期，可以是外部厂商提供用户数据的更新周期，也可以是用户信用评估装置中自身设置的采集更新周期。由于大数据涉及庞大的用户基数，离线特征信息中可以包括用户所有的历史特征信息，数据量庞大，因此该预设的更新周期一般较长，通常至少为一周至一个月的时间。在可选实例中，所述离线特征信息可以为用户较为稳定的特征信息，例如性别、年龄、籍贯、职业、收入情况等属性，还可以包括所有的历史契约信用记录，对于此类通常较为稳定的用户特征信息，只需要按照预设更新周期进行更新即可，因此将这些特征类别的信息作为离线特征信息。
在可选实例中，所述离线特征信息可以是经过筛选的特征类别的离线特征信息，即第三方提供的用户数据或业务平台采集得到的用户数据中可能包括多个特征类别的离线特征信息，用户信用评估装置可以从中 筛选出指定特征类别的离线特征信息。所述指定的特征类别，可以是用户信用评估装置根据预设的训练样本数据获取的，所述训练样本数据包括多个用户的信用评分结果样本和各个用户的多个特征类别的离线特征信息样本，所述训练样本数据也称为用户信用数据。用户信用评估装置根据所述用户信用样本数据中多个用户的信用评分结果样本和各个用户的多个特征类别的离线特征信息样本，计算各个特征类别与信用评分结果之间的相关度，从而将与信用评分结果之间的相关度达到预设阈值的特征类别确定为指定的特征类别。
S102，根据目标用户的离线特征信息以及预设的离线预测模型，计算目标用户的离线信用评分。
所述离线预测模型，可以是经过训练的逻辑回归分类模型，也可以是经过训练的集成学习模型、深度学习模型、随机森林模型等。用户信用评估装置将目标用户的离线特征信息代入到所述预设的离线预测模型中，即可计算得到目标用户的离线信用评分。
所述离线预测模型可以是用户信用评估装置根据预设的训练样本数据训练得到的，所述训练样本数据可以包括多个用户的信用评分结果样本和各个用户的离线特征信息；所述离线预测模型还可以是用户信用评估装置从外部获取的经过训练的离线预测模型。
S103，获取目标用户的实时特征信息，所述实时特征信息为距离当前预设时间范围内采集到的用户的特征信息，所述预设时间范围小于所述预设更新周期。
所述实时特征信息如图2所示，用户信用评估装置可以通过业务平台采集得到的用户数据中获取得到。用户信用评估装置可以通过对得到的用户数据进行特征计算，将用户数据中的用户属性、用户行为或用户属性/行为的变化转换为统一格式的实时特征信息，例如数字化的特征 信息。所述业务平台可以采集用户最新的特征信息，所述预设时间范围小于所述预设更新周期，例如最近一天、两天或一周内采集到的用户的特征信息。在可选实例中，用户信用评估装置可以预先设定一些特征类别作为高风险特征，当用户这些高风险特征对应的特征信息发生变化时，将会对用户的信用评分带来很大的影响，例如用户办理了特定平台借贷业务、办理出国签证业务、所在地理位置发生变化或发生了特定领域的大额消费等，对于这些需要实时关注的高风险特征，用户信用评估装置可以将对应的特征信息作为用户的实时特征信息进行实时收集并录入，对于其他的特征信息作为离线特征信息进行预设更新周期的更新。
同样的，所述实时特征信息可以是经过筛选的特征类别的实时特征信息，即业务平台采集得到的用户数据中可能包括多个特征类别的实时特征信息，用户信用评估装置可以从中筛选出指定特征类别的实时特征信息。所述指定的特征类别，可以是用户信用评估装置根据预设的训练样本数据，所述训练样本数据包括多个用户的信用评分结果样本和各个用户的多个特征类别的实时特征信息样本，所述训练样本数据也称为用户信用样本数据。用户信用评估装置根据所述用户信用样本数据中多个用户的信用评分结果样本和各个用户的多个特征类别的实时特征信息样本，计算各个特征类别与信用评分结果之间的相关度，从而将与信用评分结果之间的相关度达到预设阈值的特征类别确定为指定的特征类别。
S104，根据目标用户的实时特征信息以及预设的实时预测模型，计算目标用户的实时信用评分。
所述实时预测模型，可以是经过训练的逻辑回归分类模型，也可以是经过训练的集成学习模型、深度学习模型、随机森林模型等。用户信 用评估装置将目标用户的实时特征信息代入到所述预设的实时预测模型中，即可计算得到目标用户的实时信用评分。
所述实时预测模型可以是用户信用评估装置根据预设的训练样本数据训练得到的，所述训练样本数据可以包括多个用户的信用评分结果样本和各个用户的离线特征信息；所述实时预测模型还可以是用户信用评估装置从外部获取的经过训练的实时预测模型。
S105，根据得到的目标用户的离线信用评分和实时信用评分结合预设的综合预测模型，计算目标用户的综合信用评分。
所述综合预测模型，可以是经过训练的逻辑回归分类模型，也可以是经过训练的集成学习模型、深度学习模型、随机森林模型、梯度提升决策树模型等。用户信用评估装置将目标用户的离线信用评分和实时信用评分代入到所述预设的实时预测模型中，即可计算得到目标用户的实时信用评分。
所述综合预测模型可以是用户信用评估装置根据预设的训练样本数据训练得到的，所述训练样本数据可以包括多个用户的信用评分结果样本和各个用户的离线特征信息和实时特征信息，用户信用评估装置在使用离线预测模型根据用户的离线特征信息得到各个用户的离线信用评分，以及使用实时预测模型根据用户的实时特征信息得到各个用户的实时信用评分后，根据所述多个用户的信用评分结果以及各个用户的离线信用评分和实时信用评分对所述综合预测模型进行训练。所述实时预测模型还可以是用户信用评估装置从外部获取的经过训练的实时预测模型。
示例性的，可以采用下式逻辑回归算法的综合预测模型计算得到目标用户的实时信用评分：
Score＝1/(1+exp(-(α*Score1+β*Score2+γ)))
其中α，β，γ为训练模型得到的参数，Score1和Score2分别为目标用户的离线信用评分和实时信用评分，结果Score为目标用户的综合信用评分。
进而在可选实例中，用户信用评估装置可以根据经过本实例上述步骤后计算得到的目标用户的综合信用评分为目标用户推送产品信息，例如推送金融产品信息、固定资产管理产品信息等；或根据目标用户的综合信用评分对目标用户的数据业务进行监控管理，例如对目标用户的借贷业务进行风控管理、对目标用户的流动资金进行管理建议等。
采用本申请提供的用户信用评估方法，通过获取用户的离线特征信息和实时特征信息，分别计算用户的离线信用评分和实时信用评分，从而计算用户的综合信用评分，实现了结合用户的长期特征数据和实时特征数据准确预测用户的信用状况，解决了现有技术中因用户信息滞后造成的信用估计不准确的问题。
图3是本申请实例中对离线预测模型进行训练的流程示意图，如图所示，本实例中的离线预测模型训练流程可以包括以下步骤：
S301，获取多个用户的信用评分结果样本和各个用户的多个特征类别的离线特征信息样本。
可选的，所述多个用户的信用评分结果样本和各个用户的多个特征类别的离线特征信息样本可以根据输入至用户信用评估装置的训练样本数据中提取得到。
或者，所述多个用户的信用评分结果样本可以通过对该多个用户的违约记录计算得到，即根据该多个用户是否违约，或违约事件的次数和严重程度等确定多个用户的信用评分结果样本。而在可选实例中，所述多个用户的信用评分结果样本也可以采用人工评分的方式得到。进而在得到上述多个用户的评分结果样本后，用户信用评估装置可以通过采集 来自第三方提供的用户数据，或可以通过业务平台采集得到的用户数据中获取得各个用户的多个特征类别的离线特征信息样本。
S302，根据所述多个用户的信用评分结果样本和各个用户的多个特征类别的离线特征信息样本，计算各个特征类别与信用评分结果之间的相关度。
用户的信用样本数据中包括多个用户的信用评分结果样本和各个用户的多个特征类别的离线特征信息样本。所述特征类别可以例如年龄、所在地、性别、职业等，特征类别与信用评分结果之间的相关度，反应的是如年龄、性别、职业等对于用户信用评分结果的影响，如果相关度较高，则表示该特征类别对于信用评分结果影响较大，反之则是对信用评分结果影响很小，可以在建立离线预测模型时，不考虑该特征类别的离线特征信息。
具体的，所述各个特征类别与信用评分结果之间的相关度，示例性地可以采用下式计算相关度r：
Figure PCTCN2017085049-appb-000001
其中x为某个特征类别的离线特征信息，y为用户的信用评分结果.下标i则表示对应不同的用户。
在其他可选实例中，还可以采用IV值、卡方值等相关度算法计算 所述各个特征类别与信用评分结果之间的相关度。
S303，将与信用评分结果之间的相关度达到预设阈值的特征类别确定为所述离线特征信息的特征类别，并从所述各个用户的多个特征类别的离线特征信息样本中筛选出对应特征类别的离线特征信息。
在计算得到各个特征类别与信用评分结果之间的相关度后，可以与相应的预设阈值进行比较后，将相关度达到要求的特征类别确定所述离线特征信息的特征类别，并从所述各个用户的多个特征类别的离线特征信息样本中筛选出对应特征类别的离线特征信息。
S304，根据经过筛选的对应特征类别的用户的离线特征信息建立离线预测模型，并根据所述多个用户的信用评分结果样本和各个用户的对应特征类别的离线特征信息对离线预测模型进行训练。
所述离线预测模型，可以是经过训练的逻辑回归分类模型，也可以是经过训练的集成学习模型、深度学习模型、随机森林模型等。所述离线预测模型可以是用户信用评估装置根据经过筛选的对应特征类别的用户的离线特征信息结合特定的模型参数计算用户的信用评分的一个预测算式，通过所述多个用户的信用评分结果样本和各个用户的对应特征类别的离线特征信息对该预测算式中的模型参数进行训练迭代，从而可以得到最接近信用评分结果样本的预测算式的模型参数，从而得到经过训练的离线预测模型。
需要指出的是，上述S302和S303为可选步骤，在可选实例中可以不经筛选的将获取到的各个用户的多个特征类别的离线特征信息样本都作为离线特征信息进行实时预测模型的训练。
图4是本申请实例中对实时预测模型进行训练的流程示意图，如图所示本实例中的实时预测模型训练流程可以包括：
S401，获取多个用户的信用评分结果样本和各个用户的多个特征类 别的实时特征信息样本。
可选的，所述多个用户的信用评分结果样本和各个用户的多个特征类别的实时特征信息样本可以根据输入至用户信用评估装置的训练样本数据中提取得到。
或者，所述多个用户的信用评分结果样本可以通过对该多个用户的违约记录计算得到，即根据该多个用户是否违约，或违约事件的次数和严重程度等确定多个用户的信用评分结果样本。而在可选实例中，所述多个用户的信用评分结果样本也可以采用人工评分的方式得到。进而在得到上述多个用户的评分结果样本后，用户信用评估装置可以在业务平台采集得到的用户数据中获取得到各个用户的多个特征类别的实时特征信息样本。
S402，根据所述多个用户的信用评分结果样本和各个用户的多个特征类别的实时特征信息样本，计算各个特征类别与信用评分结果之间的相关度。
用户的信用样本数据中包括多个用户的信用评分结果样本和各个用户的多个特征类别的离线特征信息样本。所述特征类别可以例如年龄、所在地、性别、职业等，特征类别与信用评分结果之间的相关度，反应的是如年龄、性别、职业等对于用户信用评分结果的影响，如果相关度较高，则表示该特征类别对于信用评分结果影响较大，反之则是对信用评分结果影响很小，可以在建立实时预测模型时，不考虑该特征类别的实时特征信息。
具体的，所述各个特征类别与信用评分结果之间的相关度，示例性地可以采用下式计算相关度s：
Figure PCTCN2017085049-appb-000002
其中z为某个特征类别的实时特征信息，y为用户的信用评分结果.下标i则表示对应不同的用户。
在其他可选实例中，还可以采用IV值、卡方值等相关度算法计算所述各个特征类别的实时特征信息与信用评分结果之间的相关度。
S403，将与信用评分结果之间的相关度达到预设阈值的特征类别确定为所述实时特征信息的特征类别，并从所述各个用户的多个特征类别的实时特征信息样本中筛选出对应特征类别的实时特征信息。
在计算得到各个特征类别与信用评分结果之间的相关度后，可以与相应的预设阈值进行比较后，将相关度达到要求的特征类别确定所述实时特征信息的特征类别，并从所述各个用户的多个特征类别的实时特征信息样本中筛选出对应特征类别的实时特征信息。
S404，根据经过筛选的对应特征类别的用户的实时特征信息建立实时预测模型，并根据所述多个用户的信用评分结果样本和各个用户的对应特征类别的实时特征信息对实时预测模型进行训练。
所述实时预测模型，可以是经过训练的逻辑回归分类模型，也可以是经过训练的集成学习模型、深度学习模型、随机森林模型等。所述实时预测模型可以是用户信用评估装置根据经过筛选的对应特征类别的 用户的实时特征信息结合特定的模型参数计算用户的信用评分的一个预测算式，通过所述多个用户的信用评分结果样本和各个用户的对应特征类别的实时特征信息对该预测算式中的模型参数进行训练迭代，从而可以得到最接近信用评分结果样本的预测算式的模型参数，从而得到经过训练的实时预测模型。
需要指出的是，上述S402和S403为可选步骤，在可选实例中可以不经筛选的将获取到的各个用户的多个特征类别的实时特征信息样本都作为实时特征信息进行实时预测模型的训练。
图5是本申请实例中对综合预测模型进行训练的流程示意图。
S501，获取多个用户的信用评分结果样本和各个用户的离线特征信息和实时特征信息。
可选的，所述多个用户的信用评分结果样本和各个用户的离线特征信息和实时特征信息可以根据输入至用户信用评估装置的训练样本数据中提取得到。
或者，所述多个用户的信用评分结果样本可以通过对该多个用户的违约记录计算得到，即根据该多个用户是否违约，或违约事件的次数和严重程度等确定多个用户的信用评分结果样本。而在可选实例中，所述多个用户的信用评分结果样本也可以采用人工评分的方式得到。进而在得到上述多个用户的评分结果样本后，用户信用评估装置可以通过采集来自第三方提供的用户数据，或可以通过业务平台采集得到的用户数据中获取得到各个用户的多个特征类别的实时特征信息样本。
S502，根据所述各个用户的离线特征信息以及预设的离线预测模型，计算所述各个用户的离线信用评分。
S503，根据所述各个用户的实时特征信息以及预设的实时预测模型，计算所述各个用户的实时信用评分。
S504，根据用户的离线信用评分和实时信用评分建立所述综合预测模型，并根据所述多个用户的信用评分结果以及各个用户的离线信用评分和实时信用评分对所述综合预测模型进行训练。
所述综合预测模型，可以是经过训练的逻辑回归分类模型，也可以是经过训练的集成学习模型、深度学习模型、随机森林模型、梯度提升决策树模型等。所述综合预测模型可以是用户信用评估装置根据用户的离线信用评分和实时信用评分结合特定的模型参数计算用户的综合信用评分的一个预测算式，通过所述多个用户的信用评分结果样本和计算得到的各个用户的离线信用评分和实时信用评分可以对该预测算式中的模型参数进行训练迭代，从而可以得到最接近信用评分结果样本的预测算式的模型参数，从而得到经过训练的综合预测模型。
示例性的，可以采用下式逻辑回归算法的综合预测模型计算得到目标用户的实时信用评分：
Score＝1/(1+exp(-(α*Score1+β*Score2+γ)))
其中α，β，γ为训练模型得到的模型参数，Score1和Score2分别为目标用户的离线信用评分和实时信用评分，结果Score为目标用户的综合信用评分。
图6是本申请一种用户信用评估装置的结构示意图，如图所示本申请实例中的用户信用评估装置可以包括：
离线特征获取模块610，用于获取目标用户的离线特征信息，所述离线特征信息为按照预设更新周期进行更新的用户的特征信息。
所述离线特征信息如图2所示，离线特征获取模块610可以通过采集来自第三方提供的用户数据得到，也可以在业务平台采集得到的用户数据中获取得到。离线特征获取模块610可以通过对上述得到的用户数据进行特征计算，将用户数据中的用户属性、用户行为或用户属性/行 为的变化转换为统一格式的离线特征信息，例如数字化的特征信息。所述预设的更新周期，可以是外部厂商提供用户数据的更新周期，也可以是离线特征获取模块610中自身设置的采集更新周期。由于大数据涉及庞大的用户基数，离线特征信息中可以包括用户所有的历史特征信息，数据量庞大，因此该预设的更新周期一般较长，通常至少为一周至一个月的时间。在可选实例中，所述离线特征信息可以为用户较为稳定的特征信息，例如性别、年龄、籍贯、职业、收入情况等属性，还可以包括所有的历史契约信用记录，对于此类通常较为稳定的用户特征信息，只需要按照预设更新周期进行更新即可，因此将这些特征类别的信息作为离线特征信息。
在可选实例中，所述离线特征信息可以是经过筛选的特征类别的离线特征信息，即第三方提供的用户数据或业务平台采集得到的用户数据中可能包括多个特征类别的离线特征信息，离线特征获取模块610可以从中筛选出指定特征类别的离线特征信息。所述指定的特征类别，可以是用户信用评估装置根据预设的训练样本数据获取的，所述训练样本数据包括多个用户的信用评分结果样本和各个用户的多个特征类别的离线特征信息样本，所述训练样本数据也称为用户信用样本数据。用户信用评估装置根据所述用户信用样本数据中多个用户的信用评分结果样本和各个用户的多个特征类别的离线特征信息样本，计算各个特征类别与信用评分结果之间的相关度，从而将与信用评分结果之间的相关度达到预设阈值的特征类别确定为指定的特征类别。
离线评分模块620，用于根据目标用户的离线特征信息以及预设的离线预测模型，计算目标用户的离线信用评分。
所述离线预测模型，可以是经过训练的逻辑回归分类模型，也可以是经过训练的集成学习模型、深度学习模型、随机森林模型等。离线评 分模块620将目标用户的离线特征信息代入到所述预设的离线预测模型中，即可计算得到目标用户的离线信用评分。
所述离线预测模型可以是用户信用评估装置根据预设的训练样本数据训练得到的，所述训练样本数据可以包括多个用户的信用评分结果样本和各个用户的离线特征信息；所述离线预测模型还可以是用户信用评估装置从外部获取的经过训练的离线预测模型。
实时特征获取模块630，用于获取目标用户的实时特征信息，所述实时特征信息为距离当前预设时间范围内采集到的用户的特征信息，所述预设时间范围小于所述预设更新周期。
所述实时特征信息如图2所示，实时特征获取模块630可以通过业务平台采集得到的用户数据中获取得到。实时特征获取模块630可以通过对得到的用户数据进行特征计算，将用户数据中的用户属性、用户行为或用户属性/行为的变化转换为统一格式的实时特征信息，例如数字化的特征信息。所述业务平台可以采集用户最新的特征信息，所述预设时间范围小于所述预设更新周期，例如最近一天、两天或一周内采集到的用户的特征信息。在可选实例中，用户信用评估装置可以预先设定一些特征类别作为高风险特征，当用户这些高风险特征对应的特征信息发生变化时，将会对用户的信用评分带来很大的影响，例如用户办理了特定平台借贷业务、办理出国签证业务、所在地理位置发生变化或发生了特定领域的大额消费等，对于这些需要实时关注的高风险特征，实时特征获取模块630可以将对应的特征信息作为用户的实时特征信息进行实时收集并录入，对于其他的特征信息作为离线特征信息进行预设更新周期的更新。
同样的，所述实时特征信息可以是经过筛选的特征类别的实时特征信息，即第三方提供的用户数据或业务平台采集得到的用户数据中可能 包括多个特征类别的实时特征信息，实时特征获取模块630可以从中筛选出指定特征类别的实时特征信息。所述指定的特征类别，可以是用户信用评估装置根据预设的训练样本数据，所述训练样本数据包括多个用户的信用评分结果样本和各个用户的多个特征类别的实时特征信息样本，所述训练样本数据也称为用户信用样本数据。用户信用评估装置根据所述用户信用样本数据中多个用户的信用评分结果样本和各个用户的多个特征类别的实时特征信息样本，计算各个特征类别与信用评分结果之间的相关度，从而将与信用评分结果之间的相关度达到预设阈值的特征类别确定为指定的特征类别。
实时评分模块640，用于根据目标用户的实时特征信息以及预设的实时预测模型，计算目标用户的实时信用评分。
所述实时预测模型，可以是经过训练的逻辑回归分类模型，也可以是经过训练的集成学习模型、深度学习模型、随机森林模型等。用户信用评估装置将目标用户的实时特征信息代入到所述预设的实时预测模型中，即可计算得到目标用户的实时信用评分。
所述实时预测模型可以是用户信用评估装置根据预设的训练样本数据训练得到的，所述训练样本数据可以包括多个用户的信用评分结果样本和各个用户的离线特征信息；所述实时预测模型还可以是用户信用评估装置从外部获取的经过训练的实时预测模型。
综合评分模块650，用于根据得到的目标用户的离线信用评分和实时信用评分结合预设的综合预测模型，计算目标用户的综合信用评分。
所述综合预测模型，可以是经过训练的逻辑回归分类模型，也可以是经过训练的集成学习模型、深度学习模型、随机森林模型、梯度提升决策树模型等。综合评分模块650将目标用户的离线信用评分和实时信用评分代入到所述预设的实时预测模型中，即可计算得到目标用户的实 时信用评分。
所述综合预测模型可以是用户信用评估装置根据预设的训练样本数据训练得到的，所述训练样本数据可以包括多个用户的信用评分结果样本和各个用户的离线特征信息和实时特征信息，用户信用评估装置在使用离线预测模型根据用户的离线特征信息得到各个用户的离线信用评分，以及使用实时预测模型根据用户的实时特征信息得到各个用户的实时信用评分后，根据所述多个用户的信用评分结果以及各个用户的离线信用评分和实时信用评分对所述综合预测模型进行训练。所述实时预测模型还可以是用户信用评估装置从外部获取的经过训练的实时预测模型。
示例性的，可以采用下式逻辑回归算法的综合预测模型计算得到目标用户的实时信用评分：
Score＝1/(1+exp(-(α*Score1+β*Score2+γ)))
其中α，β，γ为训练模型得到的参数，Score1和Score2分别为目标用户的离线信用评分和实时信用评分，结果Score为目标用户的综合信用评分。
采用本申请提供的用户信用评估装置，通过获取用户的离线特征信息和实时特征信息，分别计算用户的离线信用评分和实时信用评分，从而计算用户的综合信用评分，实现了结合用户的长期特征数据和实时特征数据准确预测用户的信用状况，解决了现有技术中因用户信息滞后造成的信用估计不准确的问题。
在可选实例中，所述用户信用评估装置还可以包括：
样本获取模块660，用于获取多个用户的信用评分结果样本和各个用户的离线特征信息。
可选的，所述多个用户的信用评分结果样本和各个用户的离线特征 信息可以根据输入至用户信用评估装置的训练样本数据中提取得到。
或者，所述多个用户的信用评分结果样本可以通过对该多个用户的违约记录计算得到，即根据该多个用户是否违约，或违约事件的次数和严重程度等确定多个用户的信用评分结果样本。而在可选实例中，所述多个用户的信用评分结果样本也可以采用人工评分的方式得到。进而在得到上述多个用户的评分结果样本后，样本获取模块660可以通过采集来自第三方提供的用户数据，或可以通过业务平台采集得到的用户数据中获取得到各个用户的离线特征信息。
离线模型训练模块670，用于根据用户的离线特征信息建立所述离线预测模型，并根据所述多个用户的信用评分结果样本和各个用户的离线特征信息对所述离线预测模型进行训练。
所述离线预测模型，可以是经过训练的逻辑回归分类模型，也可以是经过训练的集成学习模型、深度学习模型、随机森林模型等。所述离线预测模型可以是根据经过筛选的对应特征类别的用户的离线特征信息结合特定的模型参数计算用户的信用评分的一个预测算式，离线模型训练模块670通过所述多个用户的信用评分结果样本和各个用户的对应特征类别的离线特征信息对该预测算式中的模型参数进行训练迭代，从而可以得到最接近信用评分结果样本的预测算式的模型参数，从而得到经过训练的离线预测模型。
进而可选的，所述样本获取模块660如图7所示进一步可以包括：
离线样本获取单元661，用于获取多个用户的信用评分结果样本和各个用户的多个特征类别的离线特征信息样本。
相关度计算单元663，用于根据所述多个用户的信用评分结果样本和各个用户的多个特征类别的离线特征信息样本，计算各个特征类别与信用评分结果之间的相关度。
所述特征类别可以例如年龄、所在地、性别、职业等，特征类别与信用评分结果之间的相关度，反应的是如年龄、性别、职业等对于用户信用评分结果的影响，如果相关度较高，则表示该特征类别对于信用评分结果影响较大，反之则是对信用评分结果影响很小，可以在建立离线预测模型时，不考虑该特征类别的离线特征信息。
具体的，所述各个特征类别与信用评分结果之间的相关度，示例性地可以采用下式计算相关度r：
Figure PCTCN2017085049-appb-000003
其中x为某个特征类别的离线特征信息，y为用户的信用评分结果.下标i则表示对应不同的用户。
在其他可选实例中，还可以采用IV值、卡方值等相关度算法计算所述各个特征类别与信用评分结果之间的相关度。
特征类别筛选单元665，用于将与信用评分结果之间的相关度达到预设阈值的特征类别确定为所述离线特征信息的特征类别，并从所述各个用户的多个特征类别的离线特征信息样本中筛选出对应特征类别的离线特征信息。
在可选实例中，所述样本获取模块660，用于获取多个用户的信用评分结果样本和各个用户的实时特征信息。
可选的，所述多个用户的信用评分结果样本和各个用户的实时特征信息可以在输入至用户信用评估装置的训练样本数据中提取得到。
或者，所述多个用户的信用评分结果样本可以通过对该多个用户的违约记录计算得到，即根据该多个用户是否违约，或违约事件的次数和严重程度等确定多个用户的信用评分结果样本。而在可选实例中，所述多个用户的信用评分结果样本也可以采用人工评分的方式得到。进而在得到上述多个用户的评分结果样本后，用户信用评估装置可以业务平台采集得到的用户数据中获取得到各个用户的实时特征信息。
所述用户信用评估装置还可以包括：实时模型训练模块680，用于根据用户的实时特征信息建立所述实时预测模型，并根据所述多个用户的信用评分结果样本和各个用户的实时特征信息对所述实时预测模型进行训练。
所述实时预测模型，可以是经过训练的逻辑回归分类模型，也可以是经过训练的集成学习模型、深度学习模型、随机森林模型等。所述实时预测模型可以是用户信用评估装置根据经过筛选的对应特征类别的用户的实时特征信息结合特定的模型参数计算用户的信用评分的一个预测算式，实时模型训练模块680通过所述多个用户的信用评分结果样本和各个用户的对应特征类别的实时特征信息对该预测算式中的模型参数进行训练迭代，从而可以得到最接近信用评分结果样本的预测算式的模型参数，从而得到经过训练的实时预测模型。
进而可选的，所述样本获取模块如图7所示进一步可以包括：
实时样本获取单元662，用于获取多个用户的信用评分结果样本和各个用户的多个特征类别的实时特征信息样本。
相关度计算单元663，用于根据所述多个用户的信用评分结果样本和各个用户的多个特征类别的实时特征信息样本，计算各个特征类别与 信用评分结果之间的相关度。
所述特征类别可以例如年龄、所在地、性别、职业等，特征类别与信用评分结果之间的相关度，反应的是如年龄、性别、职业等对于用户信用评分结果的影响，如果相关度较高，则表示该特征类别对于信用评分结果影响较大，反之则是对信用评分结果影响很小，可以在建立实时预测模型时，不考虑该特征类别的实时特征信息。
具体的，所述各个特征类别与信用评分结果之间的相关度，示例性地可以采用下式计算相关度s：
Figure PCTCN2017085049-appb-000004
其中z为某个特征类别的实时特征信息，y为用户的信用评分结果.下标i则表示对应不同的用户。
在其他可选实例中，还可以采用IV值、卡方值等相关度算法计算所述各个特征类别的实时特征信息与信用评分结果之间的相关度。
特征类别筛选单元665，用于将与信用评分结果之间的相关度达到预设阈值的特征类别确定为所述实时特征信息的特征类别，并从所述各个用户的多个特征类别的实时特征信息样本中筛选出对应特征类别的实时特征信息。
在计算得到各个特征类别与信用评分结果之间的相关度后，特征类 别筛选单元665可以与相应的预设阈值进行比较后，将相关度达到要求的特征类别确定所述实时特征信息的特征类别，并从所述各个用户的多个特征类别的实时特征信息样本中筛选出对应特征类别的实时特征信息。
在可选实例中，所述样本获取模块660，用于获取多个用户的信用评分结果样本和各个用户的离线特征信息和实时特征信息。
可选的，所述多个用户的信用评分结果样本和各个用户的离线特征信息和实时特征信息可以根据输入至用户信用评估装置的训练样本数据中提取得到。
或者，所述多个用户的信用评分结果样本可以通过对该多个用户的违约记录计算得到，即根据该多个用户是否违约，或违约事件的次数和严重程度等确定多个用户的信用评分结果样本。而在可选实例中，所述多个用户的信用评分结果样本也可以采用人工评分的方式得到。进而在得到上述多个用户的评分结果样本后，所述样本获取模块660可以通过采集来自第三方提供的用户数据，或可以通过业务平台采集得到的用户数据中获取得到各个用户的多个特征类别的实时特征信息样本。
所述离线评分模块620，还用于根据所述各个用户的离线特征信息以及预设的离线预测模型，计算所述各个用户的离线信用评分。
所述实时评分模块640，还用于根据所述各个用户的实时特征信息以及预设的实时预测模型，计算所述各个用户的实时信用评分。
所述用户信用评估装置还可以包括：
综合模型训练模块690，用于根据用户的离线信用评分和实时信用评分建立所述综合预测模型，并根据所述多个用户的信用评分结果以及各个用户的离线信用评分和实时信用评分对所述综合预测模型进行训练。
所述综合预测模型，可以是经过训练的逻辑回归分类模型，也可以是经过训练的集成学习模型、深度学习模型、随机森林模型、梯度提升决策树模型等。所述综合预测模型可以是用户信用评估装置根据用户的离线信用评分和实时信用评分结合特定的模型参数计算用户的综合信用评分的一个预测算式，通过所述多个用户的信用评分结果样本和计算得到的各个用户的离线信用评分和实时信用评分可以对该预测算式中的模型参数进行训练迭代，从而可以得到最接近信用评分结果样本的预测算式的模型参数，从而得到经过训练的综合预测模型。
示例性的，可以采用下式逻辑回归算法的综合预测模型计算得到目标用户的实时信用评分：
Score＝1/(1+exp(-(α*Score1+β*Score2+γ)))
其中α，β，γ为训练模型得到的模型参数，Score1和Score2分别为目标用户的离线信用评分和实时信用评分，结果Score为目标用户的综合信用评分。
进而在可选实例中，所述用户信用评估装置还可以包括信息推送模块6100或业务监控模块6110，其中：
信息推送模块6100，用于根据目标用户的综合信用评分为目标用户推送产品信息，即根据经过本申请实例的综合评分模块650计算得到的目标用户的综合信用评分为目标用户推送产品信息，例如推送金融产品信息、固定资产管理产品信息等。
业务监控模块6110，用于根据目标用户的综合信用评分对目标用户的数据业务进行监控管理，例如对目标用户的借贷业务进行风控管理、对目标用户的流动资金进行管理建议等。
从而，本申请实例中的用户信用评估装置通过获取用户的离线特征信息和实时特征信息，分别计算用户的离线信用评分和实时信用评分， 从而计算用户的综合信用评分，实现了结合用户的长期特征数据和实时特征数据准确预测用户的信用状况，解决了现有技术中因用户信息滞后造成的信用估计不准确的问题。
本领域普通技术人员可以理解实现上述实例方法中的全部或部分流程，是可以通过计算机程序来指令相关的硬件来完成，所述的程序可存储于一计算机可读取存储介质中，该程序在执行时，可包括如上述各方法的实例的流程。其中，所述的存储介质可为磁碟、光盘、只读存储记忆体(Read-Only Memory，ROM)或随机存储记忆体(Random Access Memory，RAM)等。
以上所揭露的仅为本申请较佳实例而已，当然不能以此来限定本申请之权利范围，因此依本申请权利要求所作的等同变化，仍属本申请所涵盖的范围。

Claims (17)

  1. 一种用户信用评估方法，其特征在于，所述方法包括：
    获取目标用户的离线特征信息，所述离线特征信息为按照预设更新周期进行更新的用户的特征信息；
    根据目标用户的离线特征信息以及预设的离线预测模型，计算目标用户的离线信用评分；
    获取目标用户的实时特征信息，所述实时特征信息为距离当前预设时间范围内采集到的用户的特征信息，所述预设时间范围小于所述预设更新周期；
    根据目标用户的实时特征信息以及预设的实时预测模型，计算目标用户的实时信用评分；
    根据得到的目标用户的离线信用评分和实时信用评分结合预设的综合预测模型，计算目标用户的综合信用评分。
  2. 如权利要求1所述的用户信用评估方法，其中，所述获取目标用户的离线特征信息之前还包括：
    获取多个用户的信用评分结果样本和各个用户的离线特征信息；
    建立所述离线预测模型，根据所述多个用户的信用评分结果样本和各个用户的离线特征信息对所述离线预测模型进行训练，获得所述离线预测模型的模型参数。
  3. 如权利要求2所述的用户信用评估方法，其中，所述获取多个用户的信用评分结果样本和各个用户的离线特征信息包括：
    获取多个用户的信用评分结果样本和各个用户的多个特征类别的 离线特征信息样本；
    根据所述多个用户的信用评分结果样本和各个用户的多个特征类别的离线特征信息样本，计算各个特征类别与信用评分结果之间的相关度；
    将与信用评分结果之间的相关度达到预设阈值的特征类别确定为所述离线特征信息的特征类别，并从所述各个用户的多个特征类别的离线特征信息样本中筛选出确定的所述特征类别的离线特征信息。
  4. 如权利要求1所述的用户信用评估方法，其中，所述获取用户的实时特征信息之前还包括：
    获取多个用户的信用评分结果样本和各个用户的实时特征信息；
    建立所述实时预测模型，并根据所述多个用户的信用评分结果样本和各个用户的实时特征信息对所述实时预测模型进行训练，获得所述实时预测模型的模型参数。
  5. 如权利要求4所述的用户信用评估方法，其中，所述获取多个用户的信用评分结果样本和各个用户的实时特征信息包括：
    获取多个用户的信用评分结果样本和各个用户的多个特征类别的实时特征信息样本；
    根据所述多个用户的信用评分结果样本和各个用户的多个特征类别的实时特征信息样本，计算各个特征类别与信用评分结果之间的相关度；
    将与信用评分结果之间的相关度达到预设阈值的特征类别确定为所述实时特征信息的特征类别，并从所述各个用户的多个特征类别的实时特征信息样本中筛选出确定的所述特征类别的实时特征信息。
  6. 如权利要求1所述的用户信用评估方法，其中，所述根据得到的目标用户的离线信用评分和实时信用评分结合预设的综合预测模型，计算目标用户的综合信用评分之前还包括：
    获取多个用户的信用评分结果样本和各个用户的离线特征信息和实时特征信息；
    根据所述各个用户的离线特征信息以及预设的离线预测模型，计算所述各个用户的离线信用评分；
    根据所述各个用户的实时特征信息以及预设的实时预测模型，计算所述各个用户的实时信用评分；
    建立所述综合预测模型，并根据所述多个用户的信用评分结果以及各个用户的离线信用评分和实时信用评分对所述综合预测模型进行训练，获得所述综合预测模型的模型参数。
  7. 如权利要求1所述的用户信用评估方法，其中，所述实时特征信息包括业务平台采集得到的用户数据；
    所述离线特征信息包括第三方提供的用户数据或业务平台采集得到的用户数据。
  8. 如权利要求1-7中任一项所述的用户信用评估方法，所述方法进一步包括：
    根据目标用户的综合信用评分为目标用户推送产品信息；或
    根据目标用户的综合信用评分对目标用户的数据业务进行监控管理。
  9. 一种用户信用评估装置，其特征在于，所述装置包括：
    离线特征获取模块，用于获取目标用户的离线特征信息，所述离线特征信息为按照预设更新周期进行更新的用户的特征信息；
    离线评分模块，用于根据目标用户的离线特征信息以及预设的离线预测模型，计算目标用户的离线信用评分；
    实时特征获取模块，用于获取目标用户的实时特征信息，所述实时特征信息为距离当前预设时间范围内采集到的用户的特征信息，所述预设时间范围小于所述预设更新周期；
    实时评分模块，用于根据目标用户的实时特征信息以及预设的实时预测模型，计算目标用户的实时信用评分；
    综合评分模块，用于根据得到的目标用户的离线信用评分和实时信用评分结合预设的综合预测模型，计算目标用户的综合信用评分。
  10. 如权利要求9所述的用户信用评估装置，进一步地所述装置还包括：
    样本获取模块，用于获取多个用户的信用评分结果样本和各个用户的离线特征信息；
    离线模型训练模块，用于建立所述离线预测模型，并根据所述多个用户的信用评分结果样本和各个用户的离线特征信息对所述离线预测模型进行训练，获得所述离线预测模型的模型参数。
  11. 如权利要求10所述的用户信用评估装置，其中，所述样本获取模块包括：
    离线样本获取单元，用于获取多个用户的信用评分结果样本和各个用户的多个特征类别的离线特征信息样本；
    相关度计算单元，用于根据所述多个用户的信用评分结果样本和各个用户的多个特征类别的离线特征信息样本，计算各个特征类别与信用评分结果之间的相关度；
    特征类别筛选单元，用于将与信用评分结果之间的相关度达到预设阈值的特征类别确定为所述离线特征信息的特征类别，并从所述各个用户的多个特征类别的离线特征信息样本中筛选出确定的所述特征类别的离线特征信息。
  12. 如权利要求9所述的用户信用评估装置，进一步地所述装置还包括：
    样本获取模块，用于获取多个用户的信用评分结果样本和各个用户的实时特征信息；
    实时模型训练模块，用于建立所述实时预测模型，并根据所述多个用户的信用评分结果样本和各个用户的实时特征信息对所述实时预测模型进行训练，获得所述实时预测模型的模型参数。
  13. 如权利要求12所述的用户信用评估装置，其中，所述样本获取模块包括：
    实时样本获取单元，用于获取多个用户的信用评分结果样本和各个用户的多个特征类别的实时特征信息样本；
    相关度计算单元，用于根据所述多个用户的信用评分结果样本和各个用户的多个特征类别的实时特征信息样本，计算各个特征类别与信用评分结果之间的相关度；
    特征类别筛选单元，用于将与信用评分结果之间的相关度达到预设阈值的特征类别确定为所述实时特征信息的特征类别，并从所述各个用 户的多个特征类别的实时特征信息样本中筛选出确定的所述特征类别的实时特征信息。
  14. 如权利要求9所述的用户信用评估装置，进一步地所述装置还包括：
    样本获取模块，获取多个用户的信用评分结果样本和各个用户的离线特征信息和实时特征信息；
    所述离线评分模块，还用于根据所述各个用户的离线特征信息以及预设的离线预测模型，计算所述各个用户的离线信用评分；
    所述实时评分模块，还用于根据所述各个用户的实时特征信息以及预设的实时预测模型，计算所述各个用户的实时信用评分；
    综合模型训练模块，用于建立所述综合预测模型，并根据所述多个用户的信用评分结果以及各个用户的离线信用评分和实时信用评分对所述综合预测模型进行训练，获得所述综合预测模型的模型参数。
  15. 如权利要求9所述的用户信用评估装置，其中，所述实时特征信息为业务平台采集得到的用户数据；
    所述离线特征信息包括第三方提供的用户数据和业务平台采集得到的用户数据。
  16. 如权利要求9-15中任一项所述的用户信用评估装置，进一步地，所述装置还包括：
    信息推送模块，用于根据目标用户的综合信用评分为目标用户推送产品信息；或
    业务监控模块，用于根据目标用户的综合信用评分对目标用户的数 据业务进行监控管理。
  17. 一种计算机可读存储介质，存储有计算机可读指令，可以使至少一个处理器执行如权利要求1-8任一项所述的方法。
PCT/CN2017/085049 2016-06-12 2017-05-19 一种用户信用评估方法、装置及存储介质 WO2017215403A1 (zh)

Priority Applications (4)

Application Number Priority Date Filing Date Title
JP2018543338A JP6732034B2 (ja) 2016-06-12 2017-05-19 ユーザー・クレジット評価方法、装置及び記憶媒体
EP17812514.2A EP3471046A1 (en) 2016-06-12 2017-05-19 Method and apparatus for assessing user credit, and storage medium
KR1020187029224A KR102178633B1 (ko) 2016-06-12 2017-05-19 사용자 신용 평가 방법 및 장치, 그리고 저장 매체
US15/954,710 US20180232805A1 (en) 2016-06-12 2018-04-17 User credit rating method and apparatus, and storage medium

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201610416661.1A CN106127363B (zh) 2016-06-12 2016-06-12 一种用户信用评估方法和装置
CN201610416661.1 2016-06-12

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US15/954,710 Continuation US20180232805A1 (en) 2016-06-12 2018-04-17 User credit rating method and apparatus, and storage medium

Publications (1)

Publication Number Publication Date
WO2017215403A1 true WO2017215403A1 (zh) 2017-12-21

Family

ID=57269931

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/085049 WO2017215403A1 (zh) 2016-06-12 2017-05-19 一种用户信用评估方法、装置及存储介质

Country Status (6)

Country Link
US (1) US20180232805A1 (zh)
EP (1) EP3471046A1 (zh)
JP (1) JP6732034B2 (zh)
KR (1) KR102178633B1 (zh)
CN (1) CN106127363B (zh)
WO (1) WO2017215403A1 (zh)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111598275A (zh) * 2020-04-03 2020-08-28 福建星云电子股份有限公司 一种电动汽车信用分评测方法、装置、设备和介质
CN111860299A (zh) * 2020-07-17 2020-10-30 北京奇艺世纪科技有限公司 目标对象的等级确定方法、装置、电子设备及存储介质
CN112862593A (zh) * 2021-01-28 2021-05-28 深圳前海微众银行股份有限公司 信用评分卡模型训练方法、装置、系统及计算机存储介质
CN113011966A (zh) * 2021-03-18 2021-06-22 中国光大银行股份有限公司 基于深度学习的信用评分方法及装置

Families Citing this family (37)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106127363B (zh) * 2016-06-12 2022-04-15 腾讯科技（深圳）有限公司 一种用户信用评估方法和装置
CN108074122A (zh) * 2016-11-18 2018-05-25 腾讯科技（深圳）有限公司 产品试用推荐方法、装置及服务器
CN108416662B (zh) * 2017-02-10 2021-09-21 腾讯科技（深圳）有限公司 一种数据验证方法及装置
CN107330445B (zh) * 2017-05-31 2020-06-05 北京京东尚科信息技术有限公司 用户属性的预测方法和装置
CN107766418A (zh) * 2017-09-08 2018-03-06 广州汪汪信息技术有限公司 一种基于融合模型的信用评估方法、电子设备和存储介质
CN109559214A (zh) 2017-09-27 2019-04-02 阿里巴巴集团控股有限公司 虚拟资源分配、模型建立、数据预测方法及装置
US11823258B2 (en) * 2017-10-11 2023-11-21 Mx Technologies, Inc. Aggregation based credit decision
CN107993140A (zh) * 2017-11-22 2018-05-04 深圳市耐飞科技有限公司 一种个人信贷风险评估方法及系统
CN107944738A (zh) * 2017-12-07 2018-04-20 税友软件集团股份有限公司 一种税务信用积分计算方法及装置
CN110322334A (zh) * 2018-03-29 2019-10-11 上海麦子资产管理集团有限公司 信用评级方法及装置、计算机可读存储介质、终端
CN108846687A (zh) * 2018-04-02 2018-11-20 平安科技（深圳）有限公司 客户分类方法、装置及存储介质
CN110634060A (zh) * 2018-06-21 2019-12-31 马上消费金融股份有限公司 一种用户信用风险的评估方法、系统、装置及存储介质
CN109191096A (zh) * 2018-08-22 2019-01-11 阿里巴巴集团控股有限公司 一种签约风险量化方法、代扣风险量化方法、装置及设备
CN109461016B (zh) * 2018-09-10 2023-05-05 平安科技（深圳）有限公司 数据评分方法、装置、计算机设备及存储介质
CN111340265A (zh) * 2018-12-19 2020-06-26 北京嘀嘀无限科技发展有限公司 司机下线干预方法、装置、电子设备和计算机存储介质
CN110060144B (zh) * 2019-03-18 2024-01-30 平安科技（深圳）有限公司 额度模型训练方法、额度评估方法、装置、设备及介质
US10977729B2 (en) * 2019-03-18 2021-04-13 Zestfinance, Inc. Systems and methods for model fairness
CN110135972A (zh) * 2019-04-23 2019-08-16 上海淇玥信息技术有限公司 一种提高用户动支率的方法、装置、系统和记录介质
CN110222894A (zh) * 2019-06-06 2019-09-10 阿里巴巴集团控股有限公司 广告投放方法、装置及设备
CN110322343A (zh) * 2019-07-02 2019-10-11 上海上湖信息技术有限公司 一种用户全生命周期信用预测方法、装置和计算机设备
KR102156757B1 (ko) * 2019-09-27 2020-09-16 (주)데이터리퍼블릭 기계 학습을 이용한 신용 평가를 위한 시스템, 방법, 및 컴퓨터 프로그램
CN110889759A (zh) * 2019-11-21 2020-03-17 北京三快在线科技有限公司 信用数据的确定方法、装置及存储介质
CN111027935A (zh) * 2019-12-10 2020-04-17 支付宝(杭州)信息技术有限公司 基于信用的电子签证申请方法以及装置
CN111339134B (zh) * 2020-02-11 2024-03-08 广州拉卡拉信息技术有限公司 一种数据查询方法及装置
CN111738549A (zh) * 2020-05-21 2020-10-02 平安国际智慧城市科技股份有限公司 食品安全风险评估方法、装置、设备及存储介质
KR102385054B1 (ko) 2020-05-26 2022-04-08 주식회사 다날 인공지능 기반의 신용등급 변동 예측 처리 장치 및 그 동작 방법
CN111666191B (zh) * 2020-06-09 2023-09-29 贝壳技术有限公司 数据质量监控方法、装置、电子设备及存储介质
CN112419050B (zh) * 2020-12-24 2022-05-24 浙江工商大学 基于电话通讯网络和社交行为的信用评估方法及装置
CN112749980B (zh) * 2021-01-13 2021-12-07 深圳市恒鑫科技服务有限公司 一种基于区块链的信用资产处理方法及系统
CN112906772A (zh) * 2021-02-04 2021-06-04 深圳前海微众银行股份有限公司 样本处理方法、装置、设备及计算机可读存储介质
CN113159924A (zh) * 2021-04-30 2021-07-23 中国银行股份有限公司 授信客户对象的确定方法及装置
CN113435764B (zh) * 2021-07-05 2023-01-31 深圳前海微众银行股份有限公司 风险因素追踪方法、装置、设备及计算机可读存储介质
KR102368010B1 (ko) * 2021-07-20 2022-02-25 리포츠 주식회사 운동 생활정보에 기초한 인공지능 기반의 대안적 신용평가정보 제공 방법 및 시스템
KR20240066782A (ko) 2022-11-08 2024-05-16 디티에쓰아이 주식회사 빅데이터와 기계학습을 이용한 대출 연체 정보 예측 시스템 및 그 방법
CN116205376B (zh) * 2023-04-27 2023-10-17 北京阿帕科蓝科技有限公司 行为预测方法、行为预测模型的训练方法和装置
CN117541318B (zh) * 2024-01-09 2024-04-02 前海超级前台(深圳)信息技术有限公司 一种离线消费智能评估监管方法、系统和介质
CN117808578A (zh) * 2024-03-01 2024-04-02 杭银消费金融股份有限公司 一种智能化人行征信数据解析方法及系统

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101996381A (zh) * 2009-08-14 2011-03-30 中国工商银行股份有限公司 一种零售资产风险的计算方法及系统
KR20120114713A (ko) * 2011-04-08 2012-10-17 나이스신용평가정보주식회사 대출 진단 시뮬레이션 서비스 시스템 및 방법
CN104866969A (zh) * 2015-05-25 2015-08-26 百度在线网络技术（北京）有限公司 个人信用数据处理方法和装置
CN105260471A (zh) * 2015-10-19 2016-01-20 广州唯品会信息科技有限公司 商品个性化排序模型训练方法及系统
CN106097043A (zh) * 2016-06-01 2016-11-09 腾讯科技（深圳）有限公司 一种信用数据的处理方法及服务器
CN106127363A (zh) * 2016-06-12 2016-11-16 腾讯科技（深圳）有限公司 一种用户信用评估方法和装置

Family Cites Families (20)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6088686A (en) * 1995-12-12 2000-07-11 Citibank, N.A. System and method to performing on-line credit reviews and approvals
JP2001229264A (ja) * 2000-02-18 2001-08-24 Dainippon Printing Co Ltd スマートカードによる与信・認証ビジネスシステムとそれに使用するスマートカード
US20040030667A1 (en) * 2002-08-02 2004-02-12 Capital One Financial Corporation Automated systems and methods for generating statistical models
CN1598831A (zh) * 2004-08-06 2005-03-23 武燕华 个人信用数据管理系统及方法
US8036979B1 (en) * 2006-10-05 2011-10-11 Experian Information Solutions, Inc. System and method for generating a finance attribute from tradeline data
US20090327120A1 (en) * 2008-06-27 2009-12-31 Eze Ike O Tagged Credit Profile System for Credit Applicants
CN101937541A (zh) * 2009-06-30 2011-01-05 商文彬 一种用于评价客户信用度的方法及设备
KR20130008130A (ko) * 2011-07-11 2013-01-22 최원국 스마트 기기의 푸시 알림 기능을 이용한 대출 중개 서비스 제공 방법
CN102346901A (zh) * 2011-11-22 2012-02-08 北京信城通数码科技有限公司 一种互联网药品交易主体信用评估系统及其方法
IN2014DN08176A (zh) * 2012-03-31 2015-05-01 Trans Union Llc
US20140074688A1 (en) * 2012-09-13 2014-03-13 Rawllin International Inc. Behavioral based score
JP2014071532A (ja) * 2012-09-27 2014-04-21 Mycredit Kk 個人信用情報提供装置
TWM488068U (zh) * 2013-12-13 2014-10-11 Global Opto Technology Corp 借貸媒合平臺系統
KR101524971B1 (ko) * 2014-02-11 2015-06-02 숭실대학교산학협력단 개인 성향 예측 방법 및 그 장치
JP5852218B1 (ja) * 2014-12-19 2016-02-03 ヤフー株式会社 生成装置、生成方法及び生成プログラム
CN104636447B (zh) * 2015-01-21 2017-12-29 上海天呈医流科技股份有限公司 一种面向医疗器械b2b网站用户的智能评价方法和系统
CN105069683A (zh) * 2015-07-24 2015-11-18 广州时韵信息科技有限公司 一种纳税风险评估系统
US10387939B2 (en) * 2015-08-21 2019-08-20 Q2 Software, Inc. Method, medium, and system for surfacing recommendations
CN105512815A (zh) * 2015-11-30 2016-04-20 安徽融信金模信息技术有限公司 一种用于企业风险评估的系统
CN105447752A (zh) * 2015-12-08 2016-03-30 安徽融信金模信息技术有限公司 一种基于信息共用的企业信用评估系统

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101996381A (zh) * 2009-08-14 2011-03-30 中国工商银行股份有限公司 一种零售资产风险的计算方法及系统
KR20120114713A (ko) * 2011-04-08 2012-10-17 나이스신용평가정보주식회사 대출 진단 시뮬레이션 서비스 시스템 및 방법
CN104866969A (zh) * 2015-05-25 2015-08-26 百度在线网络技术（北京）有限公司 个人信用数据处理方法和装置
CN105260471A (zh) * 2015-10-19 2016-01-20 广州唯品会信息科技有限公司 商品个性化排序模型训练方法及系统
CN106097043A (zh) * 2016-06-01 2016-11-09 腾讯科技（深圳）有限公司 一种信用数据的处理方法及服务器
CN106127363A (zh) * 2016-06-12 2016-11-16 腾讯科技（深圳）有限公司 一种用户信用评估方法和装置

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP3471046A4 *

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111598275A (zh) * 2020-04-03 2020-08-28 福建星云电子股份有限公司 一种电动汽车信用分评测方法、装置、设备和介质
CN111598275B (zh) * 2020-04-03 2022-11-11 福建星云电子股份有限公司 一种电动汽车信用分评测方法、装置、设备和介质
CN111860299A (zh) * 2020-07-17 2020-10-30 北京奇艺世纪科技有限公司 目标对象的等级确定方法、装置、电子设备及存储介质
CN111860299B (zh) * 2020-07-17 2023-09-08 北京奇艺世纪科技有限公司 目标对象的等级确定方法、装置、电子设备及存储介质
CN112862593A (zh) * 2021-01-28 2021-05-28 深圳前海微众银行股份有限公司 信用评分卡模型训练方法、装置、系统及计算机存储介质
CN112862593B (zh) * 2021-01-28 2024-05-03 深圳前海微众银行股份有限公司 信用评分卡模型训练方法、装置、系统及计算机存储介质
CN113011966A (zh) * 2021-03-18 2021-06-22 中国光大银行股份有限公司 基于深度学习的信用评分方法及装置

Also Published As

Publication number Publication date
EP3471046A4 (en) 2019-04-17
EP3471046A1 (en) 2019-04-17
KR20180119674A (ko) 2018-11-02
CN106127363B (zh) 2022-04-15
KR102178633B1 (ko) 2020-11-13
US20180232805A1 (en) 2018-08-16
CN106127363A (zh) 2016-11-16
JP2019509556A (ja) 2019-04-04
JP6732034B2 (ja) 2020-07-29

Similar Documents

Publication Publication Date Title
WO2017215403A1 (zh) 一种用户信用评估方法、装置及存储介质
JP6930013B2 (ja) パフォーマンスモデル悪影響補正
CN110070391B (zh) 数据处理方法、装置、计算机可读介质及电子设备
WO2020037942A1 (zh) 风险预测处理方法、装置、计算机设备和介质
CN105488539B (zh) 分类模型的生成方法及装置、系统容量的预估方法及装置
AU2017258946A1 (en) Automatic interview question recommendation and analysis
CA2935281C (en) A multidimensional recursive learning process and system used to discover complex dyadic or multiple counterparty relationships
CN111160959B (zh) 一种用户点击转化预估方法及装置
CN114238033B (zh) 板卡运行状态预警方法、装置、设备及可读存储介质
CN111708682B (zh) 数据预测方法、装置、设备及存储介质
WO2017071369A1 (zh) 一种预测用户离网的方法和设备
CN111754241A (zh) 一种用户行为感知方法、装置、设备及介质
CN112632179A (zh) 模型构建方法、装置、存储介质及设备
CN107451249B (zh) 事件发展趋势的预测方法及装置
CN115063035A (zh) 基于神经网络的客户评估方法、系统、设备及存储介质
CN117132383A (zh) 一种信贷数据处理方法、装置、设备及可读存储介质
TWI639091B (zh) Big data based automated analysis processing system
KR101884907B1 (ko) 빅데이터 분석 기반 신뢰도 예측 방법
CN116167454A (zh) 智能化的二分类模型训练方法及装置
US20220101061A1 (en) Automatically identifying and generating machine learning prediction models for data input fields
KR101613397B1 (ko) 시계열 텍스트 데이터 및 시계열 수치 데이터의 연관 방법 및 그 장치
CN113610225A (zh) 质量评估模型训练方法、装置、电子设备及存储介质
CN113032643B (zh) 目标行为识别系统
CN117493937A (zh) 一种归因模型构建方法、系统、设备及存储介质
Luby Strengthening analyses of line-up procedures: a log-linear model framework

Legal Events

Date Code Title Description
ENP Entry into the national phase

Ref document number: 2018543338

Country of ref document: JP

Kind code of ref document: A

ENP Entry into the national phase

Ref document number: 20187029224

Country of ref document: KR

Kind code of ref document: A

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

Ref document number: 17812514

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

ENP Entry into the national phase

Ref document number: 2017812514

Country of ref document: EP

Effective date: 20190114