WO2019080407A1 - 信贷评估方法、装置、设备及计算机可读存储介质 - Google Patents

信贷评估方法、装置、设备及计算机可读存储介质

Info

Publication number
WO2019080407A1
WO2019080407A1 PCT/CN2018/075663 CN2018075663W WO2019080407A1 WO 2019080407 A1 WO2019080407 A1 WO 2019080407A1 CN 2018075663 W CN2018075663 W CN 2018075663W WO 2019080407 A1 WO2019080407 A1 WO 2019080407A1
Authority
WO
WIPO (PCT)
Prior art keywords
data
credit
evaluation
user
expert
Prior art date
Application number
PCT/CN2018/075663
Other languages
English (en)
French (fr)
Inventor
李天平
Original Assignee
深圳壹账通智能科技有限公司
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 深圳壹账通智能科技有限公司 filed Critical 深圳壹账通智能科技有限公司
Publication of WO2019080407A1 publication Critical patent/WO2019080407A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION 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
    • G06Q40/00Finance; Insurance; Tax strategies; Processing of corporate or income taxes
    • G06Q40/03Credit; Loans; Processing thereof

Definitions

  • the present application relates to the field of financial credit, and in particular, to a credit evaluation method, apparatus, device, and computer readable storage medium.
  • the traditional method of credit approval is to manually review the borrowing request of the borrowing user by the staff of the financial institution.
  • the staff needs to check the large amount of data information of the borrowing user to determine the risk of the loan transaction, for example, the borrowing user borrows How much money can be paid, whether it can be repaid within the loan period, etc.
  • financial institutions began to use computer risk control systems to evaluate credit requests.
  • the wind control model is an important part of the entire risk control system. It is mainly used to identify, analyze and evaluate user data. Thereby the conclusion of the evaluation is obtained.
  • the current wind control model mainly uses the static rule engine of drools, and introduces expert rules to evaluate the credit request (that is, convert the human experience into a computer logic algorithm, and then use the computer to identify the credit request).
  • the human experience is not necessarily correct.
  • the human experience is susceptible to subjective factors, which affects the accuracy of the risk control model assessment, making the wind control model unable to give a reasonable loan plan, or even difficult to identify conscious fraud. Thereby increasing the bad debt rate of the loan.
  • the main purpose of the present application is to provide a credit evaluation method, apparatus, device and computer readable storage medium, which aim to improve the evaluation capability of the wind control model and reduce the loan bad debt rate.
  • the present application provides a credit evaluation method, and the credit evaluation method includes the following steps:
  • the historical credit data includes historical pre-lending data and historical post-loan data.
  • the step of acquiring historical credit data and constructing an AI wind control engine according to the historical credit data includes:
  • the historical pre-lending data and the historical post-lost data are respectively used as an input training set and an output training set, and the initial machine learning model is analyzed and trained based on the input training set and the output training set to obtain an AI wind control engine.
  • the initial machine learning model is cold initiated by expert rules.
  • the input element in the input training set includes a multi-dimensional vector x of dimension m
  • the output element in the output training set includes a loan non-performing rate y
  • the analyzing and training the initial machine learning model based on the input training set and the output training set, and the steps of obtaining the AI wind control engine include:
  • ⁇ m is a parameter of the decision tree
  • h(x, D) is a hypothesis function, which is a decision result of the decision tree, and the first-order expansion of the Taylor formula of the cost function f(x) is:
  • the AI wind control engine is obtained according to the linear combination tree and the cost function.
  • the steps of the assessment include:
  • the step of acquiring the expert evaluation conclusion of the preset expert engine and the AI evaluation conclusion of the AI wind control engine, and generating corresponding decision suggestions based on the expert evaluation conclusion and the AI evaluation conclusion includes:
  • a corresponding loan recommendation is generated according to the credit line, wherein the loan recommendation includes a loan amount and a loan period.
  • the step of acquiring the expert evaluation result of the preset expert engine and the AI evaluation result of the AI wind control engine, and generating a corresponding decision suggestion based on the expert evaluation conclusion and the AI evaluation conclusion include:
  • the AI wind control engine is iteratively trained by the iterative training samples to adjust the AI wind control engine.
  • the present application further provides a credit evaluation apparatus, where the credit evaluation apparatus includes:
  • An engine construction module configured to acquire historical credit data, and construct an AI wind control engine according to the historical credit data
  • a credit evaluation module configured to acquire, before receiving the loan request, the pre-lending data of the borrowing user corresponding to the borrowing request, and input the pre-lending data of the user into a preset expert engine and the AI wind control engine for credit Evaluation
  • the suggestion generating module is configured to obtain an expert evaluation conclusion of the preset expert engine and an AI evaluation conclusion of the AI wind control engine, and generate corresponding decision suggestions based on the expert evaluation conclusion and the AI evaluation conclusion.
  • the present application further provides a credit evaluation apparatus including a processor, a memory, and a credit evaluation program stored on the memory and executable by the processor, wherein When the credit evaluation program is executed by the processor, the following steps are implemented:
  • the present application further provides a computer readable storage medium having a credit evaluation program stored thereon, wherein when the credit evaluation program is executed by a processor, the following steps are implemented:
  • FIG. 1 is a schematic structural diagram of hardware of a credit evaluation apparatus involved in an embodiment of the present application
  • FIG. 2 is a schematic flow chart of a first embodiment of a credit evaluation method according to the present application.
  • 3 is the user pre-credit data of the borrowing user corresponding to the borrowing request, and input the user pre-lending data into the preset expert engine and the AI wind control engine respectively when receiving the borrowing request.
  • FIG. 4 is a schematic flow chart of a second embodiment of a credit evaluation method according to the present application.
  • FIG. 5 is a schematic diagram of functional modules of a first embodiment of a credit evaluation apparatus according to the present application.
  • FIG. 1 is a schematic structural diagram of hardware of a credit evaluation apparatus involved in an embodiment of the present application.
  • the credit evaluation apparatus in the embodiment of the present application may include a processor 1001 (for example, a CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005.
  • the communication bus 1002 is configured to implement connection communication between the components;
  • the user interface 1003 may include a display, an input unit such as a keyboard;
  • the network interface 1004 may optionally include a standard wired interface and a wireless interface.
  • the memory 1005 may be a high-speed RAM memory, or may be a non-volatile memory, such as a disk storage, and the memory 1005 may alternatively be a storage device independent of the processor 1001 described above.
  • the hardware structure of the credit evaluation device shown in FIG. 1 does not constitute a limitation of the credit evaluation device, and may include more or less components than those illustrated, or combine some components, or different. Parts layout.
  • the memory 1005 as a computer readable storage medium of FIG. 1 can include an operating system, a network communication module, and a credit evaluation program.
  • the network communication module is mainly used to connect to the database and perform data communication with the database; and the processor 1001 can call the credit evaluation program stored in the memory 1005 and execute various embodiments of the credit evaluation method of the embodiment of the present application.
  • the application provides a credit evaluation method.
  • FIG. 2 is a schematic flowchart of a first embodiment of a credit evaluation method according to the present application.
  • the credit evaluation method includes the following steps:
  • Step S10 Obtain historical credit data, and construct an AI wind control engine according to the historical credit data;
  • a credit evaluation method which adopts a dual-core wind control engine to evaluate and make decisions on borrowing requests, reduces the adverse effects of human thinking limitations on the wind control model, and improves the accuracy of the risk control system, thereby reducing credit bad debts. rate.
  • an AI wind control engine which is different from a conventional expert engine.
  • the risk control expert or related staff will analyze the experience (rule) after analyzing a large amount of loan data, and then convert these experience summaries into computer logic algorithms, thus forming the evaluation rules of the expert engine.
  • It is called expert rule); but human experience is not necessarily correct. Human experience is easily influenced by subjective factors, which makes the expert rules have certain defects and reduces the evaluation ability of the risk control engine using expert rules. Therefore, the AI wind control engine construction process in this embodiment does not completely adopt the expert rule, but constructs the AI wind control engine by means of machine learning. Among them, machine learning means not relying on humans to sum up experience and input logic.
  • the database is first queried to obtain historical credit data, and the historical credit data is a historical lending record of the financial institution, including behavior data provided by the user before the loan (such as credit status, personal financial resources, bank transaction records, social security payment status).
  • the process of constructing the AI wind control engine in this embodiment adopts algorithms including, but not limited to, logistic regression, decision tree, GBDT, SVM, and the like.
  • the sample set includes real estate, income, age, education, etc.
  • a combination of m vector groups can be defined, where x is a multidimensional vector and y is the non-performing rate of the loan. That is
  • T(x; ⁇ m ) represents a decision tree, which can be regarded as a regression tree
  • ⁇ m is a parameter of the decision tree
  • n is the number of decision trees.
  • Equation 3 h(x, D) is a hypothesis function, which is the decision result of the decision tree
  • the staff can make manual adjustments to improve the AI wind control engine.
  • Step S20 Upon receiving the loan request, obtain the user loan pre-lending data corresponding to the borrowing user, and input the user pre-lending data into the preset expert engine and the AI wind control engine for credit evaluation;
  • the AI wind control engine can be used for loan evaluation.
  • the AI wind control engine and the preset expert engine are simultaneously used, and the credit evaluation is performed by using the dual-core wind control method.
  • the meaning of the preset expert engine is as described above. It is the experience (regular) summary of the analysis of a large amount of loan data by the risk control expert or related staff, and then the summary of these experiences is converted into the logic algorithm of the computer, thus constituting the Expert engine evaluation rules.
  • the financial institution's risk control model obtains the relevant pre-lending data of the borrowing user when receiving the borrowing request from the borrowing user (the borrowing request includes the borrowing amount and the borrowing period, etc.) (pre-loan data includes the borrower's Credit status, personal financial resources, bank record, social security payment, call data, etc., and input these risk control data into the preset expert engine and AI wind control engine respectively, and respectively through the preset expert engine and AI wind control engine Evaluate the loan request.
  • Step S30 Obtain an expert evaluation conclusion of the preset expert engine and an AI evaluation conclusion of the AI wind control engine, and generate a corresponding decision suggestion based on the expert evaluation conclusion and the AI evaluation conclusion.
  • the corresponding expert evaluation conclusion and the AI evaluation conclusion are obtained, and the combination of the expert evaluation and the AI evaluation conclusion can be used to obtain the decision suggestion (such as the amount of the loan, the loan period, etc.).
  • the expert evaluation conclusion and the AI evaluation conclusion may be represented by a score, and the score ranges from 0 to 100; the higher the score, the lower the bad debt rate of the borrowing user's loan request;
  • the loan request is evaluated by the preset expert engine and the AI wind control engine, and the expert score and the AI score are obtained, the expert score and the AI score need to be weighted and run.
  • a comprehensive evaluation value the meaning of the comprehensive evaluation value is similar to the evaluation score obtained by the single-core evaluation, and the range is also between 0 and 100.
  • the credit limit represents the recommended maximum amount and maximum loan period for lending to the borrower; by calculating the comprehensive evaluation value of the borrower, the borrower can be determined.
  • the credit line which generates the corresponding loan proposal (including the loan amount and the loan period).
  • the evaluation conclusion can be divided into two points (ie, agreeing to release the money and rejecting the two points of the loan), and the generation process of the decision suggestion can also be implemented in other ways. It does not constitute a limitation on the credit evaluation method of the present embodiment.
  • the wind control model may also generate a corresponding credit evaluation report.
  • the credit evaluation report includes the pre-lending data of the borrowing users used in the evaluation process, as well as the expert evaluation conclusions obtained from the pre-lending data evaluation, the AI evaluation conclusions, and the final decision-making recommendations.
  • the display manner of the data can be customized.
  • the pre-lending data includes the total expenditure of the borrower in the first three months, and the monthly expenditure includes hydropower expenditure, telephone business expenditure, credit card consumption, etc. For each monthly expenditure, a pie chart can be used.
  • the historical credit data is obtained, and the AI wind control engine is constructed according to the historical credit data; when the borrowing request is received, the borrowing request is obtained corresponding to the user's pre-lending data of the borrowing user, and the user is The pre-lending data is respectively input into a preset expert engine and the AI wind control engine for credit evaluation; obtaining an expert evaluation conclusion of the preset expert engine and an AI evaluation conclusion of the AI wind control engine, and based on the expert evaluation conclusion And the AI assessment conclusions generate corresponding decision suggestions.
  • the embodiment constructs the AI wind control engine by the machine learning method, and simultaneously evaluates the loan request through the AI wind control engine and the expert rule engine, and obtains the decision suggestion by using the dual engine evaluation method, thereby greatly reducing the limitation of human thinking.
  • the adverse effects of sex on the engine construction and credit evaluation process improve the assessment ability of the risk control model, and effectively reduce the bad debt rate of loans.
  • FIG. 3 is the user pre-credit data of the borrowing user corresponding to the borrowing request when the borrowing request is received, and the user pre-lending data is input into the preset expert engine and the AI respectively.
  • step S20 includes:
  • Step S21 When receiving the loan request, generate a corresponding data acquisition request according to the user identifier and the authorization permission included in the loan request, and send the data acquisition request to the data management system to obtain the corresponding user loan data;
  • Step S22 Receive user pre-lending data fed back by the data management system, and input the user pre-lending data into the preset expert engine and the AI wind control engine for credit evaluation.
  • the pre-lending data of the borrowing user often involves the privacy of the borrowing user, and the pre-lending data needs to be obtained from other data sources.
  • the borrowing user's bank card consumption record needs to be obtained from the banking system.
  • the borrower's billing data needs to be obtained from the telecommunications company; if the risk control model wants to obtain these pre-lending data, it needs to obtain the license and authorization of the borrowing user first.
  • the borrowing request includes the identity of the borrowing user (such as identity information such as the ID number), and the license information of the borrowing user (such as an electronic signature) to indicate the borrowing.
  • Users allow the risk control model (financial institution) to obtain their own relevant pre-lending data for credit evaluation.
  • the wind control model Upon receiving the loan request, the wind control model generates a corresponding data acquisition request according to the user identification and the authorization, and sends the data acquisition request to the relevant data management system (such as a banking system, a telecommunications company, etc.).
  • the data management system When receiving the data acquisition request, the data management system first obtains the user identifier therein, and determines which data acquisition request is related data of which user needs to be acquired; then, the authorization license is verified, and the authorized license is determined. The authenticity determines whether the risk control model requesting the data acquisition is a legal acquirer; when the authorization verification passes, the data management system will query the corresponding user pre-lending data and return the data to the wind control model.
  • the wind control model receives the user pre-loan data fed back by the data management system, the user pre-lending data can be separately input into the preset expert engine and the AI wind control engine for credit evaluation.
  • FIG. 4 is a schematic flowchart of a second embodiment of a credit evaluation method according to the present application.
  • the method further includes:
  • Step S40 acquiring user loan data of the borrowing user, and combining the user pre-lending data and the user loan data to obtain an iterative training sample;
  • Step S50 Perform iterative training on the AI wind control engine through the iterative training samples to adjust the AI wind control engine.
  • the AI rule evaluation logic
  • the AI engine needs to be constantly revised and adjusted. Specifically, after determining the loan to the borrowing user, the risk control model will continue to obtain the post-loan data of the borrowing user (such as repayment records, consumption records, etc.); and then combine the pre-lending data obtained in step S20 with the post-loan data.
  • the iterative training sample is continuously adjusted and adjusted by the AI engine of the iterative training sample pair, and the adjusted content includes changing the threshold of the parameter, changing the priority of some wind control data or evaluation factor, changing the weight of the data, etc.; for example, the AI engine In the strategy algorithm, the data weighting level is 1 to 5. For the data provided by the user when borrowing, the A data weighting level is 5, which means that the A data has a large reference value for judging whether the user can repay the loan on time; Some post-loan data found that the reference value of A data is not as big as expected. At this time, the AI engine will correct the original policy algorithm and reduce the weight level of A data.
  • the AI engine After adjusting the AI engine, the possibility of overdue or bad debts of the borrowing user may be given in the new loan cycle, and the actual post-loan data of the new cycle will be used for feedback correction.
  • the AI engine is continuously improved, and the evaluation accuracy is gradually improved.
  • the present application also provides a credit evaluation device.
  • FIG. 5 is a schematic diagram of functional modules of a first embodiment of a credit evaluation apparatus according to the present application.
  • the credit evaluation apparatus includes:
  • An engine construction module 10 configured to acquire historical credit data, and construct an AI wind control engine according to the historical credit data;
  • the credit evaluation module 20 is configured to acquire, before receiving the loan request, the pre-lending data of the borrowing user corresponding to the borrowing request, and input the pre-lending data of the user into the preset expert engine and the AI wind control engine respectively. Credit evaluation;
  • the suggestion generating module 30 is configured to obtain an expert evaluation conclusion of the preset expert engine and an AI evaluation result of the AI wind control engine, and generate corresponding decision suggestions based on the expert evaluation conclusion and the AI evaluation conclusion
  • the historical credit data includes historical pre-lending data and historical post-loan data
  • the engine construction module 10 is further configured to separately use the historical pre-lending data and historical post-loan data as an input training set and an output training set. And based on the above input training set and output training set, the initial machine learning model is analyzed and trained to obtain an AI wind control engine.
  • the initial machine learning model is cold initiated by expert rules.
  • the input element in the input training set includes a multi-dimensional vector x of dimension m
  • the output element in the output training set includes a loan non-performing rate y.
  • the engine construction module 10 is further configured to combine the input element and the output element to define m multi-dimensional training vector groups.
  • ⁇ m is a parameter of the decision tree
  • h(x, D) is a hypothesis function, which is a decision result of the decision tree, and the first-order expansion of the Taylor formula of the cost function f(x) is:
  • the AI wind control engine is obtained according to the linear combination tree and the cost function.
  • the credit evaluation module 20 includes:
  • a data obtaining unit configured to generate a corresponding data acquisition request according to the user identifier and the authorization permission included in the loan request, and send the data acquisition request to the data management system to obtain a corresponding User pre-lending data;
  • the credit evaluation unit is configured to receive user pre-lending data fed back by the data management system, and input the user pre-lending data into the preset expert engine and the AI wind control engine for credit evaluation.
  • the suggestion generating module 30 includes:
  • a weighting operation unit configured to obtain an expert evaluation conclusion of the preset expert engine and an AI evaluation result of the AI wind control engine, and perform a weighting operation based on the expert evaluation conclusion and the AI evaluation conclusion to obtain the borrowing user credits;
  • the suggestion generating unit is configured to generate a corresponding loan proposal according to the credit limit, wherein the loan recommendation includes a loan amount and a loan period.
  • the credit evaluation device further includes:
  • a data combination module configured to obtain data of the user's credits of the borrowing user, and combine the user pre-lending data and the user's post-loan data to obtain an iterative training sample
  • An iterative training module is configured to iteratively train the AI wind control engine through the iterative training samples to adjust the AI wind control engine.
  • the present application also provides a computer readable storage medium.
  • a credit evaluation program is stored on the computer readable storage medium of the present application, and the computer readable storage medium stores a credit evaluation program, wherein when the credit evaluation program is executed by the processor, the steps of the credit evaluation method as described above are implemented.

Landscapes

  • Business, Economics & Management (AREA)
  • Accounting & Taxation (AREA)
  • Finance (AREA)
  • Engineering & Computer Science (AREA)
  • Development Economics (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • Strategic Management (AREA)
  • Technology Law (AREA)
  • Physics & Mathematics (AREA)
  • General Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Financial Or Insurance-Related Operations Such As Payment And Settlement (AREA)

Abstract

本申请公开了一种信贷评估方法,该方法包括:获取历史信贷数据,并根据所述历史信贷数据构造AI风控引擎;在接收到借款请求时,获取所述借款请求对应借款用户的用户贷前数据,并将所述用户贷前数据分别输入预设专家引擎和所述AI风控引擎进行信贷评估;获取所述预设专家引擎的专家评估结论和所述AI风控引擎的AI评估结论,并基于所述专家评估结论和AI评估结论生成对应的决策建议。本申请还公开了一种信贷评估装置、设备和计算机可读存储介质。本申请采用双引擎评估的方式获得信贷决策建议,大大降低人类思维局限性对引擎构造和信贷评估过程的不利影响,提高风控模型的评估能力,有效降低了贷款的坏账率。

Description

信贷评估方法、装置、设备及计算机可读存储介质
本申请要求于2017年10月25日提交中国专利局、申请号为201711010536.1、发明名称为“信贷评估方法、装置、设备及计算机可读存储介质”的中国专利申请的优先权,其全部内容通过引用结合在申请中。
技术领域
本申请涉及金融信贷领域,尤其涉及一种信贷评估方法、装置、设备及计算机可读存储介质。
背景技术
传统的信贷审批方法是由金融机构的工作人员对借款用户的借款请求进行人工审核,审核的过程中需要工作人员核对借款用户的大量数据信息,以判别贷款交易的风险情况,例如借款用户借得起多少钱、贷款周期内能否偿还等。而随着科技的发展,金融机构开始使用计算机风控***来对信贷请求进行评估,其中风控模型是整个风控***的重要组成部分,其主要用于对用户数据进行识别、分析和评估,从而得出评估结论。
目前的风控模型主要是采用drools的静态规则引擎,引入专家规则对信贷请求进行评估(即将人类的经验转化为计算机的逻辑算法,然后用计算机对信贷请求进行识别)。但是人类的经验不一定是正确的,人类的经验容易受到主观因素的影响,从而影响风控模型评估的准确度,使得风控模型无法给出合理的放款方案,甚至难以识别有意识的诈骗行为,从而提高了贷款的坏账率。
申请内容
本申请的主要目的在于提供一种信贷评估方法、装置、设备及计算机可读存储介质,旨在提高风控模型的评估能力,降低贷款坏账率。
为实现上述目的,本申请提供一种信贷评估方法,所述信贷评估方法包括以下步骤:
获取历史信贷数据,并根据所述历史信贷数据构造AI风控引擎;
在接收到借款请求时,获取所述借款请求对应借款用户的用户贷前数据,并将所述用户贷前数据分别输入预设专家引擎和所述AI风控引擎进行信贷评估;
获取所述预设专家引擎的专家评估结论和所述AI风控引擎的AI评估结论,并基于所述专家评估结论和AI评估结论生成对应的决策建议。
可选地,所述历史信贷数据包括历史贷前数据和历史贷后数据,
所述获取历史信贷数据,并根据所述历史信贷数据构造AI风控引擎的步骤包括:
分别将所述历史贷前数据和历史贷后数据作为输入训练集和输出训练集,并基于所述输入训练集和输出训练集对初始机器学习模型进行分析训练,得到AI风控引擎。
可选地,所述初始机器学习模型通过专家规则冷启动。
可选地,所述输入训练集中的输入元素包括维度为m的多维向量x,所述输出训练集中的输出元素包括贷款不良率y,
所述基于所述输入训练集和输出训练集对初始机器学习模型进行分析训练,得到AI风控引擎的步骤包括:
将所述输入元素和输出元素进行组合,定义出m个多维训练向量组
{(x 1,y 1),(x 2,y 2),…,(x m,y m)};
基于梯度提升树GBDT算法对所述多维训练向量组中的多维训练向量进行学习,获得对应的决策树
T(x;θ m),
其中,θ m为所述决策树的参数;
将所述决策树进行线性组合,得到线性组合树
Figure PCTCN2018075663-appb-000001
所述线性组合数的代价函数为
f(x)=l(h(x,D),Y),
其中,所述h(x,D)为假设函数,为所述决策树的决策结果,所述代价函数f(x)的泰勒公式一阶展开为:
Figure PCTCN2018075663-appb-000002
其中,
Figure PCTCN2018075663-appb-000003
为前t颗决策树的加权和,
Figure PCTCN2018075663-appb-000004
为最速下降迭代公式,
Figure PCTCN2018075663-appb-000005
为假设函数的梯度;
根据所述线性组合树和代价函数获得AI风控引擎。
可选地,所述在接收到借款请求时,获取所述借款请求对应借款用户的用户贷前数据,并将所述用户贷前数据分别输入预设专家引擎和所述AI风控引擎进行信贷评估的步骤包括:
在接收到借款请求时,根据所述借款请求中包括的用户标识和授权许可生成对应的数据获取请求,并将所述数据获取请求发送至数据管理***,以获取对应的用户贷前数据;
接收所述数据管理***反馈的用户贷前数据,并将所述用户贷前数据分别输入预设专家引擎和所述AI风控引擎进行信贷评估。
可选地,所述获取所述预设专家引擎的专家评估结论和所述AI风控引擎的AI评估结论,并基于所述专家评估结论和AI评估结论生成对应的决策建议的步骤包括:
获取所述预设专家引擎的专家评估结论和所述AI风控引擎的AI评估结论,并基于所述专家评估结论和AI评估结论进行加权运算,获得所述借款用户的信用额度;
根据所述信用额度生成对应的放款建议,其中所述放款建议包括放款额度和放款周期。
可选地,所述获取所述预设专家引擎的专家评估结论和所述AI风控引擎的AI评估结论,并基于所述专家评估结论和AI评估结论生成对应的决策建议的步骤之后,还包括:
获取所述借款用户的用户贷后数据,并将所述用户贷前数据和用户贷后数据组合得到迭代训练样本;
通过所述迭代训练样本对所述AI风控引擎进行迭代训练,以对所述AI风控引擎进行调整。
此外,为实现上述目的,本申请还提供一种信贷评估装置,所述信贷评估装置包括:
引擎构造模块,用于获取历史信贷数据,并根据所述历史信贷数据构造AI风控引擎;
信贷评估模块,用于在接收到借款请求时,获取所述借款请求对应借款用户的用户贷前数据,并将所述用户贷前数据分别输入预设专家引擎和所述AI风控引擎进行信贷评估;
建议生成模块,用于获取所述预设专家引擎的专家评估结论和所述AI风控引擎的AI评估结论,并基于所述专家评估结论和AI评估结论生成对应的决策建议。
此外,为实现上述目的,本申请还提供一种信贷评估设备,所述信贷评估设备包括处理器、存储器、以及存储在所述存储器上并可被所述处理器执行的信贷评估程序,其中所述信贷评估程序被所述处理器执行时,实现以下步骤:
获取历史信贷数据,并根据所述历史信贷数据构造AI风控引擎;
在接收到借款请求时,获取所述借款请求对应借款用户的用户贷前数据,并将所述用户贷前数据分别输入预设专家引擎和所述AI风控引擎进行信贷评估;
获取所述预设专家引擎的专家评估结论和所述AI风控引擎的AI评估结论,并基于所述专家评估结论和AI评估结论生成对应的决策建议。
此外,为实现上述目的,本申请还提供一种计算机可读存储介质,所述计算机可读存储介质上存储有信贷评估程序,其中所述信贷评估程序被处理器执行时,实现以下步骤:
获取历史信贷数据,并根据所述历史信贷数据构造AI风控引擎;
在接收到借款请求时,获取所述借款请求对应借款用户的用户贷 前数据,并将所述用户贷前数据分别输入预设专家引擎和所述AI风控引擎进行信贷评估;
获取所述预设专家引擎的专家评估结论和所述AI风控引擎的AI评估结论,并基于所述专家评估结论和AI评估结论生成对应的决策建议。
附图说明
图1为本申请实施例方案中涉及的信贷评估设备的硬件结构示意图;
图2为本申请信贷评估方法第一实施例的流程示意图;
图3为图2所述在接收到借款请求时,获取所述借款请求对应借款用户的用户贷前数据,并将所述用户贷前数据分别输入预设专家引擎和所述AI风控引擎进行信贷评估的细化流程示意图;
图4为本申请信贷评估方法第二实施例的流程示意图;
图5为本申请信贷评估装置第一实施例的功能模块示意图。
本申请目的的实现、功能特点及优点将结合实施例,参照附图做进一步说明。
具体实施方式
应当理解,此处所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。
本申请实施例涉及的信贷评估方法主要应用于信贷评估设备。参照图1,图1为本申请实施例方案中涉及的信贷评估设备的硬件结构示意图。本申请实施例中信贷评估设备可以包括处理器1001(例如CPU),通信总线1002,用户接口1003,网络接口1004,存储器1005。其中,通信总线1002用于实现这些组件之间的连接通信;用户接口1003可以包括显示屏(Display)、输入单元比如键盘(Keyboard);网络接口1004可选的可以包括标准的有线接口、无线接口(如WI-FI接口);存储器1005可以是高速RAM存储器,也可以是稳定的存储器(non-volatile memory),例如磁盘存储器,存储器1005可选的还 可以是独立于前述处理器1001的存储装置。本领域技术人员可以理解,图1中示出的信贷评估设备的硬件结构并不构成对信贷评估设备的限定,可包括比图示更多或更少的部件,或者组合某些部件,或者不同的部件布置。
继续参照图1,图1中作为一种计算机可读存储介质的存储器1005可以包括操作***、网络通信模块以及信贷评估程序。网络通信模块主要用于连接数据库,与数据库进行数据通信;而处理器1001可以调用存储器1005中存储的信贷评估程序,并执行本申请实施例的信贷评估方法的各个实施例。
本申请提供一种信贷评估方法。
参照图2,图2为本申请信贷评估方法第一实施例的流程示意图。
本实施例中,所述信贷评估方法包括以下步骤:
步骤S10,获取历史信贷数据,并根据所述历史信贷数据构造AI风控引擎;
本实施例中提出一种信贷评估方法,采用双核风控引擎对借款请求进行评估和决策,降低人类思维局限性对风控模型的不利影响,提升风控***的准确率,从而降低信贷的坏账率。
本实施例中首先需要构造AI风控引擎,该AI风控引擎与传统的专家引擎是不同。对于专家引擎,是由风控专家或是相关工作人员对大量贷款数据分析后进行经验(规律)总结,再将这些经验总结转化为计算机的逻辑算法,从而构成了该专家引擎的评估规则(可称为专家规则);但是人类的经验不一定是正确的,人类的经验容易受到主观因素的影响,从而使得该专家规则具有一定缺陷,降低了采用专家规则的风控引擎的评估能力。因此,本实施例中的AI风控引擎构造过程不完全采用专家规则,而是通过机器学习的方式构造AI风控引擎。其中,机器学习是指不依赖人类来总结经验、输入逻辑,人类只需要把大量样本数据(包括输入样本数据和输出样本数据)输入给计算机,然后由计算机自己总结出其中的数据转换逻辑或映射转换关系,归纳出相应的逻辑代码,从而得到一个数据转换规则,形成对应的评估引擎。具体的,本实施例中,先查询数据库获取历史信贷数据, 这些历史信贷数据是金融机构历史放贷记录,包括用户贷前提供的行为数据(如信用情况、个人财力、银行往来记录、社保缴纳情况、通话数据等)、以及该用户贷后的还款记录和贷后行为数据;然后将用户贷前的行为数据作为输入训练集、将贷后的还款记录和贷后行为数据作为输出训练集,并将该输入训练集和输出训练集输入至初始计算机学习模型,基于该输入训练集和输出训练集对初始机器学习模型进行分析训练,由计算机自动对这两种数据结构形式和数据内容等方面进行对比和分析,总结出转换规律,归纳出转换逻辑,构造得到AI风控引擎。
进一步的,由于机器学习往往需要大量训练数据的基础上才能展现出优秀的学习能力,但是对于信贷数据(尤其是在用于评价个人信用方面),用户数量是有限的,在用户上可获取的有效信息更加有限,因此可以通过专家规则作为初始机器学习模型的冷启动零点,也就是说可以采用专家规则作为初始机器学习模型的初始逻辑,然后再用历史信贷数据对其不断进行训练,构造得到AI风控引擎。
值得说明的是,本实施例中的构造AI风控引擎过程采用了包括但不限于逻辑回归、决策树、GBDT、SVM等算法。例如,对于GBDT(梯度提升树)而言,样本集包括房产、收入、年龄、学历等内容,可采用其组合定义出m个向量组,其中x是多维向量,y是贷款的不良率指,即有
{(x 1,y 1),(x 2,y 2),…,(x m,y m)}   ①
根据不同的组合,可学习出不同的弱模型(矮的决策树),然后将其线性组合成一个函数:
Figure PCTCN2018075663-appb-000006
在②式中,T(x;θ m)表示决策树,可看成是一棵回归树;
θ m为决策树的参数;
m为决策树的棵数。
这株GBDT的代价函数是:
f(x)=l(h(x,D),Y)    ③
在③式中,h(x,D)是假设函数,是决策树的决策结果;
f(x)的泰勒公式一阶展开是:
Figure PCTCN2018075663-appb-000007
在④式中,
Figure PCTCN2018075663-appb-000008
是前t颗决策树的加权和;
Figure PCTCN2018075663-appb-000009
是最速下降迭代公式;
Figure PCTCN2018075663-appb-000010
假设函数的梯度。
通过上述可知GBDT的损失函数所采用的思想与逻辑回归是类似的,只不过后者的基函数采用双射sigmoid函数,而本实施例在此处采用了CART。由此可知计算GBDT参数的方法与逻辑回归类似,也可以采用梯度下降法。根据上述可求得一株以CART为基函数的线性组合树GDBT(如式②),以及对应的代价函数(如式③),即AI风控引擎。
当然对于训练得到的AI风控引擎,工作人员可以进行手动调整,以对AI风控引擎进行完善。
步骤S20,在接收到借款请求时,获取所述借款请求对应借款用户的用户贷前数据,并将所述用户贷前数据分别输入预设专家引擎和所述AI风控引擎进行信贷评估;
本实施例中,在训练得到AI风控引擎后,即可使用该AI风控引擎进行借贷评估了。值得说明的是,本实施例中,并非只使用AI风控引擎进行评估,而是同时使用AI风控引擎和预设的专家引擎、采用双核风控的方法进行信贷评估。其中,预设专家引擎的含义如上述,是由风控专家或是相关工作人员对大量贷款数据分析后进行经验(规律)总结,再将这些经验总结转化为计算机的逻辑算法,从而构成了该专家引擎的评估规则。
具体的,金融机构的风控模型在接收到借款用户的提出借款请求时(该借款请求包括了借款金额和借款周期等),会获取借款用户的相关贷前数据(贷前数据包括借款方的信用情况、个人财力、银行往来记录、社保缴纳情况、通话数据等),并将这些风控数据分别输入 至预设专家引擎和AI风控引擎,并通过预设专家引擎和AI风控引擎分别对借款请求进行评估。
步骤S30,获取所述预设专家引擎的专家评估结论和所述AI风控引擎的AI评估结论,并基于所述专家评估结论和AI评估结论生成对应的决策建议。
本实施例中,通过预设专家引擎和AI风控引擎分别对借款请求进行评估后,将得到对应的专家评估结论和AI评估结论,并结合专家评估结合和AI评估结论可得出决策建议(如借款金额,借款周期等)。
在具体实施中,对于专家评估结论和AI评估结论,可以是以分值的方式进行体现,分值范围为0到100;分值越高,则表明借款用户的借款请求的坏账率越低;在通过预设专家引擎和AI风控引擎分别对借款请求进行评估、得到专家分值和AI分值两个评估分值时,还需将专家分值和AI分值进行加权运行后,可得到一个综合评估值,该综合评估值的含义与单核评估所得的评估分值类似,范围也是0到100之间,分值越高,则表明借款用户的借款请求的坏账率越低;而对于不同的分值段,其对应了不同的信用额度,其信用额度即代表了向该借款用户放款的建议最大额度和最大放款周期;通过计算该借款用户的综合评估值,即可确定该借款用户的信用额度,从而生成了对应的放款建议(包括放款额度和放款周期)。
当然在具体实施中,对于专家评估结论和AI评估结论,也可以两分的评估结论(即同意放款和拒绝放款两分),而决策建议的生成过程也可以是以其它方式实现,上述举例并不构成对本实施例信贷评估方法的限定。
进一步的,本实施例中,在生成决策建议后,风控模型还可生成对应的信贷评估报告。该信贷评估报告中包括了评估过程使用到的借款用户贷前数据,还包括对贷前数据评估得到的专家评估结论、AI评估结论,以及最终的决策建议等。而在信贷评估报告中,由于涉及到大量数据,因此可自定义数据的显示方式。例如,贷前数据中包括了借款用户前3个月的总支出,每个月的支出又包括水电支出、电话 业务支出、***等内容;对于每个月的支出,可以用饼状图的方式进行显示,方便决策人员了解借款用户每个月对不同事物的资金投入情况;而对于3个月总支出情况,则可以在用一副柱状图进行显示,方便决策人员了解借款用户每个月的支出趋势。当然不同类型的数据的显示方式可以由决策人员根据实际情况进行配置。
本实施例中,通过获取历史信贷数据,并根据所述历史信贷数据构造AI风控引擎;在接收到借款请求时,获取所述借款请求对应借款用户的用户贷前数据,并将所述用户贷前数据分别输入预设专家引擎和所述AI风控引擎进行信贷评估;获取所述预设专家引擎的专家评估结论和所述AI风控引擎的AI评估结论,并基于所述专家评估结论和AI评估结论生成对应的决策建议。通过以上方式,本实施例以机器学习的方法构造AI风控引擎,并通过AI风控引擎和专家规则引擎同时对借款请求进行评估,采用双引擎评估的方式获得决策建议,大大降低人类思维局限性对引擎构造和信贷评估过程的不利影响,提高风控模型的评估能力,有效降低了贷款的坏账率。
参照图3,图3为图2所述在接收到借款请求时,获取所述借款请求对应借款用户的用户贷前数据,并将所述用户贷前数据分别输入预设专家引擎和所述AI风控引擎进行信贷评估的细化流程示意图。
基于上述图2所示实施例,本实施例中,步骤S20包括:
步骤S21,在接收到借款请求时,根据所述借款请求中包括的用户标识和授权许可生成对应的数据获取请求,并将所述数据获取请求发送至数据管理***,以获取对应的用户贷前数据;
步骤S22,接收所述数据管理***反馈的用户贷前数据,并将所述用户贷前数据分别输入所述预设专家引擎和AI风控引擎进行信贷评估。
本实施例中,对于借款用户的贷前数据,往往会涉及到借款用户的隐私,而这些贷前数据是需要从其它数据源获取的,例如,借款用户的银行卡消费记录需要从银行***获得的,借款用户的话费账单数据需要从电信公司获得的;若风控模型要获取这些贷前数据,需要先得到借款用户的许可和授权。具体的,借款用户在发送借款请求时, 该借款请求中包括了借款用户的身份标识(如身份证号等身份信息),还包括借款用户的授权许可信息(如电子签名等),以表明借款用户允许风控模型(金融机构)获取自己的相关贷前数据进行信贷评估。风控模型在接收到该借款请求时,将根据其中的用户标识和授权许可生成对应的数据获取请求,并将该数据获取请求发送至相关的数据管理***(如银行***、电信公司等)。数据管理***在接收到该数据获取请求时,首先会获取其中的用户标识,确定本次数据获取请求是需要获取哪位用户的相关数据;然后将对该授权许可进行验证,判断该授权许可的真伪性,从而确定请求获取数据的风控模型是否为合法获取者;当授权验证通过时,数据管理***将查询对应的用户贷前数据,并将该数据返回至风控模型。风控模型在接收到数据管理***反馈的用户贷前数据时,即可将该用户贷前数据分别输入所述预设专家引擎和AI风控引擎进行信贷评估。
参照图4,图4为本申请信贷评估方法第二实施例的流程示意图。
基于上述图2所示实施例,本实施例中,步骤S30之后,还包括:
步骤S40,获取所述借款用户的用户贷后数据,并将所述用户贷前数据和用户贷后数据组合得到迭代训练样本;
步骤S50,通过所述迭代训练样本对所述AI风控引擎进行迭代训练,以对所述AI风控引擎进行调整。
本实施例中,考虑到随着时间的发展,不同的经济周期及社会因素可能会对AI引擎的AI规则(评估逻辑)造成影响,使得某些因子的有效区间产生了摆动,例如月收入6000元,在不同的物价条件下,其数据的权重因子是不一样的;为了使得AI引擎能够适应经济周期和社会实际状况,需要对AI引擎不断进行修正和调整。具体的,在确定向借款用户放款后,风控模型还将持续获取借款用户贷后数据(如还款记录,消费记录等);然后将步骤S20所得的贷前数据和该贷后数据组合得到迭代训练样本,通过该迭代训练样本对的AI引擎不断进行修正调整,调整的内容包括改变参数的阈值、更改某些风控数据或评估因子的优先级、改变数据的权重等;例如,AI引擎的策略算法中,数据权重等级为1到5,对于用户借款时提供数据中,A 数据权重等级为5,也就是说A数据对于判断用户是否能按时偿还贷款具有较大的参考价值;而通过若干贷后数据发现,A数据的参考价值并没有意想中的大,此时AI引擎将对原有的策略算法进行修正,降低A数据的权重等级。对AI引擎调整后,再在新的贷款周期中给出借款用户可能出现逾期或坏账的可能性,再用新的周期的实际的贷后数据进行反馈修正。通过这种迭代的训练方式,不断对AI引擎进行改进,逐步提升评估准确率。
此外,本申请还提供一种信贷评估装置。
参照图5,图5为本申请信贷评估装置第一实施例的功能模块示意图。
本实施例中,所述信贷评估装置包括:
引擎构造模块10,用于获取历史信贷数据,并根据所述历史信贷数据构造AI风控引擎;
信贷评估模块20,用于在接收到借款请求时,获取所述借款请求对应借款用户的用户贷前数据,并将所述用户贷前数据分别输入预设专家引擎和所述AI风控引擎进行信贷评估;
建议生成模块30,用于获取所述预设专家引擎的专家评估结论和所述AI风控引擎的AI评估结论,并基于所述专家评估结论和AI评估结论生成对应的决策建议
进一步的,所述历史信贷数据包括历史贷前数据和历史贷后数据,所述引擎构造模块10,还用于分别将所述历史贷前数据和历史贷后数据作为输入训练集和输出训练集,并基于上述输入训练集和输出训练集对初始机器学习模型进行分析训练,得到AI风控引擎。
进一步的,所述初始机器学习模型通过专家规则冷启动。
进一步的,所述输入训练集中的输入元素包括维度为m的多维向量x,所述输出训练集中的输出元素包括贷款不良率y,
所述引擎构造模块10,还用于将所述输入元素和输出元素进行组合,定义出m个多维训练向量组
{(x 1,y 1),(x 2,y 2),…,(x m,y m)};
基于梯度提升树GBDT算法对所述多维训练向量组中的多维训 练向量进行学习,获得对应的决策树
T(x;θ m),
其中,θ m为所述决策树的参数;
将所述决策树进行线性组合,得到线性组合树
Figure PCTCN2018075663-appb-000011
所述线性组合数的代价函数为
f(x)=l(h(x,D),Y),
其中,所述h(x,D)为假设函数,为所述决策树的决策结果,所述代价函数f(x)的泰勒公式一阶展开为:
Figure PCTCN2018075663-appb-000012
其中,
Figure PCTCN2018075663-appb-000013
为前t颗决策树的加权和,
Figure PCTCN2018075663-appb-000014
为最速下降迭代公式,
Figure PCTCN2018075663-appb-000015
为假设函数的梯度;
根据所述线性组合树和代价函数获得AI风控引擎。
进一步的,所述信贷评估模块20包括:
数据获取单元,用于在接收到借款请求时,根据所述借款请求中包括的用户标识和授权许可生成对应的数据获取请求,并将所述数据获取请求发送至数据管理***,以获取对应的用户贷前数据;
信贷评估单元,用于接收所述数据管理***反馈的用户贷前数据,并将所述用户贷前数据分别输入所述预设专家引擎和AI风控引擎进行信贷评估。
进一步的,所述建议生成模块30包括:
加权运算单元,用于获取所述预设专家引擎的专家评估结论和所述AI风控引擎的AI评估结论,并基于所述专家评估结论和AI评估结论进行加权运算,获得所述借款用户的信用额度;
建议生成单元,用于根据所述信用额度生成对应的放款建议,其中所述放款建议包括放款额度和放款周期。
进一步的,所述信贷评估装置还包括:
数据组合模块,用于获取所述借款用户的用户贷后数据,并将所述用户贷前数据和用户贷后数据组合得到迭代训练样本;
迭代训练模块,用于通过所述迭代训练样本对所述AI风控引擎进行迭代训练,以对所述AI风控引擎进行调整。
其中,上述信贷评估装置中各个模块与上述信贷评估方法实施例中各步骤相对应,其功能和实现过程在此处不再一一赘述。
此外,本申请还提供一种计算机可读存储介质。
本申请计算机可读存储介质上存储有信贷评估程序,所述计算机可读存储介质上存储有信贷评估程序,其中所述信贷评估程序被处理器执行时,实现如上述的信贷评估方法的步骤。
其中,信贷评估程序被执行时所实现的方法可参照本申请信贷评估方法的各个实施例,此处不再赘述。
以上仅为本申请的优选实施例,并非因此限制本申请的专利范围,凡是利用本申请说明书及附图内容所作的等效结构或等效流程变换,或直接或间接运用在其他相关的技术领域,均同理包括在本申请的专利保护范围内。

Claims (20)

  1. 一种信贷评估方法,其特征在于,所述信贷评估方法包括以下步骤:
    获取历史信贷数据,并根据所述历史信贷数据构造AI风控引擎;
    在接收到借款请求时,获取所述借款请求对应借款用户的用户贷前数据,并将所述用户贷前数据分别输入预设专家引擎和所述AI风控引擎进行信贷评估;
    获取所述预设专家引擎的专家评估结论和所述AI风控引擎的AI评估结论,并基于所述专家评估结论和AI评估结论生成对应的决策建议。
  2. 如权利要求1所述的信贷评估方法,其特征在于,所述历史信贷数据包括历史贷前数据和历史贷后数据,
    所述获取历史信贷数据,并根据所述历史信贷数据构造AI风控引擎的步骤包括:
    分别将所述历史贷前数据和历史贷后数据作为输入训练集和输出训练集,并基于所述输入训练集和输出训练集对初始机器学习模型进行分析训练,得到AI风控引擎。
  3. 如权利要求2所述的信贷评估方法,其特征在于,所述初始机器学习模型通过专家规则冷启动。
  4. 如权利要求2所述的信贷评估方法,其特征在于,所述输入训练集中的输入元素包括维度为m的多维向量x,所述输出训练集中的输出元素包括贷款不良率y,
    所述基于所述输入训练集和输出训练集对初始机器学习模型进行分析训练,得到AI风控引擎的步骤包括:
    将所述输入元素和输出元素进行组合,定义出m个多维训练向量组
    {(x 1,y 1),(x 2,y 2),…,(x m,y m)};
    基于梯度提升树GBDT算法对所述多维训练向量组中的多维训练向量进行学习,获得对应的决策树
    T(x;θ m),
    其中,θ m为所述决策树的参数;
    将所述决策树进行线性组合,得到线性组合树
    Figure PCTCN2018075663-appb-100001
    所述线性组合数的代价函数为
    f(x)=l(h(x,D),Y),
    其中,所述h(x,D)为假设函数,为所述决策树的决策结果,所述代价函数f(x)的泰勒公式一阶展开为:
    Figure PCTCN2018075663-appb-100002
    其中,
    Figure PCTCN2018075663-appb-100003
    为前t颗决策树的加权和,
    Figure PCTCN2018075663-appb-100004
    为最速下降迭代公式,
    Figure PCTCN2018075663-appb-100005
    为假设函数的梯度;
    根据所述线性组合树和代价函数获得AI风控引擎。
  5. 如权利要求1所述的信贷评估方法,其特征在于,所述在接收到借款请求时,获取所述借款请求对应借款用户的用户贷前数据,并将所述用户贷前数据分别输入预设专家引擎和所述AI风控引擎进行信贷评估的步骤包括:
    在接收到借款请求时,根据所述借款请求中包括的用户标识和授权许可生成对应的数据获取请求,并将所述数据获取请求发送至数据管理***,以获取对应的用户贷前数据;
    接收所述数据管理***反馈的用户贷前数据,并将所述用户贷前数据分别输入预设专家引擎和所述AI风控引擎进行信贷评估。
  6. 如权利要求1所述的信贷评估方法,其特征在于,所述获取所述预设专家引擎的专家评估结论和所述AI风控引擎的AI评估结论,并基于所述专家评估结论和AI评估结论生成对应的决策建议的步骤包括:
    获取所述预设专家引擎的专家评估结论和所述AI风控引擎的AI 评估结论,并基于所述专家评估结论和AI评估结论进行加权运算,获得所述借款用户的信用额度;
    根据所述信用额度生成对应的放款建议,其中所述放款建议包括放款额度和放款周期。
  7. 如权利要求1所述的信贷评估方法,其特征在于,所述获取所述预设专家引擎的专家评估结论和所述AI风控引擎的AI评估结论,并基于所述专家评估结论和AI评估结论生成对应的决策建议的步骤之后,还包括:
    获取所述借款用户的用户贷后数据,并将所述用户贷前数据和用户贷后数据组合得到迭代训练样本;
    通过所述迭代训练样本对所述AI风控引擎进行迭代训练,以对所述AI风控引擎进行调整。
  8. 一种信贷评估装置,其特征在于,所述信贷评估装置包括:
    引擎构造模块,用于获取历史信贷数据,并根据所述历史信贷数据构造AI风控引擎;
    信贷评估模块,用于在接收到借款请求时,获取所述借款请求对应借款用户的用户贷前数据,并将所述用户贷前数据分别输入预设专家引擎和所述AI风控引擎进行信贷评估;
    建议生成模块,用于获取所述预设专家引擎的专家评估结论和所述AI风控引擎的AI评估结论,并基于所述专家评估结论和AI评估结论生成对应的决策建议。
  9. 如权利要求8所述的信贷评估装置,其特征在于,所述历史信贷数据包括历史贷前数据和历史贷后数据,
    所述引擎构造模块,还用于分别将所述历史贷前数据和历史贷后数据作为输入训练集和输出训练集,并基于所述输入训练集和输出训练集对初始机器学习模型进行分析训练,得到AI风控引擎。
  10. 如权利要求9所述的信贷评估装置,其特征在于,所述初始机器学习模型通过专家规则冷启动。
  11. 如权利要求9所述的信贷评估装置,其特征在于,所述输入训练集中的输入元素包括维度为m的多维向量x,所述输出训练集中 的输出元素包括贷款不良率y,
    所述引擎构造模块,还用于将所述输入元素和输出元素进行组合,定义出m个多维训练向量组
    {(x 1,y 1),(x 2,y 2),…,(x m,y m)};
    基于梯度提升树GBDT算法对所述多维训练向量组中的多维训练向量进行学习,获得对应的决策树
    T(x;θ m),
    其中,θ m为所述决策树的参数;
    将所述决策树进行线性组合,得到线性组合树
    Figure PCTCN2018075663-appb-100006
    所述线性组合数的代价函数为
    f(x)=l(h(x,D),Y),
    其中,所述h(x,D)为假设函数,为所述决策树的决策结果,所述代价函数f(x)的泰勒公式一阶展开为:
    Figure PCTCN2018075663-appb-100007
    其中,
    Figure PCTCN2018075663-appb-100008
    为前t颗决策树的加权和,
    Figure PCTCN2018075663-appb-100009
    为最速下降迭代公式,
    Figure PCTCN2018075663-appb-100010
    为假设函数的梯度;
    根据所述线性组合树和代价函数获得AI风控引擎。
  12. 如权利要求8所述的信贷评估装置,其特征在于,所述信贷评估模块包括:
    数据获取单元,用于在接收到借款请求时,根据所述借款请求中包括的用户标识和授权许可生成对应的数据获取请求,并将所述数据获取请求发送至数据管理***,以获取对应的用户贷前数据;
    信贷评估单元,用于接收所述数据管理***反馈的用户贷前数据,并将所述用户贷前数据分别输入所述预设专家引擎和AI风控引擎进行信贷评估。
  13. 如权利要求8所述的信贷评估装置,其特征在于,所述建议生成模块包括:
    加权运算单元,用于获取所述预设专家引擎的专家评估结论和所述AI风控引擎的AI评估结论,并基于所述专家评估结论和AI评估结论进行加权运算,获得所述借款用户的信用额度;
    建议生成单元,用于根据所述信用额度生成对应的放款建议,其中所述放款建议包括放款额度和放款周期。
  14. 如权利要求8所述的信贷评估装置,其特征在于,所述信贷评估装置还包括:
    数据组合模块,用于获取所述借款用户的用户贷后数据,并将所述用户贷前数据和用户贷后数据组合得到迭代训练样本;
    迭代训练模块,用于通过所述迭代训练样本对所述AI风控引擎进行迭代训练,以对所述AI风控引擎进行调整。
  15. 一种信贷评估设备,其特征在于,所述信贷评估设备包括处理器、存储器、以及存储在所述存储器上并可被所述处理器执行的信贷评估程序,其中所述信贷评估程序被所述处理器执行时,实现以下步骤:
    获取历史信贷数据,并根据所述历史信贷数据构造AI风控引擎;
    在接收到借款请求时,获取所述借款请求对应借款用户的用户贷前数据,并将所述用户贷前数据分别输入预设专家引擎和所述AI风控引擎进行信贷评估;
    获取所述预设专家引擎的专家评估结论和所述AI风控引擎的AI评估结论,并基于所述专家评估结论和AI评估结论生成对应的决策建议。
  16. 如权利要求15所述的信贷评估设备,其特征在于,所述历史信贷数据包括历史贷前数据和历史贷后数据,
    所述获取历史信贷数据,并根据所述历史信贷数据构造AI风控引擎的步骤包括:
    分别将所述历史贷前数据和历史贷后数据作为输入训练集和输出训练集,并基于所述输入训练集和输出训练集对初始机器学习模型 进行分析训练,得到AI风控引擎。
  17. 如权利要求16所述的信贷评估设备,其特征在于,所述初始机器学习模型通过专家规则冷启动。
  18. 如权利要求16所述的信贷评估方法,其特征在于,所述输入训练集中的输入元素包括维度为m的多维向量x,所述输出训练集中的输出元素包括贷款不良率y,
    所述基于所述输入训练集和输出训练集对初始机器学习模型进行分析训练,得到AI风控引擎的步骤包括:
    将所述输入元素和输出元素进行组合,定义出m个多维训练向量组
    {(x 1,y 1),(x 2,y 2),…,(x m,y m)};
    基于梯度提升树GBDT算法对所述多维训练向量组中的多维训练向量进行学习,获得对应的决策树
    T(x;θ m),
    其中,θ m为所述决策树的参数;
    将所述决策树进行线性组合,得到线性组合树
    Figure PCTCN2018075663-appb-100011
    所述线性组合数的代价函数为
    f(x)=l(h(x,D),Y),
    其中,所述h(x,D)为假设函数,为所述决策树的决策结果,所述代价函数f(x)的泰勒公式一阶展开为:
    Figure PCTCN2018075663-appb-100012
    其中,
    Figure PCTCN2018075663-appb-100013
    为前t颗决策树的加权和,
    Figure PCTCN2018075663-appb-100014
    为最速下降迭代公式,
    Figure PCTCN2018075663-appb-100015
    为假设函数的梯度;
    根据所述线性组合树和代价函数获得AI风控引擎。
  19. 如权利要求15所述的信贷评估方法,其特征在于,所述在 接收到借款请求时,获取所述借款请求对应借款用户的用户贷前数据,并将所述用户贷前数据分别输入预设专家引擎和所述AI风控引擎进行信贷评估的步骤包括:
    在接收到借款请求时,根据所述借款请求中包括的用户标识和授权许可生成对应的数据获取请求,并将所述数据获取请求发送至数据管理***,以获取对应的用户贷前数据;
    接收所述数据管理***反馈的用户贷前数据,并将所述用户贷前数据分别输入预设专家引擎和所述AI风控引擎进行信贷评估。
  20. 一种计算机可读存储介质,其特征在于,所述计算机可读存储介质上存储有信贷评估程序,其中所述信贷评估程序被处理器执行时,实现以下步骤:
    获取历史信贷数据,并根据所述历史信贷数据构造AI风控引擎;
    在接收到借款请求时,获取所述借款请求对应借款用户的用户贷前数据,并将所述用户贷前数据分别输入预设专家引擎和所述AI风控引擎进行信贷评估;
    获取所述预设专家引擎的专家评估结论和所述AI风控引擎的AI评估结论,并基于所述专家评估结论和AI评估结论生成对应的决策建议。
PCT/CN2018/075663 2017-10-25 2018-02-07 信贷评估方法、装置、设备及计算机可读存储介质 WO2019080407A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201711010536.1 2017-10-25
CN201711010536.1A CN107886425A (zh) 2017-10-25 2017-10-25 信贷评估方法、装置、设备及计算机可读存储介质

Publications (1)

Publication Number Publication Date
WO2019080407A1 true WO2019080407A1 (zh) 2019-05-02

Family

ID=61782284

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/075663 WO2019080407A1 (zh) 2017-10-25 2018-02-07 信贷评估方法、装置、设备及计算机可读存储介质

Country Status (2)

Country Link
CN (1) CN107886425A (zh)
WO (1) WO2019080407A1 (zh)

Cited By (23)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110377619A (zh) * 2019-06-19 2019-10-25 深圳壹账通智能科技有限公司 风控报告的生成方法、装置、设备及存储介质
CN110415036A (zh) * 2019-07-30 2019-11-05 深圳市珍爱捷云信息技术有限公司 用户等级的确定方法、装置、计算机设备和存储介质
CN110717817A (zh) * 2019-08-14 2020-01-21 深圳壹账通智能科技有限公司 贷前审核方法及装置、电子设备和计算机可读存储介质
CN111429260A (zh) * 2020-03-19 2020-07-17 重庆富民银行股份有限公司 用于风控***的用户体验提升方法及***
CN111461874A (zh) * 2020-04-13 2020-07-28 浙江大学 一种基于联邦模式的信贷风险控制***及方法
CN111652713A (zh) * 2020-07-01 2020-09-11 中国银行股份有限公司 权益风控建模方法和装置
CN111652711A (zh) * 2020-06-09 2020-09-11 深圳前海微众银行股份有限公司 基于催收数据的贷前分析方法、装置、设备与存储介质
CN112017060A (zh) * 2020-07-15 2020-12-01 北京淇瑀信息科技有限公司 为目标用户进行资源分配的方法、装置及电子设备
CN112348654A (zh) * 2020-09-23 2021-02-09 民生科技有限责任公司 企业授信额度自动评定方法、***及可读存储介质
CN112446777A (zh) * 2019-09-03 2021-03-05 腾讯科技(深圳)有限公司 一种信用评估方法、装置、设备及存储介质
CN112785086A (zh) * 2021-02-10 2021-05-11 中国工商银行股份有限公司 信贷逾期风险预测方法及装置
CN112785415A (zh) * 2021-01-20 2021-05-11 深圳前海微众银行股份有限公司 评分卡模型构建方法、装置、设备及计算机可读存储介质
CN112991077A (zh) * 2021-02-18 2021-06-18 深圳无域科技技术有限公司 金融风险控制***及方法
CN113362072A (zh) * 2021-06-30 2021-09-07 平安普惠企业管理有限公司 风控数据处理方法、装置、电子设备及存储介质
CN113592623A (zh) * 2021-07-20 2021-11-02 浙江惠瀜网络科技有限公司 车贷贷前风险评估体系的构建方法和风险评估方法
CN113657724A (zh) * 2021-07-29 2021-11-16 上海淇玥信息技术有限公司 基于多源异构数据的资源分配方法、装置和电子设备
CN113657779A (zh) * 2021-08-20 2021-11-16 杭州时趣信息技术有限公司 一种动态配置化的风控决策方法、***、设备和存储介质
CN113761908A (zh) * 2020-11-26 2021-12-07 北京沃东天骏信息技术有限公司 一种存量用户信息的处理方法和装置
CN115994821A (zh) * 2023-01-09 2023-04-21 中云融拓数据科技发展(深圳)有限公司 基于产业链数字化场景金融模型建立金融风控体系的方法
CN116701153A (zh) * 2023-08-09 2023-09-05 云账户技术(天津)有限公司 结算服务性能的评估方法、装置、电子设备及存储介质
CN116739742A (zh) * 2023-06-02 2023-09-12 北京百度网讯科技有限公司 信贷风控模型的监控方法、装置、设备及存储介质
CN117556225A (zh) * 2024-01-12 2024-02-13 杭银消费金融股份有限公司 一种人行征信数据风险管理***
CN118035750A (zh) * 2024-04-11 2024-05-14 杭银消费金融股份有限公司 一种基于panel data的信贷模型样本构建方法

Families Citing this family (22)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108615190A (zh) * 2018-04-27 2018-10-02 深圳市分期乐网络科技有限公司 风控模型验证方法、装置、设备和存储介质
CN108765131B (zh) * 2018-05-18 2023-06-09 深圳壹账通智能科技有限公司 基于微表情的信贷审核方法、装置、终端及可读存储介质
CN108960672B (zh) * 2018-07-23 2023-01-03 上海品唯信息技术有限公司 限额限次的风控方法、装置及计算机可读存储介质
CN109727116A (zh) * 2018-08-17 2019-05-07 平安普惠企业管理有限公司 信贷分析方法、装置、设备及计算机可读存储介质
CN109447384A (zh) * 2018-08-27 2019-03-08 深圳壹账通智能科技有限公司 风控***的验证方法、装置、设备及存储介质
CN109685635A (zh) * 2018-09-11 2019-04-26 深圳平安财富宝投资咨询有限公司 金融业务的风险评估方法、风控服务端及存储介质
CN109670929A (zh) * 2018-09-13 2019-04-23 深圳壹账通智能科技有限公司 贷款预警的控制方法、装置、设备及计算机可读存储介质
CN109656774A (zh) * 2018-09-27 2019-04-19 深圳壹账通智能科技有限公司 开放式规则流引擎分析方法、装置、终端设备及存储介质
CN109598479B (zh) * 2018-10-25 2024-04-05 三六零科技集团有限公司 一种账单提取方法、装置、电子设备及介质
CN109461070A (zh) * 2018-10-25 2019-03-12 深圳壹账通智能科技有限公司 一种风险审批方法、装置、存储介质和服务器
CN109670940A (zh) * 2018-11-12 2019-04-23 深圳壹账通智能科技有限公司 基于机器学习的信用风险评估模型生成方法及相关设备
CN109859029A (zh) * 2019-01-04 2019-06-07 深圳壹账通智能科技有限公司 异常申请检测方法、装置、计算机设备以及存储介质
CN110276679B (zh) * 2019-05-23 2021-05-04 武汉大学 一种面向深度学习的网络个人信贷欺诈行为检测方法
CN110335139B (zh) * 2019-06-21 2022-10-14 深圳前海微众银行股份有限公司 基于相似度的评估方法、装置、设备及可读存储介质
CN110727770A (zh) * 2019-08-30 2020-01-24 深圳壹账通智能科技有限公司 请求的智能筛选方法、装置、计算机设备和存储介质
CN110688316A (zh) * 2019-09-27 2020-01-14 国家计算机网络与信息安全管理中心 基于规则引擎的移动应用风险等级推理评估方法及装置
CN111047435A (zh) * 2019-12-23 2020-04-21 支付宝(杭州)信息技术有限公司 信贷数据处理、信贷分配方法、装置及电子设备
CN111127195B (zh) * 2019-12-31 2023-08-11 中信百信银行股份有限公司 基于仿真的数据风控价值评估方法、装置、设备、介质
CN111898879A (zh) * 2020-07-15 2020-11-06 北京海恩炼鑫台信息技术有限责任公司 一种ai智能风控建模方法
CN112365265B (zh) * 2020-10-26 2021-07-02 建投数据科技(山东)有限公司 一种互联网金融智能风控***
CN113506164B (zh) * 2021-07-05 2023-05-26 普洛斯科技(重庆)有限公司 一种风控决策方法、装置、电子设备及机器可读存储介质
CN113919937B (zh) * 2021-09-22 2023-06-23 北京睿知图远科技有限公司 一种基于贷款评估风控的ks监控***

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2012012623A1 (en) * 2010-07-23 2012-01-26 Thomson Reuters Global Resources Credit risk mining
CN104881783A (zh) * 2015-05-14 2015-09-02 中国科学院信息工程研究所 电子银行账户欺诈行为及风险检测方法与***
CN106096657A (zh) * 2016-06-13 2016-11-09 北京物思创想科技有限公司 基于机器学习来预测数据审核目标的方法及***
CN106897918A (zh) * 2017-02-24 2017-06-27 上海易贷网金融信息服务有限公司 一种混合式机器学习信用评分模型构建方法
CN106952155A (zh) * 2017-03-08 2017-07-14 深圳前海纵腾金融科技服务有限公司 一种基于信用评分的催收方法及装置

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101046876A (zh) * 2006-03-31 2007-10-03 探宇科技股份有限公司 应用数据挖掘方法的信用评分***及信用评分方法
AU2015100760A4 (en) * 2015-06-08 2015-07-09 Insfin Insurance & Finance Group Pty Ltd A software and business method to evaluate redundancy and unemployment risk
CN105678619A (zh) * 2016-01-07 2016-06-15 中国联合网络通信集团有限公司 贷款额度评估方法及装置
CN106097094A (zh) * 2016-06-12 2016-11-09 桂林电子科技大学 一种面向中小企业的人机结合信贷评估新模型
CN107292743A (zh) * 2017-06-07 2017-10-24 前海梧桐(深圳)数据有限公司 针对企业投融资的智能化决策方法及其***

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2012012623A1 (en) * 2010-07-23 2012-01-26 Thomson Reuters Global Resources Credit risk mining
CN104881783A (zh) * 2015-05-14 2015-09-02 中国科学院信息工程研究所 电子银行账户欺诈行为及风险检测方法与***
CN106096657A (zh) * 2016-06-13 2016-11-09 北京物思创想科技有限公司 基于机器学习来预测数据审核目标的方法及***
CN106897918A (zh) * 2017-02-24 2017-06-27 上海易贷网金融信息服务有限公司 一种混合式机器学习信用评分模型构建方法
CN106952155A (zh) * 2017-03-08 2017-07-14 深圳前海纵腾金融科技服务有限公司 一种基于信用评分的催收方法及装置

Cited By (34)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110377619A (zh) * 2019-06-19 2019-10-25 深圳壹账通智能科技有限公司 风控报告的生成方法、装置、设备及存储介质
CN110415036A (zh) * 2019-07-30 2019-11-05 深圳市珍爱捷云信息技术有限公司 用户等级的确定方法、装置、计算机设备和存储介质
CN110415036B (zh) * 2019-07-30 2023-05-12 深圳市珍爱捷云信息技术有限公司 用户等级的确定方法、装置、计算机设备和存储介质
CN110717817A (zh) * 2019-08-14 2020-01-21 深圳壹账通智能科技有限公司 贷前审核方法及装置、电子设备和计算机可读存储介质
CN112446777A (zh) * 2019-09-03 2021-03-05 腾讯科技(深圳)有限公司 一种信用评估方法、装置、设备及存储介质
CN112446777B (zh) * 2019-09-03 2023-11-17 腾讯科技(深圳)有限公司 一种信用评估方法、装置、设备及存储介质
CN111429260A (zh) * 2020-03-19 2020-07-17 重庆富民银行股份有限公司 用于风控***的用户体验提升方法及***
CN111429260B (zh) * 2020-03-19 2023-06-06 重庆富民银行股份有限公司 用于风控***的用户体验提升方法及***
CN111461874A (zh) * 2020-04-13 2020-07-28 浙江大学 一种基于联邦模式的信贷风险控制***及方法
CN111652711A (zh) * 2020-06-09 2020-09-11 深圳前海微众银行股份有限公司 基于催收数据的贷前分析方法、装置、设备与存储介质
CN111652713A (zh) * 2020-07-01 2020-09-11 中国银行股份有限公司 权益风控建模方法和装置
CN111652713B (zh) * 2020-07-01 2024-02-27 中国银行股份有限公司 权益风控建模方法和装置
CN112017060B (zh) * 2020-07-15 2024-04-26 北京淇瑀信息科技有限公司 为目标用户进行资源分配的方法、装置及电子设备
CN112017060A (zh) * 2020-07-15 2020-12-01 北京淇瑀信息科技有限公司 为目标用户进行资源分配的方法、装置及电子设备
CN112348654A (zh) * 2020-09-23 2021-02-09 民生科技有限责任公司 企业授信额度自动评定方法、***及可读存储介质
CN113761908A (zh) * 2020-11-26 2021-12-07 北京沃东天骏信息技术有限公司 一种存量用户信息的处理方法和装置
CN112785415A (zh) * 2021-01-20 2021-05-11 深圳前海微众银行股份有限公司 评分卡模型构建方法、装置、设备及计算机可读存储介质
CN112785415B (zh) * 2021-01-20 2024-01-12 深圳前海微众银行股份有限公司 评分卡模型构建方法、装置、设备及计算机可读存储介质
CN112785086A (zh) * 2021-02-10 2021-05-11 中国工商银行股份有限公司 信贷逾期风险预测方法及装置
CN112991077A (zh) * 2021-02-18 2021-06-18 深圳无域科技技术有限公司 金融风险控制***及方法
CN112991077B (zh) * 2021-02-18 2024-03-19 深圳无域科技技术有限公司 金融风险控制***及方法
CN113362072A (zh) * 2021-06-30 2021-09-07 平安普惠企业管理有限公司 风控数据处理方法、装置、电子设备及存储介质
CN113362072B (zh) * 2021-06-30 2023-09-08 成都一蟹科技有限公司 风控数据处理方法、装置、电子设备及存储介质
CN113592623A (zh) * 2021-07-20 2021-11-02 浙江惠瀜网络科技有限公司 车贷贷前风险评估体系的构建方法和风险评估方法
CN113657724A (zh) * 2021-07-29 2021-11-16 上海淇玥信息技术有限公司 基于多源异构数据的资源分配方法、装置和电子设备
CN113657779A (zh) * 2021-08-20 2021-11-16 杭州时趣信息技术有限公司 一种动态配置化的风控决策方法、***、设备和存储介质
CN113657779B (zh) * 2021-08-20 2024-01-09 杭州时趣信息技术有限公司 一种动态配置化的风控决策方法、***、设备和存储介质
CN115994821A (zh) * 2023-01-09 2023-04-21 中云融拓数据科技发展(深圳)有限公司 基于产业链数字化场景金融模型建立金融风控体系的方法
CN116739742A (zh) * 2023-06-02 2023-09-12 北京百度网讯科技有限公司 信贷风控模型的监控方法、装置、设备及存储介质
CN116701153B (zh) * 2023-08-09 2023-10-27 云账户技术(天津)有限公司 结算服务性能的评估方法、装置、电子设备及存储介质
CN116701153A (zh) * 2023-08-09 2023-09-05 云账户技术(天津)有限公司 结算服务性能的评估方法、装置、电子设备及存储介质
CN117556225A (zh) * 2024-01-12 2024-02-13 杭银消费金融股份有限公司 一种人行征信数据风险管理***
CN117556225B (zh) * 2024-01-12 2024-04-05 杭银消费金融股份有限公司 一种人行征信数据风险管理***
CN118035750A (zh) * 2024-04-11 2024-05-14 杭银消费金融股份有限公司 一种基于panel data的信贷模型样本构建方法

Also Published As

Publication number Publication date
CN107886425A (zh) 2018-04-06

Similar Documents

Publication Publication Date Title
WO2019080407A1 (zh) 信贷评估方法、装置、设备及计算机可读存储介质
TWI712981B (zh) 風險辨識模型訓練方法、裝置及伺服器
García-Meca et al. Does managerial ability influence the quality of financial reporting?
CN106529729A (zh) 基于BP_Adaboost模型的***用户违约的预测方法及***
WO2019109523A1 (zh) 经营数据审核方法、装置、设备及计算机可读存储介质
US8751338B2 (en) System and method of intra-cycle payment of accrued employee wages
CN103150672A (zh) 一种银行快速授信***及方法
CN108898476A (zh) 一种贷款客户信用评分方法和装置
US20220351284A1 (en) System and method for the rapid, flexible approval and disbursement of a loan
Bai et al. Testing panel cointegration with unobservable dynamic common factors that are correlated with the regressors
Li et al. Performance management of university financial budget execution relying on comprehensive budget management strengthening model
Khedmatgozar The impact of perceived risks on internet banking adoption in Iran: a longitudinal survey
CN107527284A (zh) 一种基于海关数据和金融数据的复合征信的评定方法
Zhang et al. Data-driven robust cost consensus model with individual adjustment willingness in group decision-making
CN103970651A (zh) 基于组件安全属性的软件体系结构安全性评估方法
Sun et al. Does financial development hamper or improve the resource curse? Analysis based on the panel threshold effect model
CN109727116A (zh) 信贷分析方法、装置、设备及计算机可读存储介质
Jin et al. Shifting from the incurred to the expected credit loss model and stock price crash risk
CN110213239B (zh) 可疑交易报文生成方法、装置及服务器
Zeng [Retracted] Research on Risk Measurement and Early Warning of Electronic Banking Business Based on GMDH Algorithm
JP2003036346A (ja) オペレーショナル・リスク評価方法及びそのシステム
US11663658B1 (en) Assessing the presence of selective omission via collaborative counterfactual interventions
CN110362981B (zh) 基于可信设备指纹判断异常行为的方法及***
CN115860889A (zh) 一种基于人工智能的金融贷款大数据管理方法及***
TWM596929U (zh) 房貸轉增貸系統

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: 18871670

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 14/09/2020)

122 Ep: pct application non-entry in european phase

Ref document number: 18871670

Country of ref document: EP

Kind code of ref document: A1