WO2017157165A1 - 信用分数模型训练方法、信用分数计算方法、装置及服务器 - Google Patents
信用分数模型训练方法、信用分数计算方法、装置及服务器 Download PDFInfo
- Publication number
- WO2017157165A1 WO2017157165A1 PCT/CN2017/075131 CN2017075131W WO2017157165A1 WO 2017157165 A1 WO2017157165 A1 WO 2017157165A1 CN 2017075131 W CN2017075131 W CN 2017075131W WO 2017157165 A1 WO2017157165 A1 WO 2017157165A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- credit score
- user
- sample
- relationship
- credit
- Prior art date
Links
Images
Classifications
-
- 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
- G06N5/04—Inference or reasoning models
- G06N5/045—Explanation of inference; Explainable artificial intelligence [XAI]; Interpretable artificial intelligence
-
- 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
- G06Q40/02—Banking, e.g. interest calculation or account maintenance
-
- 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
- Embodiments of the present invention relate to the field of credit evaluation, and in particular, to a credit score model training method, a credit score calculation method, apparatus, and server.
- the credit score refers to the score obtained by the credit evaluation agency using the credit score model to quantitatively analyze the personal credit information.
- the credit score is used to characterize the user's personal credit.
- the model for calculating the credit score is mainly the FICO (Fair Isaac Company) credit score model.
- the FICO credit score model first takes the user's personal credit, morality and ability to pay as indicators; according to the grade of each indicator, the scores of each indicator are obtained; different grades correspond to different scores; the credit scores of users are calculated according to the weights and scores of each indicator .
- the data used in the FICO credit score model is for the user's personal information. When the user's personal information is missing or there is an error, it is difficult to accurately calculate the user's credit score.
- the embodiment of the present invention provides a credit score model training method, a credit score calculation method, a device, and a server. .
- the technical solution is as follows:
- a credit score model training method comprising:
- the sample For each of the sample users, based on the i-th credit score of the sample user, the sample The user's i-th relationship credit score and the sample user's default annotation information, training the credit score model, and calculating the i+1st credit score of the sample user;
- the i-th relationship credit score of the sample user is calculated according to an i-th credit score of each other user having a social relationship with the sample user; the sample user's default annotation information is used for the The classification of the sample users.
- a credit score calculation method is provided, which is applied to a credit score model provided by the first aspect, the method comprising:
- the first credit score of the user and the first relationship credit score of the user are input into the trained credit score model, and the target credit score of the user is calculated.
- a credit score model training apparatus comprising:
- An information obtaining module configured to acquire an i-th credit score of the plurality of sample users, an i-th relationship credit score of each of the sample users, and default label information of each of the sample users, where i is a starting value of 1 Positive integer
- An information input module configured, for each of the sample users, to train a credit score according to an i-th credit score of the sample user, an i-th relationship credit score of the sample user, and default information of the sample user Modeling, and calculating an i+1th credit score of the sample user;
- a model training module configured to obtain the trained credit score model when an i+1th credit score of each sample user satisfies a training end condition
- the i-th relationship credit score of the sample user is calculated according to an i-th credit score of each other user having a social relationship with the sample user; the sample user's default annotation information is used for the The classification of the sample users.
- a credit score calculation apparatus which is applied to a credit score model provided by the third aspect, the apparatus comprising:
- a credit acquisition module configured to acquire a first credit score of each user and a number of each of the users a relationship credit score, the first relationship credit score of the user is calculated according to a first credit score of each other user having a social relationship with the user;
- a target calculation module configured, for each user, inputting the first credit score of the user and the first relationship credit score of the user into the trained credit score model, and calculating the target credit of the user fraction.
- a credit score model training server comprising:
- One or more processors are One or more processors.
- the memory stores one or more programs, the one or more programs being configured to be executed by the one or more processors, the one or more programs including instructions for:
- the credit score model is trained according to the i-th credit score of the sample user, the i-th relationship credit score of the sample user, and the default annotation information of the sample user, and the calculation is performed.
- the i-th relationship credit score of the sample user is calculated according to an i-th credit score of each other user having a social relationship with the sample user; the sample user's default annotation information is used for the The classification of the sample users.
- the sixth aspect provides a credit score calculation server, which is applied to the credit score model provided by the fifth aspect, where the server includes:
- One or more processors are One or more processors.
- the memory stores one or more programs, the one or more programs being configured to be executed by the one or more processors, the one or more programs including instructions for:
- For each user input the first credit score of the user and the first relationship credit score of the user into the trained credit score model, and calculate a target credit score of the user. number.
- the trained credit score model is obtained; it is difficult to accurately calculate the user's credit score when the user's personal information is missing or there is an error;
- the credit scores of other users who have a social relationship with the user are comprehensively calculated to calculate the target credit score of the user, which expands the coverage of the credit score model and improves the accuracy of calculating the credit score of the user.
- FIG. 1 is a flowchart of a credit score model training method provided by an exemplary embodiment of the present invention
- FIG. 2A is a flowchart of a credit score model training method provided by another exemplary embodiment of the present invention.
- 2B is a flowchart of a method for calculating a relationship credit score provided by an exemplary embodiment of the present invention
- 2C is a schematic structural diagram of credit score model training provided by an exemplary embodiment of the present invention.
- FIG. 3 is a flowchart of a credit score calculation method provided by an exemplary embodiment of the present invention.
- 4A is a flowchart of a credit score calculation method provided by another exemplary embodiment of the present invention.
- 4B is a flowchart of an end condition of a loop calculation provided by an exemplary embodiment of the present invention.
- FIG. 5 is a structural block diagram of a credit score model training apparatus according to an embodiment of the present invention.
- FIG. 6 is a structural block diagram of a credit score model training apparatus according to another embodiment of the present invention.
- FIG. 7 is a structural block diagram of a credit score calculation apparatus according to an embodiment of the present invention.
- FIG. 8 is a structural block diagram of a credit score calculation apparatus according to another embodiment of the present invention.
- FIG. 9 is a structural block diagram of a server according to an embodiment of the present invention.
- a credit score is a score obtained by modeling a person's credit information using a mathematical model. This score can be used in the financial sector for risk assessment of credit business.
- the personal credit information includes information items related to the user's personal credit.
- the credit score model disclosed in the embodiment of the present invention mainly calculates the target credit score of the user by using the credit score of the user and the relationship credit score calculated based on the social relationship of the user, thereby realizing that when the user's personal information is missing or wrong.
- the effect of the user's target credit score can still be calculated by the relationship credit score calculated based on the user's social relationship.
- Embodiments of the present invention include an embodiment of a credit score model training method and an embodiment of calculating a user credit score based on a credit score model.
- For the credit score model training method please refer to the embodiment shown in FIG. 1 and FIG. 2A.
- For the credit score calculation method please refer to the embodiment shown in FIG. 3 and FIG. 4A.
- FIG. 1 shows a flowchart of a credit score model training method provided by an exemplary embodiment of the present invention.
- the method includes:
- Step 101 Acquire an i-th credit score of a plurality of sample users, an i-th relationship credit score of each sample user, and default label information of each sample user, where i is a positive integer with a starting value of 1.
- a sample set for training the credit score model is first obtained.
- the sample set includes: a first credit score of a plurality of sample users, a first relationship credit score corresponding to each sample user, and default label information of each sample user.
- the sample user's relationship credit score is calculated based on the credit scores of various other users who have a social relationship with the sample user.
- each other user who has a social relationship with the sample user may be a sample user in the sample set or a user other than the sample set.
- the sample user's default label information is used to indicate the sample user's default.
- Step 102 For each sample user, according to the i-th credit score of the sample user, the i-th relationship credit score of the sample user, and the default annotation information of the sample user, the credit score model is trained, and the i+ of the sample user is calculated. 1 credit score.
- the step of calculating the i+1th credit score of the sample user may include: when i ⁇ 2, the i-th credit score of the sample user and the i-th relationship information score of the sample user for each sample user Enter the preset credit score model and calculate the i+1st credit score of the sample user.
- Step 103 When the i+1th credit score of each sample user satisfies the training end condition, the trained credit score model is obtained.
- the credit score model training method obtains the i-th credit score of a plurality of sample users, the i-th relationship credit score of each sample user, and the default annotation information of each sample user; According to the i-th credit score of the sample user, the i-th relationship credit score of the sample user, and the default annotation information of the sample user, the credit score model is trained to calculate the i+1th credit score of the sample user; When the i+1th credit score satisfies the training end condition, the trained credit score model is obtained; and it is difficult to accurately calculate the user's credit score when the user's personal information is missing or has errors; By simultaneously counting the user's personal credit score and the credit scores of other users who have social relationships with the user, the target credit score of the user is comprehensively calculated, the coverage of the credit score model is expanded, and the accuracy of calculating the credit score of the user is improved. Sexual effect.
- FIG. 2A shows a flowchart of a credit score model training method provided by another exemplary embodiment of the present invention.
- the method includes:
- Step 201 Acquire a sample set.
- the sample set includes: an identifier of a plurality of sample users and a default labeling information of each sample user.
- the identifier of the sample user is used to distinguish between sample users.
- the sample set contains 100 sample users, and the identifiers of each sample user are user1, user2, user3, ... user100.
- the default annotation information of the sample user is used to indicate the default situation of the sample user.
- the default annotation information of the sample user included in the sample set is: user1(1), user2(0), user3(0), ...user100(1) , where 1 in parentheses indicates that the sample user has a default condition, which belongs to the default category; 0 in the parentheses indicates that the sample user does not have a default condition, and belongs to the non-default category.
- User1(1) represents the sample user user1 belongs to the default class
- user2(0) represents the sample user user2 belongs to the non-default class
- user3(0) represents the sample user user3 belongs to the non-default class
- user100(1) represents the sample user user100 belongs to the default class.
- the sample user's default annotation information may be based on the sample user's credit score. If the user's default condition is marked, it can also be marked according to the actual default situation of the sample user, which is not limited in this embodiment.
- an example of labeling a sample user's default based on a sample user's credit score is exemplified.
- the FICO credit score model in the related technology is used to calculate the credit score of each sample user, and the default situation of the sample user is marked according to the calculated credit score of the sample user.
- the FICO credit score model has a scoring range of 300-800 points, a sample user with a credit score greater than 550 points is not defaulted, and a sample user with a credit score less than 550 points is marked as default.
- sample user 1 is 580, the credit score of sample user 2 is 450, the credit score of sample user 3 is 480, the credit score of sample user 4 is 600, and the credit score of sample user 5 is 700, which can be marked.
- Sample user 1 is non-default, sample user 2 is default, sample user 3 is default, sample user 4 is non-default, and sample user 5 is non-default. Therefore, the sample user 1, the sample user 4, and the sample user 5 can be divided into one category, that is, the non-default category; the sample user 2 and the sample user 3 are divided into another category, that is, the default category.
- Step 202 Acquire an ith credit score of a plurality of sample users, where i is a positive integer with a starting value of 1.
- the step may include: calculating, according to the personal information of each sample user, the first credit score of each sample user; for example, calculating the first time of the sample user according to the credit, morality and payment ability of the sample user. Credit scores, such as the calculation process of the FICO credit score model.
- the first relationship credit score of the sample user is determined as the first credit score of the sample user. The first relationship credit score is calculated based on the first credit score of each other user who has a social relationship with the sample user.
- Step 203 Acquire an i-th relationship credit score of each sample user.
- the sample user's first relationship credit score is calculated based on the first credit score of each other user who has a social relationship with the sample user.
- each other user who has a social relationship with the sample user may be another sample user in the sample set or a user other than the sample set.
- the number of other users who have a social relationship with each sample user may be the same or different. For example, there are 50 other users who have a social relationship with the sample user A, and 50 users are sample users in the sample set; there is a social relationship with the sample user B. There are 60 other users, of which 20 are sample users in the sample set, 40 are users outside the sample set; 60 other users have social relationships with sample user C, but 40 are sample sets. For the sample users, there are 20 users outside the sample set. In this embodiment, the number of other users who have a social relationship with each sample user is not specifically limited.
- obtaining the ith relationship credit score of each sample user in step 203 can be obtained by the following steps, as shown in FIG. 2B.
- step 203a a social relationship chain of each sample user is obtained.
- a social relationship chain is a social relationship between a sample user and other users.
- the social relationship may include social relationships in a real society, such as: kinship, colleague relationship, and classmate relationship; and may be a friend relationship or a concern relationship in a social tool, such as: WeChat friend relationship, QQ friend relationship, Weibo Friends relationship, friend relationship in the same group, and Weibo concern relationship.
- the social relationship chain corresponding to each sample user is obtained. That is, the social relationship existing between each sample user and other users is obtained.
- Step 203b Acquire an ith credit score of each other user corresponding to each sample user according to the social relationship chain.
- the i-th credit score of each other user corresponding to the sample user is obtained according to the social relationship chain of the sample user.
- Step 203c Calculate an i-th relationship credit score of each sample user according to the obtained i-th credit score of each other user.
- calculating the relationship credit score for each sample user can include the following two implementations:
- the step includes: calculating, for each sample user, an average value of the i-th credit scores of each other user corresponding to the sample user according to the obtained i-th credit score of each other user, The calculated average value is determined as the i-th relationship credit score of the sample user.
- the i-th relationship credit score of sample user 1 passes through various other social relationships with sample user 1. The user's ith credit score is calculated.
- the i-th relationship credit score of each sample user can be calculated.
- the relationship credit score of each sample user may be calculated by using the i-th credit score of each other user corresponding to the sample user and the weight corresponding to each other user, and specifically includes the following steps.
- the first sub-step obtains the weights corresponding to each other user according to the social relationship chain.
- the weight corresponding to each other user refers to the proportion of the first credit score of each other user in calculating the relationship credit score of the sample user.
- the weights of the other users may be determined according to the type of the social relationship. For example, the weight of the other users who have a kinship relationship with the sample user is greater than the weight of other users who have the relationship of the classmates; and may also be determined by intimacy, such as:
- the weight corresponding to other users with high frequency of communication with the sample user is greater than the weight of other users with low communication frequency; it can also be determined by the length of time that the social relationship exists, for example, other social interaction with the sample user is longer.
- the weight corresponding to the user is greater than the weight corresponding to other users who have a short social relationship time.
- the manner of determining the weight corresponding to each other user who has a social relationship with the sample user is not specifically limited.
- the i-th relationship credit score of the sample user is calculated according to the i-th credit score of each other user and the weight corresponding to each other user.
- the specific calculation formula is as follows:
- score_fri_avg refers to the i-th relationship credit score of the sample user
- friend_score_j refers to the i-th credit score of the j-th other user who has a social relationship with the sample user
- op_j refers to the j-th social relationship with the sample user.
- the i-th relationship credit score of the sample user 1 is calculated by the i-th credit score of each other user who has a social relationship with the sample user 1, and the weight corresponding to each other user. In the same way, you can calculate the ith time of each sample user. Credit score.
- Step 204 For each sample user, according to the i-th credit score of the sample user, the i-th relationship credit score of the sample user, and the default annotation information of the sample user, the credit score model is trained, and the i+ of the sample user is calculated. 1 credit score.
- the credit score model is trained according to the first credit score of the sample user, the first relationship credit score of the sample user, and the default annotation information of the sample user, and the sample user is calculated.
- the second credit score is calculated for each sample user.
- each will be The sample user's default label information is used as the target variable and trained to obtain the credit score model.
- the step of training the credit score model may be performed by a method such as a logistic regression, a neural network, or a decision tree, which is not limited in this embodiment.
- the second credit score of each sample user can be calculated by the credit score model.
- the credit score model is trained according to the i-th credit score of the sample user and the i-th credit score of the sample user, and the i+1th credit score of the sample user is calculated. .
- the credit score model is trained according to the obtained i-th credit score of each sample user and the i-th credit score of each sample user as a characteristic variable. And the i+1th credit score of each sample user is calculated by the algorithm in the credit score model, and there is no need to input the default annotation information of each sample user again.
- the formula for calculating the i+1th credit score of the sample user according to the i-th credit score of the sample user and the i-th relationship credit score of the sample user is as follows:
- new_score refers to the i+1th credit score of the sample user
- score refers to the i-th credit score of the sample user
- score_fri_avg refers to the i-th relationship credit score of the sample user.
- the f function is the credit score model obtained by training.
- the sample user may be classified according to the calculated i+1th credit score of each sample user.
- the default annotation information of each sample user and the calculated i+1 credit score of each sample user By using the default annotation information of each sample user and the calculated i+1 credit score of each sample user, a connection is established between the credit score of each sample user and the default annotation information, so as to use the credit score model. When the user is accurately classified.
- Step 205 calculating an i+1th credit score of each sample user and an i-th credit score of each sample user The correlation coefficient between the sub-credit scores.
- the correlation coefficient between the i+1th credit score of each sample user and the i-th credit score of each sample user is calculated.
- the correlation coefficient between the i+1th credit score of the sample user and the i-th credit score of the sample user is calculated as follows:
- cor is the correlation coefficient between the i+1th credit score of each sample user and the i-th credit score of each sample user, and score refers to the i-th credit score of each sample user, and new_score refers to each sample user.
- the i+1th credit score is the correlation coefficient between the i+1th credit score of each sample user and the i-th credit score of each sample user, and score refers to the i-th credit score of each sample user, and new_score refers to each sample user.
- the i+1th credit score is the correlation coefficient between the i+1th credit score of each sample user and the i-th credit score of each sample user, and score refers to the i-th credit score of each sample user, and new_score refers to each sample user.
- the i+1th credit score is the correlation coefficient between the i+1th credit score of each sample user and the i-th credit score of each sample user, and score refers to the i-th credit score of each sample user, and new_score refer
- the correlation coefficient between the i+1 credit score and the i-th credit score of each sample user means that the correlation coefficient between the set X and the set Y is calculated by the correlation coefficient calculation formula.
- Step 206 Detect whether the correlation coefficient is greater than a first threshold.
- the correlation coefficient is compared with the first threshold, and then the i+ of each sample user is detected. The correlation between the 1st credit score and the i-th credit score of each sample user.
- Step 207 If the correlation coefficient is greater than the first threshold, the trained credit score model is obtained.
- the model training process ends, the current credit score model is the trained credit score model, and the i+1th credit score of each sample user is the target credit score calculated by the credit score model.
- the i+1th credit score of each sample user is taken as input, and the i+1th time of each other user corresponding to each sample user is obtained again according to the social relationship chain. Credit score.
- FIG. 2C a schematic structural diagram of the credit score model training shown in steps 201 to 208 in the embodiment of FIG. 2 is shown in FIG. 2C.
- the credit score model training method obtains the i-th credit score of a plurality of sample users, the i-th relationship credit score of each sample user, and the default annotation information of each sample user;
- the credit score model is trained according to the i-th credit score of the sample user, the i-th relationship credit score of the sample user, and the default annotation information of the sample user, and the i+1th credit score of the sample user is calculated;
- the trained credit score model is obtained; and it is difficult to accurately calculate the user's credit score when the user's personal information is missing or has errors;
- the user's target credit score is calculated by simultaneously counting the user's personal credit score and the credit scores of other users who have social relationships with the user, which expands the coverage of the credit score model and improves the accuracy of calculating the credit score of the user. Effect.
- the first relationship credit score of the sample user is determined as the first credit score of the sample user, and the personal credit score of the user is reached.
- the effect of the user's credit score can also be calculated based on the user's relationship credit score.
- step 203b the ith credit score of each other user corresponding to each sample user is obtained according to the social relationship chain, which can be understood as:
- the first credit score of each other user of each sample user is obtained according to the social relationship chain.
- the first credit score of each other user refers to a credit score of each other user calculated based on the credit score model of the personal information before training the credit score model provided by the embodiment.
- the first credit score of each other user can also be used as the first credit score of other users, which is not limited in this embodiment.
- the i-th credit score of other users may change during the i-th iteration. Then, in the ith iteration, the ith credit score of each other user corresponding to each sample user needs to be obtained according to the social relationship chain.
- the i-th credit score of other users does not change during the i-th iteration, and therefore, during the ith iteration No need
- the i-th credit score of each other user corresponding to the sample user is obtained according to the social relationship chain, and the first credit score of each other user acquired according to the social relationship chain is directly used.
- each friend corresponding to the borrower 1 includes: friend 1, the corresponding credit score is 600, the weight of the friend 1 is 0.4; the friend 2, the corresponding credit score is 400, the weight of the friend 1 is 0.4; the friend 3, the corresponding The credit score is 500, the weight of friend 1 is 0.2, the credit score of friend 4 is 700, and the weight of friend 1 is 0.5; wherein friend 1 is the borrower 3 of the sample user, and friend 2 is the borrower in the sample user. person 2.
- the credit scores of each friend corresponding to each borrower and the weights corresponding to each friend can be obtained.
- the first relationship credit score corresponding to each borrower is calculated according to the credit scores of each of the borrowers corresponding to each borrower.
- the first relationship credit score corresponding to the borrower 1 can be calculated by the credit score of each friend corresponding to the borrower and the weight corresponding to each friend.
- the first credit score of 100 borrowers, the first relationship credit score corresponding to 100 borrowers, and the default bid information corresponding to 100 borrowers are input into a preset credit score model, and the credit score model is based on internal
- the algorithm calculates the second credit score corresponding to each borrower. For example, the first credit score of the borrower 1, the first relationship credit score and the default markup information, the first credit score of the borrower 2, the first relationship credit score and the default markup information until the borrower 100
- the first credit score, the first relationship credit score and the default bid information are simultaneously input into a preset credit score model, and the second credit score corresponding to each of the borrower 1 to the borrower 100 is calculated by the credit score model.
- the correlation coefficient between the second credit score corresponding to 100 borrowers and the first credit score corresponding to 100 borrowers is calculated.
- the trained credit score model is obtained; when the correlation coefficient is less than the first threshold, the calculated second credit scores of the borrower 1 to the borrower 100 are respectively calculated and recalculated The corresponding second relationship credit scores of the respective ones continue to be input into the preset credit score model, and the third credit score corresponding to each of the borrower 1 to the borrower 100 is continuously calculated.
- the trained credit score model is obtained until the calculated correlation coefficient between the i+1st credit score and the i th credit score corresponding to each borrower is greater than the first threshold.
- the user's target credit score can be calculated by inputting the user's credit score and the user's relationship credit score into the credit score model. Please refer to the embodiment of FIG. 3 for specific steps.
- FIG. 3 shows a flowchart of a credit score calculation method provided by an exemplary embodiment of the present invention.
- the method includes:
- Step 301 Acquire a first credit score of each user and a first relationship credit score of each user.
- the user's first relationship credit score is calculated based on the first credit score of each other user who has a social relationship with the user.
- For detailed calculation process of the user's first relationship credit score please refer to step 203 in the embodiment of FIG. 2A.
- the credit score model needs to be used to calculate the target credit score of the user, firstly, the first credit score of each user who needs to calculate the target credit score and the first relationship credit of each user are acquired. fraction.
- Step 302 For each user, input the first credit score of the user and the first relationship credit score of the user into the trained credit score model, and calculate the target credit score of the user.
- the credit score calculation method obtains the first credit score of each user and the first relationship credit score of each user; the first credit score of each user and the first The secondary relationship credit score is input into the credit score model, and the target credit score of the user is calculated; the problem that it is difficult to accurately calculate the credit score of the user when the user's personal information is missing or has an error is solved; Calculating the credit score of the user's personal credit score and other users' social relationships with the user calculates the target credit score of the user, expands the coverage of the credit score model, and improves the accuracy of calculating the credit score of the user.
- FIG. 4A shows a flowchart of a credit score calculation method provided by another exemplary embodiment of the present invention.
- the method includes:
- Step 401 Acquire a first credit score of each user and a first relationship credit score of each user.
- the user's first relationship credit score is calculated based on the first credit score of each other user who has a social relationship with the user.
- For detailed calculation process of the user's first relationship credit score please refer to step 203 in the embodiment of FIG. 2A.
- the first credit score of each user who needs to calculate the target credit score and the first relationship credit score of each user are acquired.
- the number of users is not specifically limited.
- Step 402 For each user, input the first credit score of the user and the first relationship credit score of the user into the trained credit score model, and calculate the xth credit score of the user.
- the first credit score of the user and the first relationship credit score of the user are input into the trained credit score model, and the first credit score of the user is determined according to an algorithm in the credit score model. And the first relationship credit score is analyzed and calculated to obtain the user's xth credit score, and x is a positive integer.
- the calculated xth credit score of each user refers to the credit score calculated after the same number of iterations in the credit score model.
- the trained credit score model is obtained after 100 iterations, when the first credit score of each user and the first relationship credit score of each user are input to After the trained credit score model, the calculated xth credit score of each user refers to the credit score of each user obtained after 100 iteration operations.
- Step 403 For each user, when the xth credit score of the user satisfies a preset condition, the xth credit score of the user is determined as the target credit score of the user.
- the calculated xth credit score of the user For each user, when the calculated xth credit score of the user satisfies a preset condition, the calculated xth credit score of the user is directly determined as the target credit score of the user.
- the step of determining whether the xth credit score of a certain user meets the preset condition may include the following substeps:
- Step 403a calculating a difference between the xth credit score of the user and the x-1st credit score of the user.
- the difference between the user's xth credit score and the x-1th credit score is calculated.
- calculating a difference between the user's xth credit score and the x-1th credit score is used to detect a correlation between the user's xth credit score and the x-1th credit score.
- a correlation coefficient between the xth credit score of each user and the x-1st credit score of each user may be calculated to determine the correlation between the two.
- Step 403b detecting whether the difference is less than a second threshold.
- each difference is compared with a second threshold, thereby detecting the xth credit score of each user and The correlation between the x-1th credit scores of individual users.
- Step 403c If the difference is less than the second threshold, determine the xth credit score of the user as the target credit score of the user.
- the difference between the xth credit score and the x-1th credit score of the user is less than the second threshold, determining that the xth credit score satisfies a preset condition, and directly determining the xth credit score of the user The target credit score for the user.
- the xth credit score of each user and the xth relationship credit score of each user are again input into the credit score model, and the x+1th credit of each user is continuously calculated. fraction.
- Step 404 For each user, when the xth credit score of the user does not satisfy the preset condition, the user's xth credit score and the xth relationship credit score are re-entered into the credit score model, and the user's number is calculated. x+1 credits.
- the user's xth relationship credit score is reacquired, and the user's xth credit score and the user's xth relationship credit score are re-entered into the credit score.
- the credit score calculation method obtains the first credit score of each user and the first relationship credit score of each user; the first credit score of each user and the first The secondary relationship credit score is input into the credit score model, and the target credit score of the user is calculated; the problem that it is difficult to accurately calculate the credit score of the user when the user's personal information is missing or has an error is solved; Calculating the credit score of the user's personal credit score and other users' social relationships with the user calculates the target credit score of the user, expands the coverage of the credit score model, and improves the accuracy of calculating the credit score of the user.
- FIG. 5 is a structural block diagram of a credit score model training apparatus according to an embodiment of the present invention.
- the credit score model training device can be implemented by software, hardware or a combination of both.
- the credit score model training device includes:
- the information obtaining module 520 is configured to obtain default labeling information of a plurality of sample users, and obtain an i-th credit score of each sample user and an i-th relationship credit score of each sample user.
- the information input module 540 is configured to, for each sample user, train the credit score model according to the i-th credit score of the sample user, the i-th relationship credit score of the sample user, and the default annotation information of the sample user, and calculate the sample user.
- the i+1th credit score is configured to, for each sample user, train the credit score model according to the i-th credit score of the sample user, the i-th relationship credit score of the sample user, and the default annotation information of the sample user, and calculate the sample user.
- the i+1th credit score is configured to, for each sample user, train the credit score model according to the i-th credit score of the sample user, the i-th relationship credit score of the sample user, and the default annotation information of the sample user.
- the model training module 560 is configured to satisfy the training node at the i+1th credit score of each sample user. When the condition is bundled, the trained credit score model is obtained.
- the sample user's i-th relationship credit score is calculated according to the i-th credit score of each other user who has a social relationship with the sample user; the sample user's default annotation information is used for classifying the sample user.
- the credit score model training apparatus obtains the i-th credit score of a plurality of sample users, the i-th relationship credit score of each sample user, and the default annotation information of each sample user;
- the credit score model is trained according to the i-th credit score of the sample user, the i-th relationship credit score of the sample user, and the default annotation information of the sample user, and the i+1th credit score of the sample user is calculated;
- the trained credit score model is obtained; and it is difficult to accurately calculate the user's credit score when the user's personal information is missing or has errors; It is achieved that the user's target credit score is calculated by simultaneously counting the user's personal credit score and the credit scores of other users who have social relationships with the user, thereby expanding the coverage of the credit score model and improving the accuracy of calculating the credit score of the user. .
- FIG. 6 is a structural block diagram of a credit score model training apparatus according to another embodiment of the present invention.
- the credit score model training device can be implemented by software, hardware or a combination of both.
- the credit score model training device includes:
- the information obtaining module 610 is configured to obtain default labeling information of a plurality of sample users, and obtain an i-th credit score of each sample user and an i-th relationship credit score of each sample user.
- the information acquiring module 610 may include: a first calculating unit 611 and a second calculating unit 612.
- the first calculating unit 611 is configured to calculate, for each sample user, a first credit score of the sample user according to the personal credit information of the sample user.
- the second calculating unit 612 is configured to determine, when the sample user that the first credit score cannot be calculated according to the personal credit information, determine the first relationship credit score of the sample user as the first credit score of the sample user, first The secondary relationship credit score is calculated based on the first credit score of each other user who has a social relationship with the sample user.
- the information obtaining module 610 may include: a social acquiring unit 613, a score obtaining unit 614, and a relationship calculating unit 615.
- the social obtaining unit 613 is configured to acquire a social relationship chain of each sample user, where the social relationship chain is The social relationship between the sample user and other users.
- the score obtaining unit 614 is configured to acquire an ith credit score of each other user corresponding to each sample user according to the social relationship chain.
- the relationship calculation unit 615 is configured to calculate an i-th relationship credit score of each sample user according to the acquired i-th credit score of each of the other users.
- the relationship calculation unit 615 may include: a third calculation unit 615a and a score determination unit 615b.
- the third calculating unit 615a is configured to calculate, for each sample user, an average value of the i-th credit scores of each other user corresponding to the sample user.
- the score determining unit 615b is configured to determine the average value as the i-th relationship credit score of the sample user.
- the relationship calculation unit 615 may include: a weight obtaining unit 615c and a fourth calculating unit 615d.
- the weight obtaining unit 615c is configured to acquire a weight corresponding to each other user according to the social relationship chain for each sample user.
- the fourth calculating unit 615d is configured to calculate a relationship credit score of the sample user according to the i-th credit score of each other user and the weight corresponding to each other user, and the specific formula is as follows:
- score_fri_avg refers to the i-th relationship credit score of the sample user
- friend_score_j refers to the i-th credit score of the j-th other user who has a social relationship with the sample user
- op_j refers to the j-th social relationship with the sample user.
- the information input module 620 is configured to input, for each sample user, the i-th credit score of the sample user, the i-th relationship credit score of the sample user, and the default annotation information of the sample user into a preset credit score model, and calculate The i+1 credit score of the sample user.
- the model training module 630 is configured to obtain a trained credit score model when the i+1th credit score of each sample user satisfies the training end condition.
- the sample user's i-th relationship credit score is calculated according to the i-th credit score of each other user who has a social relationship with the sample user; the sample user's default annotation information is used for classifying the sample user.
- the model training module 630 may include: a coefficient calculation unit 631, a coefficient detection unit 632, and a training completion unit 633.
- the coefficient calculation unit 631 is configured to calculate a correlation coefficient between the i+1th credit score of each sample user and the ith credit score of each sample user.
- the coefficient detecting unit 632 is configured to detect whether the correlation coefficient is greater than a first threshold.
- the training completion unit 633 is configured to obtain a trained credit score model if the correlation coefficient is greater than the first threshold.
- the credit score model includes a logistic regression classification model or a decision tree classification model.
- the credit score model training apparatus obtains the i-th credit score of a plurality of sample users, the i-th relationship credit score of each sample user, and the default annotation information of each sample user;
- the credit score model is trained according to the i-th credit score of the sample user, the i-th relationship credit score of the sample user, and the default annotation information of the sample user, and the i+1th credit score of the sample user is calculated;
- the trained credit score model is obtained; and it is difficult to accurately calculate the user's credit score when the user's personal information is missing or has errors; It is achieved that the user's target credit score is calculated by simultaneously counting the user's personal credit score and the credit scores of other users who have social relationships with the user, thereby expanding the coverage of the credit score model and improving the accuracy of calculating the credit score of the user. .
- the first relationship credit score of the sample user is determined as the first credit score of the sample user, and the personal credit score of the user is reached.
- the effect of the user's credit score can also be calculated based on the user's relationship credit score.
- FIG. 7 is a structural block diagram of a credit score calculation apparatus according to an embodiment of the present invention.
- the credit score calculation device can be implemented by software, hardware, or a combination of both.
- the credit score calculation device includes:
- the credit obtaining module 720 is configured to acquire a first credit score of each user and a first relationship credit score of each user, where the first relationship credit score of the user is first according to each other user who has a social relationship with the user. The credit score is calculated.
- the target calculation module 740 is configured to input, for each user, the first credit score of the user and the first relationship credit score of the user into the trained credit score model, and calculate a target credit score of the user.
- the credit score calculation apparatus obtains the first of each user by acquiring The sub-credit score and the first relationship credit score of each user; input the first credit score of each user and the first relationship credit score into the credit score model, calculate the target credit score of the user; and solve the problem when the user When the personal information is missing or there is an error, it is difficult to accurately calculate the user's credit score; the user's goal is calculated by simultaneously counting the user's personal credit score and the credit scores of other users who have social relationships with the user.
- the credit score expands the coverage of the credit score model and improves the accuracy of calculating the credit score of the user.
- FIG. 8 is a structural block diagram of a credit score calculation apparatus according to another embodiment of the present invention.
- the credit score calculation device can be implemented by software, hardware, or a combination of both.
- the credit score calculation device includes:
- the credit obtaining module 820 is configured to acquire a first credit score of each user and a first relationship credit score of each user, where the first relationship credit score of the user is first according to each other user who has a social relationship with the user. The credit score is calculated.
- the target calculation module 840 is configured to input, for each user, the first credit score of the user and the first relationship credit score of the user into the trained credit score model, and calculate a target credit score of the user.
- the target calculation module 840 may include: a pending calculation unit 841 and a target determination unit 842.
- the pending calculation unit 841 is configured to input the first credit score of the user and the first relationship credit score of the user into the trained credit score model, and calculate the xth credit score of the user.
- the target determining unit 842 is configured to determine the user's xth credit score as the user's target credit score when the user's xth credit score satisfies the preset condition.
- the target determining unit 842 may include: a difference calculating unit 843, a difference detecting unit 844, and a credit determining unit 845.
- the difference calculation unit 843 is configured to calculate a difference between the xth credit score of the user and the x-1th credit score of the user.
- the difference detecting unit 844 is configured to detect whether the difference is less than a second threshold.
- the credit determining unit 845 is configured to determine the xth credit score of the user as the target credit score of the user when the difference is less than the second threshold.
- the device may further include: a loop input module 860.
- the loop input module 860 is configured to: when the user's xth credit score does not meet the preset condition, The user's xth credit score and the xth relationship credit score are again input into the credit score model, and the user's x+1th credit score is calculated;
- the credit score calculation apparatus obtains the first credit score of each user and the first relationship credit score of each user; the first credit score of each user and the first The secondary relationship credit score is input into the credit score model, and the target credit score of the user is calculated; the problem that it is difficult to accurately calculate the credit score of the user when the user's personal information is missing or has an error is solved; Calculating the credit score of the user's personal credit score and other users' social relationships with the user calculates the target credit score of the user, expands the coverage of the credit score model, and improves the accuracy of calculating the credit score of the user.
- FIG. 9 is a schematic structural diagram of a server according to an embodiment of the present invention.
- the server 900 includes a central processing unit (CPU) 901, a system memory 904 including a random access memory (RAM) 902 and a read only memory (ROM) 903, and a system bus 905 that connects the system memory 904 and the central processing unit 901.
- the server 900 also includes a basic input/output system (I/O system) 906 that facilitates transfer of information between various devices within the computer, and mass storage for storing the operating system 913, applications 914, and other program modules 915.
- I/O system basic input/output system
- the basic input/output system 906 includes a display 908 for displaying information and an input device 909 such as a mouse or keyboard for user input of information.
- the display 908 and the input device 907 are both connected to the central processing unit 901 via an input and output controller 910 connected to the system bus 905.
- the basic input/output system 906 can also include an input output controller 910 for receiving and processing input from a plurality of other devices, such as a keyboard, mouse, or electronic stylus.
- input and output controller 910 also provides output to a display screen, printer, or other type of output device.
- the mass storage device 907 is connected to the central processing unit 901 by a mass storage controller (not shown) connected to the system bus 905.
- the mass storage device 907 and its associated computer readable medium provide non-volatile storage for the server 900. That is, the mass storage device 907 can include a computer readable medium (not shown) such as a hard disk or a CD-ROM drive.
- the computer readable medium can include computer storage media and communication media.
- Computer storage media includes for storage of computer readable instructions, data structures, program modules or Volatile and non-volatile, removable and non-removable media implemented by any method or technology of information such as other data.
- Computer storage media include RAM, ROM, EPROM, EEPROM, flash memory or other solid state storage technologies, CD-ROM, DVD or other optical storage, tape cartridges, magnetic tape, magnetic disk storage or other magnetic storage devices.
- RAM random access memory
- ROM read only memory
- EPROM Erasable programmable read-only memory
- EEPROM electrically erasable programmable read-only memory
- the server 900 may also be operated by a remote computer connected to the network through a network such as the Internet. That is, the server 900 can be connected to the network 912 through a network interface unit 911 connected to the system bus 905, or can also be connected to other types of networks or remote computer systems (not shown) using the network interface unit 911. .
- the memory further includes one or more programs, the one or more programs being stored in a memory, the one or more programs for performing the credit score model training method or credit score calculation method provided by the above embodiments.
- non-transitory computer readable storage medium comprising instructions, such as a memory comprising instructions executable by a processor of a mobile terminal to perform the credit model training method or credit score calculation described above method.
- the non-transitory computer readable storage medium may be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, and an optical data storage device.
- the device for training the credit score model in the above embodiment is only illustrated by the division of the above functional modules. In actual applications, the functions may be assigned differently according to needs.
- the function module is completed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above.
- the apparatus for training the credit score model provided by the above embodiment is the same as the method embodiment of the credit score model training, and the specific implementation process is described in detail in the method embodiment, and details are not described herein again.
- the device for calculating the credit score provided by the above embodiment is only exemplified by the division of each functional module described above in the calculation of the credit score model. In actual applications, the functions may be assigned differently according to needs.
- the function module is completed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above.
- the device for calculating the credit score provided by the above embodiment belongs to the same concept as the method for calculating the credit score, and the specific The implementation process is detailed in the method embodiment, and will not be described here.
- a person skilled in the art may understand that all or part of the steps of implementing the above embodiments may be completed by hardware, or may be instructed by a program to execute related hardware, and the program may be stored in a computer readable storage medium.
- the storage medium mentioned may be a read only memory, a magnetic disk or an optical disk or the like.
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Accounting & Taxation (AREA)
- Finance (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Technology Law (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- Marketing (AREA)
- Economics (AREA)
- Development Economics (AREA)
- Evolutionary Computation (AREA)
- Data Mining & Analysis (AREA)
- Medical Informatics (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computational Linguistics (AREA)
- Financial Or Insurance-Related Operations Such As Payment And Settlement (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
一种信用分数模型训练方法、信用分数计算方法、装置及服务器,属于信用评估领域。所述方法包括:获取样本用户的第i次信用分数、第i次关系信用分数和违约标注信息(101);根据样本用户的第i次信用分数、第i次关系信用分数和违约标注信息,训练信用分数模型,并计算得到样本用户的第i+1次信用分数(102);在满足训练结束条件时,得到训练完毕的信用分数模型(103);该方法解决了当用户个人的信息存在缺失或错误时,很难准确的计算出用户的信用分数的问题;通过用户的信用分数和基于用户的社交关系所计算出的关系信用分数综合计算出该用户的目标信用分数,达到了扩大信用分数模型覆盖范围,提高计算用户信用分数准确性的效果。
Description
本申请要求于2016年3月15日提交中国专利局、申请号为201610145843X、发明名称为“信用分数模型训练方法、信用分数计算方法及装置”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。
本发明实施例涉及信用评估领域,特别涉及一种信用分数模型训练方法、信用分数计算方法、装置及服务器。
信用分数是指信用评估机构利用信用分数模型对个人信用信息进行量化分析得出的分值。信用分数用于表征用户的个人信用。
相关技术中,计算信用分数的模型主要为FICO(Fair Isaac Company)信用分数模型。FICO信用分数模型首先将用户的个人信用、品德和支付能力作为指标;根据每个指标的档次获取各个指标的分数;不同档次对应不同的分数;根据每个指标的权重和分数计算用户的信用分数。
在实现本发明实施例的过程中,发明人发现相关技术至少存在以下问题:
FICO信用分数模型中采用的数据都是针对用户的个人信息的,当用户的个人信息有所缺失或存在错误时,很难准确的计算出用户的信用分数。
发明内容
为了解决当用户个人的信息有所缺失或存在错误时,很难准确的计算出用户的信用分数的问题,本发明实施例提供了一种信用分数模型训练方法、信用分数计算方法、装置及服务器。所述技术方案如下:
根据本发明实施例的第一方面,提供了一种信用分数模型训练方法,所述方法包括:
获取若干个样本用户的第i次信用分数、每个所述样本用户的第i次关系信用分数和每个所述样本用户的违约标注信息,i为起始值为1的正整数;
对于每个所述样本用户,根据所述样本用户的第i次信用分数、所述样本
用户的第i次关系信用分数和所述样本用户的违约标注信息,训练信用分数模型,并计算得到所述样本用户的第i+1次信用分数;
在各个所述样本用户的第i+1次信用分数满足训练结束条件时,得到训练完毕的所述信用分数模型;
其中,所述样本用户的第i次关系信用分数是根据与所述样本用户存在社交关系的各个其他用户的第i次信用分数计算得到的;所述样本用户的违约标注信息用于对所述样本用户的分类。
根据本发明实施例的第二方面,提供了一种信用分数计算方法,应用于第一方面提供的信用分数模型中,所述方法包括:
获取各个用户的第一次信用分数和每个所述用户的第一次关系信用分数,所述用户的第一次关系信用分数是根据与所述用户存在社交关系的各个其他用户的第一次信用分数计算得到的;
对于每个用户,将所述用户的第一次信用分数和所述用户的第一次关系信用分数输入训练完毕的所述信用分数模型,计算得到所述用户的目标信用分数。
根据本发明实施例的第三方面,提供了一种信用分数模型训练装置,所述装置包括:
信息获取模块,用于获取若干个样本用户的第i次信用分数、每个所述样本用户的第i次关系信用分数和每个所述样本用户的违约标注信息,i为起始值为1的正整数;
信息输入模块,用于对于每个所述样本用户,根据所述样本用户的第i次信用分数、所述样本用户的第i次关系信用分数和所述样本用户的违约标注信息,训练信用分数模型,并计算得到所述样本用户的第i+1次信用分数;
模型训练模块,用于在各个所述样本用户的第i+1次信用分数满足训练结束条件时,得到训练完毕的所述信用分数模型;
其中,所述样本用户的第i次关系信用分数是根据与所述样本用户存在社交关系的各个其他用户的第i次信用分数计算得到的;所述样本用户的违约标注信息用于对所述样本用户的分类。
根据本发明实施例的第四方面,提供了一种信用分数计算装置,应用于第三方面提供的信用分数模型中,所述装置包括:
信用获取模块,用于获取各个用户的第一次信用分数和每个所述用户的第
一次关系信用分数,所述用户的第一次关系信用分数是根据与所述用户存在社交关系的各个其他用户的第一次信用分数计算得到的;
目标计算模块,用于对于每个用户,将所述用户的第一次信用分数和所述用户的第一次关系信用分数输入训练完毕的所述信用分数模型,计算得到所述用户的目标信用分数。
第五方面,提供了一种信用分数模型训练服务器,所述服务器包括:
一个或多个处理器;和
存储器;
所述存储器存储有一个或多个程序,所述一个或多个程序被配置成由所述一个或多个处理器执行,所述一个或多个程序包含用于进行以下操作的指令:
获取若干个样本用户的第i次信用分数、每个所述样本用户的第i次关系信用分数和每个所述样本用户的违约标注信息,i为起始值为1的正整数;
对于每个所述样本用户,根据所述样本用户的第i次信用分数、所述样本用户的第i次关系信用分数和所述样本用户的违约标注信息,训练信用分数模型,并计算得到所述样本用户的第i+1次信用分数;
在各个所述样本用户的第i+1次信用分数满足训练结束条件时,得到训练完毕的所述信用分数模型;
其中,所述样本用户的第i次关系信用分数是根据与所述样本用户存在社交关系的各个其他用户的第i次信用分数计算得到的;所述样本用户的违约标注信息用于对所述样本用户的分类。
第六方面提供了一种信用分数计算服务器,应用于第五方面提供的信用分数模型中,所述服务器包括:
一个或多个处理器;和
存储器;
所述存储器存储有一个或多个程序,所述一个或多个程序被配置成由所述一个或多个处理器执行,所述一个或多个程序包含用于进行以下操作的指令:
获取各个用户的第一次信用分数和每个所述用户的第一次关系信用分数,所述用户的第一次关系信用分数是根据与所述用户存在社交关系的各个其他用户的第一次信用分数计算得到的;
对于每个用户,将所述用户的第一次信用分数和所述用户的第一次关系信用分数输入训练完毕的所述信用分数模型,计算得到所述用户的目标信用分
数。
本发明实施例提供的技术方案带来的有益效果是:
通过获取若干个样本用户的第i次信用分数、每个样本用户的第i次关系信用分数和每个样本用户的违约标注信息;对于每个样本用户,根据样本用户的第i次信用分数、样本用户的第i次关系信用分数和样本用户的违约标注信息,训练信用分数模型,并计算得到样本用户的第i+1次信用分数;在各个样本用户的第i+1次信用分数满足训练结束条件时,得到训练完毕的信用分数模型;解决了当用户个人的信息有所缺失或存在错误时,很难准确的计算出用户的信用分数的问题;达到了通过同时统计用户个人的信用分数和与用户存在社交关系的其他用户的信用分数来综合计算出该用户的目标信用分数,扩大了信用分数模型的覆盖范围,提高了计算用户的信用分数的准确性的效果。
为了更清楚地说明本发明实施例中的技术方案,下面将对实施例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发明的一些实施例,对于本领域普通技术人员来讲,在不付出创造性劳动的前提下,还可以根据这些附图获得其他的附图。
图1是本发明一个示例性实施例提供的信用分数模型训练方法的流程图;
图2A是本发明另一个示例性实施例提供的信用分数模型训练方法的流程图;
图2B是本发明一个示例性实施例提供的关系信用分数计算方法的流程图;
图2C是本发明一个示例性实施例提供的信用分数模型训练的结构示意图;
图3是本发明一个示例性实施例提供的信用分数计算方法的流程图;
图4A是本发明另一个示例性实施例提供的信用分数计算方法的流程图;
图4B是本发明一个示例性实施例提供的循环计算的结束条件的流程图;
图5是本发明一个实施例提供的信用分数模型训练装置的结构方框图;
图6是本发明另一个实施例提供的信用分数模型训练装置的结构方框图;
图7是本发明一个实施例提供的信用分数计算装置的结构方框图;
图8是本发明另一个实施例提供的信用分数计算装置的结构方框图;
图9是本发明一个实施例提供的服务器的结构方框图。
为使本发明的目的、技术方案和优点更加清楚,下面将结合附图对本发明实施方式作进一步地详细描述。
信用分数是指利用数学模型对个人的信用信息进行建模得出的分数。该分数可在金融领域中用于对信贷类业务进行风险评估。个人信用信息包括针对用户个人与信用有关的信息项。
本发明实施例公开的信用分数模型主要通过用户的信用分数和基于用户的社交关系所计算出的关系信用分数综合计算出该用户的目标信用分数,实现了当用户个人的信息存在缺失或错误时,通过基于用户的社交关系计算出的关系信用分数依然可以计算出该用户的目标信用分数的效果。本发明实施例包括:信用分数模型训练方法的实施例和基于信用分数模型计算用户信用分数的实施例。其中,信用分数模型训练方法请参考图1和图2A所示的实施例,信用分数计算方法请参考图3和图4A所示的实施例。
请参考图1,其示出了本发明一个示例性实施例提供的信用分数模型训练方法的流程图。该方法包括:
步骤101,获取若干个样本用户的第i次信用分数、每个样本用户的第i次关系信用分数和每个样本用户的违约标注信息,i为起始值为1的正整数。
在训练信用分数模型之前,首先获取用于训练信用分数模型的样本集。
其中,样本集包括:若干个样本用户的第一次信用分数、每个样本用户对应的第一次关系信用分数以及每个样本用户的违约标注信息。
样本用户的关系信用分数是根据与样本用户存在社交关系的各个其他用户的信用分数计算得到的。可选的,与样本用户存在社交关系的各个其他用户可以是样本集中的样本用户,也可以是样本集以外的用户。
样本用户的违约标注信息用于表示样本用户的违约情况。
步骤102,对于每个样本用户,根据样本用户的第i次信用分数、样本用户的第i次关系信用分数和样本用户的违约标注信息,训练信用分数模型,并计算得到样本用户的第i+1次信用分数。
可选的,计算样本用户的第i+1次信用分数的步骤可以包括:当i≥2时,对于每个样本用户,将样本用户的第i次信用分数和样本用户第i次关系信息分数输入预设的信用分数模型中,计算得到样本用户的第i+1次信用分数。
步骤103,在各个样本用户的第i+1次信用分数满足训练结束条件时,得到训练完毕的信用分数模型。
综上所述,本实施例提供的信用分数模型训练方法,通过获取若干个样本用户的第i次信用分数、每个样本用户的第i次关系信用分数和每个样本用户的违约标注信息;根据样本用户的第i次信用分数、样本用户的第i次关系信用分数和样本用户的违约标注信息,训练信用分数模型,计算得到样本用户的第i+1次信用分数;在各个样本用户的第i+1次信用分数满足训练结束条件时,得到训练完毕的信用分数模型;解决了当用户个人的信息有所缺失或存在错误时,很难准确的计算出用户的信用分数的问题;达到了通过同时统计用户个人的信用分数和与用户存在社交关系的其他用户的信用分数来综合计算出该用户的目标信用分数,扩大了信用分数模型的覆盖范围,提高了计算用户的信用分数的准确性的效果。
请参考图2A,其示出了本发明另一个示例性实施例提供的信用分数模型训练方法的流程图。该方法包括:
步骤201,获取样本集。
其中,样本集包括:若干个样本用户的标识和每个样本用户的违约标注信息。
样本用户的标识用于对样本用户之间的区分,比如:样本集中包含有100个样本用户,每个样本用户的标识为user1、user2、user3、……user100。
样本用户的违约标注信息用于表示样本用户的违约情况,比如:样本集中包含的样本用户的违约标注信息为:user1(1)、user2(0)、user3(0)、……user100(1),其中,括号中的1表示样本用户存在违约情况,属于违约类;括号中的0表示样本用户不存在违约情况,属于未违约类。则user1(1)代表样本用户user1属于违约类,user2(0)代表样本用户user2属于未违约类,user3(0)代表样本用户user3属于未违约类,user100(1)代表样本用户user100属于违约类。
可选的,样本用户的违约标注信息可以是根据样本用户的信用分数对样本
用户的违约情况进行标注的,也可以根据样本用户的实际违约情况进行标注的,本实施例对此不作限定。
示例性地,以根据样本用户的信用分数对样本用户的违约情况进行标注举例说明。比如:利用相关技术中的FICO信用分数模型计算每个样本用户的信用分数,根据计算出的样本用户的信用分数,对样本用户的违约情况进行标注。假定FICO信用分数模型的打分范围为300-800分之间,对信用分数大于550分的样本用户标注未违约,对信用分数小于550分的样本用户标注违约。以样本用户1的信用分数为580,样本用户2的信用分数为450,样本用户3的信用分数为480,样本用户4的信用分数为600,样本用户5的信用分数为700为例,可以标注样本用户1为未违约,样本用户2为违约,样本用户3为违约,样本用户4为未违约,样本用户5为未违约。因此,可以将样本用户1、样本用户4和样本用户5分为一类即未违约类;将样本用户2和样本用户3分为另一类即违约类。
步骤202,获取若干个样本用户的第i次信用分数,i为起始值为1的正整数。
当i=1时,本步骤可以包括:根据各个样本用户的个人信息计算得到各个样本用户的第一次信用分数;比如:根据样本用户的信用、品德和支付能力计算得到样本用户的第一次信用分数,如FICO信用分数模型的计算过程。可选的,当存在根据个人信用信息无法计算得到第一次信用分数的样本用户时,将样本用户的第一次关系信用分数确定为样本用户的第一信用分数。其中,第一次关系信用分数是根据与样本用户存在社交关系的各个其他用户的第一次信用分数计算得到的。
当i≥2时,获取步骤208中计算得到的各个样本用户的第i次信用分数。
步骤203,获取每个样本用户的第i次关系信用分数。
样本用户的第一次关系信用分数是根据与样本用户存在社交关系的各个其他用户的第一次信用分数计算得到的。
可选的,与样本用户存在社交关系的各个其他用户可以是样本集中的其他样本用户,也可以是样本集以外的用户。
可选的,对于每个样本用户,与每个样本用户存在社交关系的其他用户的个数可以相同,也可以不相同。比如:与样本用户A存在社交关系的其他用户有50个,且50个用户都为样本集中的样本用户;与样本用户B存在社交关系
的其他用户有60个,其中,有20个为样本集中的样本用户,有40个为样本集以外的用户;与样本用户C存在社交关系的其他用户也有60个,但有40个为样本集中的样本用户,有20个为样本集以外的用户。本实施例中,对与每个样本用户存在社交关系的其他用户的个数不作具体限定。
可选的,步骤203中获取每个样本用户的第i次关系信用分数可以通过下述的步骤得到,如图2B所示。
步骤203a,获取每个样本用户的社交关系链。
社交关系链是样本用户与其他用户之间存在的社交关系。其中,社交关系可以包括真实社会中的社交关系,如:亲属关系、同事关系和同学关系等;还可以是社交工具中的好友关系或关注关系,比如:微信好友关系、QQ好友关系、微博好友关系、同一群组中的好友关系和微博关注关系等。
根据样本集中的样本用户的标识,获取每个样本用户对应的社交关系链。也即,获取每个样本用户与其他用户之间存在的社交关系。
步骤203b,根据社交关系链获取每个样本用户对应的各个其他用户的第i次信用分数。
对于每个样本用户,根据样本用户的社交关系链获取与样本用户对应的各个其他用户的第i次信用分数。
步骤203c,根据获取到的各个其他用户的第i次信用分数计算各个样本用户的第i次关系信用分数。
通常情况下,计算各个样本用户的关系信用分数可以包括以下两种实现方式:
作为一种可能的实现方式,本步骤包括:对于每个样本用户,根据获取到的各个其他用户的第i次信用分数计算与样本用户对应的各个其他用户的第i次信用分数的平均值,将计算得到的平均值确定为样本用户的第i次关系信用分数。
比如:样本用户1的第一次信用分数为350,与样本用户1存在社交关系的各个其他用户包括:其他用户1,对应的第一次信用分数为400;其他用户2,对应的第一次信用分数为400;其他用户3,对应的第一次信用分数为500;其他用户4,对应的第一次信用分数为600;则样本用户1的第一次关系信用分数为475,具体计算公式为:(400+400+500+600)/4=1900/4=475。以此类推,样本用户1的第i次关系信用分数通过与样本用户1存在社交关系的各个其他
用户的第i次信用分数计算得到。同理,可以计算得出每个样本用户的第i次关系信用分数。
作为另一种可能的实现方式,可以通过与样本用户对应的各个其他用户的第i次信用分数和各个其他用户所对应的权重计算得到各个样本用户的关系信用分数,具体包括如下步骤。
第一子步骤,对于每个样本用户,根据社交关系链获取各个其他用户所对应的权重。其中,各个其他用户所对应的权重是指各个其他用户的第一次信用分数在计算样本用户的关系信用分数时所占的比重。各个其他用户所对应的权重可以根据社交关系类型确定,比如:与样本用户存在亲属关系的其他用户所对应的权重大于存在同学关系的其他用户所对应的权重;还可以通过亲密度确定,比如:与样本用户通信频率高的其他用户所对应的权重大于通信频率低的其他用户所对应的权重;还可以通过社交关系存在的时间长短来确定,比如:与样本用户存在社交关系时间较长的其他用户所对应的权重大于存在社交关系时间较短的其他用户所对应的权重。本实施例中,对与样本用户存在社交关系的各个其他用户所对应的权重的确定方式不作具体限定。
第二子步骤,根据各个其他用户的第i次信用分数和各个其他用户所对应的权重计算样本用户的第i次关系信用分数。具体的计算公式如下:
其中,score_fri_avg是指样本用户的第i次关系信用分数,friend_score_j是指与样本用户存在社交关系的第j个其他用户的第i次信用分数,op_j是指与样本用户存在社交关系的第j个其他用户所对应的权重。
比如:样本用户1的第一次信用分数为350,与样本用户1存在社交关系的各个其他用户包括:其他用户1,对应的第一次信用分数为400,所对应的权重为0.2;其他用户2,对应的第一次信用分数为400,所对应的权重为0.4;其他用户3,对应的第一次信用分数为500,所对应的权重为0.3;其他用户4,对应的第一次信用分数为600,所对应的权重为0.3;则样本用户1的第一次关系信用分数为475,具体计算公式为:(400*0.2+400*0.4+500*0.3+600*0.3)/(0.2+0.4+0.3+0.3)=570/1.2=475。以此类推,样本用户1的第i次关系信用分数通过与样本用户1存在社交关系的各个其他用户的第i次信用分数和各个其他用户所对应的权重计算得到。同理,可以计算得出每个样本用户的第i次关
系信用分数。
步骤204,对于每个样本用户,根据样本用户的第i次信用分数、样本用户的第i次关系信用分数和样本用户的违约标注信息,训练信用分数模型,并计算得到样本用户的第i+1次信用分数。
当i=1时,对于每个样本用户,根据样本用户的第一次信用分数、样本用户的第一次关系信用分数和样本用户的违约标注信息,训练信用分数模型,并计算得到样本用户的第2次信用分数。
在进行第一次迭代时,根据步骤202中获取到的每个样本用户的第一次信用分数和步骤203中获取到的每个样本用户的第一次关系信用分数作为特征变量,将每个样本用户的违约标注信息作为目标变量,训练得到信用分数模型中。其中,训练信用分数模型的步骤可以通过逻辑回归、神经网络或者决策树等方法进行训练,本实施例对此并不做限定。在训练得到信用分数模型之后,可以通过信用分数模型计算每个样本用户的第二次信用分数。
当i≥2时,对于每个样本用户,根据样本用户的第i次信用分数和样本用户的第i次关系信用分数,训练信用分数模型,并计算得到样本用户的第i+1次信用分数。
可选的,在进行第i次迭代时,只需根据获取到的每个样本用户的第i次信用分数和每个样本用户的第i次关系信用分数作为特征变量,进而训练信用分数模型,并通过信用分数模型中的算法计算得到每个样本用户的第i+1次信用分数,无需再次输入每个样本用户的违约标注信息。
可选的,根据样本用户的第i次信用分数和样本用户的第i次关系信用分数计算得到该样本用户的第i+1次信用分数的公式如下:
new_score=f(score,score_fri_avg)
其中,new_score是指样本用户的第i+1次信用分数,score是指样本用户的第i次信用分数,score_fri_avg是指样本用户的第i次关系信用分数。f函数即为训练得到的信用分数模型。
可选的,根据计算得到的每个样本用户的第i+1次信用分数可以对样本用户进行分类。通过每个样本用户的违约标注信息和计算得到的每个样本用户的第i+1次信用分数,将每个样本用户的信用分数与违约标注信息之间建立连接,以便于在使用信用分数模型时,对用户进行准确的分类。
步骤205,计算各个样本用户的第i+1次信用分数和各个样本用户的第i
次信用分数之间的相关系数。
在步骤204计算得到各个样本用户的第i+1次信用分数后,计算各个样本用户的第i+1次信用分数和各个样本用户的第i次信用分数之间的相关系数。
可选的,样本用户第i+1次信用分数和样本用户第i次信用分数之间相关系数的计算公式如下:
cor=correlation_coefficient(score,new_score)
其中,cor是指各个样本用户第i+1次信用分数和各个样本用户第i次信用分数之间的相关系数,score是指各个样本用户的第i次信用分数,new_score是指各个样本用户的第i+1次信用分数。
比如:参加训练信用分数模型的样本用户共有100个,所有样本用户的第i+1次信用分数的集合为X,所有样本用户的第i次信用分数的集合为Y,则各个样本用户的第i+1次信用分数和各个样本用户的第i次信用分数之间的相关系数是指通过相关系数计算公式计算集合X和集合Y之间的相关系数。
步骤206,检测相关系数是否大于第一阈值。
计算得到各个样本用户的第i+1次信用分数和各个样本用户的第i次信用分数之间的相关系数后,将该相关系数与第一阈值进行比较,进而检测各个样本用户的第i+1次信用分数与各个样本用户的第i次信用分数之间的相关性。
步骤207,若相关系数大于第一阈值,则得到训练完毕的信用分数模型。
若各个样本用户的第i+1次信用分数和各个样本用户的第i次信用分数之间的相关系数大于第一阈值,说明各个样本用户的第i次信用分数和第i+1次信用分数之间趋于相同,则模型训练过程结束,当前的信用分数模型即为训练完毕的信用分数模型,各个样本用户的第i+1次信用分数即为信用分数模型计算出的目标信用分数。
步骤208,若相关系数小于第一阈值,则令i=i+1,再次执行步骤202。
可选的,当相关系数小于第一阈值时,则将各个样本用户的第i+1次信用分数作为输入,再次根据社交关系链获取每个样本用户对应的各个其他用户的第i+1次信用分数。但是,在循环执行过程中,针对步骤204,只需输入各个样本用户的第i+1次信用分数和各个样本用户的第i+1次关系信用分数即可,各个样本用户的违约标注信息不作输入。
为了便于读者的理解,将图2实施例中的步骤201至步骤208所示的信用分数模型训练的结构示意图如图2C所示。
综上所述,本实施例提供的信用分数模型训练方法,通过获取若干个样本用户的第i次信用分数、每个样本用户的第i次关系信用分数和每个样本用户的违约标注信息;根据样本用户的第i次信用分数、样本用户的第i次关系信用分数和样本用户的违约标注信息,训练信用分数模型,并计算得到样本用户的第i+1次信用分数;在各个样本用户的第i+1次信用分数满足训练结束条件时,得到训练完毕的信用分数模型;解决了当用户个人的信息有所缺失或存在错误时,很难准确的计算出用户的信用分数的问题;达到了通过同时统计用户个人的信用分数和与用户存在社交关系的其他用户的信用分数综合计算出用户的目标信用分数,扩大了信用分数模型的覆盖范围,提高了计算用户的信用分数的准确性的效果。
另外,当存在根据个人信用信息无法计算得到第一次信用分数的样本用户时,将样本用户的第一次关系信用分数确定为样本用户的第一次信用分数,达到了在用户的个人信用分数缺失的情况下也可以根据用户的关系信用分数计算得到用户的信用分数的效果。
需要补充说明的是,在步骤203b中,根据社交关系链获取每个样本用户对应的各个其他用户的第i次信用分数,可以理解为:
当i=1时,根据社交关系链,获取每个样本用户的各个其他用户的第一次信用分数。
可选地,各个其他用户的第一次信用分数是指:在训练本实施例提供的信用分数模型之前,基于个人信息的信用分数模型所计算出的各个其他用户的信用分数。当然,若存在其它已有的信用分数模型所计算出的信用分数,也可以作为其他用户的第一次信用分数,本实施例对此不加以限定。
当i≥2时,根据社交关系链,获取每个样本用户对应的各个其他用户的第i次信用分数。
在训练信用分数模型过程中,由于与样本用户存在社交关系的其他用户可能是样本集中的其他样本用户,因此,在第i次迭代过程中,其他用户的第i次信用分数可能会发生变化,则在进行第i次迭代时,需要根据社交关系链获取每个样本用户对应的各个其他用户的第i次信用分数。当与样本用户存在社交关系的其他用户不是样本集中的其他样本用户时,则在第i次迭代过程中,其他用户的第i次信用分数不会发生变化,因此,在进行第i次迭代时,不需
根据社交关系链获取该样本用户对应的各个其他用户的第i次信用分数,直接使用根据社交关系链获取到的各个其他用户的第一次信用分数即可。
比如:在训练信用分数模型过程中,获取到与样本用户A存在社交关系的其他用户1和其他用户2都是样本集中的其他样本用户;且其他用户1的第一次信用分数为350,其他用户2的第一次信用分数为400;则在经过第一次迭代后,其他用户1的第二次信用分数为380,其他用户2的第二次信用分数为390,也即,由于其他用户1和其他用户2是样本集中的其他样本用户,因此在第一次迭代过程中,其他用户1和其他用户2的第二次信用分数发生了变化。在进行第二次迭代时,需要根据社交关系链获取样本用户A对应的各个其他用户的第二次信用分数。以此类推,在进行第i次迭代时,需要根据社交关系链获取样本用户A对应的各个其他用户的第i次信用分数。
在一个具体的例子中,假定将100个借款人作为样本用户。
首先获取每个借款人的标识、与每个借款人的标识对应的第一次信用分数以及每个借款人的违约标注信息。比如:借款人1,对应的信用分数350,标注违约;借款人2,对应的信用分数400,标注违约;借款人3,对应的信用分数600,标注未违约;借款人4,对应的信用分数650,标注未违约;…;借款人100,对应的信用分数440,标注违约。
再获取与每个借款人的标识对应的各个好友的信用分数。比如:借款人1对应的各个好友包括:好友1,对应的信用分数为600,好友1的权重为0.4;好友2,对应的信用分数为400,好友1的权重为0.4;好友3,对应的信用分数为500,好友1的权重为0.2;好友4,对应的信用分数为700,好友1的权重为0.5;其中,好友1为样本用户中的借款人3,好友2为样本用户中的借款人2。同理,可以获取各个借款人对应的各个好友的信用分数和各个好友对应的权重。
根据每个借款人对应的各个好友的信用分数,计算每个借款人对应的第一次关系信用分数。比如:借款人1对应的第一次关系信用分数可以通过借款人对应的各个好友的信用分数和各个好友对应的权重计算得到,具体的计算公式为:(600*0.4+400*0.4+500*0.2+700*0.5)/(0.4+0.4+0.2+0.5)=567。因此,借款人1对应的第一次关系信用分数为567。同理,可以计算得到各个借款人对应的第一次关系信用分数。
将100个借款人的第一次信用分数、100个借款人对应的第一次关系信用分数和100个借款人对应的违约标注信息输入到预设的信用分数模型中,信用分数模型根据内部的算法计算得到每个借款人对应的第二次信用分数。比如:将借款人1的第一次信用分数、第一次关系信用分数和违约标注信息、借款人2的第一次信用分数、第一次关系信用分数和违约标注信息,一直到借款人100的第一次信用分数、第一次关系信用分数和违约标注信息同时输入到预设的信用分数模型中,通过信用分数模型计算得到借款人1至借款人100各自对应的第二次信用分数。
通过计算100个借款人对应的第二次信用分数和100个借款人对应的第一次信用分数之间相关系数。比如:通过步骤206中所示的公式,计算出借款人1至借款人100各自对应的第二次信用分数和各自对应的第一次信用分数之间的相关系数。
当相关系数大于第一阈值时,得到训练完毕的信用分数模型;当相关系数小于第一阈值时,则将计算得到的借款人1至借款人100各自对应的第二次信用分数和重新计算得到的各自对应的第二次关系信用分数继续输入到预设的信用分数模型中,继续计算借款人1至借款人100各自对应的第三次信用分数。直到计算得到的每个借款人对应的第i+1次信用分数和第i次信用分数之间的相关系数大于第一阈值时,得到训练完毕的信用分数模型。
基于图2A所示的实施例,得到训练完毕的信用分数模型后,通过向信用分数模型中输入用户的信用分数和用户的关系信用分数即可计算用户的目标信用分数。具体步骤请参考图3实施例。
请参考图3,其示出了本发明一个示例性实施例提供的信用分数计算方法的流程图。该方法包括:
步骤301,获取各个用户的第一次信用分数和每个用户的第一次关系信用分数。
用户的第一次关系信用分数是根据与用户存在社交关系的各个其他用户的第一次信用分数计算得到的。用户的第一次关系信用分数的详细计算过程请参考图2A实施例中的步骤203所示。
在需要利用信用分数模型计算用户的目标信用分数时,首先获取需要计算目标信用分数的各个用户的第一次信用分数以及每个用户的第一次关系信用
分数。
步骤302,对于每个用户,将该用户的第一次信用分数和该用户的第一次关系信用分数输入训练完毕的信用分数模型,计算得到该用户的目标信用分数。
综上所述,本实施例提供的信用分数计算方法,通过获取各个用户的第一次信用分数和每个用户的第一次关系信用分数;将每个用户的第一次信用分数和第一次关系信用分数输入信用分数模型,计算得到该用户的目标信用分数;解决了当用户个人的信息有所缺失或存在错误时,很难准确的计算出用户的信用分数的问题;达到了通过同时统计用户个人的信用分数和与用户存在社交关系的其他用户的信用分数计算用户的目标信用分数,扩大了信用分数模型的覆盖范围,提高了计算用户的信用分数的准确性的效果。
请参考图4A,其示出了本发明另一个示例性实施例提供的信用分数计算方法的流程图。该方法包括:
步骤401,获取各个用户的第一次信用分数和每个用户的第一次关系信用分数。
用户的第一次关系信用分数是根据与用户存在社交关系的各个其他用户的第一次信用分数计算得到的。用户的第一次关系信用分数的详细计算过程请参考图2A实施例中的步骤203所示。
在需要利用信用分数模型计算用户的目标信用分数时,首先获取需要计算目标信用分数的各个用户的第一次信用分数以及每个用户的第一次关系信用分数。
本实施例中,对用户的个数不作具体限定。
步骤402,对于每个用户,将该用户的第一次信用分数和该用户的第一次关系信用分数输入训练完毕的信用分数模型,计算得到该用户的第x次信用分数。
对于每个用户,将该用户的第一次信用分数和该用户的第一次关系信用分数输入到训练完毕的信用分数模型中,根据信用分数模型中的算法对该用户的第一次信用分数和第一次关系信用分数进行分析和计算,得到该用户的第x次信用分数,x为正整数。
可选的,将各个用户的第一次信用分数和每个用户的第一次关系信用分数
输入到训练完毕的信用分数模型中,计算得到的各个用户的第x次信用分数是指经过与信用分数模型中相同次数迭代后计算得到的信用分数。
比如:图2A所示的信用分数模型训练方法中,经过100次迭代得到了训练完毕的信用分数模型,当将各个用户的第一次信用分数和每个用户的第一次关系信用分数输入到训练完毕的信用分数模型后,计算得到的各个用户的第x次信用分数是指经过100次迭代运算后得到的各个用户的信用分数。
步骤403,对于每个用户,在该用户的第x次信用分数满足预设条件时,将该用户的第x次信用分数确定为该用户的目标信用分数。
对于每个用户,当计算得到的该用户的第x次信用分数满足预设条件时,则直接将计算得到的用户的第x次信用分数确定为用户的目标信用分数。
可选的,请参考图4B,确定某一用户的第x次信用分数是否满足预设条件的步骤可以包括如下子步骤:
步骤403a,计算该用户的第x次信用分数和该用户的第x-1次信用分数之间的差值。
通过信用分数模型计算得到用户的第x次信用分数后,计算用户的第x次信用分数和第x-1次信用分数之间的差值。
可选的,计算用户的第x次信用分数和第x-1次信用分数之间的差值用于检测用户的第x次信用分数和第x-1次信用分数之间的相关性。可选地,当用户个数较少时可以采用计算各个用户的第x次信用分数和各个用户的第x-1次信用分数之间的相关系数来确定两者之间的相关性。
本实施例中,仅以计算用户的第x次信用分数和第x-1次信用分数之间的差值进行举例说明,可选地,还可以通过计算用户的第x次信用分数和第x-1次信用分数之间的相关系数来进行确定。
步骤403b,检测差值是否小于第二阈值。
在计算得到每个用户的第x次信用分数和第x-1次信用分数之间的差值后,将每个差值与第二阈值进行比较,进而检测各个用户的第x次信用分数与各个用户的第x-1次信用分数之间的相关性。
步骤403c,若差值小于第二阈值,则将该用户的第x次信用分数确定为该用户的目标信用分数。
若该用户的第x次信用分数和第x-1次信用分数的差值小于第二阈值,则确定第x次信用分数满足预设条件,此时直接将该用户的第x次信用分数确定
为用户的目标信用分数。
而若该用户的第x次信用分数和第x-1次信用分数的差值不小于第二阈值,则确定第x次信用分数不满足预设条件。令x=x+1,继续执行步骤402。
也即,若差值大于第二阈值,则将各个用户的第x次信用分数和各个用户的第x次关系信用分数再次输入到信用分数模型中,继续计算各个用户的第x+1次信用分数。
步骤404,对于每个用户,在该用户的第x次信用分数不满足预设条件时,将用户的第x次信用分数和第x次关系信用分数再次输入信用分数模型,计算得到用户的第x+1次信用分数。
当用户的第x次信用分数不满足预设条件时,则重新获取用户的第x次关系信用分数,并将用户的第x次信用分数和用户的第x次关系信用分数再次输入到信用分数模型中,继续计算用户的第x+1次信用分数。也即令x=x+1,再次执行计算各个用户的第x次信用分数和各个用户的第x-1次信用分数之间的差值的步骤。
综上所述,本实施例提供的信用分数计算方法,通过获取各个用户的第一次信用分数和每个用户的第一次关系信用分数;将每个用户的第一次信用分数和第一次关系信用分数输入信用分数模型,计算得到该用户的目标信用分数;解决了当用户个人的信息有所缺失或存在错误时,很难准确的计算出用户的信用分数的问题;达到了通过同时统计用户个人的信用分数和与用户存在社交关系的其他用户的信用分数计算用户的目标信用分数,扩大了信用分数模型的覆盖范围,提高了计算用户的信用分数的准确性的效果。
请参考图5,其示出了本发明一个实施例提供的信用分数模型训练装置的结构方框图。该信用分数模型训练装置可以通过软件、硬件或者两者的结合实现。该信用分数模型训练装置包括:
信息获取模块520,用于获取若干样本用户的违约标注信息,并获取每个样本用户的第i次信用分数以及每个样本用户的第i次关系信用分数。
信息输入模块540,用于对于每个样本用户,根据样本用户的第i次信用分数、样本用户的第i次关系信用分数和样本用户的违约标注信息,训练信用分数模型,并计算得到样本用户的第i+1次信用分数。
模型训练模块560,用于在各个样本用户的第i+1次信用分数满足训练结
束条件时,得到训练完毕的信用分数模型。
其中,样本用户的第i次关系信用分数是根据与样本用户存在社交关系的各个其他用户的第i次信用分数计算得到的;样本用户的违约标注信息用于对样本用户的分类。
综上所述,本实施例提供的信用分数模型训练装置,通过获取若干个样本用户的第i次信用分数、每个样本用户的第i次关系信用分数和每个样本用户的违约标注信息;根据样本用户的第i次信用分数、样本用户的第i次关系信用分数和样本用户的违约标注信息,训练信用分数模型,并计算得到样本用户的第i+1次信用分数;在各个样本用户的第i+1次信用分数满足训练结束条件时,得到训练完毕的信用分数模型;解决了当用户个人的信息有所缺失或存在错误时,很难准确的计算出用户的信用分数的问题;达到了通过同时统计用户个人的信用分数和与用户存在社交关系的其他用户的信用分数计算用户的目标信用分数,扩大了信用分数模型的覆盖范围,提高了计算用户的信用分数的准确性的效果。
请参考图6,其示出了本发明另一个实施例提供的信用分数模型训练装置的结构方框图。该信用分数模型训练装置可以通过软件、硬件或者两者的结合实现。该信用分数模型训练装置包括:
信息获取模块610,用于获取若干样本用户的违约标注信息,并获取每个样本用户的第i次信用分数以及每个样本用户的第i次关系信用分数。
可选的,当i=1时,信息获取模块610可以包括:第一计算单元611和第二计算单元612。
第一计算单元611,用于对于每个样本用户,根据该样本用户的个人信用信息计算得到样本用户的第一次信用分数。
第二计算单元612,用于当存在根据个人信用信息无法计算得到第一次信用分数的样本用户时,将样本用户的第一次关系信用分数确定为样本用户的第一次信用分数,第一次关系信用分数是根据与样本用户存在社交关系的各个其他用户的第一次信用分数计算得到的。
可选的,信息获取模块610可以包括:社交获取单元613、分数获取单元614和关系计算单元615。
社交获取单元613,用于获取每个样本用户的社交关系链,社交关系链是
样本用户与其他用户之间存在的社交关系。
分数获取单元614,用于根据社交关系链获取每个样本用户对应的各个其他用户的第i次信用分数。
关系计算单元615,用于根据获取到的各个所述其他用户的第i次信用分数计算各个样本用户的第i次关系信用分数。
可选的,关系计算单元615,可以包括:第三计算单元615a和分数确定单元615b。
第三计算单元615a,用于对于每个样本用户,计算与样本用户对应的各个其他用户的第i次信用分数的平均值。
分数确定单元615b,用于将平均值确定为样本用户的第i次关系信用分数。
可选的,关系计算单元615,可以包括:权重获取单元615c和第四计算单元615d。
权重获取单元615c,用于对于每个样本用户,根据社交关系链获取各个其他用户所对应的权重。
第四计算单元615d,用于根据各个其他用户的第i次信用分数和各个其他用户所对应的权重计算样本用户的关系信用分数,具体公式如下:
其中,score_fri_avg是指样本用户的第i次关系信用分数,friend_score_j是指与样本用户存在社交关系的第j个其他用户的第i次信用分数,op_j是指与样本用户存在社交关系的第j个其他用户所对应的权重。
信息输入模块620,用于对于每个样本用户,将样本用户的第i次信用分数、样本用户的第i次关系信用分数和样本用户的违约标注信息,输入预设的信用分数模型,计算得到样本用户的第i+1次信用分数。
模型训练模块630,用于在各个样本用户的第i+1次信用分数满足训练结束条件时,得到训练完毕的信用分数模型。
其中,样本用户的第i次关系信用分数是根据与样本用户存在社交关系的各个其他用户的第i次信用分数计算得到的;样本用户的违约标注信息用于对样本用户的分类。
可选的,模型训练模块630可以包括:系数计算单元631、系数检测单元632和训练完毕单元633。
系数计算单元631,用于计算各个样本用户的第i+1次信用分数和各个样本用户的第i次信用分数之间的相关系数。
系数检测单元632,用于检测相关系数是否大于第一阈值。
训练完毕单元633,用于若相关系数大于第一阈值,则得到训练完毕的信用分数模型。
可选的,信用分数模型包括逻辑回归分类模型或决策树分类模型。
综上所述,本实施例提供的信用分数模型训练装置,通过获取若干个样本用户的第i次信用分数、每个样本用户的第i次关系信用分数和每个样本用户的违约标注信息;根据样本用户的第i次信用分数、样本用户的第i次关系信用分数和样本用户的违约标注信息,训练信用分数模型,并计算得到样本用户的第i+1次信用分数;在各个样本用户的第i+1次信用分数满足训练结束条件时,得到训练完毕的信用分数模型;解决了当用户个人的信息有所缺失或存在错误时,很难准确的计算出用户的信用分数的问题;达到了通过同时统计用户个人的信用分数和与用户存在社交关系的其他用户的信用分数计算用户的目标信用分数,扩大了信用分数模型的覆盖范围,提高了计算用户的信用分数的准确性的效果。
另外,当存在根据个人信用信息无法计算得到第一次信用分数的样本用户时,将样本用户的第一次关系信用分数确定为样本用户的第一次信用分数,达到了在用户的个人信用分数缺失的情况下也可以根据用户的关系信用分数计算得到用户的信用分数的效果。
请参考图7,其示出了本发明一个实施例提供的信用分数计算装置的结构方框图。该信用分数计算装置可以通过软件、硬件或者两者的结合实现。该信用分数计算装置包括:
信用获取模块720,用于获取各个用户的第一次信用分数和每个用户的第一次关系信用分数,用户的第一次关系信用分数是根据与用户存在社交关系的各个其他用户的第一次信用分数计算得到的。
目标计算模块740,用于对于每个用户,将用户的第一次信用分数和用户的第一次关系信用分数输入训练完毕的信用分数模型,计算得到用户的目标信用分数。
综上所述,本实施例提供的信用分数计算装置,通过获取各个用户的第一
次信用分数和每个用户的第一次关系信用分数;将每个用户的第一次信用分数和第一次关系信用分数输入信用分数模型,计算得到该用户的目标信用分数;解决了当用户个人的信息有所缺失或存在错误时,很难准确的计算出用户的信用分数的问题;达到了通过同时统计用户个人的信用分数和与用户存在社交关系的其他用户的信用分数计算用户的目标信用分数,扩大了信用分数模型的覆盖范围,提高了计算用户的信用分数的准确性的效果。
请参考图8,其示出了本发明另一个实施例提供的信用分数计算装置的结构方框图。该信用分数计算装置可以通过软件、硬件或者两者的结合实现。该信用分数计算装置包括:
信用获取模块820,用于获取各个用户的第一次信用分数和每个用户的第一次关系信用分数,用户的第一次关系信用分数是根据与用户存在社交关系的各个其他用户的第一次信用分数计算得到的。
目标计算模块840,用于对于每个用户,将用户的第一次信用分数和用户的第一次关系信用分数输入训练完毕的信用分数模型,计算得到用户的目标信用分数。
可选的,目标计算模块840,可以包括:待定计算单元841和目标确定单元842。
待定计算单元841,用于将用户的第一次信用分数和用户的第一次关系信用分数输入训练完毕的信用分数模型,计算得到用户的第x次信用分数。
目标确定单元842,用于在用户的第x次信用分数满足预设条件时,将用户的第x次信用分数确定为用户的目标信用分数。
可选的,目标确定单元842可以包括:差值计算单元843、差值检测单元844和信用确定单元845。
差值计算单元843,用于计算用户的第x次信用分数和用户的第x-1次信用分数之间的差值。
差值检测单元844,用于检测差值是否小于第二阈值。
信用确定单元845,用于在差值小于第二阈值时,将用户的第x次信用分数确定为用户的目标信用分数。
可选的,该装置,还可以包括:循环输入模块860。
循环输入模块860,用于在用户的第x次信用分数不满足预设条件时,将
用户的第x次信用分数和第x次关系信用分数再次输入信用分数模型,计算得到用户的第x+1次信用分数;
令x=x+1,再次执行计算各个用户的第x次信用分数和各个用户的第x-1次信用分数之间的差值的步骤。
综上所述,本实施例提供的信用分数计算装置,通过获取各个用户的第一次信用分数和每个用户的第一次关系信用分数;将每个用户的第一次信用分数和第一次关系信用分数输入信用分数模型,计算得到该用户的目标信用分数;解决了当用户个人的信息有所缺失或存在错误时,很难准确的计算出用户的信用分数的问题;达到了通过同时统计用户个人的信用分数和与用户存在社交关系的其他用户的信用分数计算用户的目标信用分数,扩大了信用分数模型的覆盖范围,提高了计算用户的信用分数的准确性的效果。
请参考图9,其示出了本发明一个实施例提供的服务器的结构示意图。所述服务器900包括中央处理单元(CPU)901、包括随机存取存储器(RAM)902和只读存储器(ROM)903的***存储器904,以及连接***存储器904和中央处理单元901的***总线905。所述服务器900还包括帮助计算机内的各个器件之间传输信息的基本输入/输出***(I/O***)906,和用于存储操作***913、应用程序914和其他程序模块915的大容量存储设备907。
所述基本输入/输出***906包括有用于显示信息的显示器908和用于用户输入信息的诸如鼠标、键盘之类的输入设备909。其中所述显示器908和输入设备907都通过连接到***总线905的输入输出控制器910连接到中央处理单元901。所述基本输入/输出***906还可以包括输入输出控制器910以用于接收和处理来自键盘、鼠标、或电子触控笔等多个其他设备的输入。类似地,输入输出控制器910还提供输出到显示屏、打印机或其他类型的输出设备。
所述大容量存储设备907通过连接到***总线905的大容量存储控制器(未示出)连接到中央处理单元901。所述大容量存储设备907及其相关联的计算机可读介质为服务器900提供非易失性存储。也就是说,所述大容量存储设备907可以包括诸如硬盘或者CD-ROM驱动器之类的计算机可读介质(未示出)。
不失一般性,所述计算机可读介质可以包括计算机存储介质和通信介质。计算机存储介质包括以用于存储诸如计算机可读指令、数据结构、程序模块或
其他数据等信息的任何方法或技术实现的易失性和非易失性、可移动和不可移动介质。计算机存储介质包括RAM、ROM、EPROM、EEPROM、闪存或其他固态存储其技术,CD-ROM、DVD或其他光学存储、磁带盒、磁带、磁盘存储或其他磁性存储设备。当然,本领域技术人员可知所述计算机存储介质不局限于上述几种。上述的***存储器904和大容量存储设备907可以统称为存储器。
根据本发明的各种实施例,所述服务器900还可以通过诸如因特网等网络连接到网络上的远程计算机运行。也即服务器900可以通过连接在所述***总线905上的网络接口单元911连接到网络912,或者说,也可以使用网络接口单元911来连接到其他类型的网络或远程计算机***(未示出)。
所述存储器还包括一个或者一个以上的程序,所述一个或者一个以上程序存储于存储器中,所述一个或者一个以上程序用于执行上述实施例提供的信用分数模型训练方法或信用分数计算方法。
在示例性实施例中,还提供了一种包括指令的非临时性计算机可读存储介质,例如包括指令的存储器,上述指令可由移动终端的处理器执行以完成上述信用模型训练方法或者信用分数计算方法。例如,所述非临时性计算机可读存储介质可以是ROM、随机存取存储器(RAM)、CD-ROM、磁带、软盘和光数据存储设备等。
需要说明的是:上述实施例提供的信用分数模型训练的装置在训练信用分数模型时,仅以上述各功能模块的划分进行举例说明,实际应用中,可以根据需要而将上述功能分配由不同的功能模块完成,即将设备的内部结构划分成不同的功能模块,以完成以上描述的全部或者部分功能。另外,上述实施例提供的信用分数模型训练的装置与信用分数模型训练的方法实施例属于同一构思,其具体实现过程详见方法实施例,这里不再赘述。
需要说明的另一点是:上述实施例提供的信用分数计算的装置在计算信用分数模型时,仅以上述各功能模块的划分进行举例说明,实际应用中,可以根据需要而将上述功能分配由不同的功能模块完成,即将设备的内部结构划分成不同的功能模块,以完成以上描述的全部或者部分功能。另外,上述实施例提供的信用分数计算的装置与信用分数计算的方法实施例属于同一构思,其具体
实现过程详见方法实施例,这里不再赘述。
上述本发明实施例序号仅仅为了描述,不代表实施例的优劣。
本领域普通技术人员可以理解实现上述实施例的全部或部分步骤可以通过硬件来完成,也可以通过程序来指令相关的硬件完成,所述的程序可以存储于一种计算机可读存储介质中,上述提到的存储介质可以是只读存储器,磁盘或光盘等。
以上所述仅为本发明的较佳实施例,并不用以限制本发明,凡在本发明的精神和原则之内,所作的任何修改、等同替换、改进等,均应包含在本发明的保护范围之内。
Claims (36)
- 一种信用分数模型训练方法,其特征在于,所述方法包括:获取若干样本用户的违约标注信息;获取每个所述样本用户的第i次信用分数以及每个所述样本用户的第i次关系信用分数,i为起始值为1的正整数;对于每个所述样本用户,根据所述样本用户的第i次信用分数、所述样本用户的第i次关系信用分数和所述样本用户的违约标注信息,训练信用分数模型,并计算得到所述样本用户的第i+1次信用分数;在各个所述样本用户的第i+1次信用分数满足训练结束条件时,得到训练完毕的所述信用分数模型;其中,所述样本用户的第i次关系信用分数是根据与所述样本用户存在社交关系的各个其他用户的第i次信用分数计算得到的;所述样本用户的违约标注信息用于对所述样本用户的分类。
- 根据权利要求1所述的方法,其特征在于,当i=1时,所述获取若干个样本用户的第i次信用分数,包括:对于每个所述样本用户,根据所述样本用户的个人信用信息计算得到所述样本用户的第一次信用分数;或者,当存在根据所述个人信用信息无法计算得到所述第一次信用分数的样本用户时,将所述样本用户的第一次关系信用分数确定为所述样本用户的第一次信用分数,所述第一次关系信用分数是根据与所述样本用户存在社交关系的各个其他用户的第一次信用分数计算得到的。
- 根据权利要求1所述的方法,其特征在于,所述在各个所述样本用户的第i+1次信用分数满足训练结束条件时,得到训练完毕的所述信用分数模型,包括:计算各个所述样本用户的第i+1次信用分数和各个所述样本用户的第i次信用分数之间的相关系数;检测所述相关系数是否大于第一阈值;若所述相关系数大于所述第一阈值,则得到训练完毕的所述信用分数模型。
- 根据权利要求1至3任一所述的方法,其特征在于,所述获取每个所述样本用户的第i次关系信用分数,包括:获取每个所述样本用户的社交关系链,所述社交关系链是所述样本用户与所述其他用户之间存在的社交关系;根据所述社交关系链获取每个所述样本用户对应的各个所述其他用户的第i次信用分数;根据获取到的各个所述其他用户的第i次信用分数计算各个所述样本用户的第i次关系信用分数。
- 根据权利要求4所述的方法,其特征在于,所述根据获取到的各个所述其他用户的第i次信用分数计算各个所述样本用户的第i次关系信用分数,包括:对于每个所述样本用户,计算与所述样本用户对应的各个所述其他用户的第i次信用分数的平均值;将所述平均值确定为所述样本用户的第i次关系信用分数。
- 根据权利要求1至3任一所述的方法,其特征在于,在各个所述样本用户的第i+1次信用分数不满足所述训练结束条件时,令i=i+1,并再次执行所述获取每个所述样本用户的第i次信用分数以及每个所述样本用户的第i次关系信用分数的步骤。
- 根据权利要求1至3任一所述的方法,其特征在于,所述信用分数模型包括逻辑回归分类模型或决策树分类模型。
- 一种信用分数计算方法,其特征在于,应用于如权利要求1至8任一所述的信用分数模型中,所述方法包括:获取各个用户的第一次信用分数和每个所述用户的第一次关系信用分数,所述用户的第一次关系信用分数是根据与所述用户存在社交关系的各个其他用户的第一次信用分数计算得到的;对于每个用户,将所述用户的第一次信用分数和所述用户的第一次关系信用分数输入训练完毕的所述信用分数模型,计算得到所述用户的目标信用分数。
- 根据权利要求9所述的方法,其特征在于,所述将所述用户的第一次信用分数和所述用户的第一次关系信用分数输入训练完毕的所述信用分数模型,计算得到所述用户的目标信用分数,包括:将所述用户的第一次信用分数和所述用户的第一次关系信用分数输入训练完毕的所述信用分数模型,计算得到所述用户的第x次信用分数;在所述用户的第x次信用分数满足预设条件时,将所述用户的第x次信用分数确定为所述用户的目标信用分数。
- 根据权利要求10所述的方法,其特征在于,所述在所述用户的第x次信用分数满足预设条件时,将所述用户的第x次信用分数确定为所述用户的目标信用分数,包括:计算所述用户的第x次信用分数和所述用户的第x-1次信用分数之间的差值;检测所述差值是否小于第二阈值;若所述差值小于所述第二阈值,则将所述用户的第x次信用分数确定为所述用户的目标信用分数。
- 根据权利要求10所述的方法,其特征在于,所述方法还包括:在所述用户的第x次信用分数不满足所述预设条件时,将所述用户的第x次信用分数和所述用户的第x次关系信用分数再次输入所述信用分数模型,计算得到所述用户的第x+1次信用分数;令x=x+1,再次执行所述计算所述用户的第x次信用分数和所述用户的第x-1次信用分数之间的差值的步骤。
- 一种信用分数模型训练装置,其特征在于,所述装置包括:信息获取模块,用于获取若干个样本用户的违约标注信息;所述信息获取模块,还用于获取每个所述样本用户的第i次信用分数以及每个所述样本用户的第i次关系信用分数,i为起始值为1的正整数;信息输入模块,用于对于每个所述样本用户,根据所述样本用户的第i次信用分数、所述样本用户的第i次关系信用分数和所述样本用户的违约标注信息,训练信用分数模型,并计算得到所述样本用户的第i+1次信用分数;模型训练模块,用于在各个所述样本用户的第i+1次信用分数满足训练结束条件时,得到训练完毕的所述信用分数模型;其中,所述样本用户的第i次关系信用分数是根据与所述样本用户存在社交关系的各个其他用户的第i次信用分数计算得到的;所述样本用户的违约标注信息用于对所述样本用户的分类。
- 根据权利要求13所述的装置,其特征在于,当i=1时,所述信息获取模块,包括:第一计算单元,用于对于每个所述样本用户,根据所述样本用户的个人信用信息计算得到所述样本用户的第一次信用分数;或者,第二计算单元,用于当存在根据所述个人信用信息无法计算得到所述第一次信用分数的样本用户时,将所述样本用户的第一次关系信用分数确定为所述样本用户的第一次信用分数,所述第一次关系信用分数是根据与所述样本用户存在社交关系的各个其他用户的第一次信用分数计算得到的。
- 根据权利要求14所述的装置,其特征在于,所述模型训练模块,包括:系数计算单元,用于计算各个所述样本用户的第i+1次信用分数和各个所述样本用户的第i次信用分数之间的相关系数;系数检测单元,用于检测所述相关系数是否大于第一阈值;训练完毕单元,用于若所述相关系数大于所述第一阈值,则得到训练完毕的所述信用分数模型。
- 根据权利要求13至15任一所述的装置,其特征在于,所述信息获取模块,还包括:社交获取单元,用于获取每个所述样本用户的社交关系链,所述社交关系链是所述样本用户与所述其他用户之间存在的社交关系;分数获取单元,用于根据所述社交关系链获取每个所述样本用户对应的各个所述其他用户的第i次信用分数;关系计算单元,用于根据获取到的各个所述其他用户的第i次信用分数计算各个所述样本用户的第i次关系信用分数。
- 根据权利要求16所述的装置,其特征在于,所述关系计算单元,包括:第三计算单元,用于对于每个所述样本用户,计算与所述样本用户对应的各个所述其他用户的第i次信用分数的平均值;分数确定单元,用于将所述平均值确定为所述样本用户的第i次关系信用分数。
- 根据权利要求13至15任一所述的装置,其特征在于,所述装置还包括用于在各个所述样本用户的第i+1次信用分数不满足所述训练结束条件时,令i=i+1,并再次执行所述获取每个所述样本用户的第i次信用分数以及每个所述样本用户的第i次关系信用分数的步骤的模块。
- 根据权利要求13至15任一所述的装置,其特征在于,所述信用分数模型包括逻辑回归分类模型或决策树分类模型。
- 一种信用分数计算装置,其特征在于,应用于如权利要求13至20任一所述的信用分数模型中,所述装置包括:信用获取模块,用于获取各个用户的第一次信用分数和每个所述用户的第一次关系信用分数,所述用户的第一次关系信用分数是根据与所述用户存在社交关系的各个其他用户的第一次信用分数计算得到的;目标计算模块,用于对于每个用户,将所述用户的第一次信用分数和所述用户的第一次关系信用分数输入训练完毕的所述信用分数模型,计算得到所述用户的目标信用分数。
- 根据权利要求21所述的装置,其特征在于,所述目标计算模块,包括:待定计算单元,用于将所述用户的第一次信用分数和所述用户的第一次关系信用分数输入训练完毕的所述信用分数模型,计算得到所述用户的第x次信用分数;目标确定单元,用于在所述用户的第x次信用分数满足预设条件时,将所述用户的第x次信用分数确定为所述用户的目标信用分数。
- 根据权利要求22所述的装置,其特征在于,所述目标确定单元,包括:差值计算单元,用于计算所述用户的第x次信用分数和所述用户的第x-1次信用分数之间的差值;差值检测单元,用于检测所述差值是否小于第二阈值;信用确定单元,用于在所述差值小于所述第二阈值时,将所述用户的第x次信用分数确定为所述用户的目标信用分数。
- 根据权利要求22所述的装置,其特征在于,所述装置,还包括:循环输入模块,用于在所述用户的第x次信用分数不满足所述预设条件时,将所述用户的第x次信用分数和所述用户的第x次关系信用分数再次输入所述信用分数模型,计算得到所述用户的第x+1次信用分数;令x=x+1,再次执行所述计算所述用户的第x次信用分数和所述用户的第x-1次信用分数之间的差值的步骤。
- 一种信用分数模型训练服务器,其特征在于,所述服务器包括:一个或多个处理器;和存储器;所述存储器存储有一个或多个程序,所述一个或多个程序被配置成由所述一个或多个处理器执行,所述一个或多个程序包含用于进行以下操作的指令:获取若干样本用户的违约标注信息;获取每个所述样本用户的第i次信用分数以及每个所述样本用户的第i次关系信用分数,i为起始值为1的正整数;对于每个所述样本用户,根据所述样本用户的第i次信用分数、所述样本用户的第i次关系信用分数和所述样本用户的违约标注信息,训练信用分数模型,并计算得到所述样本用户的第i+1次信用分数;在各个所述样本用户的第i+1次信用分数满足训练结束条件时,得到训练完毕的所述信用分数模型;其中,所述样本用户的第i次关系信用分数是根据与所述样本用户存在社交关系的各个其他用户的第i次信用分数计算得到的;所述样本用户的违约标注信息用于对所述样本用户的分类。
- 根据权利要求25所述的服务器,其特征在于,当i=1时,所述一个或多个程序还包含用于进行以下操作的指令:对于每个所述样本用户,根据所述样本用户的个人信用信息计算得到所述样本用户的第一次信用分数;或者,当存在根据所述个人信用信息无法计算得到所述第一次信用分数的样本用户时,将所述样本用户的第一次关系信用分数确定为所述样本用户的第一次信用分数,所述第一次关系信用分数是根据与所述样本用户存在社交关系的各个其他用户的第一次信用分数计算得到的。
- 根据权利要求25所述的服务器,其特征在于,所述一个或多个程序还包含用于进行以下操作的指令:计算各个所述样本用户的第i+1次信用分数和各个所述样本用户的第i次信用分数之间的相关系数;检测所述相关系数是否大于第一阈值;若所述相关系数大于所述第一阈值,则得到训练完毕的所述信用分数模型。
- 根据权利要求25至27任一所述的服务器,其特征在于,所述一个或多个程序还包含用于进行以下操作的指令:获取每个所述样本用户的社交关系链,所述社交关系链是所述样本用户与所述其他用户之间存在的社交关系;根据所述社交关系链获取每个所述样本用户对应的各个所述其他用户的第i次信用分数;根据获取到的各个所述其他用户的第i次信用分数计算各个所述样本用户的第i次关系信用分数。
- 根据权利要求28所述的服务器,其特征在于,所述一个或多个程序还包含用于进行以下操作的指令:对于每个所述样本用户,计算与所述样本用户对应的各个所述其他用户的第i次信用分数的平均值;将所述平均值确定为所述样本用户的第i次关系信用分数。
- 根据权利要求25至27任一所述的服务器,其特征在于,所述一个或多个程序还包含用于进行以下操作的指令:在各个所述样本用户的第i+1次信用分数不满足所述训练结束条件时,令i=i+1,并再次执行所述获取每个所述样本用户的第i次信用分数以及每个所述样本用户的第i次关系信用分数的步骤。
- 根据权利要求25至27任一所述的服务器,其特征在于,所述信用分数模型包括逻辑回归分类模型或决策树分类模型。
- 一种信用分数计算服务器,其特征在于,应用于如权利要求25至32任一所述的信用分数模型中,所述服务器包括:一个或多个处理器;和存储器;所述存储器存储有一个或多个程序,所述一个或多个程序被配置成由所述一个或多个处理器执行,所述一个或多个程序包含用于进行以下操作的指令:获取各个用户的第一次信用分数和每个所述用户的第一次关系信用分数,所述用户的第一次关系信用分数是根据与所述用户存在社交关系的各个其他用户的第一次信用分数计算得到的;对于每个用户,将所述用户的第一次信用分数和所述用户的第一次关系信用分数输入训练完毕的所述信用分数模型,计算得到所述用户的目标信用分数。
- 根据权利要求33所述的服务器,其特征在于,所述一个或多个程序还包含用于进行以下操作的指令:将所述用户的第一次信用分数和所述用户的第一次关系信用分数输入训练完毕的所述信用分数模型,计算得到所述用户的第x次信用分数;在所述用户的第x次信用分数满足预设条件时,将所述用户的第x次信用分数确定为所述用户的目标信用分数。
- 根据权利要求34所述的服务器,其特征在于,所述一个或多个程序还包含用于进行以下操作的指令:计算所述用户的第x次信用分数和所述用户的第x-1次信用分数之间的差值;检测所述差值是否小于第二阈值;若所述差值小于所述第二阈值,则将所述用户的第x次信用分数确定为所述用户的目标信用分数。
- 根据权利要求34所述的服务器,其特征在于,所述一个或多个程序还包含用于进行以下操作的指令:在所述用户的第x次信用分数不满足所述预设条件时,将所述用户的第x次信用分数和所述用户的第x次关系信用分数再次输入所述信用分数模型,计算得到所述用户的第x+1次信用分数;令x=x+1,再次执行所述计算所述用户的第x次信用分数和所述用户的第x-1次信用分数之间的差值的步骤。
Priority Applications (4)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
KR1020187007645A KR102038237B1 (ko) | 2016-03-15 | 2017-02-28 | 신용점수 모델 훈련 방법,신용 점수 계산 방법,장치 및 서버 |
EP17765707.9A EP3349171A4 (en) | 2016-03-15 | 2017-02-28 | Credit-score model training method, and credit-score calculation method, device, and server |
JP2018535228A JP6595718B2 (ja) | 2016-03-15 | 2017-02-28 | クレジット点数モデルトレーニング方法、クレジット点数計算方法、装置及びサーバー |
US15/954,378 US11915311B2 (en) | 2016-03-15 | 2018-04-16 | User score model training and calculation |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610145843.X | 2016-03-15 | ||
CN201610145843.XA CN107194795A (zh) | 2016-03-15 | 2016-03-15 | 信用分数模型训练方法、信用分数计算方法及装置 |
Related Child Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
US15/954,378 Continuation US11915311B2 (en) | 2016-03-15 | 2018-04-16 | User score model training and calculation |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2017157165A1 true WO2017157165A1 (zh) | 2017-09-21 |
Family
ID=59851922
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/CN2017/075131 WO2017157165A1 (zh) | 2016-03-15 | 2017-02-28 | 信用分数模型训练方法、信用分数计算方法、装置及服务器 |
Country Status (6)
Country | Link |
---|---|
US (1) | US11915311B2 (zh) |
EP (1) | EP3349171A4 (zh) |
JP (1) | JP6595718B2 (zh) |
KR (1) | KR102038237B1 (zh) |
CN (1) | CN107194795A (zh) |
WO (1) | WO2017157165A1 (zh) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20190108585A1 (en) * | 2017-10-11 | 2019-04-11 | Mx Technologies, Inc. | Aggregation based credit decision |
Families Citing this family (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107862339B (zh) * | 2017-11-15 | 2022-04-29 | 百度在线网络技术(北京)有限公司 | 用于输出信息的方法和装置 |
CN108898476A (zh) * | 2018-06-14 | 2018-11-27 | 中国银行股份有限公司 | 一种贷款客户信用评分方法和装置 |
CN109461043A (zh) * | 2018-09-18 | 2019-03-12 | 深圳壹账通智能科技有限公司 | 产品推送方法、装置、计算机设备和存储介质 |
KR102400804B1 (ko) * | 2019-09-05 | 2022-05-24 | 고려대학교 산학협력단 | 기계학습 기반 신용평가 분류 모델의 평가 이유를 설명하는 장치 및 방법 |
CN110599329A (zh) * | 2019-09-09 | 2019-12-20 | 腾讯科技(深圳)有限公司 | 信用评估方法、信用评估装置及电子设备 |
CN112017062B (zh) * | 2020-07-15 | 2024-06-07 | 北京淇瑀信息科技有限公司 | 基于客群细分的资源额度分配方法、装置及电子设备 |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20020138333A1 (en) * | 2001-03-22 | 2002-09-26 | Decotiis Allen R. | System, method and article of manufacture for a weighted model to conduct propensity studies |
CN103514566A (zh) * | 2013-10-15 | 2014-01-15 | 国家电网公司 | 一种风险控制***及方法 |
CN104463603A (zh) * | 2014-12-05 | 2015-03-25 | 中国联合网络通信集团有限公司 | 一种信用评估方法及*** |
CN104574110A (zh) * | 2014-12-29 | 2015-04-29 | 张伟平 | 一种信用数字认证方法 |
Family Cites Families (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2002109208A (ja) * | 2000-09-29 | 2002-04-12 | Acom Co Ltd | 信用リスク管理方法、分析モデル決定方法、分析サーバ及び分析モデル決定装置 |
JP2002123649A (ja) * | 2000-10-13 | 2002-04-26 | Nippon Telegr & Teleph Corp <Ntt> | 信用調査装置、信用調査方法及び信用調査装置における信用調査プログラムを記録した記録媒体 |
US8302164B2 (en) * | 2004-07-22 | 2012-10-30 | Facebook, Inc. | Authorization and authentication based on an individual's social network |
JP5071475B2 (ja) * | 2007-03-27 | 2012-11-14 | 富士通株式会社 | 重回帰分析による予測モデルの作成方法、作成装置、作成プログラム |
US8694401B2 (en) * | 2011-01-13 | 2014-04-08 | Lenddo, Limited | Systems and methods for using online social footprint for affecting lending performance and credit scoring |
US20140081832A1 (en) * | 2012-09-18 | 2014-03-20 | Douglas Merrill | System and method for building and validating a credit scoring function |
KR20130065778A (ko) * | 2011-11-29 | 2013-06-20 | 코리아크레딧뷰로 (주) | 신용정보 관리 방법 및 시스템 |
WO2013126713A1 (en) * | 2012-02-22 | 2013-08-29 | Pave, Inc. | Apparatuses, methods and systems for facilitating communities of social based investment |
JP2014071532A (ja) * | 2012-09-27 | 2014-04-21 | Mycredit Kk | 個人信用情報提供装置 |
US20160267587A1 (en) * | 2015-03-12 | 2016-09-15 | Backed Inc. | Systems and methods for online guarantorship of loans |
US10387939B2 (en) * | 2015-08-21 | 2019-08-20 | Q2 Software, Inc. | Method, medium, and system for surfacing recommendations |
-
2016
- 2016-03-15 CN CN201610145843.XA patent/CN107194795A/zh active Pending
-
2017
- 2017-02-28 JP JP2018535228A patent/JP6595718B2/ja active Active
- 2017-02-28 WO PCT/CN2017/075131 patent/WO2017157165A1/zh active Application Filing
- 2017-02-28 KR KR1020187007645A patent/KR102038237B1/ko active IP Right Grant
- 2017-02-28 EP EP17765707.9A patent/EP3349171A4/en not_active Ceased
-
2018
- 2018-04-16 US US15/954,378 patent/US11915311B2/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20020138333A1 (en) * | 2001-03-22 | 2002-09-26 | Decotiis Allen R. | System, method and article of manufacture for a weighted model to conduct propensity studies |
CN103514566A (zh) * | 2013-10-15 | 2014-01-15 | 国家电网公司 | 一种风险控制***及方法 |
CN104463603A (zh) * | 2014-12-05 | 2015-03-25 | 中国联合网络通信集团有限公司 | 一种信用评估方法及*** |
CN104574110A (zh) * | 2014-12-29 | 2015-04-29 | 张伟平 | 一种信用数字认证方法 |
Non-Patent Citations (1)
Title |
---|
See also references of EP3349171A4 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20190108585A1 (en) * | 2017-10-11 | 2019-04-11 | Mx Technologies, Inc. | Aggregation based credit decision |
US11823258B2 (en) * | 2017-10-11 | 2023-11-21 | Mx Technologies, Inc. | Aggregation based credit decision |
Also Published As
Publication number | Publication date |
---|---|
KR102038237B1 (ko) | 2019-10-29 |
EP3349171A4 (en) | 2018-10-17 |
JP2018530093A (ja) | 2018-10-11 |
EP3349171A1 (en) | 2018-07-18 |
US20180232665A1 (en) | 2018-08-16 |
CN107194795A (zh) | 2017-09-22 |
JP6595718B2 (ja) | 2019-10-23 |
KR20180041727A (ko) | 2018-04-24 |
US11915311B2 (en) | 2024-02-27 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2017157165A1 (zh) | 信用分数模型训练方法、信用分数计算方法、装置及服务器 | |
TWI712981B (zh) | 風險辨識模型訓練方法、裝置及伺服器 | |
Paun et al. | Comparing Bayesian models of annotation | |
US11259718B1 (en) | Systems and methods for automated body mass index calculation to determine value | |
WO2020215571A1 (zh) | 一种识别敏感数据的方法、装置、存储介质及计算机设备 | |
US20170018030A1 (en) | System and Method for Determining Credit Worthiness of a User | |
WO2020082734A1 (zh) | 文本情感识别方法、装置、电子设备及计算机非易失性可读存储介质 | |
Wang et al. | Cost-effective quality assurance in crowd labeling | |
WO2021143478A1 (zh) | 识别对抗样本以保护模型安全的方法及装置 | |
Baba et al. | Leveraging non-expert crowdsourcing workers for improper task detection in crowdsourcing marketplaces | |
US11042880B1 (en) | Authenticating users in the presence of small transaction volumes | |
US20170046346A1 (en) | Method and System for Characterizing a User's Reputation | |
WO2023236538A1 (zh) | 风险代码事前检测方法及装置、电子设备、计算机可读存储介质和计算机程序产品 | |
Matsui et al. | Crowdordering | |
KR102195629B1 (ko) | 인공지능 학습데이터 생성을 위한 크라우드소싱 기반 프로젝트에서의 작업 능력 지수에 기반한 작업자 선정 방법 | |
CN111598122B (zh) | 数据校验方法、装置、电子设备和存储介质 | |
Chen et al. | Fairness Improvement with Multiple Protected Attributes: How Far Are We? | |
CN108985755A (zh) | 一种账号状态识别方法、装置及服务器 | |
Li et al. | Using candlestick charts to predict adolescent stress trend on micro-blog | |
US20220327450A1 (en) | Method for increasing or decreasing number of workers and inspectors in crowdsourcing-based project for creating artificial intelligence learning data | |
CN113222609B (zh) | 风险识别方法和装置 | |
CN110008972A (zh) | 用于数据增强的方法和装置 | |
US11574150B1 (en) | Data interpretation analysis | |
KR102244699B1 (ko) | 인공지능 학습데이터 생성을 위한 크라우드소싱 기반 프로젝트의 문장 유사도를 이용한 감정 라벨링 방법 | |
CN110362981B (zh) | 基于可信设备指纹判断异常行为的方法及*** |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
ENP | Entry into the national phase |
Ref document number: 20187007645 Country of ref document: KR Kind code of ref document: A |
|
WWE | Wipo information: entry into national phase |
Ref document number: 2018535228 Country of ref document: JP |
|
NENP | Non-entry into the national phase |
Ref country code: DE |