CN111275541A

CN111275541A - Borrower quality evaluation method and system based on multi-dimensional information, electronic device and computer readable storage medium

Info

Publication number: CN111275541A
Application number: CN202010037649.6A
Authority: CN
Inventors: 高星
Original assignee: CITIC Aibank Corp Ltd
Current assignee: CITIC Aibank Corp Ltd
Priority date: 2020-01-14
Filing date: 2020-01-14
Publication date: 2020-06-12

Abstract

The invention discloses a method and a system for evaluating the quality of a borrower based on multi-dimensional information, electronic equipment and a computer readable storage medium, wherein the quality evaluation method comprises the following steps: acquiring relevant information of the borrower, wherein the relevant information comprises at least one of buried point log information, public opinion information, service information, the borrower and information of an associated party of the borrower; determining index data corresponding to the related information of the borrower respectively; and determining quality evaluation information of the borrower through a random forest algorithm or a support vector machine model based on the determined index data corresponding to the relevant information of the borrower. The method utilizes the behavioral expression data of each borrower in the deposit and management bank on various platforms to carry out multi-dimensional analysis, and solves the problems that the data analysis dimension is insufficient, the analysis of a related party is insufficient, and a model is not effective in the quality analysis of the borrowers on the online loan platform, so that risk control is carried out, and early warning is carried out in advance.

Description

Borrower quality evaluation method and system based on multi-dimensional information, electronic device and computer readable storage medium

Technical Field

The invention relates to the technical field of financial information, in particular to a method and a system for evaluating the quality of a borrower based on multi-dimensional information, electronic equipment and a computer-readable storage medium.

Background

At present, financial supervision and technology is in the initial development stage, and at present, a big data technology, a commercial intelligence technology, an artificial intelligence technology and a block chain technology are mainly applied to complete data integration processing and deep mining on various types of data, so that asset quality, user behavior tracking and fraud occurrence analysis are realized. Aiming at user quality analysis, various user portrait technologies are mostly used for analysis, and only classification algorithms are used for analyzing various labels.

The existing technology aims at the quality analysis of borrowers, only can be limited to the behavior of the borrowers on a certain platform for analysis, and the dimensionality of data which can be analyzed is very small because the data source is narrow in width and information cannot be shared among the platforms. The dimensions of the excavation are not deep enough when analyzing the user quality. Most platforms cannot perform periodic behavior analysis on the user and cannot analyze changes of the user in social association.

The technology utilizes the behavior of borrowers on each network credit platform and utilizes the existing data resources to carry out effective index analysis, and can carry out effective early warning aiming at suspicious behaviors of customers. The behavior, and hence the performance, of the party associated with the user is analyzed for its adverse or beneficial effects on the user itself. A model of the user's quality is established and automatically optimized, thereby effectively estimating the user's quality. The risk prevention and control capability of the deposit and management bank and the online loan platform is improved.

Disclosure of Invention

The invention aims to provide a borrower quality evaluation method, a borrower quality evaluation system, electronic equipment and a computer-readable storage medium based on multi-dimensional information, which utilize behavior performance data of each borrower in a deposit bank on various platforms to carry out multi-dimensional analysis and solve the problems of insufficient data analysis dimensionality, insufficient analysis of related parties and insufficient effectiveness of models in the borrower quality analysis of a network loan platform, thereby controlling risks and early warning in advance.

In order to achieve the purpose, the invention provides a borrower quality evaluation method based on multi-dimensional information, which comprises the following steps: acquiring relevant information of the borrower, wherein the relevant information comprises at least one of buried point log information, public opinion information, service information, the borrower and information of an associated party of the borrower; determining index data corresponding to the related information of the borrower respectively; and determining quality evaluation information of the borrower through a random forest algorithm or a support vector machine model based on the determined index data corresponding to the relevant information of the borrower.

Optionally or preferably, the step of determining the index data corresponding to the relevant information of the borrower respectively includes at least one of the following: when the related information is the buried point log information, analyzing the behavior of a client, counting various operations of the user and analyzing the operation flow of the user by deploying a logic tracking code, and providing a data source for emotion analysis of a specific user; when the relevant information is public opinion information, relevant parameters or labels of specific borrowers are output by analyzing the relevant public opinions of each network loan platform and the overall situation of the industry; when the relevant information is the service information, analyzing the period of each behavior of the borrower according to the service data, and providing a core modeling index; when the related information is the information of the borrower and the related party thereof, the natural semantic analysis is carried out by retrieving the important change of the borrower and the related party thereof and the recently announced retrieval content, and the related label of the quality analysis of the borrower is supplemented.

Optionally or preferably, the screening further comprises index data before determining the quality evaluation information of the lender.

Alternatively or preferably, it is determined whether to credit the lender based on the determined quality assessment information of the lender.

Optionally or preferably, when the borrower and the related party are analyzed, the emotion analysis is performed on various information, and the information is converted into an index of emotion preference of the borrower.

Optionally or preferably, when analyzing the service data, contents of charging, withdrawal, borrowing, repayment, opening, flow mark and closing mark, which contain the amount information, are converted into the amount indexes, various kinds of running records of the user are converted into the behavior related indexes, and the user authorization information is used as the enumeration indexes.

Optionally or preferably, when analyzing the related public opinion information, the related state and activity of the borrower are analyzed by counting the times of clicking various buttons and accessing urls by the borrower on various pages and the input information, and then the related state and activity are converted into various indexes related to behaviors, emotions and preferences.

The invention also provides a system for analyzing the quality of the network loan borrowers in a multi-dimensional way, which comprises the following steps: the embedded point log system analyzes the behaviors of the clients, counts various operations of the users and analyzes the operation flow of the users by deploying logic tracking codes, and provides a data source for emotion analysis of specific users; the public opinion analysis system outputs related parameters or specific borrower labels by analyzing the related public opinions of each network credit platform and the overall situation of the industry; the business system analyzes the period of each behavior of the borrower according to the business data and provides core modeling indexes; the system for analyzing the borrower and the related party analyzes natural semantics by retrieving the great change of the borrower and the related party and recently noticing retrieval contents, and supplements related labels for quality analysis of the borrower. And a decision support system for determining whether to loan the borrower based on the determined quality evaluation information of the borrower.

The invention also provides an electronic device comprising a processor and a memory; a memory for storing operating instructions; and the processor is used for executing the borrower quality evaluation method based on the multi-dimensional information by calling the operation instruction.

The present invention also provides a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements the above-described method for assessing the quality of a borrower based on multidimensional information.

The technical scheme provided by the embodiment of the invention has the following beneficial effects: by deploying and implementing the method or the system, operators can be helped to identify the customers with high net value, the users who are protected against default risks play an effective role, the customers with high net value are converted into the customers of other products in the row through a series of measures, and benefits are brought to the row. The positive effect is caused by the fact that the dimension of the analysis network loan borrower is wide, the life cycle of data is long, the value of a customer can be accurately presented, and at least more than 3000 thousands of procedure expenses and interest income are brought to a row. Meanwhile, the loss caused by bad customers is avoided to be about 2000 ten thousand yuan, and the comprehensive income reaches 5000 ten thousand yuan. The income evaluation is mainly carried out through specific transaction behaviors of recharging, cash withdrawal, repayment and the like of the online credit platform. In addition, the method saves a large amount of manpower (about 20 people, the manpower cost is about 500 thousands), operators do not need to consume guide energy to estimate various data, the method provides intermediate result data for the borrower to score, the intermediate result data is also an index directly needed by the operators, and the operators can conveniently study and judge various trends.

Drawings

FIG. 1 is a flow chart of a multidimensional analysis network loan borrower quality provided by an embodiment of the invention;

fig. 2 is another flowchart of the method for evaluating the quality of a borrower based on multi-dimensional information according to the embodiment of the present invention;

fig. 3 is a partial flowchart of a method for evaluating the quality of a borrower based on multi-dimensional information according to an embodiment of the present invention;

fig. 4 is a schematic diagram of the relationship between borrowers in the method for evaluating the quality of a borrower based on multi-dimensional information according to the embodiment of the present invention;

fig. 5 is a schematic diagram illustrating the generation of negative indexes and positive indexes of all levels of associated parties in the method for evaluating the quality of a borrower based on multi-dimensional information according to the embodiment of the present invention;

fig. 6 is a block diagram of a system for analyzing the quality of a lender using a multidimensional network according to an embodiment of the invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, embodiments of the present invention will be described in detail with reference to the accompanying drawings. It is to be understood that the described embodiments are merely illustrative or exemplary in nature and are in no way intended to limit the invention, its application, or uses. In the following detailed description, numerous specific details are set forth in order to provide a thorough understanding of the present invention. It will be apparent, however, to one skilled in the art that the present invention may be practiced without some or all of these specific details. The following description of the embodiments is merely intended to provide a better understanding of the present invention by presenting examples of the invention. The present invention is in no way limited to any specific configuration and algorithm set forth below, but rather covers any modification, replacement or improvement of elements, components or algorithms without departing from the spirit of the invention.

The present invention will be described in further detail with reference to specific examples, but the embodiments of the present invention are not limited thereto.

As shown in fig. 1, the method for evaluating the quality of a borrower based on multi-dimensional information according to the present embodiment includes the following steps:

step 110, acquiring relevant information of the borrower, wherein the relevant information comprises at least one of buried point log information, public opinion information, service information, the borrower and information of an associated party of the borrower;

step 120, determining index data corresponding to the relevant information of the borrower respectively;

and step 130, determining quality evaluation information of the borrower through a random forest algorithm or a support vector machine model based on the determined index data corresponding to the relevant information of the borrower.

As shown in fig. 6, fig. 6 is a diagram illustrating a software and hardware deployment architecture of the system, and the present invention is mainly implemented by deploying various systems in the diagram.

The center of fig. 6 is a network loan borrower analysis system, specifically, the corresponding processing flow is fig. 2, data modeling is mainly performed through various indexes provided by various other systems, and the work of scoring a borrower user is completed. The random forest and the support vector machine are common classification algorithms in the field of data mining, the classification result is that the probability of the sample approaching to a good sample, for example, the probability is 0.1243, can be converted into 124.3 points, and then according to the actual experience value, the standard of the good-and-bad score is set, for example, 550 points, the sample with the score less than 550 points is a bad sample, and the sample with the score more than or equal to 550 points is a good sample. The probability value is typically multiplied by 1000 to obtain a specific score, and the sum grid of good and bad samples is set 550.

The embedded point log system is mainly used for deploying codes for logic tracking through related WeChat small programs, mobile phone applications and H5 pages of an inline memory management system, wherein the codes are mainly used for analyzing behaviors of customers, counting various operations of users, analyzing operation processes of the users and providing data sources for emotion analysis of specific users.

The public opinion analysis system mainly analyzes the relevant public opinions of each network loan platform and the overall situation of the industry, and the relevant public opinions of the borrowers mainly analyze various messages of the borrowers in the form of relevant legal persons and output relevant parameters or labels of specific borrowers to the network loan borrower analysis system.

An analysis system for a borrower and an associated party thereof, as shown in fig. 4, mainly aims at retrieval at a web end, major change of the borrower and the associated party thereof, recent announcement, analysis of natural semantics, analysis of a factor of interest or interest of related retrieval contents, and supplement of related tags of quality analysis of the borrower.

The external system is mainly a core system, an accounting management system, a fund storage management system and related business data of a payment system in a row, and is used as a core modeling index for analyzing actions of a borrower client such as recharging, cash withdrawal, borrowing and repayment, analyzing the period of each action and providing the core modeling index.

The decision support system mainly utilizes the network loan borrower analysis system to obtain the indexes of user quality, platform quality and the like, combines the current public opinion indexes to finish the judgment and estimation of the whole risk and support the decision of future behavior.

As shown in fig. 2, fig. 2 is a main flow chart of the network lender analysis system in fig. 6:

in step 101, emotion analysis is performed on various types of information in an analysis system for a borrower and an associated party thereof, and the information is converted into an index of emotion preference of the borrower.

In step 102, various transaction systems in the bank, such as fund deposit management, payment systems, core systems, financial management systems, various types of recharging, withdrawal, borrowing, repayment, opening, flow marks, closing marks and the like, which contain the content of the amount information are converted into the indexes of the amount types, various types of flow records of the user are converted into the relevant indexes of the behaviors, and in addition, the information of user authorization and the like is used as the indexes of the enumeration types.

In step 103, a point-embedded log system is utilized to count the times of clicking various buttons and accessing url by the borrower on each page, and the input information is counted and used for analyzing the related state and activity degree of the borrower, so as to be converted into various indexes related to behavior, emotion and preference.

In step 104, the feature engineering module analyzes the indexes according to the index data sent by each subsystem, completes the screening of the indexes according to the result of the training set, eliminates the improper old indexes and adds the good new indexes to the indexes such as the IV value, the KS value and the like of the indexes and other effects.

In step 105, the quality 'good or bad' identification is carried out on the borrower by using the screened indexes aiming at the support vector machine and a random classification algorithm, and the quality is converted into the quality score of the borrower and the platform quality score according to the obtained probability value.

As shown in fig. 3, fig. 3 shows a detailed flow chart of step 104 in fig. 2:

the analyzed indexes mainly comprise continuous indexes and category indexes, the indexes are specific numerical values of certain data, and recharging times can be obtained by counting recharging details in a database; target information is registered through a database to obtain indexes such as the borrowed amount of an investor, the interest rate value and the like; through the record in the log, the behavior of the user operation can be known, and the times of user query and the times of system login can be known through program statistics. By inquiring the database and counting the log information, various indexes can be obtained, and whether the indexes are used or not needs to be processed in a subsequent process. The continuous indexes comprise indexes of money amount classes, indexes of interest rate classes and the like, the distribution of the indexes on the range is continuous, the possible numerical values are infinite, and the method performs characteristic binning processing on the numerical values. The so-called feature binning is discretization of continuous variables, the discretized features are more stable, and the risk of overfitting of the model is reduced.

Indexes of operation behavior classes, such as login times, inquiry credit times and other indexes; and the indexes of the business behavior class such as the number of recharging, the number of withdrawal, the number of lending, the number of the streaming marks and the like are discretized, so that the WOE value and the IV value are directly calculated in the subsequent steps without special processing.

For the indexes of the enumeration types, such as the client type, the transaction authorization type and the like, if the enumeration types search a certain threshold (10 is suggested in the method), the base reduction processing is carried out, and the enumeration types with similar business meanings are merged, such as the bidding authority, the automatic bidding authority and the like.

And after all indexes are pretreated, screening the indexes by using a scoring card model.

The scorecard model is evaluated primarily using the WOE value, which is a form of encoding of the original independent variables, as well as the IV value. To perform WOE encoding on a variable, it is necessary to first perform grouping/discretization on the variable (equal width cutting, equal height cutting, or cutting using a decision tree). WOE represents the difference between "bad clients in the current packet to all bad clients" and "good clients in the current packet to all bad clients". WOE can be understood as: the ratio of bad customers to good customers in the current group, and the difference in this ratio across all samples. The difference is expressed logarithmically as the ratio of the two ratios. The larger the WOE, the greater the difference, the more likely the bad samples of the sample in the packet are, and the smaller the WOE, the smaller the difference, the less likely the bad samples of the packet are.

The IV is called Information Value, and Chinese means Information Value or Information amount. V is a prediction capability used to measure the independent variable. Similar indicators are information gain, kini coefficient, etc. The calculation of IV is based on WOE and can be seen as a weighted sum of WOE pairs.

The WOE values and the IV values of all the indexes are calculated, the average value avg _ WOE of the absolute values of the WOE values is calculated, and the average value avg _ IV of the IV values is calculated. Features with absolute values less than 2 avg _ woe are picked, and then of these feature values, features with avg _ iv values less than avg _ iv/2 are picked. Features that do not meet the criteria are discarded.

The results of the binning process for checking continuity features are then performed. The binned data is discarded if a bin is found to exceed 90% for a particular index. And finally, checking whether the box separation result and the bad sample rate are in a monotonous relation, and if not, combining the two boxes with the bad sample rates close to each other.

The system for analyzing the borrower and the associated parties in fig. 6 provides various features by analyzing recent related conditions of the borrower, the enterprise and the organization associated with the borrower, and outputs the features to the online loan borrower analysis system, wherein the main indexes include a positive change index of a first-level associated party, a negative change index of the first-level associated party, a positive notice index of the first-level associated party, a negative notice index of the first-level associated party, a positive change index of a second-level associated party, a negative change index of the second-level associated party, a positive notice index of the second-level associated party, a negative notice index of the second-level associated party, a positive change index of a third-level associated party, a negative change index of the third-level associated party, a positive notice index of the third-level associated party, and a negative notice index of the third-level associated party.

As shown in fig. 4, the related parties of the borrower who is a legal person mainly include the first ten stockholders, the related parent company, the subsidiary company, other subsidiary companies of the parent company, the important investors, the affiliated enterprises, the pool enterprises, and the like. The second level of relatedness refers to the self-relatedness of the first level of relatedness, such as a subsidiary of the borrower. The third level of affiliation refers to an affiliate of the second level of affiliation, such as a subsidiary of a borrower's subsidiary.

And scoring each index of all related parties by using a web crawler or an information database of a third party and combining an emotion analysis system, wherein the score interval is between 0 and 100. The specific process is shown in FIG. 5.

Fig. 5 is a process of generating negative indexes and positive indexes of all levels of related parties, in which various types of information of all levels of related parties are searched for analysis, and negative change indexes of a first level of related parties, positive announcement indexes of the first level of related parties, negative announcement indexes of the first level of related parties, positive change indexes of a second level of related parties, negative change indexes of the second level of related parties, positive announcement indexes of the second level of related parties, negative announcement indexes of the second level of related parties, positive change indexes of the third level of related parties, negative change indexes of the third level of related parties, positive announcement indexes of the third level of related parties, and negative announcement indexes of the third level of related parties are obtained after the analysis.

1. Retrieving information of all first-level associated parties of the relevant borrower through a database of enterprise information (which may be obtained from a third-party database or a third-party API)

2. Searching out the information of the second level related party by the related party information of the first level through the database of the enterprise information

3. Searching out the information of the third season related party by using the related party information of the second level through the database of the enterprise information

4. Obtaining related company bulletin and changed information through information database (which may be third party database)

5. Extracting information such as bulletin, change and the like of company on internet through webpage retrieval tool

6. All announcements and altered information are analyzed for emotion (either semantic emotion analysis using a third party interface or self modeling)

7. And calculating an average of all 30-day announcements of a certain related party and the index of the changed emotion analysis, and storing the average in a database as a characteristic value.

It should be noted that:

when the borrower is a natural person, the characteristic analysis work is mainly completed by knowing the conditions of the company and the industry where the borrower is located, and related characteristics are the same as those of a legal person and are not repeated.

The embedded point log system completes the statistics of user behaviors mainly by deploying recording codes on each webpage of the system, and the system is common and is not repeated.

The public opinion analysis system mainly carries out semantic emotion analysis by retrieving information of all main portal websites to finish the final output of public opinion emotion characteristics, and related systems are mature and are not repeated.

The external data is mainly the companies for data analysis, and various indexes provided by the external data can help for analysis.

In the scheme of this embodiment, the embedded point log system, the public opinion analysis system, the borrower and the analysis system of the related party thereof, and the external systems are all mutually independent and equal systems, and the operation between each other has no precedence.

Unless defined otherwise, technical or scientific terms used herein shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention belongs. The use of "first," "second," and similar terms in the description and claims of this patent does not denote any order, quantity, or importance, but rather the terms are used to distinguish one element from another. Also, the use of the terms "a" or "an" and the like do not denote a limitation of quantity, but rather denote the presence of at least one.

The above description is only exemplary embodiments of the present invention and should not be taken as limiting the invention, and any modifications, equivalents, improvements and the like that are within the spirit and principle of the present invention should be included in the scope of the present invention.

Claims

1. A quality evaluation method for borrowers based on multi-dimensional information is characterized by comprising the following steps:

acquiring related information of the borrower, wherein the related information comprises at least one of buried point log information, public opinion information, service information, the borrower and related party information thereof;

determining index data corresponding to the related information of the borrower respectively;

and determining the quality evaluation information of the lender through a random forest algorithm or a support vector machine model based on the determined index data corresponding to the relevant information of the borrower.

2. A method for assessing the quality of a borrower based on multidimensional information according to claim 1, wherein the step of determining index data corresponding to each piece of the borrower-related information comprises at least one of:

when the related information is the buried point log information, analyzing the behavior of a client, counting various operations of the user and analyzing the operation flow of the user by deploying a logic tracking code, and providing a data source for emotion analysis of a specific user;

when the related information is public opinion information, outputting related parameters or a label of a specific borrower by analyzing the related public opinions of each network loan platform and the overall situation of the industry;

when the related information is service information, analyzing the period of each behavior of the borrower according to the service data, and providing core modeling indexes;

and when the related information is the information of the borrower and the related party thereof, performing natural semantic analysis by retrieving the major change of the borrower and the related party thereof and the recently announced retrieval content, and supplementing related labels for quality analysis of the borrower.

3. A method as claimed in claim 2, wherein the method further comprises screening the index data prior to determining the lender's quality assessment information.

4. A method of assessing the quality of a borrower from multiple dimensions according to claim 3, wherein said determining whether to credit said borrower is based on said determined quality assessment information of said borrower.

5. A borrower quality assessment method based on multi-dimensional information according to claim 4, wherein when analyzing the borrower and its associated parties, the method performs emotion analysis on various types of information and converts them into indicators of emotional preference of the borrower.

6. A borrower quality evaluation method based on multi-dimensional information according to claim 5, wherein when analyzing the business data, the contents of charging, withdrawal, borrowing, repayment, opening, streaming bidding and closing bidding containing amount information are converted into amount-class indexes, various types of running records of the user are converted into behavior-related indexes, and user authorization information is used as an enumeration-class index.

7. The method as claimed in claim 6, wherein when analyzing the related public opinion information, the borrower's related status and activity are analyzed by counting the number of times the borrower clicks various buttons on each page, visits url, and inputs information, and then converted into various indexes related to behavior, emotion, and preference.

8. A system for analyzing the quality of a network lender in multiple dimensions, comprising:

the embedded point log system analyzes the behaviors of the clients, counts various operations of the users and analyzes the operation flow of the users by deploying logic tracking codes, and provides a data source for emotion analysis of specific users;

the public opinion analysis system outputs related parameters or specific borrower labels by analyzing the related public opinions of each network credit platform and the overall situation of the industry;

the business system analyzes the period of each behavior of the borrower according to the business data and provides core modeling indexes;

the system for analyzing the borrower and the related party analyzes natural semantics by retrieving the great change of the borrower and the related party and recently noticing retrieval contents, and supplements related labels for quality analysis of the borrower.

A decision support system that determines whether to loan the borrower based on the determined quality assessment information for the borrower.

9. An electronic device, comprising a processor and a memory;

the memory is used for storing operation instructions;

the processor is used for executing the borrower quality evaluation method based on the multi-dimensional information according to any one of claims 1 to 7 by calling the operation instruction.

10. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, implements the method for assessing the quality of a borrower based on multidimensional information according to any one of claims 1 to 7.