CN113011961B - Method, device, equipment and storage medium for monitoring risk of company-related information - Google Patents
Method, device, equipment and storage medium for monitoring risk of company-related information Download PDFInfo
- Publication number
- CN113011961B CN113011961B CN202110133551.5A CN202110133551A CN113011961B CN 113011961 B CN113011961 B CN 113011961B CN 202110133551 A CN202110133551 A CN 202110133551A CN 113011961 B CN113011961 B CN 113011961B
- Authority
- CN
- China
- Prior art keywords
- information
- company
- feature
- network
- dimension
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 61
- 238000012544 monitoring process Methods 0.000 title claims abstract description 55
- 239000013598 vector Substances 0.000 claims abstract description 94
- 238000013135 deep learning Methods 0.000 claims abstract description 80
- 238000012549 training Methods 0.000 claims abstract description 68
- 238000012545 processing Methods 0.000 claims description 21
- 238000006243 chemical reaction Methods 0.000 claims description 11
- 238000012806 monitoring device Methods 0.000 claims description 10
- 238000013507 mapping Methods 0.000 claims description 9
- 238000004364 calculation method Methods 0.000 claims description 7
- 238000000605 extraction Methods 0.000 claims description 5
- 230000003247 decreasing effect Effects 0.000 claims description 4
- 125000004122 cyclic group Chemical group 0.000 claims 4
- 238000011156 evaluation Methods 0.000 abstract description 16
- 238000012502 risk assessment Methods 0.000 description 8
- 238000004891 communication Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 4
- 238000007670 refining Methods 0.000 description 4
- RWSOTUBLDIXVET-UHFFFAOYSA-N Dihydrogen sulfide Chemical compound S RWSOTUBLDIXVET-UHFFFAOYSA-N 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000011144 upstream manufacturing Methods 0.000 description 2
- 238000013075 data extraction Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000002349 favourable effect Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
- G06Q40/03—Credit; Loans; Processing thereof
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
- G06F16/367—Ontology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Strategic Management (AREA)
- Economics (AREA)
- Data Mining & Analysis (AREA)
- General Business, Economics & Management (AREA)
- Marketing (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Development Economics (AREA)
- Human Resources & Organizations (AREA)
- Accounting & Taxation (AREA)
- Finance (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Technology Law (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Animal Behavior & Ethology (AREA)
- Computational Linguistics (AREA)
- Databases & Information Systems (AREA)
- Game Theory and Decision Science (AREA)
- Artificial Intelligence (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a method, a device, equipment and a storage medium for monitoring risk of company-related information, which are used for acquiring customer characteristic information of a company and related characteristic information of a company-related party; extracting feature embedding vectors based on the customer feature information and the associated feature information; training a multi-head attention module of a preset deep learning network and the feature embedding network according to the feature embedding vector to obtain a target model; and acquiring feature information to be evaluated of the company to be evaluated, and performing risk prediction on the feature information to be evaluated through the target model to obtain a prediction result. When the training is applied to obtain the target model for risk prediction, the feature information to be evaluated is predicted through the multi-head attention mechanism of the target model, so that the accuracy of an evaluation result in the process of performing company risk evaluation is effectively improved.
Description
Technical Field
The present invention relates to the field of risk monitoring technologies, and in particular, to a method, an apparatus, a device, and a storage medium for risk monitoring of company-related information.
Background
At present, when the risk assessment of a company is carried out, risk detection can be carried out by manually refining a risk signal according to expert rules or business experience, and the refining mode is simple index statistics. Or constructing a knowledge graph according to the credit data of the company only, utilizing graph deep learning to learn characteristic embedding of the knowledge graph of the company, and carrying out risk assessment on the characteristic embedding through a risk assessment model.
However, the accuracy of the evaluation result is low whether the risk evaluation is performed by manually refining the risk signal or by the credit data of the company itself.
Disclosure of Invention
The invention mainly aims to provide a method, a device, equipment and a storage medium for monitoring company-related information risk, which aim to solve the technical problem that the accuracy of an evaluation result is lower when the company risk evaluation is carried out currently.
In order to achieve the above object, an embodiment of the present invention provides a method for monitoring risk of company-related information, where the method for monitoring risk of company-related information includes:
acquiring customer characteristic information of a company and associated characteristic information of an associated party of the company;
extracting feature embedding vectors based on the customer feature information and the associated feature information;
Training a multi-head attention module of a preset deep learning network and the feature embedding network according to the feature embedding vector to obtain a target model;
And acquiring feature information to be evaluated of the to-be-evaluated company, and carrying out risk prediction on the to-be-evaluated company through the target model and the feature information to be evaluated to obtain a prediction result.
Preferably, the training the multi-head attention module of the preset deep learning network and the feature embedding network according to the feature embedding vector to obtain the target model includes:
Processing the feature embedded vector through a multi-head attention module of a preset deep learning network to obtain a hidden layer vector;
Mapping the hidden layer vector through a linear two-classifier of the preset deep learning network to obtain a classification result;
and training the characteristic embedded network of the preset deep learning network and the multi-head attention module according to the classification result to obtain a target model.
Preferably, the step of processing the feature embedded vector by a multi-head attention module of a preset deep learning network to obtain a hidden layer vector includes:
acquiring attention weight parameters of a multi-head attention module in a preset deep learning network;
and carrying out preset operation on the feature embedded vector according to the attention weight parameter to obtain a hidden layer vector.
Preferably, the step of extracting a feature embedding vector based on the client feature information and the associated feature information includes:
inputting the client characteristic information and the associated characteristic information into a characteristic embedded network of a preset deep learning network;
and carrying out data conversion on the client characteristic information and the associated characteristic information through the characteristic embedding network to obtain a characteristic embedding vector.
Preferably, the step of obtaining customer characteristic information of a company and associated characteristic information of an associated party of the company includes:
Acquiring historical credit violation information of a company, creating a training tag according to the historical credit violation information, acquiring tag characteristic information of the company, and forming customer characteristic information by the training tag and the tag characteristic information;
And acquiring the association graph information of the company, and acquiring the association characteristic information of the company association party according to the association graph information.
Preferably, the step of acquiring the association characteristic information of the company association party according to the association map information includes:
identifying the association party of the company according to the association map information;
And acquiring the association characteristic information of the association party.
Preferably, after the step of obtaining the feature information to be evaluated of the company to be evaluated, the method further includes:
And performing contribution calculation on the feature information to be evaluated through a model interpretation module to generate associated risk monitoring information.
In order to achieve the above object, the present invention further provides a company-related information risk monitoring apparatus, including:
The acquisition module is used for acquiring the client characteristic information of the company and the associated characteristic information of the company associated party;
the extracting module is used for extracting feature embedding vectors based on the client feature information and the associated feature information;
The training module is used for training the multi-head attention module of the preset deep learning network and the feature embedding network according to the feature embedding vector to obtain a target model;
The prediction module is used for obtaining feature information to be evaluated of a company to be evaluated, and performing risk prediction on the feature information to be evaluated through the target model to obtain a prediction result.
Further, in order to achieve the above object, the present invention also provides a company-related information risk monitoring device, which includes a memory, a processor, and a company-related information risk monitoring program stored on the memory and executable on the processor, wherein the company-related information risk monitoring program, when executed by the processor, implements the steps of the company-related information risk monitoring method described above.
Further, in order to achieve the above object, the present invention further provides a storage medium, on which a company-related information risk monitoring program is stored, which when executed by a processor, implements the steps of the above-described company-related information risk monitoring method.
The embodiment of the invention provides a method, a device, equipment and a storage medium for monitoring risk of company-related information, which are used for acquiring customer characteristic information of a company and related characteristic information of a company-related party; extracting feature embedding vectors based on the customer feature information and the associated feature information; training a multi-head attention module of a preset deep learning network and the feature embedding network according to the feature embedding vector to obtain a target model; and acquiring feature information to be evaluated of the company to be evaluated, and performing risk prediction on the feature information to be evaluated through the target model to obtain a prediction result. According to the method, the feature embedding vector is extracted from the customer feature information of the company and the associated feature information of the company associated party and is used as a training basis, and the multi-head attention module of the preset deep learning network and the feature embedding network are trained, so that when the target model obtained through training is applied to risk prediction, the feature information to be evaluated is predicted through the multi-head attention mechanism of the target model, and the accuracy of an evaluation result in the process of carrying out company risk evaluation is effectively improved.
Drawings
FIG. 1 is a schematic diagram of a hardware operating environment according to an embodiment of a method for risk monitoring company-related information;
FIG. 2 is a flowchart illustrating a first embodiment of a method for risk monitoring company-related information according to the present invention;
FIG. 3 is a flowchart illustrating a method for risk monitoring company-related information according to a second embodiment of the present invention;
FIG. 4 is a flowchart illustrating a third embodiment of a method for risk monitoring company-related information according to the present invention;
FIG. 5 is a flowchart illustrating a method for risk monitoring company-related information according to a fourth embodiment of the present invention;
fig. 6 is a schematic functional block diagram of a company-related information risk monitoring device according to a preferred embodiment of the present invention.
The achievement of the objects, functional features and advantages of the present invention will be further described with reference to the accompanying drawings, in conjunction with the embodiments.
Detailed Description
It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.
The embodiment of the invention provides a method, a device, equipment and a storage medium for monitoring risk of company-related information, which are used for acquiring customer characteristic information of a company and related characteristic information of a company-related party; extracting feature embedding vectors based on the customer feature information and the associated feature information; training a multi-head attention module of a preset deep learning network and the feature embedding network according to the feature embedding vector to obtain a target model; and acquiring feature information to be evaluated of the company to be evaluated, and performing risk prediction on the feature information to be evaluated through the target model to obtain a prediction result. According to the method, the feature embedding vector is extracted from the customer feature information of the company and the associated feature information of the company associated party and is used as a training basis, and the multi-head attention module of the preset deep learning network and the feature embedding network are trained, so that when the target model obtained through training is applied to risk prediction, the feature information to be evaluated is predicted through the multi-head attention mechanism of the target model, and the accuracy of an evaluation result in the process of carrying out company risk evaluation is effectively improved.
Referring to fig. 1, fig. 1 is a schematic structural diagram of a company-related information risk monitoring device of a hardware running environment according to an embodiment of the present invention.
In the following description, suffixes such as "module", "component", or "unit" for representing elements are used only for facilitating the description of the present invention, and have no specific meaning per se. Thus, "module," "component," or "unit" may be used in combination.
The company associated information risk monitoring device in the embodiment of the invention can be a PC, or can be a mobile terminal device such as a tablet computer, a portable computer and the like.
As shown in fig. 1, the company-associated information risk monitoring apparatus may include: a processor 1001, such as a CPU, a network interface 1004, a user interface 1003, a memory 1005, a communication bus 1002. Wherein the communication bus 1002 is used to enable connected communication between these components. The user interface 1003 may include a Display, an input unit such as a Keyboard (Keyboard), and the optional user interface 1003 may further include a standard wired interface, a wireless interface. The network interface 1004 may optionally include a standard wired interface, a wireless interface (e.g., WI-FI interface). The memory 1005 may be a high-speed RAM memory or a stable memory (non-volatile memory), such as a disk memory. The memory 1005 may also optionally be a storage device separate from the processor 1001 described above.
It will be appreciated by those skilled in the art that the corporate linkage information risk monitoring device structure shown in fig. 1 does not constitute a limitation of the corporate linkage information risk monitoring device and may include more or fewer components than shown, or may combine certain components, or a different arrangement of components.
As shown in fig. 1, an operating system, a network communication module, a user interface module, and a company-related information risk monitoring program may be included in the memory 1005 as one type of storage medium.
In the device shown in fig. 1, the network interface 1004 is mainly used for connecting to a background server, and performing data communication with the background server; the user interface 1003 is mainly used for connecting a client (user side) and performing data communication with the client; and the processor 1001 may be configured to call the company-associated information risk monitoring program stored in the memory 1005 and perform the following operations:
acquiring customer characteristic information of a company and associated characteristic information of an associated party of the company;
extracting feature embedding vectors based on the customer feature information and the associated feature information;
Training a multi-head attention module of a preset deep learning network and the feature embedding network according to the feature embedding vector to obtain a target model;
And acquiring feature information to be evaluated of the to-be-evaluated company, and carrying out risk prediction on the to-be-evaluated company through the target model and the feature information to be evaluated to obtain a prediction result.
Further, the training the multi-head attention module of the preset deep learning network and the feature embedding network according to the feature embedding vector to obtain the target model includes:
Processing the feature embedded vector through a multi-head attention module of a preset deep learning network to obtain a hidden layer vector;
Mapping the hidden layer vector through a linear two-classifier of the preset deep learning network to obtain a classification result;
and training the characteristic embedded network of the preset deep learning network and the multi-head attention module according to the classification result to obtain a target model.
Further, the step of processing the feature embedded vector by the multi-head attention module of the preset deep learning network to obtain a hidden layer vector includes:
acquiring attention weight parameters of a multi-head attention module in a preset deep learning network;
and carrying out preset operation on the feature embedded vector according to the attention weight parameter to obtain a hidden layer vector.
Further, the step of extracting a feature embedding vector based on the client feature information and the associated feature information includes:
inputting the client characteristic information and the associated characteristic information into a characteristic embedded network of a preset deep learning network;
and carrying out data conversion on the client characteristic information and the associated characteristic information through the characteristic embedding network to obtain a characteristic embedding vector.
Further, the step of obtaining the customer characteristic information of the company and the associated characteristic information of the company associated party includes:
Acquiring historical credit violation information of a company, creating a training tag according to the historical credit violation information, acquiring tag characteristic information of the company, and forming customer characteristic information by the training tag and the tag characteristic information;
And acquiring the association graph information of the company, and acquiring the association characteristic information of the company association party according to the association graph information.
Further, the step of obtaining the association characteristic information of the company association party according to the association map information includes:
identifying the association party of the company according to the association map information;
And acquiring the association characteristic information of the association party.
Further, after the step of acquiring the feature information to be evaluated of the company to be evaluated, the processor 1001 may be configured to call the company-related information risk monitoring program stored in the memory 1005, and perform the following operations:
And performing contribution calculation on the feature information to be evaluated through a model interpretation module to generate associated risk monitoring information.
In order that the above-described aspects may be better understood, exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.
In order to better understand the above technical solutions, the following detailed description will refer to the accompanying drawings and specific embodiments.
Referring to fig. 2, a first embodiment of the present invention provides a flow chart of a method for monitoring risk of company-related information. In this embodiment, the method for monitoring risk of company-related information includes the steps of:
Step S10, acquiring customer characteristic information of a company and associated characteristic information of an associated party of the company;
The method for monitoring the risk of the company-related information is applied to a system for monitoring the risk of the company-related information, and understandably, for convenience of description, the system for monitoring the risk of the company-related information is simply referred to as a system; the system comprises a preset deep learning network and a model explanatory module, wherein the preset deep learning network is preferably a graph deep learning network in the embodiment, the graph deep learning network consists of a feature embedding network, a linear two-classifier and a multi-head attention module, and the feature embedding network is used for converting data into features with fixed sizes for representation; linear bi-classifiers are used in machine learning to separate objects or events into two or more classifications using quadrics; the multi-head attention module is used for parallelly calculating and selecting a plurality of pieces of information from input information by utilizing a plurality of queries; the model interpretation module is used to calculate how the plurality of input variables in the model work and how the predictions of the model change according to the values of the input variables.
It can be appreciated that, currently, when performing risk assessment of a company, risk signals can be manually refined according to expert rules or business experience to perform risk detection, and the refinement mode is simple index statistics. Or constructing a knowledge graph according to the credit data of the company only, utilizing graph deep learning to learn characteristic embedding of the knowledge graph of the company, and carrying out risk assessment on the characteristic embedding through a risk assessment model. However, the accuracy of the evaluation result is low whether the risk evaluation is performed by manually refining the risk signal or by the credit data of the company itself. On the basis, the application provides a company associated information risk monitoring method, which is characterized in that feature embedded vectors are extracted from customer feature information of a company and associated feature information of a company associated party to serve as training basis, and a multi-head attention module of a preset deep learning network and the feature embedded network are trained, so that when a target model obtained by training is applied to risk prediction, the feature information to be evaluated is predicted through a multi-head attention mechanism of the target model, and the accuracy of an evaluation result in the process of carrying out company risk evaluation is effectively improved.
Further, the system acquires historical credit violation information of at least one company, creates a training tag according to the historical credit violation information, and acquires tag feature information of the company, wherein the tag feature information refers to various feature information corresponding to the company with the training tag, and as can be understood, in machine learning or deep learning, the more data are used for training an initial learning network, the higher the accuracy of a final model obtained through training is, so that when the historical credit violation information is acquired, the historical credit violation information of a plurality of formulas can be acquired as much as possible, and the number is not capped. In order to perform risk assessment more accurately, the system also needs to acquire feature information of other companies or enterprises related to the current company, and if historical credit violation information of a plurality of companies is acquired, feature information of each company's associated company or enterprise needs to be acquired respectively, so as to obtain associated feature information of company's associated party.
Step S20, extracting a feature embedding vector based on the client feature information and the associated feature information;
Further, in order to process the customer feature information and the associated feature information more conveniently, the system needs to extract the feature information from the customer feature information and the associated feature information, specifically, the system inputs the customer feature information and the associated feature information into a feature embedding network of a preset deep learning network, performs data conversion on the input customer feature information and associated feature information through the feature embedding network, and obtains a feature embedding vector after the data conversion is completed.
Further, the step of extracting a feature embedding vector based on the client feature information and the associated feature information includes:
Step S21, inputting the customer characteristic information and the associated characteristic information into a characteristic embedded network of a preset deep learning network;
and S22, performing data conversion on the client feature information and the associated feature information through the feature embedding network to obtain a feature embedding vector.
Further, in order to be able to conveniently train the preset deep learning network through the client feature information and the associated feature information, data format conversion is required to be carried out on the client feature information and the associated feature information, specifically, the system inputs the client feature information and the associated feature information into a feature embedding network of the preset deep learning network, data conversion is carried out on the client feature information and the associated feature information through the feature embedding network, specifically, as the data dimension of the client feature information and the associated feature information is possibly different from the data dimension which can be identified by the preset deep learning network, dimension information of identifiable data is arranged in the feature embedding network of the preset deep learning network, after the client feature information and the associated feature information are received by the preset deep learning network, the data dimension of the client feature information and the associated feature information is identified by the feature embedding network, and then dimension increasing or decreasing processing is carried out on the client feature information and the associated feature information, so that the data dimension of the client feature information and the associated feature information after dimension decreasing processing is identical with the dimension information of the identifiable data in the feature embedding network, and dimension information of the identifiable data in the feature embedding network is converted to obtain feature embedding vectors; the system takes care of calculating the feature embedding vectors of various types of correspondents of the company and finds out each type of correspondents node set of the company by circularly executing the process of obtaining the feature embedding vectors. It can be appreciated that when the client feature information and the associated feature information are input into the feature embedded network of the preset deep learning network, the information such as the number of attention heads, the attention parameters, the linear classifier parameters, the maximum iteration parameters and the like needs to be input into the heterogram formed by the node convolution kernel and the edge set. The feature embedded vectors of the customer feature information and the associated feature information are extracted through the preset deep learning network, so that the feature embedded vectors are prevented from being manually extracted, the efficiency of data processing is improved, errors generated during manual extraction can be reduced, and the accuracy of data extraction is improved.
Step S30, training a multi-head attention module of a preset deep learning network and the feature embedding network according to the feature embedding vector to obtain a target model;
Further, the system firstly calculates the hidden layer vector by calculating the feature embedded vector through a multi-head attention module of a preset deep learning network; mapping hidden layer vectors through a linear two-classifier in a preset deep learning network to obtain a classification result composed of a plurality of classifications; and finally training a multi-head attention module of a preset deep learning network and a characteristic embedding network according to the generated classification result to obtain a target model.
And S40, obtaining feature information to be evaluated of the company to be evaluated, and carrying out risk prediction on the feature information to be evaluated through the target model to obtain a prediction result.
Further, when a company needs to transact risk business under the system, a business transacting request can be sent to the system, when the system receives the business transacting request sent by the company, the company needing to transact risk business is firstly determined to be the company to be assessed, then the characteristic information to be assessed of the company to be assessed is acquired, the risk information to be assessed of the company to be assessed is input into a target model obtained through training, risk prediction is carried out on the characteristic information to be assessed through the target model, finally a prediction result for carrying out risk prediction on the company is obtained, wherein the prediction result is a default probability between 0 and 1, the system can output the prediction result through a display device, so that staff can make a decision on the business transacting request of the company according to the default probability in the prediction result, specifically, whether the business transacting (such as loan) request of the company to be assessed is carried out according to the default probability decision whether the default probability is higher than the default probability set by a financial institution (such as a bank), loan is carried out on the company to be assessed, and if the default probability of the business transacting request cannot be carried out by the bank is not favorable for the bank.
The embodiment provides a method, a device, equipment and a storage medium for monitoring risk of company-related information, wherein customer characteristic information of a company and related characteristic information of a company-related party are obtained; extracting feature embedding vectors based on the customer feature information and the associated feature information; training a multi-head attention module of a preset deep learning network and the feature embedding network according to the feature embedding vector to obtain a target model; and acquiring feature information to be evaluated of the company to be evaluated, and performing risk prediction on the feature information to be evaluated through the target model to obtain a prediction result. According to the method, the feature embedding vector is extracted from the customer feature information of the company and the associated feature information of the company associated party and is used as a training basis, and the multi-head attention module of the preset deep learning network and the feature embedding network are trained, so that when the target model obtained through training is applied to risk prediction, the feature information to be evaluated is predicted through the multi-head attention mechanism of the target model, and the accuracy of an evaluation result in the process of carrying out company risk evaluation is effectively improved.
Further, referring to fig. 3, based on the first embodiment of the method for monitoring risk of company-related information, a second embodiment of the method for monitoring risk of company-related information is proposed, in the second embodiment, the training the multi-head attention module of the preset deep learning network and the feature embedding network according to the feature embedding vector, and the step of obtaining the target model includes:
Step S31, processing the feature embedded vector through a multi-head attention module of a preset deep learning network to obtain a hidden layer vector;
Step S32, mapping the hidden layer vector through a linear two-classifier of the preset deep learning network to obtain a classification result;
And step S33, training the feature embedded network of the preset deep learning network and the multi-head attention module according to the classification result to obtain a target model.
Further, the system inputs the feature embedded vector into a preset deep learning network, the input feature embedded vector is processed through a multi-head attention module in the preset deep learning network, specifically, the multi-head attention module performs weighted average operation on the input feature embedded vector, and a hidden layer vector is obtained after the operation is completed. Further, the system maps the hidden layer vector obtained by operation through a linear two-classifier of a preset deep learning network, maps the hidden layer vector into a plurality of classifications through the linear two-classifier, and forms a mapped classification result through the plurality of classifications. Further, the system trains the feature embedded network of the preset deep learning network and the multi-head attention module according to the classification result obtained by mapping, specifically trains parameters in the feature embedded network and the multi-head attention module, judges that the training is completed when the parameters are trained to be converged, ends the circulation process and obtains a final target model. After the final target model is obtained through training, only the feature information to be evaluated of the to-be-evaluated company is required to be input into the target model, risk prediction can be carried out on the to-be-evaluated company through the target model according to the feature information to be evaluated, a prediction result is generated, and the accuracy of the prediction result of the target model obtained through training is improved through training the feature embedded network of the preset deep learning network and the multi-head attention module.
Further, the step of processing the feature embedded vector by the multi-head attention module of the preset deep learning network to obtain a hidden layer vector includes:
step S311, attention weight parameters of a multi-head attention module in a preset deep learning network are obtained;
step S312, performing a preset operation on the feature embedding vector according to the attention weight parameter to obtain a hidden layer vector.
It can be understood that, attention weight parameters are allocated by the system in the multi-head attention module of the preset deep learning network, and are used for representing weights occupied by all heads in the overall module in the multi-head attention module, and the weights of all heads are not fixed and are allocated after being updated by the system according to a certain updating rule. Therefore, the system acquires attention weight parameters composed of all the head weights from the multi-head attention module of the preset deep learning network, and then performs preset weighted average operation on the feature embedded vectors according to the attention weight parameters, specifically, performs product operation on the feature embedded vectors and the weights of all the head respectively to obtain a plurality of product results, and then splices the plurality of product results through addition operation to form hidden layer vectors of the feature embedded vectors, so that the feature embedded network of the preset deep learning network and the multi-head attention module are trained after classification results are mapped through the hidden layer vectors, and the accuracy of prediction results of prediction of a target model obtained through training is improved.
According to the embodiment, the multi-head attention module and the linear two-classifier are used for processing the feature embedded vector in sequence, and the classification result obtained through processing is used for training the multi-head attention module and the feature embedded network of the preset deep learning network, so that the accuracy of the prediction result of the target model obtained through training is higher.
Further, referring to fig. 4, based on the first embodiment of the method for monitoring risk of company-related information of the present invention, a third embodiment of the method for monitoring risk of company-related information of the present invention is provided, in the third embodiment, the step of obtaining customer characteristic information of a company and related characteristic information of a party associated with the company includes:
step S11, historical credit violation information of a company is obtained, a training tag is created according to the historical credit violation information, tag characteristic information of the company is obtained, and customer characteristic information is formed by the training tag and the tag characteristic information;
Step S12, acquiring the association graph information of the company, and acquiring the association characteristic information of the company association party according to the association graph information.
Further, since the company needs to provide feature information such as credit, business, law, transaction, credit, finance, business owner, early warning, risk and the like for the bank when conducting business transaction such as loan to the financial institution such as the bank, or the authorized financial institution can directly obtain feature information such as credit, business, law, transaction, credit, finance, business owner, early warning, risk and the like of the company through the system, the system can obtain feature information such as credit, business, law, transaction, credit, finance, business owner, early warning, risk and the like of the company as tag feature information from the risk information database of the financial institution or the risk information storage unit of the company, and after obtaining the history credit violation information, the system creates a training tag according to the history credit violation condition, for example, if the violation condition exists, the tag 1 is set, and if the violation condition does not exist, the tag 0 is set. And after the label characteristic information and the training label are obtained, the training label and the label characteristic information form customer characteristic information together so as to input the customer characteristic information into a preset deep learning network in the follow-up process, and the association relationship between the training label and the label characteristic information is obtained through training.
It will be appreciated that in the case of evaluating a financial related business transaction, it is not accurate to evaluate the business according to the characteristic information of the current company, and if the company has a subsidiary company or a parent company, the characteristic information of the related company will also affect the credit ability of the current company. Therefore, the system also needs to acquire the association graph information of the current company, wherein the association graph information is used for displaying the relationship between the current company and other companies with association relationship with the current company, and after acquiring the association graph information, the system acquires the association party of the current company through the association graph information and acquires the association characteristic information of the association party.
Further, the step of obtaining the association characteristic information of the company association party according to the association map information includes:
step S121, identifying the association party of the company according to the association map information;
step S122, obtaining the association characteristic information of the association party.
Further, after the association graph information is obtained, the system can identify other companies associated with the current company through the association graph information, take the other companies associated with the current company as association parties, obtain feature information such as guarantee, upstream and downstream, same real control person, same legal person representative, litigation, trade, external investment, stock right and the like of the association parties, obtain feature information such as credit, industry, law, trade, credit, finance, enterprise owner, early warning, risk and the like of the association parties, and jointly form feature information such as guarantee, upstream and downstream, same real control person, same legal person representative, litigation, trade, external investment, stock right and the like of the association parties, and feature information such as credit, industry, law, trade, credit, finance, enterprise owner, early warning, risk and the like of the association parties.
According to the method and the device for obtaining the client feature information of the company and the associated feature information of the company associated party, the client feature information generated by the training tag and the associated feature information of the associated party enable feature embedding extraction of risk information of the company and the associated party to be more accurate through a preset image deep learning network. And moreover, the risk influence factors with interpretability are given through the model interpretability module, so that the convincing power of company-associated risk monitoring conclusion can be improved.
Further, referring to fig. 5, based on the first embodiment of the method for monitoring risk of company-related information, a fourth embodiment of the method for monitoring risk of company-related information according to the present invention is provided, in the fourth embodiment, after the step of obtaining the feature information to be evaluated of the company to be evaluated, the method further includes:
and step S100, performing contribution calculation on the feature information to be evaluated through a model interpretation module to generate associated risk monitoring information.
Further, a model interpretation module is also provided in the system, which can calculate how a plurality of input variables in the model work and how predictions of the model change according to the values of the input variables. Therefore, after training the multi-head attention module of the preset deep learning network and the feature embedding network through the feature embedding vector to obtain a target model, and when carrying out risk prediction on the to-be-evaluated company through the target model and the acquired to-be-evaluated feature information, carrying out contribution degree calculation on the to-be-evaluated feature information input into the target model through the model interpretation module, determining the contribution degree of each feature in the to-be-evaluated feature information in risk assessment, so as to carry out model interpretation on the to-be-evaluated feature information, generating associated risk monitoring information of the to-be-evaluated company after calculation is completed, and outputting the associated risk monitoring information through a display screen, specifically, outputting the associated risk monitoring information together when outputting a prediction result, so as to achieve the purpose of carrying out overall risk monitoring on the to-be-evaluated company.
According to the method, the contribution degree calculation is carried out on the feature information to be evaluated through the model explanatory module, the associated risk monitoring information is generated, and the comprehensiveness of enterprise risk monitoring is effectively improved.
Furthermore, the invention also provides a company-related information risk monitoring device.
Referring to fig. 6, fig. 6 is a schematic functional block diagram of a first embodiment of a company related information risk monitoring apparatus according to the present invention.
The company-associated information risk monitoring device includes:
an acquiring module 10, configured to acquire customer characteristic information of a company and associated characteristic information of an associated party of the company;
An extraction module 20 for extracting feature embedding vectors based on the customer feature information and the associated feature information;
The training module 30 is configured to train the multi-head attention module of the preset deep learning network and the feature embedding network according to the feature embedding vector to obtain a target model;
and the prediction module 40 is configured to obtain feature information to be evaluated of a company to be evaluated, and perform risk prediction on the feature information to be evaluated through the target model to obtain a prediction result.
Further, the acquisition module 10 includes:
the first acquisition unit is used for acquiring historical credit violation information of a company, creating a training tag according to the historical credit violation information, acquiring tag characteristic information of the company, and forming customer characteristic information by the training tag and the tag characteristic information;
The second acquisition unit is used for acquiring the association graph information of the company and acquiring the association characteristic information of the company association party according to the association graph information.
Further, the acquisition module 10 further includes:
An identification unit, configured to identify an association party of the company according to the association map information;
and the third acquisition unit is used for acquiring the association characteristic information of the association party.
Further, the extraction module 20 includes:
the input unit is used for inputting the client characteristic information and the associated characteristic information into a characteristic embedded network of a preset deep learning network;
And the conversion unit is used for carrying out data conversion on the client characteristic information and the associated characteristic information through the characteristic embedding network to obtain a characteristic embedding vector.
Further, the training module 30 includes:
The data processing unit is used for processing the feature embedded vector through a multi-head attention module of a preset deep learning network to obtain a hidden layer vector;
The mapping unit is used for mapping the hidden layer vector through a linear two-classifier of the preset deep learning network to obtain a classification result;
And the training unit is used for training the characteristic embedded network of the preset deep learning network and the multi-head attention module according to the classification result to obtain a target model.
Further, the training module 30 further includes:
the fourth acquisition unit is used for acquiring attention weight parameters of a multi-head attention module in a preset deep learning network;
And the operation unit is used for carrying out preset operation on the feature embedded vector according to the attention weight parameter to obtain a hidden layer vector.
Further, the prediction module 40 includes:
And the calculating unit is used for calculating the contribution degree of the feature information to be evaluated through the model interpretation module and generating associated risk monitoring information.
In addition, the invention further provides a storage medium, preferably a computer readable storage medium, on which a company-related information risk monitoring program is stored, wherein the company-related information risk monitoring program, when executed by a processor, implements the steps of the embodiments of the company-related information risk monitoring method.
In the embodiments of the company-related information risk monitoring device and the computer-readable medium of the present invention, all technical features of each embodiment of the company-related information risk monitoring method are included, and description and explanation contents are substantially the same as those of each embodiment of the company-related information risk monitoring method, which are not described herein in detail.
It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.
The foregoing embodiment numbers of the present invention are merely for the purpose of description, and do not represent the advantages or disadvantages of the embodiments.
From the above description of the embodiments, it will be clear to those skilled in the art that the above-described embodiment method may be implemented by means of software plus a necessary general hardware platform, but of course may also be implemented by means of hardware, but in many cases the former is a preferred embodiment. Based on such understanding, the technical solution of the present invention may be embodied essentially or partly in the form of a software product stored in a storage medium (such as ROM/RAM, magnetic disk, optical disk), comprising instructions for causing a terminal device (which may be a fixed terminal, such as an intelligent device for internet of things, including intelligent home such as an intelligent air conditioner, an intelligent lamp, an intelligent power supply, an intelligent router, or a mobile terminal, including a smart phone, a wearable internet-of-a-r/VR device, an intelligent sound box, an automatic car, or the like) to perform the method according to the embodiments of the present invention.
The foregoing description is only of the preferred embodiments of the present invention, and is not intended to limit the scope of the invention, but rather is intended to cover any equivalents of the structures or equivalent processes disclosed herein or in the alternative, which may be employed directly or indirectly in other related arts.
Claims (7)
1. The method for monitoring the risk of the company-related information is characterized by comprising the following steps of:
Acquiring customer characteristic information of a company and associated characteristic information of an associated party of the company; the customer characteristic information includes historical credit violation information, and the company association party at least comprises an associated company or enterprise of the company;
extracting feature embedding vectors based on the customer feature information and the associated feature information;
Training a multi-head attention module of a preset deep learning network and the feature embedding network according to the feature embedding vector to obtain a target model;
Acquiring feature information to be evaluated of a company to be evaluated, and carrying out risk prediction on the company to be evaluated through the target model and the feature information to be evaluated to obtain a prediction result;
Wherein the step of extracting feature embedding vectors based on the client feature information and the associated feature information comprises:
inputting the client characteristic information and the associated characteristic information into a characteristic embedded network of a preset deep learning network;
Identifying the data dimension of the client feature information and the associated feature information through the feature embedding network, and carrying out dimension processing on the data dimension so that the data dimension is in a preset dimension; wherein the dimension processing comprises dimension increasing processing or dimension decreasing processing; the preset dimension is a dimension in which the characteristics are embedded in the network and data can be identified;
The client feature information and the associated feature information which are in a preset dimension are subjected to data conversion through the feature embedding network to obtain a feature embedding vector;
The step of training the multi-head attention module of the preset deep learning network and the feature embedding network according to the feature embedding vector to obtain a target model comprises the following steps:
acquiring attention weight parameters of a multi-head attention module in a preset deep learning network;
Weighting operation is carried out on the feature embedded vector according to the attention weight parameter, and a hidden layer vector is obtained;
mapping the hidden layer vector into a plurality of classifications by a linear two-classifier of the preset deep learning network, and forming a mapped classification result by the plurality of classifications;
and performing cyclic training on the parameters in the feature embedded network of the preset deep learning network and the multi-head attention module according to the classification result, and ending the cyclic training to obtain a target model when the parameters are trained to be converged.
2. The method for risk monitoring company-related information according to claim 1, wherein the step of acquiring customer characteristic information of a company and related characteristic information of a party associated with the company comprises:
Acquiring historical credit violation information of a company, creating a training tag according to the historical credit violation information, acquiring tag characteristic information of the company, and forming customer characteristic information by the training tag and the tag characteristic information;
And acquiring the association graph information of the company, and acquiring the association characteristic information of the company association party according to the association graph information.
3. The company association information risk monitoring method according to claim 2, wherein the step of acquiring association characteristic information of the company association party according to the association map information comprises:
identifying the association party of the company according to the association map information;
And acquiring the association characteristic information of the association party.
4. The method for risk monitoring company-related information according to claim 1, further comprising, after the step of acquiring the feature information to be evaluated of the company to be evaluated:
And performing contribution calculation on the feature information to be evaluated through a model interpretation module to generate associated risk monitoring information.
5. A company-associated information risk monitoring apparatus, characterized by comprising:
The acquisition module is used for acquiring the client characteristic information of the company and the associated characteristic information of the company associated party; the customer characteristic information includes historical credit violation information, and the company association party at least comprises an associated company or enterprise of the company;
the extracting module is used for extracting feature embedding vectors based on the client feature information and the associated feature information;
The training module is used for training the multi-head attention module of the preset deep learning network and the feature embedding network according to the feature embedding vector to obtain a target model;
The prediction module is used for acquiring feature information to be evaluated of a company to be evaluated, and performing risk prediction on the feature information to be evaluated through the target model to obtain a prediction result;
The extraction module is further used for inputting the client characteristic information and the associated characteristic information into a characteristic embedded network of a preset deep learning network; identifying the data dimension of the client feature information and the associated feature information through the feature embedding network, and carrying out dimension processing on the data dimension so that the data dimension is in a preset dimension; wherein the dimension processing comprises dimension increasing processing or dimension decreasing processing; the preset dimension is a dimension in which the characteristics are embedded in the network and data can be identified; the client feature information and the associated feature information which are in a preset dimension are subjected to data conversion through the feature embedding network to obtain a feature embedding vector;
The training module is also used for acquiring attention weight parameters of a multi-head attention module in a preset deep learning network; weighting operation is carried out on the feature embedded vector according to the attention weight parameter, and a hidden layer vector is obtained; mapping the hidden layer vector into a plurality of classifications by a linear two-classifier of the preset deep learning network, and forming a mapped classification result by the plurality of classifications; and performing cyclic training on the parameters in the feature embedded network of the preset deep learning network and the multi-head attention module according to the classification result, and ending the cyclic training to obtain a target model when the parameters are trained to be converged.
6. A company-associated information risk monitoring device, characterized in that it comprises a memory, a processor and a company-associated information risk monitoring program stored on the memory and executable on the processor, which company-associated information risk monitoring program, when executed by the processor, implements the steps of the company-associated information risk monitoring method according to any of claims 1-4.
7. A storage medium having stored thereon a company-associated information risk monitoring program which, when executed by a processor, implements the steps of the company-associated information risk monitoring method of any of claims 1-4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110133551.5A CN113011961B (en) | 2021-01-29 | 2021-01-29 | Method, device, equipment and storage medium for monitoring risk of company-related information |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110133551.5A CN113011961B (en) | 2021-01-29 | 2021-01-29 | Method, device, equipment and storage medium for monitoring risk of company-related information |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113011961A CN113011961A (en) | 2021-06-22 |
| CN113011961B true CN113011961B (en) | 2024-04-19 |
Family
ID=76384700
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110133551.5A Active CN113011961B (en) | 2021-01-29 | 2021-01-29 | Method, device, equipment and storage medium for monitoring risk of company-related information |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113011961B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113989043A (en) * | 2021-10-28 | 2022-01-28 | 支付宝(杭州)信息技术有限公司 | Event risk identification method, device and equipment |
| CN116796909A (en) * | 2023-08-16 | 2023-09-22 | 浙江同信企业征信服务有限公司 | Judicial litigation risk prediction method, device, equipment and storage medium |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109657918A (en) * | 2018-11-19 | 2019-04-19 | 平安科技（深圳）有限公司 | Method for prewarning risk, device and the computer equipment of association assessment object |
| CN111724083A (en) * | 2020-07-21 | 2020-09-29 | 腾讯科技（深圳）有限公司 | Training method and device for financial risk recognition model, computer equipment and medium |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20200357060A1 (en) * | 2019-05-10 | 2020-11-12 | Fair Ip, Llc | Rules/model-based data processing system for intelligent default risk prediction |
-
2021
- 2021-01-29 CN CN202110133551.5A patent/CN113011961B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109657918A (en) * | 2018-11-19 | 2019-04-19 | 平安科技（深圳）有限公司 | Method for prewarning risk, device and the computer equipment of association assessment object |
| CN111724083A (en) * | 2020-07-21 | 2020-09-29 | 腾讯科技（深圳）有限公司 | Training method and device for financial risk recognition model, computer equipment and medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113011961A (en) | 2021-06-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110009174B (en) | Risk recognition model training method and device and server | |
| CN108648074A (en) | Loan valuation method, apparatus based on support vector machines and equipment | |
| CN111460250B (en) | Image data cleaning method, image data cleaning device, image data cleaning medium, and electronic apparatus | |
| CN109739844B (en) | Data classification method based on attenuation weight | |
| CN113011961B (en) | Method, device, equipment and storage medium for monitoring risk of company-related information | |
| CN108229485A (en) | For testing the method and apparatus of user interface | |
| CN112527972A (en) | Intelligent customer service chat robot implementation method and system based on deep learning | |
| CN111145006A (en) | Automobile financial anti-fraud model training method and device based on user portrait | |
| CN111159417A (en) | Method, device and equipment for extracting key information of text content and storage medium | |
| CN116630100B (en) | Travel data processing method, device, equipment and storage medium | |
| CN112487284A (en) | Bank customer portrait generation method, equipment, storage medium and device | |
| CN110704803A (en) | Target object evaluation value calculation method and device, storage medium and electronic device | |
| CN112950347B (en) | Resource data processing optimization method and device, storage medium and terminal | |
| CN114418748A (en) | Vehicle credit evaluation method, device, equipment and storage medium | |
| CN111353728A (en) | Risk analysis method and system | |
| CN117132383A (en) | Credit data processing method, device, equipment and readable storage medium | |
| CN116777646A (en) | Artificial intelligence-based risk identification method, apparatus, device and storage medium | |
| CN111582647A (en) | User data processing method and device and electronic equipment | |
| CN110335145A (en) | A kind of influence factor Dynamic Display method, apparatus, electronic equipment and storage medium | |
| CN115293867A (en) | Financial reimbursement user portrait optimization method, device, equipment and storage medium | |
| CN113177603B (en) | Training method of classification model, video classification method and related equipment | |
| CN111400413B (en) | Method and system for determining category of knowledge points in knowledge base | |
| CN115099988A (en) | Model training method, data processing method, device and computer medium | |
| CN111881008B (en) | Data evaluation method, data evaluation device, model training method, model evaluation device, model training equipment and storage medium | |
| CN114154556A (en) | Training method and device of sample prediction model, electronic equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |