Method and apparatus for generating a risk assessment model and a risk assessment method and apparatus

Abstract

Embodiments of the present disclosure disclose a method for generating a risk assessment model, comprising: preprocessing a plurality of declaration information to obtain a plurality of first entries included in the plurality of declaration information; performing graph network analysis on information related to the plurality of first entries to generate a first result; performing feature engineering processing on the plurality of first entries to generate a second result; and training a neural network using the first result and the second result to obtain a risk assessment model.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of data processing, and in particular, to a method and device for generating a risk assessment model and a risk assessment method and device. BACKGROUND

[0002] Under the background of the increasing international trade, customs data is gradually large. How to analyze the declaration information and estimate the risk value to screen out high-risk declaration information is an important direction of intelligent inspection at present. With the in-depth development of computer technology, there are "software" and "hardware" support for mining high-risk declaration information based on big data.

[0003] At present, there is little research on risk prediction of declaration information. In the existing analysis means, a risk rule model is mainly constructed, that is, based on historical data, a feature combination sensitive to risk value is screened out by experts and artificial experience, and then a risk value is obtained through a machine learning model. The existing risk rule model has the following disadvantages: first, it depends on the experience of experts and artificial, but high-risk declaration information is not constant, but evolves in real time. Once the rule model is determined, the model is difficult to adapt to the new high-risk mode evolved; second, it is difficult to mine deep correlation information, so it is also difficult to capture gang crime and group risk.

[0004] DISCLOSURE

[0005] According to an aspect of an embodiment of the present disclosure, a method for generating a risk assessment model is provided, comprising:

[0006] preprocessing a plurality of declaration information to obtain a plurality of first entries included in the plurality of declaration information;

[0007] performing graph network analysis on information related to the plurality of first entries to generate a first result;

[0008] performing feature engineering processing on the plurality of first entries to generate a second result; and

[0009] training a neural network using the first result and the second result to obtain a risk assessment model.

[0010] In an embodiment, performing graph network analysis on information related to the plurality of first entries to generate a first result comprises:

[0011] selecting, from all first entries included in the plurality of declaration information, first entries meeting a pre-defined association category as second entries;

[0012] inputting the second entries as node information and association relations between the second entries as edge information into a graph network to generate a plurality of first vectors respectively corresponding to the second entries as the first result.

[0013] In an embodiment, performing feature engineering processing on the plurality of first entries to generate a second result comprises:

[0014] performing feature engineering processing on the plurality of first entries to obtain a plurality of second vectors corresponding to the plurality of first entries as the second result.

[0015] In an embodiment, training a neural network using the first result and the second result to obtain a risk assessment model comprises:

[0016] generating a plurality of third vectors respectively corresponding to the plurality of declaration information according to the plurality of first vectors and the plurality of second vectors; and

[0017] training the neural network using the plurality of third vectors to obtain a risk assessment model.

[0018] In an embodiment, selecting, from all first entries included in the plurality of declaration information, first entries conforming to a predefined association category as second entries comprises:

[0019] selecting, from a plurality of categories to which a plurality of first entries in the plurality of declaration information belong, a category capable of enabling the declaration information to be associated and mutually influence risk prediction probabilities as the predefined association category.

[0020] According to another aspect of an embodiment of the present disclosure, there is provided a risk assessment method, comprising:

[0021] preprocessing declaration information to be evaluated to obtain a plurality of first entries included in the declaration information to be evaluated;

[0022] performing graph network analysis on information related to the plurality of first entries to generate a first result;

[0023] performing feature engineering processing on the plurality of first entries to generate a second result; and

[0024] obtaining, using a risk assessment model, a risk assessment result of the declaration information to be evaluated according to the first result and the second result,

[0025] wherein the risk assessment model is obtained by a plurality of declaration information using a method for generating a risk assessment model according to claim 1.

[0026] In an embodiment, performing graph network analysis on information related to the plurality of first entries to generate a first result comprises:

[0027] selecting, from the plurality of first entries, first entries that meet a pre-defined association category as second entries;

[0028] inputting the second entries as node information and association relationships between the second entries as edge information into a graph network to generate a first vector corresponding to the second entries as the first result.

[0029] In an embodiment, performing feature engineering processing on the plurality of first entries to generate a second result comprises:

[0030] performing feature engineering processing on the plurality of first entries to obtain a second vector corresponding to the plurality of first entries as the second result.

[0031] In an embodiment, obtaining a risk assessment result of the to-be-evaluated declaration information according to the first result and the second result by using a risk assessment model comprises:

[0032] generating a third vector corresponding to the to-be-evaluated declaration information according to the first vector and the second vector; and

[0033] inputting the third vector into the risk assessment model to obtain the risk assessment result of the to-be-evaluated declaration information.

[0034] In an embodiment, selecting, from the plurality of first entries, first entries that meet a pre-defined association category as second entries comprises:

[0035] selecting, from a plurality of categories to which the plurality of first entries belong, a category that can cause the to-be-evaluated declaration information to be associated with another declaration information and to influence risk prediction probabilities of each other as the pre-defined association category.

[0036] According to yet another aspect of embodiments of the present disclosure, there is provided an apparatus for generating a risk assessment model, comprising:

[0037] a preprocessing module configured to preprocess a plurality of declaration information to obtain a plurality of first entries included in the plurality of declaration information;

[0038] a graph network analysis module configured to perform graph network analysis on information related to the plurality of first entries to generate a first result;

[0039] a feature engineering processing module configured to perform feature engineering processing on the plurality of first entries to generate a second result; and

[0040] a training module configured to train a neural network to obtain a risk assessment model by using the first result and the second result.

[0041] In embodiments, the graph network analysis module is further configured to:

[0042] select, from all the first entries included in the plurality of declaration information, first entries that meet a predefined association category as second entries;

[0043] input the second entries as node information and association relationships between the second entries as edge information into a graph network to generate a plurality of first vectors respectively corresponding to the second entries as the first result.

[0044] In embodiments, the feature engineering processing module is further configured to:

[0045] perform feature engineering processing on the plurality of first entries to obtain a plurality of second vectors corresponding to the plurality of first entries as the second result.

[0046] In embodiments, the training module is further configured to:

[0047] generate a plurality of third vectors respectively corresponding to the plurality of declaration information according to the plurality of first vectors and the plurality of second vectors; and

[0048] train the neural network by using the plurality of third vectors to obtain a risk assessment model.

[0049] In embodiments, the graph network analysis module is further configured to:

[0050] select, from a plurality of categories to which the plurality of first entries in the plurality of declaration information belong, categories that can enable the declaration information to be associated and mutually affect risk prediction probabilities as the predefined association category.

[0051] According to still another aspect of embodiments of the present disclosure, there is provided a risk assessment device, comprising:

[0052] a preprocessing module configured to preprocess declaration information to be evaluated to obtain a plurality of first entries included in the declaration information to be evaluated;

[0053] a graph network analysis module configured to perform graph network analysis on information related to the plurality of first entries to generate a first result;

[0054] a feature engineering processing module configured to perform feature engineering processing on the plurality of first entries to generate a second result; and

[0055] an evaluation module configured to obtain, according to the first result and the second result, a risk evaluation result of the to-be-evaluated declaration information by using a risk evaluation model,

[0056] The risk evaluation model is obtained by using the method for generating a risk evaluation model according to claim 1 on a plurality of declaration information.

[0057] In an embodiment, the graph network analysis module is further configured to:

[0058] select, from the plurality of first entries, first entries that meet a predefined association category as second entries;

[0059] input the second entries as node information and the association relationship between the second entries as edge information into a graph network to generate a first vector corresponding to the second entries as the first result.

[0060] In an embodiment, the feature engineering processing module is further configured to:

[0061] perform feature engineering processing on the plurality of first entries to obtain a second vector corresponding to the plurality of first entries as the second result.

[0062] In an embodiment, the evaluation module is further configured to:

[0063] generate a third vector corresponding to the to-be-evaluated declaration information according to the first vector and the second vector; and

[0064] input the third vector into the risk evaluation model to obtain the risk evaluation result of the to-be-evaluated declaration information.

[0065] In an embodiment, the graph network analysis module is further configured to:

[0066] select, from a plurality of categories to which the plurality of first entries belong, a category that can cause the to-be-evaluated declaration information to be associated with another declaration information and to influence risk prediction probabilities of each other as the predefined association category. BRIEF DESCRIPTION OF DRAWINGS

[0067] The above and other objects, features and advantages of the present disclosure will become more apparent from the following description when taken in conjunction with the accompanying drawings, in which:

[0068] Figure 1 a flowchart of a method for generating a risk evaluation model according to an embodiment of the present disclosure is shown;

[0069] Figure 2 a flowchart of a risk evaluation method according to an embodiment of the present disclosure is shown.

[0070] Figure 3 A block diagram of an apparatus for generating a risk assessment model according to an embodiment of the disclosure is shown;

[0071] Figure 4 A block diagram of a risk assessment apparatus according to an embodiment of the disclosure is shown; and

[0072] Figure 5 A schematic diagram of a computer system according to an embodiment of the disclosure is shown.

[0073] The accompanying drawings are not meant to limit embodiments to all of the circuits or structures shown. Like numbers refer to like or similar elements throughout. DETAILED DESCRIPTION

[0074] Hereinafter, embodiments of the disclosure will be described with reference to the accompanying drawings. It is to be understood, however, that the description is merely exemplary of the disclosure, but is not intended to limit the disclosure. Also in the following description, description of well-known functions and constructions will be omitted so as not to obscure the concept of the disclosure.

[0075] The terminology used herein is for the purpose of describing embodiments only and is not intended to limit the disclosure. As used herein, the articles "a", "an", and "the" are intended to include one or more items, unless the context clearly indicates otherwise. Furthermore, to the extent that the terms "including", "includes", "having", "has", "contain", "contains", or variants thereof to be limited to an inclusion means to an exclusion, such that no other, additional, and / or other items are included in the described compositions, methods, and / or processes.

[0076] All terms used herein, including technical and scientific terms, have the meanings commonly understood by one of ordinary skill in the art unless otherwise defined. It should be noted that the terms used herein are to be interpreted as having a meaning that is consistent with the context of the specification, and should not be interpreted in an idealized or overly formal way.

[0077] Figure 1 A schematic diagram of a method 100 for generating a risk assessment model according to an embodiment of the disclosure is shown. The method 100 for generating a risk assessment model can include the following steps.

[0078] In step S110, a plurality of declaration information can be pre-processed to obtain a plurality of first entries included in the plurality of declaration information. Step S110 can include data cleaning on the plurality of declaration information to obtain a plurality of first entries included in the plurality of declaration information. After pre-processing (e.g., data cleaning, etc.), the number of first entries included in each declaration information is the same.

[0079] In step S120, graph network analysis can be performed on information related to the plurality of first entries to generate first results. Step S120 can include selecting, as second entries, first entries conforming to a predefined association category from all the first entries included in the plurality of declaration information, and inputting the second entries as node information and the association relationships between the second entries as edge information into a graph network to generate a plurality of first vectors respectively corresponding to the second entries as the first results. The graph network can be a graph convolution, a graph embedding, or the like. Selecting, as second entries, first entries conforming to a predefined association category from all the first entries included in the plurality of declaration information can include selecting, as the predefined association category, a category from a plurality of categories to which the plurality of first entries in the plurality of declaration information belong, which is capable of causing the declaration information to produce an association and mutually affect risk prediction probabilities. For example, the predefined association category can include a shipper, a consignee, a logistics party, goods, a customs broker, or the like. Taking the category to which an entry belongs as an example, a shipper can be selected as the association category because when one declaration information has a risk, other declaration information involving the shipper in the declaration information can all have a high risk.

[0080] In step S130, feature engineering processing can be performed on the plurality of first entries to generate second results. Step S130 can include performing feature engineering processing on the plurality of first entries to obtain a plurality of second vectors corresponding to the plurality of first entries as the second results.

[0081] In step S140, a neural network can be trained using the first results and the second results to obtain a risk assessment model. Step S140 can include generating a plurality of third vectors respectively corresponding to the plurality of declaration information according to the plurality of first vectors and the plurality of second vectors, and training the neural network using the plurality of third vectors to obtain the risk assessment model. Generating a plurality of third vectors respectively corresponding to the plurality of declaration information according to the plurality of first vectors and the plurality of second vectors can include determining a first vector and a second vector corresponding to each of the plurality of declaration information according to the plurality of first vectors and the plurality of second vectors, and splicing the first vector and the second vector corresponding to each of the plurality of declaration information to generate a third vector corresponding to each of the plurality of declaration information.

[0082] In the following, the method for generating a risk assessment model described above will be described by taking three declaration information as an example. The three declaration information can include a first declaration information, a second declaration information, and a third declaration information, wherein the first declaration information is 100 computers from user A to user B, the second declaration information is 200 mainboards from user C to user B, and the third declaration information is 50 displays from user C to user D.

[0083] First, the above three declaration information is pre-processed (e.g., data cleaning, etc.) to obtain a plurality of first entries. For example, the entries included in the first declaration information are Lenovo, 100, computer, user A and user B, the entries included in the second declaration information are Intel, 200, motherboard, user C and user B, and the entries included in the third declaration information are Lenovo, 50, display, user C and user D. The plurality of first entries obtained by pre-processing (e.g., data cleaning, etc.) the above three declaration information are Lenovo, 100, computer, user A, user B, Intel, 200, motherboard, user C, 50, display, and user D.

[0084] Then, assuming that the pre-defined association categories are brand, goods type, shipper and consignee, the second entries selected from the first entries included in the three declaration information that meet the pre-defined association categories are Lenovo, Intel, computer, motherboard, display, user A, user B, user C and user D. Lenovo, Intel, computer, motherboard, display, user A, user B, user C and user D are input as node information and the association relationships between these node information are input as edge information into a graph network (e.g., graph convolution, graph embedding, etc.) to generate a plurality of first vectors respectively corresponding to the second entries. For example, the first vector corresponding to Lenovo is s1, the first vector corresponding to Intel is s2, the first vector corresponding to computer is s3, the first vector corresponding to motherboard is s4, the first vector corresponding to display is s5, the first vector corresponding to user A is s6, the first vector corresponding to user B is s7, the first vector corresponding to user C is s8, and the first vector corresponding to user D is s9.

[0085] Next, feature engineering processing is performed on the plurality of first entries to obtain second vectors respectively corresponding to the first entries. For example, the second vector corresponding to the plurality of first entries in the first declaration information is [q11, q12, q13,..., q1k], where k is a positive integer, the second vector corresponding to the plurality of first entries in the second declaration information is [q21, q22, q23,..., q2k], and the second vector corresponding to the plurality of first entries in the third declaration information is [q31, q32, q33,..., q3k].

[0086] Then, a plurality of third vectors respectively corresponding to the three declaration information is generated according to the plurality of first vectors and the plurality of second vectors. For example, the first vector corresponding to the first declaration information is [s1, s3, s6, s7], the second vector corresponding to the first declaration information is [q11, q12, q13,..., q1k], and then the third vector corresponding to the first declaration information is [s1, s3, s6, s7, q11, q12, q13,..., q1k] through vector splicing; the first vector corresponding to the second declaration information is [s2, s4, s8, s7], the second vector corresponding to the second declaration information is [q21, q22, q23,..., q2k], and then the third vector corresponding to the second declaration information is [s2, s4, s8, s7, q21, q22, q23,..., q2k] through vector splicing; and the first vector corresponding to the third declaration information is [s1, s5, s8, s9], the second vector corresponding to the third declaration information is [q31, q32, q33,..., q3k], and then the third vector corresponding to the third declaration information is [s1, s5, s8, s9, q31, q32, q33,..., q3k] through vector splicing.

[0087] Finally, the neural network is trained by using the three third vectors [s1, s3, s6, s7, q11, q12, q13,..., q1k], [s2, s4, s8, s7, q21, q22, q23,..., q2k] and [s1, s5, s8, s9, q31, q32, q33,..., q3k] to obtain a risk assessment model. The output of the risk assessment model can be risky or non-risky, or can be the probability of being risky, etc.

[0088] The above describes the process of training the neural network by using a plurality of declaration information to obtain a risk assessment model, and the process of evaluating the to-be-evaluated declaration information by using the obtained risk assessment model will be described below. Figure 2 A flowchart of a risk assessment method 200 according to an embodiment of the present disclosure is shown. The risk assessment method 200 can include the following steps.

[0089] In step S210, the to-be-evaluated declaration information can be preprocessed to obtain a plurality of first entries included in the to-be-evaluated declaration information. Step S210 can include data cleaning of the to-be-evaluated declaration information to obtain a plurality of first entries included in the to-be-evaluated declaration information.

[0090] In step S220, graph network analysis can be performed on information related to the plurality of first entries to generate a first result. Step S220 can include: selecting, from the plurality of first entries, first entries that meet a predefined association category as second entries; and inputting the second entries as node information and the association relationship between the second entries as edge information into a graph network to generate a first vector corresponding to the second entries as the first result. The graph network can be a graph convolution, a graph embedding, or the like. Selecting, from the plurality of first entries, first entries that meet a predefined association category as second entries can include: selecting, from a plurality of categories to which the plurality of first entries belong, a category that can enable the to-be-evaluated declaration information to be associated with another declaration information and influence the risk prediction probability of each other as the predefined association category. For example, the predefined association category can include a shipper, a consignee, a logistics party, goods, a customs broker, or the like.

[0091] In step S230, feature engineering processing can be performed on the plurality of first entries to generate a second result. Step S230 can include: performing feature engineering processing on the plurality of first entries to obtain a second vector corresponding to the plurality of first entries as the second result.

[0092] In step S240, a risk evaluation result of the to-be-evaluated declaration information can be obtained according to the first result and the second result by using a risk evaluation model, wherein the risk evaluation model is obtained by using the method 100 for generating a risk evaluation model on a plurality of declaration information. Step S240 can include: generating a third vector corresponding to the to-be-evaluated declaration information according to the first vector and the second vector, and inputting the third vector into the risk evaluation model to obtain the risk evaluation result of the to-be-evaluated declaration information. Generating the third vector corresponding to the to-be-evaluated declaration information according to the first vector and the second vector can include: splicing the first vector and the second vector to generate the third vector corresponding to the to-be-evaluated declaration information.

[0093] The present disclosure uses deep learning and graph network to perform risk evaluation on declaration information. Compared with the prior art, on the one hand, the risk evaluation model is obtained by combining entry features through feature engineering processing and inputting the entry feature combination into a deep learning network for deep learning, thereby providing an intelligent risk evaluation method. This intelligent method integrates human experience and increases the completeness of entry feature selection. On the other hand, the risk diffusion influence is considered by analyzing the associated entry features through the graph network, which is beneficial to analyzing gang crimes and group risks.

[0094] Figure 3 A schematic diagram of an apparatus 300 for generating a risk evaluation model according to an embodiment of the present disclosure is shown. The apparatus 300 for generating a risk evaluation model can include a preprocessing module 310, a graph network analysis module 320, a feature engineering processing module 330, and a training module 340.

[0095] The preprocessing module 310 can be configured to preprocess the plurality of declaration information to obtain a plurality of first entries included in the plurality of declaration information. The preprocessing module 310 can be further configured to perform data cleaning on the plurality of declaration information to obtain the plurality of first entries included in the plurality of declaration information. After preprocessing (for example, data cleaning), the number of first entries included in each declaration information is the same.

[0096] The graph network analysis module 320 can be configured to perform graph network analysis on information related to the plurality of first entries to generate first results. The graph network analysis module 320 is further configured to: select, from all first entries included in the plurality of declaration information, first entries that meet a predefined association category as second entries; and input the second entries as node information and the association relationship between the second entries as edge information into a graph network to generate a plurality of first vectors corresponding to the second entries respectively as the first results. The graph network can be graph convolution, graph embedding, etc. The graph network analysis module 320 is further configured to: select, from a plurality of categories to which the plurality of first entries in the plurality of declaration information belong, a category that can cause declaration information to be associated and mutually affect risk prediction probability as the predefined association category. For example, the predefined association category can include the shipper, the consignee, the logistics party, the goods, the customs broker, etc. Taking the category to which the entry belongs as an example, the shipper can be selected as the association category, because when a declaration information has a risk, other declaration information involving the shipper in the declaration information can have a high risk.

[0097] The feature engineering processing module 330 can be configured to perform feature engineering processing on the plurality of first entries to generate second results. The feature engineering processing module 340 can be further configured to perform feature engineering processing on the plurality of first entries to obtain a plurality of second vectors corresponding to the plurality of first entries as the second results.

[0098] The training module 340 can be configured to train a neural network using the first results and the second results to obtain a risk assessment model. The training module 340 can be further configured to: generate a plurality of third vectors corresponding to the plurality of declaration information respectively according to the plurality of first vectors and the plurality of second vectors; and train the neural network using the plurality of third vectors to obtain the risk assessment model. The training module 340 can be further configured to: determine a first vector and a second vector corresponding to each of the plurality of declaration information according to the plurality of first vectors and the plurality of second vectors; and splice the first vector and the second vector corresponding to each of the declaration information to generate a third vector corresponding to each of the declaration information.

[0099] Figure 4A block diagram of a risk assessment device 400 according to an embodiment of the present disclosure is shown. The risk assessment device 400 can include a preprocessing module 410, a graph network analysis module 420, a feature engineering processing module 430, and an evaluation module 440.

[0100] The preprocessing module 410 can be configured to preprocess the declaration information to be evaluated to obtain a plurality of first entries included in the declaration information to be evaluated. The preprocessing module 410 can be further configured to perform data cleaning on the declaration information to be evaluated to obtain the plurality of first entries included in the declaration information to be evaluated.

[0101] The graph network analysis module 420 can be configured to perform graph network analysis on information related to the plurality of first entries to generate a first result. The graph network analysis module 420 can be further configured to: select, from the plurality of first entries, first entries that meet a predefined association category as second entries; input the second entries as node information and association relationships between the second entries as edge information into a graph network to generate a first vector corresponding to the second entries as the first result. The graph network can be a graph convolution, a graph embedding, or the like. The graph network analysis module 420 can be further configured to: select, from a plurality of categories to which the plurality of first entries belong, a category that can cause the declaration information to be evaluated to be associated with another declaration information and influence risk prediction probabilities of each other as the predefined association category. For example, the predefined association category can include a shipper, a consignee, a logistics party, goods, a customs broker, and the like.

[0102] The feature engineering processing module 430 can be configured to perform feature engineering processing on the plurality of first entries to generate a second result. The feature engineering processing module 430 can be further configured to perform feature engineering processing on the plurality of first entries to obtain a second vector corresponding to the plurality of first entries as the second result.

[0103] The evaluation module 440 can be configured to obtain a risk assessment result of the declaration information to be evaluated according to the first result and the second result by using a risk assessment model, wherein the risk assessment model is obtained by using the method for generating a risk assessment model 100 on a plurality of declaration information. The evaluation module 440 can be further configured to: generate a third vector corresponding to the declaration information to be evaluated according to the first vector and the second vector; and input the third vector into the risk assessment model to obtain the risk assessment result of the declaration information to be evaluated. The evaluation module 440 can be further configured to: splice the first vector and the second vector to generate the third vector corresponding to the declaration information to be evaluated.

[0104] The present disclosure utilizes deep learning, graph network to make risk estimation on declaration information. Compared with prior art, on the one hand, the item features are combined through feature engineering processing and the item feature combination is input into the deep learning network for deep learning to obtain a risk evaluation model, thereby providing an intelligent risk evaluation method. This intelligent method integrates artificial experience and increases the completeness of item feature selection. On the other hand, the graph network analyzes the associated item features to consider the risk diffusion influence, which is conducive to analyzing gang crime and group risk.

[0105] Figure 5 A schematic diagram of a computer system according to an embodiment of the present disclosure is shown. The system 500 can include a processor 510, for example, a digital signal processor (DSP). The processor 510 can be a single device or multiple devices for performing different actions of the processes described herein. The system 500 can also include an input / output (I / O) device 530 for receiving signals from or sending signals to other entities.

[0106] In addition, the system 500 can include a memory 520, which can have the following forms: non-volatile or volatile memory, for example, electrically erasable programmable read-only memory (EEPROM), flash memory, etc. The memory 520 can store computer readable instructions, which, when executed by the processor 510, can cause the processor to perform the actions described herein.

[0107] Some block diagrams and / or flowcharts are shown in the drawings. It should be understood that some blocks or combinations thereof in the block diagrams and / or flowcharts can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus, so that these instructions, when executed by the processor, can create a device for implementing the functions / operations described in the block diagrams and / or flowcharts.

[0108] Accordingly, the technology of the present disclosure can be embodied in the form of hardware and / or software (including firmware, microcode, etc.). Additionally, the technology of the present disclosure can take the form of a computer program product on a computer-readable medium having instructions executable by an instruction execution system (e.g., one or more processors). In the context of this disclosure, a computer-readable medium can be any medium that can contain, store, communicate, propagate, or transport instructions. For example, the computer-readable medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, device, or propagation medium. Specific examples of a computer-readable medium include magnetic storage devices such as a magnetic tape or hard disk (HDD), optical storage devices such as a compact disc (CD-ROM), memory such as random access memory (RAM) or flash memory, and / or wired / wireless communication links.

[0109] The above detailed description has set forth various embodiments of the methods and apparatuses for generating a risk assessment model, the risk assessment methods and apparatuses, and the computer systems via the use of illustrative figures, flow diagrams, and / or examples. In particular, those skilled in the art will appreciate that the herein disclosed embodiments can be practiced with either software, firmware, hardware, or any combination thereof. In one embodiment, the subject matter described herein can be implemented as a special purpose computer, a programmed computer, a computer module, a programmed processing device, a programmed microprocessor, a microprocessor, a microprocessor-based computer, a computer module, a computer-based memory, a programmed processing device, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory, a programmed computer, a programmed memory,

Claims

1. A method for generating a risk assessment model, comprising: preprocessing a plurality of declaration information to obtain a plurality of first entries included in the plurality of declaration information; performing graph network analysis on information related to the plurality of first entries to generate a first result; performing feature engineering processing on the plurality of first entries to generate a second result; and training a neural network using the first result and the second result to obtain a risk assessment model, wherein performing graph network analysis on information related to the plurality of first entries to generate a first result comprises: selecting, from a plurality of categories to which the plurality of first entries belong in the plurality of declaration information, a category capable of causing the plurality of declaration information to be associated and influence risk prediction probability on each other as a predefined association category; selecting, from all first entries included in the plurality of declaration information, a first entry conforming to the predefined association category as a second entry; and inputting the second entry as node information and an association relationship between the second entries as edge information into a graph network to generate a plurality of first vectors corresponding to the second entries respectively as the first result, wherein a risk assessment result of a declaration information to be assessed is obtained by the risk assessment model. performing feature engineering processing on the plurality of first entries to generate a second result comprises:

2. The method of claim 1, wherein, performing feature engineering processing on the plurality of first entries to obtain a plurality of second vectors corresponding to the plurality of first entries as the second result. training a neural network using the first result and the second result to obtain a risk assessment model comprises:

3. The method of claim 2, wherein, generating a plurality of third vectors corresponding to the plurality of declaration information respectively according to the plurality of first vectors and the plurality of second vectors; and training the neural network using the plurality of third vectors to obtain a risk assessment model.

4. A risk assessment method, comprising: preprocessing a declaration information to be assessed to obtain a plurality of first entries included in the declaration information to be assessed; performing graph network analysis on information related to the plurality of first entries to generate a first result; performing feature engineering processing on the plurality of first entries to generate a second result; and obtaining a risk assessment result of the declaration information to be assessed according to the first result and the second result using a risk assessment model, wherein the risk assessment model is obtained by a plurality of declaration information using the method for generating a risk assessment model according to claim 1, wherein performing graph network analysis on information related to the plurality of first entries to generate a first result comprises: selecting, from a plurality of categories to which the plurality of first entries belong, a category capable of causing the declaration information to be assessed to be associated with another declaration information and influence risk prediction probability on each other as a predefined association category; selecting, from the plurality of first entries, a first entry conforming to the predefined association category as a second entry; and inputting the second entry as node information and an association relationship between the second entries as edge information into a graph network to generate a first vector corresponding to the second entry as the first result. ​ ​ 5. The method of claim 4, wherein, performing feature engineering processing on the plurality of first entries to generate a second result comprises: performing feature engineering processing on the plurality of first entries to obtain a second vector corresponding to the plurality of first entries as the second result.

6. The method of claim 5, wherein, obtaining a risk assessment result of the to-be-evaluated declaration information according to the first result and the second result using a risk assessment model comprises: generating a third vector corresponding to the to-be-evaluated declaration information according to the first vector and the second vector; and inputting the third vector into the risk assessment model to obtain the risk assessment result of the to-be-evaluated declaration information.

7. An apparatus for generating a risk assessment model, comprising: a preprocessing module configured to preprocess a plurality of declaration information to obtain a plurality of first entries included in the plurality of declaration information; a graph network analysis module configured to perform graph network analysis on information related to the plurality of first entries to generate a first result; a feature engineering processing module configured to perform feature engineering processing on the plurality of first entries to generate a second result; and a training module configured to train a neural network using the first result and the second result to obtain a risk assessment model, wherein the graph network analysis module is further configured to: select, from a plurality of categories to which the plurality of first entries in the plurality of declaration information belong, a category capable of causing the plurality of declaration information to be associated and mutually influence risk prediction probabilities as a predefined association category; select, from all first entries included in the plurality of declaration information, first entries conforming to the predefined association category as second entries; and input the second entries as node information and an association relationship between the second entries as edge information into a graph network to generate a plurality of first vectors respectively corresponding to the second entries as the first result, wherein a risk assessment result of to-be-evaluated declaration information is obtained through the risk assessment model.

8. The apparatus of claim 7, wherein, The feature engineering processing module is further configured to: perform feature engineering processing on the plurality of first entries to obtain a plurality of second vectors corresponding to the plurality of first entries as the second result.

9. The apparatus of claim 8, wherein, The training module is further configured to: generate a plurality of third vectors respectively corresponding to the plurality of declaration information according to the plurality of first vectors and the plurality of second vectors; and train the neural network using the plurality of third vectors to obtain a risk assessment model.

10. A risk assessment apparatus, comprising: a preprocessing module configured to preprocess to-be-evaluated declaration information to obtain a plurality of first entries included in the to-be-evaluated declaration information; a graph network analysis module configured to perform graph network analysis on information related to the plurality of first entries to generate a first result; a feature engineering processing module configured to perform feature engineering processing on the plurality of first entries to generate a second result; and an evaluation module configured to obtain a risk assessment result of the to-be-evaluated declaration information according to the first result and the second result using a risk assessment model, The risk assessment model is obtained by using the method for generating a risk assessment model according to claim 1 on a plurality of declaration information, The graph network analysis module is further configured to: select, from a plurality of categories to which the plurality of first entries belong, a category capable of causing the to-be-evaluated declaration information and another declaration information to be associated and to influence risk prediction probabilities of each other as a predefined association category; select, from the plurality of first entries, a first entry conforming to the predefined association category as a second entry; and input the second entry as node information and an association relationship between the second entry as edge information into a graph network to generate a first vector corresponding to the second entry as the first result.

11. The risk assessment apparatus according to claim 10, wherein, The feature engineering processing module is further configured to: perform feature engineering processing on the plurality of first entries to obtain a second vector corresponding to the plurality of first entries as the second result.

12. The risk assessment apparatus according to claim 11, wherein, The evaluation module is further configured to: generate a third vector corresponding to the to-be-evaluated declaration information according to the first vector and the second vector; and input the third vector into the risk assessment model to obtain a risk assessment result of the to-be-evaluated declaration information.

Images