Object evaluation method and device

Abstract

The embodiments of the present specification provide an object evaluation method and apparatus, wherein an object evaluation method comprises: obtaining sample data of multiple sample objects in each dimension; labeling the evaluation scaling parameters of the sample data in each dimension based on prior knowledge; setting the constraints of the loss function of the evaluation model based on the labeling results; training the evaluation model based on the sample data and constraints in multiple dimensions to obtain a trained evaluation model for evaluating the target object; the parameters to be trained in the evaluation model include the evaluation scaling parameters of each dimension and the neutral evaluation parameters of each dimension; the neutral evaluation parameters represent the data of the neutral evaluation of the sample object in the dimension to which it corresponds.

Description

Technical Field

[0001] This document relates to the field of data processing technology, and in particular to an object evaluation method and device. Background Art

[0002] In real life, an object is often not evaluated from a single dimension, but from a combination of multiple dimensions. In some dimensions, when most of the objects to be evaluated perform similarly, even if an object receives a very good or very bad evaluation in this dimension, it is difficult to objectively reflect whether the object has an advantage over other objects. Therefore, the comprehensive evaluation of an object often requires taking into account the differences in each dimension. Summary of the invention

[0003] One or more embodiments of the present specification provide an object evaluation method. The object evaluation method includes: obtaining sample data of multiple sample objects in various dimensions. Annotating the evaluation scaling parameters of the sample data in each dimension based on prior knowledge. Setting the constraints of the loss function of the evaluation model based on the annotation results. Training the evaluation model based on the sample data and the constraints in multiple dimensions to obtain the trained evaluation model for evaluating the target object. The parameters to be trained in the evaluation model include the evaluation scaling parameters of each dimension and the neutral evaluation parameters of each dimension. The neutral evaluation parameters represent the data of the neutral evaluation of the sample object in the dimension to which it corresponds.

[0004] One or more embodiments of the present specification provide a risk assessment method. The risk assessment method includes: obtaining sub-model scores of at least two dimensions of a target object. Inputting the obtained sub-model scores into the trained assessment model described in the aforementioned object assessment method, and outputting the total risk assessment score of the target object.

[0005] One or more embodiments of the present specification provide an object evaluation device, including: a sample acquisition module, configured to obtain sample data of multiple sample objects in various dimensions. A parameter labeling module, configured to label the evaluation scaling parameters of the sample data in each dimension based on prior knowledge. A constraint setting module, configured to set the constraint conditions of the loss function of the evaluation model according to the labeling results. A model training module, configured to train the evaluation model according to the sample data and the constraints in multiple dimensions to obtain the trained evaluation model for evaluating the target object. The parameters to be trained in the evaluation model include the evaluation scaling parameters of each dimension and the neutral evaluation parameters of each dimension. The neutral evaluation parameters represent the data of the neutral evaluation of the sample object in the dimension to which it corresponds.

[0006] One or more embodiments of the present specification provide a risk assessment device, including: a score acquisition module, configured to acquire sub-model scores of at least two dimensions of a target object; and a total score output module, configured to input the acquired sub-model scores into the trained assessment model described in the aforementioned object assessment method, and output the total risk assessment score of the target object.

[0007] One or more embodiments of the present specification provide an object evaluation device, comprising: a processor; and a memory configured to store computer executable instructions, wherein the computer executable instructions, when executed, cause the processor to: obtain sample data of multiple sample objects in various dimensions. Annotate the evaluation scaling parameters of the sample data of each dimension based on prior knowledge. Set the constraints of the loss function of the evaluation model based on the annotation results. Train the evaluation model based on the sample data of multiple dimensions and the constraints to obtain the trained evaluation model for evaluating the target object. The parameters to be trained in the evaluation model include the evaluation scaling parameters of each dimension and the neutral evaluation parameters of each dimension. The neutral evaluation parameters represent the data of the neutral evaluation of the sample object in the dimension to which it corresponds.

[0008] One or more embodiments of the present specification provide a risk assessment device, including: a processor; and a memory configured to store computer executable instructions, wherein when the computer executable instructions are executed, the processor: obtains sub-model scores of at least two dimensions of a target object, inputs the obtained sub-model scores into the trained assessment model described in the aforementioned object assessment method, and outputs a total risk assessment score of the target object.

[0009] One or more embodiments of the present specification provide a storage medium for storing computer-executable instructions, which implement the following process when executed by a processor: obtain sample data of multiple sample objects in various dimensions. Label the evaluation scaling parameters of the sample data in each dimension based on prior knowledge. Set the constraints of the loss function of the evaluation model based on the labeling results. Train the evaluation model based on the sample data and the constraints in multiple dimensions to obtain the trained evaluation model for evaluating the target object. The parameters to be trained in the evaluation model include the evaluation scaling parameters of each dimension and the neutral evaluation parameters of each dimension. The neutral evaluation parameters represent the data of the neutral evaluation of the sample object in the dimension to which it corresponds.

[0010] One or more embodiments of the present specification provide a storage medium for storing computer executable instructions, which implement the following process when executed by a processor: obtaining sub-model scores of at least two dimensions of a target object. Inputting the obtained sub-model scores into the trained evaluation model described in the aforementioned object evaluation method, and outputting the total risk evaluation score of the target object. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate one or more embodiments of the present specification or the technical solutions in the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present specification. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative labor.

[0012] Figure 1 A process flow chart of an object evaluation method provided for one or more embodiments of this specification;

[0013] Figure 2 A first schematic diagram of an evaluation model in an object evaluation method provided in one or more embodiments of this specification;

[0014] Figure 3 A second schematic diagram of an evaluation model in an object evaluation method provided in one or more embodiments of this specification;

[0015] Figure 4 A processing flow chart of an object evaluation method applied to a blockchain risk control scenario provided by one or more embodiments of this specification;

[0016] Figure 5 A risk assessment method processing flow chart provided for one or more embodiments of this specification;

[0017] Figure 6 A score display interface diagram of a risk assessment method provided in one or more embodiments of this specification;

[0018] Figure 7 A schematic diagram of an object evaluation device provided for one or more embodiments of this specification;

[0019] Figure 8 A schematic diagram of a risk assessment device provided for one or more embodiments of this specification;

[0020] Fig. 9 A schematic diagram of the structure of an object evaluation device provided for one or more embodiments of this specification;

[0021] Fig.10A schematic diagram of the structure of a risk assessment device provided in one or more embodiments of this specification. DETAILED DESCRIPTION

[0022] In order to enable those skilled in the art to better understand the technical solutions in one or more embodiments of this specification, the technical solutions in one or more embodiments of this specification will be clearly and completely described below in conjunction with the drawings in one or more embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of them. Based on one or more embodiments of this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this document.

[0023] An object evaluation method embodiment provided in this specification:

[0024] Reference Figure 1 The object evaluation method provided in this embodiment specifically includes steps S102 to S108.

[0025] Step S102, obtaining sample data of multiple sample objects in various dimensions.

[0026] In this specification, the objects to be evaluated by the object evaluation method provided in this embodiment include but are not limited to: institutions, organizations, commodities, non-commercial items, personnel, places, etc. This specification focuses on institutions as an example, and other objects are similar to institutions and will not be repeated.

[0027] The sample object may be an object used for model training, for example, an institution used for model training.

[0028] During specific implementation, multiple dimensions may be pre-set to evaluate the object from different perspectives to obtain evaluation scores of different dimensions.

[0029] For example, if the object to be evaluated is an institution, at least five dimensions can be set to evaluate the institution from different perspectives. Exemplarily, the at least five dimensions include: experience status dimension, association relationship dimension, institution qualification dimension, public opinion information dimension, and violation information dimension. The use process of other dimensions not listed can refer to these five dimensions and will not be repeated below.

[0030] Obtaining sample data of multiple sample objects in each dimension may be to obtain the evaluation score of each sample object in each dimension as sample data. The following takes the sample object as an institution and the sample data as a percentage evaluation score as an example to specifically describe the sample data of the sample object in each dimension. The sample data of other sample objects in each dimension are similar to the evaluation scores of the institution in each dimension, and will not be described in detail below.

[0031] For example, the evaluation scores of 1,000 institutions in five dimensions, namely, experience status, relationship, institution qualification, public opinion information, and violation information, are obtained on a percentage basis.

[0032] Institution 1:

[0033] Experience dimension: 42 points

[0034] Relationship dimension: 20 points

[0035] Institutional qualification dimension: 68 points

[0036] Public opinion information dimension: 39 points

[0037] Violation information dimension: 79 points

[0038] Institution 2:

[0039] Experience dimension: 52 points

[0040] Relationship dimension: 73 points

[0041] Institutional qualification dimension: 27 points

[0042] Public opinion information dimension: missing

[0043] Violation information dimension: 49 points

[0044] Institution 3:

[0045] Experience dimension: 25 points

[0046] Relationship dimension: 28 points

[0047] Institutional qualification dimension: 63 points

[0048] Public opinion information dimension: 70 points

[0049] Violation information dimension: 60 points

[0050] …

[0051] On the one hand, there are differences in the scale of the evaluation scores of different dimensions. For example, in the scenario where the higher the score, the better, a score of 60 in the experience dimension means that the institution is good, and the institution ranks in the top 10% among all the institutions participating in the evaluation; a score of 60 in the public opinion information dimension means that the institution is average, and the institution ranks in the top 50% among all the institutions participating in the evaluation; a score of 60 in the violation information dimension means that the institution is poor, and the institution ranks in the top 80% among all the institutions participating in the evaluation.

[0052] On the other hand, in the process of obtaining sample data of multiple sample objects in various dimensions, there may be missing data of individual dimensions of individual sample objects.

[0053] It should be noted that, when the number of sample objects is large enough, data missing will not have a negative impact on the object evaluation method provided in this embodiment. A sufficient number may be a pre-set object number threshold. When the number of sample objects is greater than the pre-set object number threshold, it is considered that the number of sample objects is large enough, and data missing will not have a negative impact on the object evaluation method provided in this embodiment.

[0054] Optionally, before obtaining sample data of multiple sample objects in each dimension, it also includes: each first blockchain node among multiple first blockchain nodes, through a pre-deployed first smart contract, evaluates the dimension corresponding to each sample object in the first blockchain node, obtains the sub-model score of the dimension corresponding to each sample object in the first blockchain node, and uploads the sub-model score in the blockchain; each dimension corresponds to at least one first blockchain node; obtaining sample data of multiple sample objects in each dimension includes: the second blockchain node obtains the sub-model score of each sample object in each dimension from the blockchain through a pre-deployed second smart contract, and determines the sub-model score as the sample data.

[0055] A smart contract is a computer protocol designed to communicate, verify or execute contracts in an information-based manner. Smart contracts allow for trusted transactions without a third party, which are traceable and irreversible. In the context of blockchain, a smart contract can be a piece of code written on the blockchain that automatically executes once an event triggers the terms of the contract.

[0056] In specific implementation, the first smart contract pre-deployed on each first blockchain node can be generated based on a pre-trained evaluation sub-model stored on the first blockchain node, and the evaluation sub-model is used to obtain an evaluation score of a dimension of the object to be evaluated.

[0057] For example, the first blockchain node 1 stores an evaluation sub-model of the first dimension. For any organization, the organization can be taken as a sample object, and the sub-model score of the organization output by the evaluation sub-model is used as the sample data of the first dimension of the organization; the first blockchain node 2 stores an evaluation sub-model of the second dimension. For any organization, the organization can be taken as a sample object, and the sub-model score of the organization output by the evaluation sub-model is used as the sample data of the second dimension of the organization… The first blockchain node 5 stores an evaluation sub-model of the fifth dimension. For any organization, the organization can be taken as a sample object, and the sub-model score of the organization output by the evaluation sub-model is used as the sample data of the fifth dimension of the organization.

[0058] Then, the first blockchain node, through the first smart contract, for each sample object, upon receiving an input parameter for evaluating a dimension of the sample object, evaluates the sample object in the dimension corresponding to the node, obtains the sub-model score of the sample object in the dimension corresponding to the node, and uploads the sub-model score to the blockchain.

[0059] Each dimension may correspond to one first blockchain node, or may correspond to multiple first blockchain nodes. Each first blockchain node may correspond to one dimension.

[0060] For example, the second blockchain node can obtain the sub-model scores of the first dimension of 1,000 sample objects uploaded by the first blockchain node 1 from the blockchain through the second smart contract; the second blockchain node can obtain the sub-model scores of the second dimension of 1,000 sample objects uploaded by the first blockchain node 2 from the blockchain through the second smart contract; the second blockchain node can obtain the sub-model scores of the third dimension of 2,000 sample objects uploaded by the first blockchain node 3 from the blockchain through the second smart contract; the second blockchain node can obtain the sub-model scores of the third dimension of 1,500 sample objects uploaded by the first blockchain node 4 from the blockchain through the second smart contract… The number of sample objects corresponding to the sub-model scores uploaded by each first blockchain node obtained by the second blockchain node can be the same or different.

[0061] For another example, the second blockchain node can obtain the sub-model scores of the first dimension of institutions 1 to 2000 uploaded by the first blockchain node 1 from the blockchain through the second smart contract; the second blockchain node can obtain the sub-model scores of the second dimension of institutions 1 to 2000 uploaded by the first blockchain node 2 from the blockchain through the second smart contract; the second blockchain node can obtain the sub-model scores of the third dimension of institutions 1500 to 3500 uploaded by the first blockchain node 3 from the blockchain through the second smart contract; the second blockchain node can obtain the sub-model scores of the third dimension of institutions 500 to 1499 uploaded by the first blockchain node 4 from the blockchain through the second smart contract… The sample objects corresponding to the sub-model scores uploaded by each first blockchain node obtained by the second blockchain node from the blockchain can be exactly the same sample objects, or completely different sample objects, or partially the same and partially different sample objects.

[0062] Blockchain technology has the advantage of privacy encryption. The following can be combined with risk control scenarios to specifically illustrate how the object evaluation method provided in this embodiment is combined with blockchain technology and utilize the advantages of privacy encryption.

[0063] Let’s take the blockchain risk control scenario as an example. Through blockchain technology, different data vendors can be linked and multiple data vendors can be integrated. The more data required in the modeling stage, the better. The richer the samples, the better the modeling effect.

[0064] Specifically, in order to make the evaluation effect of the evaluation model for risk control of institutions more accurate, the producer of risk control products needs a large amount of data to participate in model training. For the producer of risk control products, the data can be purchased from the data platform or provided by the customer of the risk control product or other third parties. Among them, the data purchased from the data platform is readable and can be used for model training for the producer of risk control products; while the data provided by the customer of the risk control product or other third parties is unreadable for the producer of risk control products but can be used for model training.

[0065] In the process of determining the total evaluation score using the sub-model scores of evaluation sub-models in multiple dimensions, the input parameters of each sample object of each evaluation sub-model may involve user privacy information, such as the user transaction history data of an institution. However, the sub-model scores of the sample objects output by the evaluation sub-model can be displayed in the display interface of the risk control product and do not constitute user privacy information.

[0066] Each first blockchain node provided by the customer or other third party of the risk control product can be deployed with an evaluation sub-model of one dimension provided by the risk control product. The input parameters of each sample object that may involve user privacy information are provided by the customer or other third party of the risk control product. The input parameters are not on the chain, so they are not readable by the producer of the risk control product. Each first blockchain node executes the input parameter into the pre-stored evaluation sub-model, outputs the corresponding sub-model score, and uploads it to the blockchain. The sub-model score is on the chain, so it is readable by the producer of the risk control product.

[0067] Therefore, the second blockchain node obtains the sub-model scores of each sample object in each dimension uploaded to the blockchain by each first blockchain node through the pre-deployed second smart contract. It can link multiple different data providers and use massive data for model training to improve the model training effect, while avoiding the leakage of user privacy information, taking into account both protecting user privacy and meeting the data quantity requirements for model training.

[0068] In another embodiment, sample data of multiple sample objects in various dimensions may also be obtained from each server in the server cluster, and each server pre-stores a pre-trained evaluation sub-model, which is used to obtain an evaluation score of a dimension of the object to be evaluated. Each dimension may correspond to a server in the server cluster, or may correspond to multiple servers in the server cluster. Each server in the server cluster may correspond to a dimension.

[0069] Step S104: labeling the evaluation scaling parameters of the sample data of each dimension according to prior knowledge.

[0070] Prior knowledge can be the accumulated "artificial experience" that can be accumulated as standard numerical values. In different business scenarios, there are different types of formula expressions.

[0071] Taking the risk control scenario as an example, the customers of risk control products have accumulated years of experience in their own scenarios and have richer scenario knowledge, that is, prior knowledge. The producers of risk control products can obtain this prior knowledge from their customers.

[0072] The prior knowledge obtained from the customer can be a subjective feeling of the customer, such as:

[0073] The importance of dimension 1 in positive evaluation is greater than the importance of dimension 2 in positive evaluation;

[0074] Alternatively, in dimension 3, when an institution has label 1, the probability that it is a bad institution is greater than the probability that it is a good institution;

[0075] Alternatively, dimension 4 may only involve positive evaluations and have nothing to do with negative evaluations, and so on.

[0076] The evaluation scaling parameter is used to determine the boundary threshold of the quantized positive evaluation or negative evaluation in combination with the neutral evaluation parameter T when performing a quantized positive or negative evaluation on the sample object. The evaluation scaling parameter includes a positive evaluation scaling parameter A and a negative evaluation scaling parameter B. The positive evaluation scaling parameter A is used to determine the boundary threshold of the quantized positive evaluation in combination with the neutral evaluation parameter T; the negative evaluation scaling parameter B is used to determine the boundary threshold of the quantized negative evaluation in combination with the neutral evaluation parameter T. The positive evaluation scaling parameter A and the negative evaluation scaling parameter B can be the parameters to be trained for the evaluation model in subsequent steps.

[0077] Each dimension corresponds to a neutral evaluation parameter T. The neutral evaluation parameter T represents the neutral evaluation of the data in the dimension on the sample object. The neutral evaluation parameter T can be the parameter to be trained for the evaluation model in the subsequent steps. For example, the neutral evaluation parameter T corresponding to dimension 1 is 55 points. The evaluation score of institution 1 in dimension 1 is 60 points, which is greater than T. Institution 1 receives a positive evaluation in dimension 1 and can be regarded as a good institution; the evaluation score of institution 2 in dimension 1 is 44 points, which is less than T. Institution 2 receives a negative evaluation in dimension 1 and can be regarded as a bad institution; the evaluation score of institution 3 in dimension 1 is 55 points. Institution 3 receives a neutral evaluation in dimension 1 and can be regarded as a medium-to-high institution.

[0078] For example, for any sample object, the sub-model score of the first dimension of the sample object is x. If x is greater than T, the sub-model score is positively scaled according to the positive evaluation scaling parameter A to obtain the target evaluation score x' after the positive scaling; if x is less than or equal to T, the sub-model score is negatively scaled according to the negative evaluation scaling parameter B to obtain x' after the negative scaling. Furthermore, when the number of sample objects is large enough, the upper limit of the value range of x' can reflect the boundary threshold of the positive evaluation of the sample object by the data of the first dimension, and the lower limit of the value range of x' can reflect the boundary threshold of the negative evaluation of the sample object by the data of the first dimension.

[0079] Exemplarily, the forward scaling process may refer to the following formula:

[0080] x'=(xT)*A+0.5 (1)

[0081] Among them, x is the sub-model score of the first dimension of sample object 1, x' is the target evaluation score of the first dimension of the sample object obtained after forward scaling the sub-model score of the first dimension of sample object 1, T is the neutral evaluation parameter, and A is the positive evaluation scaling parameter.

[0082] Negative scaling can refer to the following formula:

[0083] x'=(xT)*B+0.5 (2)

[0084] Among them, x is the sub-model score of the first dimension of sample object 1, x' is the target evaluation score of the first dimension of the sample object obtained by negatively scaling the sub-model score of the first dimension of sample object 1, T is the neutral evaluation parameter, and B is the negative evaluation scaling parameter.

[0085] The sufficient number may be a preset object number threshold. When the number of sample objects is greater than the preset object number threshold, it is considered that the number of sample objects is sufficient.

[0086] Optionally, the evaluation scaling parameters include a positive evaluation scaling factor and a negative evaluation scaling factor; the evaluation scaling parameters of the sample data of each dimension are labeled according to prior knowledge, including: converting the prior knowledge into a size comparison result of the positive evaluation scaling factors of any two dimensions in each dimension, to obtain a first labeling result; and / or, converting the prior knowledge into a size comparison result of the negative evaluation scaling factors of any two dimensions in each dimension, to obtain a second labeling result; and / or, converting the prior knowledge into a size comparison result of the positive evaluation scaling factors and the negative evaluation scaling factors of each dimension, to obtain a third labeling result.

[0087] For example, the prior knowledge includes that the importance of dimension 1 in positive evaluation is greater than the importance of dimension 2 in positive evaluation. Then, the prior knowledge can be converted into a size comparison relationship between the positive evaluation scaling parameter A_{dimension 1} of dimension 1 and the positive evaluation scaling parameter A_{dimension 2} of dimension 2: A_{dimension 1}>A_{dimension 2}. The size comparison result can also be expressed by the following formula:

[0088] A_{dimension 1}=A_{dimension 2}+w A _{12}, (3)

[0089] Among them, w A _{12} is used to represent the difference in positive evaluation between dimension 1 and dimension 2, specifically, it is used to represent the difference between A_{dimension 1} and A_{dimension 2}. A _{12} can be a free search parameter greater than 0, the initial value of the free search parameter can be set in advance, and the free search parameter can be trained during the training process of the evaluation model in the subsequent steps.

[0090] The above formula (3) can be used as the first annotation result.

[0091] For another example, the prior knowledge includes that the positive boundary threshold of dimension 3 is greater than the negative boundary threshold, that is, the higher the evaluation score of the sample object in dimension 3, the more it tends to be positively evaluated; the lower the evaluation score of the sample object in dimension 3, the more it tends to be negatively evaluated. Then the prior knowledge can be converted into a size comparison relationship between the positive evaluation scaling parameter A_{dimension 3} of dimension 3 and the negative evaluation scaling parameter B_{dimension 3} of dimension 3: A_{dimension 3}>B_{dimension 3}, and the size comparison result can also be expressed by the following formula:

[0092] A_{dimension 3}=B_{dimension 3}+w_{33}, (4)

[0093] Among them, w_{33} is used to represent that on dimension 3, the higher the evaluation score, the more positive the evaluation tends to be, and the lower the evaluation score, the more negative the evaluation tends to be. Specifically, w_{33} is used to represent the difference between A_{dimension 3} and B_{dimension 3}, which can be another free search parameter greater than 0. The initial value of the free search parameter can be pre-set, and w_{33} can be obtained through evaluation model training.

[0094] The above formula (4) can be used as the third annotation result.

[0095] Optionally, the evaluation scaling parameters include a positive evaluation scaling factor and a negative evaluation scaling factor; the evaluation scaling parameters of the sample data of each dimension are labeled according to prior knowledge, including: if it is determined according to prior knowledge that any dimension is not related to negative evaluations, the negative evaluation scaling factor of the dimension is set to a preset neutral value; if it is determined according to prior knowledge that any dimension is not related to positive evaluations, the positive evaluation scaling factor of the dimension is set to a preset neutral value.

[0096] The preset neutral value may be 0.5. If a dimension is determined to be irrelevant to negative evaluations based on prior knowledge, the negative evaluation scaling factor of the dimension is set to 0.5. If a dimension is determined to be irrelevant to positive evaluations based on prior knowledge, the positive evaluation scaling factor of the dimension is set to 0.5.

[0097] Step S106, setting the constraint conditions of the loss function of the evaluation model according to the labeling results.

[0098] The loss function reflects the degree of fit of the model to the data. Specifically, it is used to measure the inconsistency between the predicted value of the evaluation model and the true value. It is a non-negative real-valued function. The smaller the loss function, the better the evaluation effect of the evaluation model. The loss function can be represented by L(f,y).

[0099] Step S106, setting the constraint conditions of the loss function of the evaluation model according to the labeling results.

[0100] In a specific implementation, a loss function under constraint conditions may be generated according to one or more of the first labeling result, the second labeling result, and the third labeling result.

[0101] The loss function of the evaluation model can be a logistic regression loss function or other types of loss functions. This specification does not impose any special restrictions on the type of loss function.

[0102] For the loss function under constraints, the smaller the constraint conditions are met and the smaller the loss function value is, the better the evaluation effect of the evaluation model is.

[0103] Exemplarily, three first annotation results, two second annotation results, and four third annotation results may be obtained through step S104. The three first annotation results are:

[0104] A_{dimension 1}=A_{dimension 2}+w A _{12},

[0105] A_{dimension 2}=A_{dimension 3}+w A _{twenty three},

[0106] A_{dimension 1}=A_{dimension 4}+w A _{14}.

[0107] Among them, A_{dimension 1} is the positive evaluation scaling parameter of dimension 1; A_{dimension 2} is the positive evaluation scaling parameter of dimension 2; A_{dimension 3} is the positive evaluation scaling parameter of dimension 3; A_{dimension 4} is the positive evaluation scaling parameter of dimension 4. A _{12} is used to represent the difference between A_{dimension 1} and A_{dimension 2} and can be a free search parameter greater than 0. A _{12} can be obtained by training the evaluation model; w A _{23} is used to represent the difference between A_{dimension 2} and A_{dimension 3} and can be a free search parameter greater than 0. A _{23} can be obtained by training the evaluation model; w A _{14} is used to represent the difference between A_{dimension 1} and A_{dimension 4} and can be a free search parameter greater than 0. A _{14} can be obtained by training the evaluation model.

[0108] The 2 second annotation results are:

[0109] B_{dimension 1}=B_{dimension 2}+w B _{12},

[0110] B_{dimension 4}=B_{dimension 1}+w B _{41},

[0111] Among them, B_{dimension1} is the negative evaluation scaling parameter of dimension 1; B_{dimension2} is the negative evaluation scaling parameter of dimension 2; B_{dimension4} is the negative evaluation scaling parameter of dimension 4. B _{12} is used to represent the difference between B_{dimension 1} and B_{dimension 2} and can be a free search parameter greater than 0. B _{12} can be obtained by training the evaluation model; w B_{41} is used to represent the difference between B_{dimension 4} and B_{dimension 1} and can be a free search parameter greater than 0. B _{41} can be obtained by training the evaluation model.

[0112] The 4 third annotation results are:

[0113] A_{dimension 1}=B_{dimension 1}+w_{11},

[0114] A_{dimension 2}=B_{dimension 2}+w_{22},

[0115] A_{dimension 3}=B_{dimension 3}+w_{33},

[0116] A_{dimension 4}=B_{dimension 4}+w_{44},

[0117] Among them, B_{dimension 3} is the negative evaluation scaling parameter of dimension 3. w_{11} is used to represent the difference between A_{dimension 1} and B_{dimension 1}, and can be a free search parameter greater than 0. This w_{11} can be obtained through evaluation model training; w_{22} is used to represent the difference between A_{dimension 2} and B_{dimension 2}, and can be a free search parameter greater than 0. This w_{22} can be obtained through evaluation model training; w_{33} is used to represent the difference between A_{dimension 3} and B_{dimension 3}, and can be a free search parameter greater than 0. This w_{33} can be obtained through evaluation model training; w_{44} is used to represent the difference between A_{dimension 4} and B_{dimension 4}, and can be a free search parameter greater than 0. This w_{44} can be obtained through evaluation model training.

[0118] Then, the constraints of the loss function of the evaluation model are set according to the annotation results, which can be:

[0119] A loss function L(f,y) is configured, and the constraints of the loss function are configured as the aforementioned three first labeling results, two second labeling results, and four third labeling results. Furthermore, when each parameter in the evaluation model conforms to the aforementioned three first labeling results, two second labeling results, and four third labeling results, the smaller the value of the loss function L(f,y), the better the evaluation effect of the evaluation model.

[0120] Step S108, training the evaluation model according to the sample data and constraints of multiple dimensions to obtain a trained evaluation model for evaluating the target object; the parameters to be trained in the evaluation model include the evaluation scaling parameters of each dimension and the neutral evaluation parameters of each dimension; the neutral evaluation parameters represent the neutral evaluation of the data of the dimension on the sample object.

[0121] The parameters to be trained in the evaluation model include the positive evaluation scaling parameter A of each dimension, the negative evaluation scaling parameter B of each dimension, and the neutral evaluation parameter T of each dimension. Other parameters may also be included, for example, the free search parameter w for characterizing the difference in positive evaluation between dimension 1 and dimension 2. A _{12}, the aforementioned free search parameter w_{33} is used to characterize that on dimension 3, the higher the evaluation score, the more positive the evaluation tends to be, the lower the evaluation score, the more negative the evaluation tends to be, and so on.

[0122] Optionally, the evaluation model is defined in the following manner: for the sub-model score of one dimension of the object to be evaluated, when the sub-model score is greater than or equal to the neutral evaluation parameter, the positive evaluation score is calculated based on the sub-model score, the positive evaluation scaling factor and the neutral evaluation parameter; when the sub-model score is less than the neutral evaluation parameter, the negative evaluation score is calculated based on the sub-model score, the negative evaluation scaling factor and the neutral evaluation parameter; the summary score of each object to be evaluated is calculated based on the sub-model scores of each dimension of each object to be evaluated, and the positive evaluation score or negative evaluation score of each object to be evaluated in each dimension.

[0123] In one embodiment, the evaluation model can be defined by the following formula:

[0124]

[0125] Among them, logit() refers to the logit transformation, which is used to indicate that the value in the brackets is converted from the numerical domain to the logit domain. In the process of optimizing the model parameters of the evaluation model, converting the values ​​to the logit domain through the logit transformation is more conducive to solving the parameters to be trained.

[0126] y can be the total evaluation score obtained by summarizing the evaluation scores of multiple dimensions of an object to be evaluated. i It can be the evaluation score of the i-th dimension of an object to be evaluated. i It can be a neutral evaluation parameter of the i-th dimension.

[0127] In formula (5), y i ∈(0,1), y∈(0,1), t i ∈(0,1)

[0128] In formula (5), the sigmoid function is used. The formula of the sigmoid function is:

[0129] σ(x)=1 / (1+e -x ) (6)

[0130] In one embodiment, a custom function as shown in formula (7) is used in formula (5), where

[0131]

[0132] In formula (6), "e" is the natural constant, x is the independent variable of the sigmoid function, and σ(x) is the dependent variable of the independent variable of the sigmoid function.

[0133] Then in formula (5), the independent variable of the sigmoid function is The dependent variable of the sigmoid function is y.

[0134] In formula (7), a i can be the positive evaluation scaling parameter of the i-th dimension of an object to be evaluated, and b i can be the negative evaluation scaling parameter of the i-th dimension of an object to be evaluated.

[0135] In formula (7), a i ∈(0, 1), b i ∈(0, 1).

[0136] For the sub-model score of a dimension of the object to be evaluated, when the sub-model score y i is greater than or equal to the neutral evaluation parameter T of this dimension, according to the sub-model score y i , the positive evaluation scaling coefficient a i and the neutral evaluation parameter T i , the positive evaluation score Y1 is calculated. The object to be evaluated can be any sample object or other objects evaluated by the evaluation model.

[0137] For example, the sub-model score y1 of the business status dimension of sample object 1 is 60, and the neutral evaluation parameter T1 of the business status dimension is 55. When y1 ≥ T1, referring to formula (5) and formula (7), the positive evaluation score Y1 of sample object 1 is (y1 - t1) * a1.

[0138] Due to a similar technical concept, when the sub-model score is less than the neutral evaluation parameter, the negative evaluation score is calculated according to the sub-model score, the negative evaluation scaling coefficient, and the neutral evaluation parameter.

[0139] For example, the sub-model score y1 of the business status dimension of sample object 2 is 24, and the neutral evaluation parameter T1 of the business status dimension is 55. When y1 < T1, referring to formula (5) and formula (7), the positive evaluation score Y1 of sample object 1 is (y1 - t1) * b1.

[0140] Optionally, a summary score of each object to be evaluated is calculated based on the sub-model scores of each dimension of each object to be evaluated, as well as the positive evaluation scores or negative evaluation scores of each object to be evaluated in each dimension, including: determining the numerical domain scores of each dimension of each object to be evaluated based on the positive evaluation scores or negative evaluation scores of each object to be evaluated in each dimension; converting the numerical domain scores into logistic regression domain scores through a logistic regression function; counting the logistic regression domain scores of each dimension of each object to be evaluated to obtain the total logistic regression domain score of each object to be evaluated; and mapping the total logistic regression domain score of each object to be evaluated to a preset threshold interval to obtain the summary score of each object to be evaluated.

[0141] According to the positive evaluation score or negative evaluation score of each object to be evaluated in each dimension, the numerical domain score of each dimension of each object to be evaluated can be determined by summing the positive evaluation score of each object to be evaluated in each dimension and a preset neutral value, and determining the numerical domain score of the object to be evaluated in each dimension. The preset neutral value can be 0.5.

[0142] The numerical domain score is converted to the logistic regression domain score through the logistic regression function. The logistic regression function can be a logit function, which is used to indicate that the value in the brackets is converted from the numerical domain to the logit domain. In the process of optimizing the model parameters of the evaluation model, converting the numerical value to the logit domain through the logit transformation is more conducive to solving the parameters to be trained.

[0143] The logistic regression domain scores of each dimension of each object to be evaluated are counted to obtain the total logistic regression domain score of each object to be evaluated. It can be understood that, for each object to be evaluated, the logistic regression domain scores of each dimension of the object to be evaluated are summed to obtain the total logistic regression domain score of the object to be evaluated.

[0144] The total score of the logistic regression domain of each object to be evaluated is mapped to a preset threshold interval to obtain a summary score of each object to be evaluated. The preset threshold interval can be between 0 and 1. Specifically, the total score of the logistic regression domain can be converted into a value greater than or equal to 0 and less than or equal to 1 by a Sigmoid function, and the converted value is used as the summary score of the object to be evaluated. The summary score of the sample object can be the total evaluation score output by inputting the sub-model scores of each dimension of the sample object into the evaluation model.

[0145] Figure 2 A first schematic diagram of an evaluation model in an object evaluation method provided in one or more embodiments of this specification.

[0146] Reference Figure 2As shown, the sub-model score S1 of the sample object 1 in the first dimension is compared with the neutral evaluation parameter T of the first dimension. If S1 is greater than or equal to the neutral evaluation parameter T of the first dimension, the difference between S1 and T is multiplied by the positive evaluation scaling parameter A; if S1 is less than the neutral evaluation parameter T of the first dimension, the difference between S1 and T is multiplied by the negative evaluation scaling parameter B, and then the corresponding numerical domain score is obtained... The sub-model score Sn of the sample object 1 in the nth dimension is compared with the neutral evaluation parameter T of the first dimension. If Sn is greater than or equal to the neutral evaluation parameter T of the nth dimension, the difference between S1 and T is multiplied by the positive evaluation scaling parameter A; if Sn is less than the neutral evaluation parameter T of the nth dimension, the difference between S1 and T is multiplied by the negative evaluation scaling parameter B, and then the corresponding numerical domain score is obtained.

[0147] Through the logistic regression function, namely the logit function, each numerical domain score is converted into a corresponding logistic regression domain score and summed up to obtain the total evaluation score.

[0148] Figure 3 A second schematic diagram of an evaluation model in an object evaluation method provided in one or more embodiments of this specification.

[0149] like Figure 3 As shown, the sub-model score S1 of the first dimension of sample object 1 and the neutral evaluation parameter T1 of the first dimension are greater than 0.5. When S1 is greater than T1, the evaluation of sample object 1 tends to be positive, when S1 is less than T1, the evaluation of sample object 1 tends to be negative, and when S1 is equal to T1, the evaluation of sample object 1 tends to be neutral, that is, neither good nor bad.

[0150] With a sufficient number of sample objects, the positive evaluation boundary threshold corresponding to A1 of the first dimension can be determined in combination with the positive scaling parameters A1 and T1. Similarly, the negative evaluation boundary threshold corresponding to B1 of the first dimension can be determined. After conversion, T1 can be considered to be moved to the 0.5 neutral line. Then, when the target evaluation score obtained after converting the sub-model score S1 of sample object 1 is greater than 0.5 and less than or equal to the positive evaluation boundary threshold corresponding to A1, the evaluation of sample object 1 tends to be positive. When S1 is less than 0.5 and greater than or equal to the negative evaluation boundary threshold corresponding to B1, the evaluation of sample object 1 tends to be negative. When S1 is equal to T1, the evaluation of sample object 1 tends to be neutral.

[0151] The sub-model score S3 of the third dimension of sample object 1, and the neutral evaluation parameter T3 of the third dimension are less than 0.5. When S3 is greater than T3, the evaluation of sample object 1 tends to be positive, when S3 is less than T3, the evaluation of sample object 1 tends to be negative, and when S3 is equal to T3, the evaluation of sample object 1 tends to be neutral, that is, neither good nor bad.

[0152] With a sufficient number of sample objects, the positive evaluation boundary threshold corresponding to A3 of the third dimension can be determined in combination with the positive scaling parameters A3 and T3. Similarly, the negative evaluation boundary threshold corresponding to B3 of the third dimension can be determined. After conversion, T3 can be regarded as being moved to the 0.5 neutral line. Then, when the target evaluation score obtained after converting the sub-model score S3 of sample object 1 is greater than 0.5 and less than or equal to the positive evaluation boundary threshold corresponding to A3, the evaluation of sample object 1 tends to be positive. When S3 is less than 0.5 and greater than or equal to the negative evaluation boundary threshold corresponding to B3, the evaluation of sample object 1 tends to be negative. When S3 is equal to T3, the evaluation of sample object 1 tends to be neutral.

[0153] like Figure 3 As shown, it can be determined based on prior knowledge that the third dimension is irrelevant to negative evaluations, so B3=0.5.

[0154] Next, the total evaluation score of sample object 1 is generated through the logistic regression function, namely the logit function, and the sigmoid function.

[0155] In the process of model training, in order to meet a i ∈(0,1), b i ∈(0,1), t i ∈(0,1), the following formula can be used:

[0156] a i =sigmoid(alpha) (8)

[0157] b i =sigmoid(beta) (9)

[0158] t i =sigmoid(tau) (10)

[0159] Formulas (8)-(10) all use the sigmoid function as in formula (6).

[0160] In formula (8), alpha is the independent variable of the sigmoid function, which can be a completely free parameter named alpha. The alpha can be obtained through evaluation model training. i is the dependent variable of the sigmoid function.

[0161] In formula (9), beta is the independent variable of the sigmoid function, which can be a completely free parameter named beta. The beta can be obtained through evaluation model training. i is the dependent variable of the sigmoid function.

[0162] In formula (10), tau is the independent variable of the sigmoid function, which can be a completely free parameter named tau. The tau can be obtained through evaluation model training. i is the dependent variable of the sigmoid function.

[0163] In summary, the sigmoid function can be used to limit the value range of the evaluation scaling parameters of each dimension and the neutral evaluation parameters of each dimension.

[0164] Optionally, the evaluation model is trained based on sample data and constraints of multiple dimensions, including: when sample data of at least one dimension of any sample object is missing, replacing the missing sample data with a neutral evaluation parameter; and training the evaluation model based on sample data of multiple dimensions, neutral evaluation parameters and constraints.

[0165] In the process of obtaining sample data of multiple sample objects in various dimensions, there may be missing data in individual dimensions of individual sample objects. In this case, when calculating the total evaluation score of the sample object, the missing data of the dimension can be replaced by the neutral evaluation parameter of the dimension with missing data in formula (7).

[0166] For example, for the sample object "Organization 2",

[0167] Experience status dimension y1 = 52 points

[0168] Relationship dimension y2 = 73 points

[0169] Institutional qualification dimension y3 = 27 points

[0170] Public opinion information dimension y4 = missing

[0171] Illegal information dimension y5 = 49 points

[0172] For y4 with missing data, when calculating the total evaluation score of "Organization 2", we can set y4=t4, and then substitute the value of y1-y5 into formula (5) to calculate the total evaluation score y of "Organization 2".

[0173] The sample scores of sample data of multiple dimensions may be sample data of multiple dimensions.

[0174] The sample data of multiple dimensions obtained in step S102 are input into the evaluation model of the aforementioned formula (5)-formula (10) to perform model training. During the model training process, the parameters to be trained all meet the constraints determined in step S106.

[0175] For any dimension, the positive evaluation scaling parameter A and the neutral evaluation parameter T of the dimension can be combined to determine the boundary threshold of the positive evaluation of the data of the dimension on the sample object; for any dimension, the negative evaluation scaling parameter B and the neutral evaluation parameter T of the dimension can be combined to determine the boundary threshold of the negative evaluation of the data of the dimension on the sample object. This can be specifically explained in combination with formula (5) and formula (7).

[0176] As shown in formula (7), for any dimension, for example, dimension 1, if the evaluation score y1 of the sample object in dimension 1 is greater than the neutral evaluation parameter t1 of the dimension, then the custom function The value of is the positive evaluation scaling parameter a1 of dimension 1; if the evaluation score y1 of the sample object in dimension 1 is less than or equal to the neutral evaluation parameter t1 of the dimension, then the custom function The value of is the negative evaluation scaling parameter b1 of dimension 1.

[0177] The sample object is institution 1, the dimension is the experience status dimension, and the evaluation score of institution 1 in the experience status dimension is y i Take this as an example to explain:

[0178] If the evaluation score of institution 1 in the experience status dimension is y i Greater than the neutral evaluation parameter t of this dimension i , then first determine the difference between the evaluation score of this dimension and the neutral evaluation parameter (y i -t i ), and then determine (y i -t i ) and a i The product of the product and then the sum of the product and 0.5 can be determined as, when the neutral evaluation score t of the experience status dimension is i After being aligned to 0.5, the evaluation score of the institution can be referred to as the target evaluation score of institution 1 in the experience status dimension for the sake of distinction.

[0179] Similarly, if the evaluation score y of institution 1 in the experience status dimension i Less than or equal to the neutral evaluation parameter t of this dimension i , then first determine the difference between the evaluation score of this dimension and the neutral evaluation parameter (y i -t i ), and then determine (y i -t i ) and b i The product of the product and then the sum of the product and 0.5 can be determined as, when the neutral evaluation score t of the experience status dimension is i After alignment to 0.5, the target evaluation score of institution 1 in the experience status dimension.

[0180] The processing process of other sample objects and sample data of each dimension of each sample object is similar to the processing process of the evaluation score of institution 1 in the experience status dimension, and will not be repeated here.

[0181] Then, for a sample object, Obtain the neutral evaluation score t of any dimension i After alignment to 0.5, the evaluation score of the sample object, and then for multiple sample objects, can be obtained by Obtain the neutral evaluation score t of any dimension i After alignment to 0.5, the value range of the evaluation score of the sample object. When the number of sample objects is large enough, the upper limit of the value range can reflect the boundary threshold of the positive evaluation of the sample object by the data of this dimension, and the lower limit of the value range can reflect the boundary threshold of the negative evaluation of the sample object by the data of this dimension. A sufficient number can be a pre-set object number threshold. When the number of sample objects is greater than the pre-set object number threshold, the number of sample objects is considered to be large enough.

[0182] Based on the same technical concept, in another embodiment, the custom function shown in formula (7) is not executed in formula (5), but a custom function as shown in formula (11) is used, where:

[0183]

[0184] And formula (11) satisfies the constraint condition 0≤b i ≤t i ≤a i ≤1.

[0185] Then, combining formula (5) and formula (11), we can get the neutral evaluation score t of any dimension: i After alignment to 0.5, the target evaluation score of the sample object in this dimension. The specific derivation process can be referred to the above description of the combination of formula (5) and formula (7). For any dimension, the positive evaluation scaling parameter A and the neutral evaluation parameter T of the dimension can be combined to determine the boundary threshold of the positive evaluation of the data of the dimension to the sample object; for any dimension, the negative evaluation scaling parameter B and the neutral evaluation parameter T of the dimension can be combined to determine the boundary threshold of the negative evaluation of the data of the dimension to the sample object.

[0186] For example, the target evaluation score of the sample object in any dimension can be obtained by combining formula (5) and formula (11):

[0187] In y i = 1, through The evaluation score y of the sample object in any dimension can be obtained i The corresponding target evaluation score is a i ;

[0188] In y i = 0, through The evaluation score y of the sample object in any dimension can be obtained i The corresponding target evaluation score is b i ;

[0189] In y i = 0.5, through The evaluation score y of the sample object in any dimension can be obtained i The corresponding target evaluation score is 0.5;

[0190] In y i >t i In the case of The evaluation score y of the sample object in any dimension can be obtained i The corresponding target evaluation score belongs to (0.5, a i ];

[0191] In y i <t i In the case of The evaluation score y of the sample object in any dimension can be obtained i The corresponding target evaluation score belongs to [b i , 0.5).

[0192] Combined with the above examples, it can be seen that in the embodiment where formula (5) uses a custom function such as formula (11), when the number of sample data of the sample objects obtained is sufficient and there is both a minimum value of 0 and a maximum value of 1, for any dimension, a i It can be the boundary threshold of the positive evaluation of the sample object by the data of this dimension, b i It can be the boundary threshold of the negative evaluation of the data of this dimension on the sample object.

[0193] Optionally, the evaluation model is trained according to sample data and constraints in multiple dimensions, including: in the process of training the evaluation model, parameter optimization processing is performed on the parameters to be trained in the evaluation model through the Adam parameter search optimization method.

[0194] During the training process of the evaluation model, the Adam parameter search optimization scheme can be combined to optimize the parameters to be trained in the evaluation model.

[0195] Adam refers to an adaptive motion estimation algorithm that uses the negative gradient of the objective function to locate the minimum of the function, automatically adjusts the learning rate for each input variable of the objective function, and updates the variables by using a moving average that exponentially decreases the gradient.

[0196] Optionally, the target object includes the institution to be risk controlled; each dimension includes at least one of an operating status dimension, an associated relationship dimension, an institution qualification dimension, a public opinion information dimension, and a violation information dimension.

[0197] After obtaining the trained evaluation model for evaluating the target object, the evaluation scores of the target object to be evaluated, for example, the risk control institution, in each dimension can be input into the trained evaluation model to obtain the total evaluation score of the target object as the evaluation result. Each dimension can include at least one of the operating status dimension, the association relationship dimension, the institution qualification dimension, the public opinion information dimension, and the violation information dimension.

[0198] For example, risk control agency X:

[0199] Experience dimension: 51 points

[0200] Relationship dimension: 63 points

[0201] Institutional qualification dimension: 23 points

[0202] Public opinion information dimension: 63 points

[0203] Violation information dimension: 74 points

[0204] After the evaluation scores of the five dimensions of the institution X to be risk controlled are input into the trained evaluation model, the total evaluation score S of the institution X is output. The total evaluation score S can accurately and objectively reflect the comprehensive evaluation result of the institution X to be risk controlled.

[0205] The following takes the application of an object evaluation method provided in this embodiment in a blockchain risk control scenario as an example to further illustrate the object evaluation method provided in this embodiment. Figure 4 A processing flow chart of an object evaluation method applied to a blockchain risk control scenario provided for one or more embodiments of this specification, specifically including steps S402 to S412.

[0206] Step S402, obtaining sample data of each dimension from the blockchain.

[0207] Step S404: annotating prior knowledge.

[0208] Step S406, generating a loss function under constraint conditions.

[0209] Step S408: establishing an evaluation model based on the constraint conditions of the evaluation scaling parameters and the neutral evaluation parameters.

[0210] Step S410, combining Adam parameter search optimization, training each parameter to be trained in the evaluation model.

[0211] Step S412: deploy the trained evaluation model to the blockchain node.

[0212] Since the technical concept of this embodiment is the same or similar to that of the aforementioned object evaluation method embodiment, the description is relatively simple, and the relevant parts may refer to the corresponding description of the aforementioned object evaluation method embodiment.

[0213] An example of a risk assessment method provided in this specification is as follows:

[0214] Figure 5 A risk assessment method processing flow chart provided for one or more embodiments of this specification.

[0215] Step S502, obtaining sub-model scores of at least two dimensions of the target object.

[0216] Target objects include but are not limited to: institutions, organizations, commodities, non-commercial items, persons, places, etc.

[0217] Taking the target object as an institution as an example, at least two dimensions may include the experience status dimension, the relationship dimension, the institution qualification dimension, the public opinion information dimension, and the violation information dimension.

[0218] Step S504: input the obtained sub-model scores into the trained evaluation model of the aforementioned object evaluation method, and output the total risk evaluation score of the target object.

[0219] The obtained sub-model scores of the N dimensions of the target object are input into the trained evaluation model, and the total risk assessment score of the target object is output. The total risk assessment score can be regarded as a comprehensive evaluation score obtained by combining the sub-model scores of the N dimensions.

[0220] Optionally, the risk assessment method also includes: converting the sub-model scores of each dimension of the target object to obtain the target evaluation scores of the target object in each dimension; the target evaluation scores are used to characterize the evaluation scores of the target object in each dimension when the neutral evaluation parameters of each dimension are aligned with the preset neutral values; and displaying the target evaluation scores of each dimension of the target object.

[0221] The specific calculation process of the target evaluation score can be referred to Figure 1The target evaluation score is used to characterize the evaluation score of the target object in each dimension when the neutral evaluation parameters of each dimension are aligned with the preset neutral value. The preset neutral value can be 0.5. When the target evaluation score is greater than 0.5, the evaluation of the target object tends to be positive. When the target evaluation score is less than 0.5, the evaluation of the target object tends to be negative. When the target evaluation score is equal to 0.5, the evaluation of the target object tends to be neutral.

[0222] Figure 6 A diagram of a score display interface in a risk assessment method provided in one or more embodiments of this specification.

[0223] like Figure 6 As shown, S1-Sn can be the sub-model scores of the target object in n dimensions. For example, n=5, and the five dimensions are experience status dimension, association relationship dimension, institution qualification dimension, public opinion information dimension, and violation information dimension.

[0224] Figure 6 The target evaluation scores obtained by converting the sub-model scores of the above five dimensions are shown, which are the thick lines in the boxes of each dimension in the figure. When the thick line is higher than the neutral line of 0.5, the evaluation of the target object tends to be positive, when the thick line is lower than 0.5, the evaluation of the target object tends to be negative, and when the thick line is equal to 0.5, the evaluation of the target object tends to be neutral.

[0225] Figure 6 The small boxes in the boxes of each dimension are used to show the upper and lower limits of the target evaluation scores of each dimension. Figure 6 The width of the small box is set smaller than the boxes of each dimension just for the sake of easy viewing. In practical applications, different colors can be set in different value areas to highlight the upper and lower limits of the target evaluation score, so as to help users intuitively feel the target evaluation scores of each dimension.

[0226] Figure 6 It also shows the total evaluation score of "90" generated based on the sub-model scores of the above five dimensions of the target object, institution A, and the trained evaluation model.

[0227] Since the technical concept of the risk assessment method embodiment is the same or similar to that of the object assessment method embodiment, the description is relatively simple, and the relevant parts may refer to the corresponding description of the object assessment method embodiment provided above.

[0228] An embodiment of an object evaluation device provided in this specification is as follows:

[0229] In the above-mentioned embodiment, an object evaluation method is provided, and correspondingly, an object evaluation device is also provided, which will be described below with reference to the accompanying drawings.

[0230] Figure 7 A schematic diagram of an object evaluation device provided for one or more embodiments of this specification.

[0231] Since the device embodiment corresponds to the method embodiment, the description is relatively simple, and the relevant parts can refer to the corresponding description of the method embodiment provided above. The device embodiment described below is only illustrative.

[0232] This embodiment provides an object evaluation device, including:

[0233] The sample acquisition module 702 is configured to acquire sample data of multiple sample objects in various dimensions;

[0234] The parameter labeling module 704 is configured to label the evaluation scaling parameters of the sample data of each dimension according to prior knowledge;

[0235] A constraint setting module 706 is configured to set the constraint conditions of the loss function of the evaluation model according to the annotation results;

[0236] The model training module 708 is configured to train the evaluation model according to sample data and constraints in multiple dimensions to obtain a trained evaluation model for evaluating the target object; the parameters to be trained in the evaluation model include evaluation scaling parameters of each dimension and neutral evaluation parameters of each dimension; the neutral evaluation parameters represent the data of neutral evaluation of the sample object in the dimension to which it corresponds.

[0237] An embodiment of a risk assessment device provided in this specification is as follows:

[0238] In the above-mentioned embodiment, a risk assessment method is provided, and correspondingly, a risk assessment device is also provided, which will be described below with reference to the accompanying drawings.

[0239] Figure 8 A schematic diagram of a risk assessment device provided for one or more embodiments of this specification.

[0240] Since the device embodiment corresponds to the method embodiment, the description is relatively simple, and the relevant parts can refer to the corresponding description of the method embodiment provided above. The device embodiment described below is only illustrative.

[0241] This embodiment provides a risk assessment device, including:

[0242] A score acquisition module 802 is configured to acquire sub-model scores of at least two dimensions of a target object;

[0243] The total score output module 804 is configured to input the obtained sub-model scores into the trained evaluation model of the aforementioned risk evaluation method, and output the total risk evaluation score of the target object.

[0244] An embodiment of an object evaluation device provided in this specification is as follows:

[0245] Corresponding to the object evaluation method described above, based on the same technical concept, one or more embodiments of this specification also provide an object evaluation device, which is used to execute the object evaluation method provided above. Fig. 9 A schematic diagram of the structure of an object evaluation device provided in one or more embodiments of this specification.

[0246] This embodiment provides an object evaluation device, including:

[0247] like Fig. 9 As shown, the object evaluation device may have relatively large differences due to different configurations or performances, and may include one or more processors 901 and memory 902, and the memory 902 may store one or more storage applications or data. Among them, the memory 902 may be a temporary storage or a permanent storage. The application stored in the memory 902 may include one or more modules (not shown in the figure), and each module may include a series of computer executable instructions in the object evaluation device. Furthermore, the processor 901 may be configured to communicate with the memory 902 and execute a series of computer executable instructions in the memory 902 on the object evaluation device. The object evaluation device may also include one or more power supplies 903, one or more wired or wireless network interfaces 904, one or more input / output interfaces 905, one or more keyboards 906, etc.

[0248] In a specific embodiment, the object evaluation device includes a memory and one or more programs, wherein the one or more programs are stored in the memory, and the one or more programs may include one or more modules, and each module may include a series of computer executable instructions in the object evaluation device, and the one or more programs configured to be executed by one or more processors include computer executable instructions for performing the following:

[0249] Get sample data of multiple sample objects in various dimensions;

[0250] The evaluation scaling parameters of sample data of each dimension are labeled based on prior knowledge;

[0251] Set the constraints of the loss function of the evaluation model according to the annotation results;

[0252] The evaluation model is trained according to sample data and constraints in multiple dimensions to obtain a trained evaluation model for evaluating the target object; the parameters to be trained in the evaluation model include evaluation scaling parameters of each dimension and neutral evaluation parameters of each dimension; the neutral evaluation parameters represent the data of neutral evaluation of the sample object in the dimension to which it corresponds.

[0253] An embodiment of a storage medium provided in this specification is as follows:

[0254] Corresponding to the object evaluation method described above, based on the same technical concept, one or more embodiments of this specification also provide a storage medium.

[0255] The storage medium provided in this embodiment is used to store computer executable instructions. When the computer executable instructions are executed by a processor, the following process is implemented:

[0256] Get sample data of multiple sample objects in various dimensions;

[0257] The evaluation scaling parameters of sample data of each dimension are labeled based on prior knowledge;

[0258] Set the constraints of the loss function of the evaluation model according to the annotation results;

[0259] The evaluation model is trained according to sample data and constraints in multiple dimensions to obtain a trained evaluation model for evaluating the target object; the parameters to be trained in the evaluation model include evaluation scaling parameters of each dimension and neutral evaluation parameters of each dimension; the neutral evaluation parameters represent the data of neutral evaluation of the sample object in the dimension to which it corresponds.

[0260] It should be noted that the embodiments of the storage medium in this specification and the embodiments of the object evaluation method in this specification are based on the same inventive concept, so the specific implementation of this embodiment can refer to the implementation of the aforementioned corresponding method, and the repeated parts will not be repeated.

[0261] An embodiment of a risk assessment device provided in this specification is as follows:

[0262] Corresponding to the risk assessment method described above, based on the same technical concept, one or more embodiments of this specification also provide a risk assessment device, which is used to execute the risk assessment method provided above. Fig.10 A schematic diagram of the structure of a risk assessment device provided in one or more embodiments of this specification.

[0263] This embodiment provides a risk assessment device, including:

[0264] like Fig.10As shown, the risk assessment device may have relatively large differences due to different configurations or performances, and may include one or more processors 1001 and memory 1002, and the memory 1002 may store one or more storage applications or data. Among them, the memory 1002 may be a short-term storage or a persistent storage. The application stored in the memory 1002 may include one or more modules (not shown in the figure), and each module may include a series of computer executable instructions in the risk assessment device. Furthermore, the processor 1001 may be configured to communicate with the memory 1002 and execute a series of computer executable instructions in the memory 1002 on the risk assessment device. The risk assessment device may also include one or more power supplies 1003, one or more wired or wireless network interfaces 1004, one or more input / output interfaces 1005, one or more keyboards 1006, etc.

[0265] In a specific embodiment, the risk assessment device includes a memory and one or more programs, wherein the one or more programs are stored in the memory, and the one or more programs may include one or more modules, and each module may include a series of computer executable instructions in the risk assessment device, and the one or more programs are configured to be executed by one or more processors, including computer executable instructions for performing the following:

[0266] Obtain sub-model scores of at least two dimensions of the target object;

[0267] The obtained sub-model scores are input into the trained evaluation model of the aforementioned object evaluation method, and the total risk evaluation score of the target object is output.

[0268] An embodiment of a storage medium provided in this specification is as follows:

[0269] Corresponding to the risk assessment method described above, based on the same technical concept, one or more embodiments of this specification also provide a storage medium.

[0270] The storage medium provided in this embodiment is used to store computer executable instructions. When the computer executable instructions are executed by a processor, the following process is implemented:

[0271] Obtain sub-model scores of at least two dimensions of the target object;

[0272] The obtained sub-model scores are input into the trained evaluation model of the aforementioned object evaluation method, and the total risk evaluation score of the target object is output.

[0273] It should be noted that the embodiments of the storage medium in this specification and the embodiments of the risk assessment method in this specification are based on the same inventive concept, so the specific implementation of this embodiment can refer to the implementation of the aforementioned corresponding method, and the repeated parts will not be repeated.

[0274] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0275] In the 1930s, improvements to a technology could be clearly distinguished as hardware improvements (for example, improvements to the circuit structure of diodes, transistors, switches, etc.) or software improvements (improvements to the method flow). However, with the development of technology, many improvements to the method flow today can be regarded as direct improvements to the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved method flow into the hardware circuit. Therefore, it cannot be said that an improvement in a method flow cannot be implemented using a hardware entity module. For example, a programmable logic device (PLD) (such as a field programmable gate array (FPGA)) is such an integrated circuit whose logical function is determined by the user's programming of the device. Designers can "integrate" a digital system on a PLD by programming it themselves, without having to ask a chip manufacturer to design and produce a dedicated integrated circuit chip. Moreover, nowadays, instead of manually making integrated circuit chips, this kind of programming is mostly implemented by "logic compiler" software, which is similar to the software compiler used when developing and writing programs, and the original code before compilation must also be written in a specific programming language, which is called hardware description language (HDL). There is not only one HDL, but many kinds, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. The most commonly used ones are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also know that it is only necessary to program the method flow slightly in the above-mentioned hardware description languages ​​and program it into the integrated circuit, and then it is easy to obtain the hardware circuit that implements the logic method flow.

[0276] The controller can be implemented in any appropriate manner, for example, the controller can take the form of a microprocessor or processor and a computer-readable medium storing a computer-readable program code (such as software or firmware) that can be executed by the (micro)processor, a logic gate, a switch, an application-specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller. Examples of controllers include, but are not limited to, the following microcontrollers: ARC625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320. The memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art also know that in addition to implementing the controller in a purely computer-readable program code manner, the controller can be implemented in the form of a logic gate, a switch, an application-specific integrated circuit, a programmable logic controller, and an embedded microcontroller by logically programming the method steps. Therefore, this controller can be considered as a hardware component, and the devices included therein for implementing various functions can also be regarded as structures within the hardware component. Or even, the devices for implementing various functions can be regarded as both software modules for implementing the method and structures within the hardware component.

[0277] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0278] For the convenience of description, the above devices are described in terms of functions and are divided into various units. Of course, when implementing the embodiments of this specification, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0279] It should be understood by those skilled in the art that one or more embodiments of this specification may be provided as a method, system or computer program product. Therefore, one or more embodiments of this specification may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0280] This specification is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of this specification. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable object evaluation device to produce a machine, so that the instructions executed by the processor of the computer or other programmable object evaluation device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0281] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable object evaluation device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0282] These computer program instructions may also be loaded onto a computer or other programmable object evaluation device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable device to implement the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0283] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0284] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0285] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0286] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0287] One or more embodiments of the present specification may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. One or more embodiments of the present specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0288] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0289] The above description is only an embodiment of this document and is not intended to limit this document. For those skilled in the art, this document may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this document should be included in the scope of the claims of this document.

Claims

1. A method for object evaluation, comprising: Get sample data of multiple sample objects in various dimensions; Annotating evaluation scaling parameters of sample data of each dimension according to prior knowledge; the evaluation scaling parameters include a positive evaluation scaling factor and a negative evaluation scaling factor; Set the constraints of the loss function of the evaluation model according to the annotation results; Training the evaluation model according to the sample data of multiple dimensions and the constraint conditions to obtain the trained evaluation model for evaluating the target object; The parameters to be trained in the evaluation model include the evaluation scaling parameters of each dimension and the neutral evaluation parameters of each dimension; the neutral evaluation parameters represent the data of the neutral evaluation of the sample object in the dimension to which they correspond; the evaluation model is defined in the following manner: for a sub-model score of a dimension of the object to be evaluated, when the sub-model score is greater than or equal to the neutral evaluation parameter, the positive evaluation score is calculated according to the sub-model score, the positive evaluation scaling coefficient and the neutral evaluation parameter; When the sub-model score is less than the neutral evaluation parameter, a negative evaluation score is calculated according to the sub-model score, the negative evaluation scaling factor and the neutral evaluation parameter; Calculate the summary score of each object to be evaluated according to the sub-model scores of each dimension of each object to be evaluated, and the positive evaluation score or the negative evaluation score of each object to be evaluated in each dimension; The step of labeling the evaluation scaling parameters of the sample data of each dimension according to prior knowledge includes: The prior knowledge is converted into a first annotation result, and / or a second annotation result, and / or a third annotation result; the first annotation result represents a comparison result of the positive evaluation scaling coefficients of any two dimensions among the dimensions, the second annotation result represents a comparison result of the negative evaluation scaling coefficients of any two dimensions among the dimensions, and the third annotation result represents a comparison result of the positive evaluation scaling coefficient and the negative evaluation scaling coefficient of each dimension.

2. The method according to claim 1, before obtaining sample data of multiple sample objects in each dimension, further comprising: Each of the plurality of first blockchain nodes evaluates the dimension corresponding to each of the sample objects in the first blockchain node through a pre-deployed first smart contract, obtains a sub-model score of the dimension corresponding to each of the sample objects in the first blockchain node, and uploads the sub-model score in the blockchain; Each dimension corresponds to at least one of the first blockchain nodes; The obtaining of sample data of multiple sample objects in various dimensions includes: The second blockchain node obtains the sub-model scores of each of the sample objects in each dimension from the blockchain through a pre-deployed second smart contract, and determines the sub-model scores as the sample data.

3. The method according to claim 1, wherein the summary score of each object to be evaluated is calculated based on the sub-model scores of each dimension of each object to be evaluated, and the positive evaluation scores or the negative evaluation scores of each object to be evaluated in each dimension, comprising: Determine the numerical domain score of each dimension of each object to be evaluated according to the positive evaluation score or the negative evaluation score of each object to be evaluated in each dimension; converting the numerical domain score into a logistic regression domain score by a logistic regression function; Counting the logistic regression domain scores of each dimension of each object to be evaluated to obtain a total logistic regression domain score of each object to be evaluated; The total score of the logistic regression domain of each object to be evaluated is mapped to a preset threshold range to obtain a summary score of each object to be evaluated.

4. The method according to claim 1, wherein the training of the evaluation model based on the sample data and the constraint conditions in multiple dimensions comprises: In the case where sample data of at least one dimension of any one of the sample objects is missing, the missing sample data is replaced by the neutral evaluation parameter; The evaluation model is trained according to the sample data of multiple dimensions, the neutral evaluation parameters and the constraint conditions.

5. According to the method of claim 1, the evaluation scaling parameters include a positive evaluation scaling factor and a negative evaluation scaling factor; the step of labeling the evaluation scaling parameters of sample data of each dimension according to prior knowledge comprises: If it is determined according to the prior knowledge that any dimension is irrelevant to negative evaluation, the negative evaluation scaling factor of the dimension is set to a preset neutral value; If it is determined according to the prior knowledge that any dimension is irrelevant to positive evaluation, the positive evaluation scaling factor of the dimension is set to a preset neutral value.

6. The method according to claim 1, characterized in that The step of training the evaluation model according to the sample data of multiple dimensions and the constraint conditions includes: In the process of training the evaluation model, the parameters to be trained in the evaluation model are optimized by using the Adam parameter search optimization method.

7. According to the method of claim 1, the target object includes the institution to be risk controlled; each dimension includes at least one of the operating status dimension, the relationship dimension, the institution qualification dimension, the public opinion information dimension and the violation information dimension.

8. A risk assessment method comprising: Obtain sub-model scores of at least two dimensions of the target object; The obtained sub-model score is input into the trained evaluation model as described in any one of claims 1 to 7, and the total risk evaluation score of the target object is output.

9. The method according to claim 8, further comprising: Converting the sub-model scores of each dimension of the target object to obtain the target evaluation scores of the target object in each dimension; The target evaluation score is used to represent the evaluation score of the target object in each dimension when the neutral evaluation parameters of each dimension are aligned with the preset neutral value; Display the target evaluation scores of various dimensions of the target object.

10. An object evaluation device, comprising: A sample acquisition module is configured to acquire sample data of multiple sample objects in various dimensions; A parameter labeling module is configured to label evaluation scaling parameters of sample data of each dimension according to prior knowledge; the evaluation scaling parameters include a positive evaluation scaling factor and a negative evaluation scaling factor; A constraint setting module, configured to set constraint conditions of a loss function of an evaluation model according to the annotation results; A model training module is configured to train the evaluation model according to the sample data of multiple dimensions and the constraint conditions to obtain the trained evaluation model for evaluating the target object; The parameters to be trained in the evaluation model include the evaluation scaling parameters of each dimension and the neutral evaluation parameters of each dimension; the neutral evaluation parameters represent the data of the neutral evaluation of the sample object in the dimension to which they correspond; the evaluation model is defined in the following manner: for a sub-model score of a dimension of the object to be evaluated, when the sub-model score is greater than or equal to the neutral evaluation parameter, the positive evaluation score is calculated according to the sub-model score, the positive evaluation scaling coefficient and the neutral evaluation parameter; When the sub-model score is less than the neutral evaluation parameter, a negative evaluation score is calculated according to the sub-model score, the negative evaluation scaling factor and the neutral evaluation parameter; Calculate the summary score of each object to be evaluated according to the sub-model scores of each dimension of each object to be evaluated, and the positive evaluation score or the negative evaluation score of each object to be evaluated in each dimension; The step of labeling the evaluation scaling parameters of the sample data of each dimension according to prior knowledge includes: The prior knowledge is converted into a first annotation result, and / or a second annotation result, and / or a third annotation result; the first annotation result represents a comparison result of the positive evaluation scaling coefficients of any two dimensions among the dimensions, the second annotation result represents a comparison result of the negative evaluation scaling coefficients of any two dimensions among the dimensions, and the third annotation result represents a comparison result of the positive evaluation scaling coefficient and the negative evaluation scaling coefficient of each dimension.

11. A risk assessment device, comprising: A score acquisition module, configured to acquire sub-model scores of at least two dimensions of a target object; The total score output module is configured to input the obtained sub-model score into the trained evaluation model as described in any one of claims 1-7, and output the total risk evaluation score of the target object.

12. An object evaluation device, comprising: processor; as well as, a memory configured to store computer executable instructions which, when executed, cause the processor to: Get sample data of multiple sample objects in various dimensions; Annotating evaluation scaling parameters of sample data of each dimension according to prior knowledge; the evaluation scaling parameters include a positive evaluation scaling factor and a negative evaluation scaling factor; Set the constraints of the loss function of the evaluation model according to the annotation results; Training the evaluation model according to the sample data of multiple dimensions and the constraint conditions to obtain the trained evaluation model for evaluating the target object; The parameters to be trained in the evaluation model include the evaluation scaling parameters of each dimension and the neutral evaluation parameters of each dimension; the neutral evaluation parameters represent the data of the neutral evaluation of the sample object in the dimension to which they correspond; the evaluation model is defined in the following manner: for a sub-model score of a dimension of the object to be evaluated, when the sub-model score is greater than or equal to the neutral evaluation parameter, the positive evaluation score is calculated according to the sub-model score, the positive evaluation scaling coefficient and the neutral evaluation parameter; When the sub-model score is less than the neutral evaluation parameter, a negative evaluation score is calculated according to the sub-model score, the negative evaluation scaling factor and the neutral evaluation parameter; Calculate the summary score of each object to be evaluated according to the sub-model scores of each dimension of each object to be evaluated, and the positive evaluation score or the negative evaluation score of each object to be evaluated in each dimension; The step of labeling the evaluation scaling parameters of the sample data of each dimension according to prior knowledge includes: The prior knowledge is converted into a first annotation result, and / or a second annotation result, and / or a third annotation result; the first annotation result represents a comparison result of the positive evaluation scaling coefficients of any two dimensions among the dimensions, the second annotation result represents a comparison result of the negative evaluation scaling coefficients of any two dimensions among the dimensions, and the third annotation result represents a comparison result of the positive evaluation scaling coefficient and the negative evaluation scaling coefficient of each dimension.

13. A risk assessment device comprising: processor; as well as, a memory configured to store computer executable instructions which, when executed, cause the processor to: Obtain sub-model scores of at least two dimensions of the target object; The obtained sub-model score is input into the trained evaluation model as described in any one of claims 1 to 7, and the total risk evaluation score of the target object is output.

14. A storage medium for storing computer executable instructions, wherein the computer executable instructions, when executed by a processor, implement the following process: Get sample data of multiple sample objects in various dimensions; Annotating evaluation scaling parameters of sample data of each dimension according to prior knowledge; the evaluation scaling parameters include a positive evaluation scaling factor and a negative evaluation scaling factor; Set the constraints of the loss function of the evaluation model according to the annotation results; Training the evaluation model according to the sample data of multiple dimensions and the constraint conditions to obtain the trained evaluation model for evaluating the target object; The parameters to be trained in the evaluation model include the evaluation scaling parameters of each dimension and the neutral evaluation parameters of each dimension; the neutral evaluation parameters represent the data of the neutral evaluation of the sample object in the dimension to which they correspond; the evaluation model is defined in the following manner: for a sub-model score of a dimension of the object to be evaluated, when the sub-model score is greater than or equal to the neutral evaluation parameter, the positive evaluation score is calculated according to the sub-model score, the positive evaluation scaling coefficient and the neutral evaluation parameter; When the sub-model score is less than the neutral evaluation parameter, a negative evaluation score is calculated according to the sub-model score, the negative evaluation scaling factor and the neutral evaluation parameter; Calculate the summary score of each object to be evaluated according to the sub-model scores of each dimension of each object to be evaluated, and the positive evaluation score or the negative evaluation score of each object to be evaluated in each dimension; The step of labeling the evaluation scaling parameters of the sample data of each dimension according to prior knowledge includes: The prior knowledge is converted into a first annotation result, and / or a second annotation result, and / or a third annotation result; the first annotation result represents a comparison result of the positive evaluation scaling coefficients of any two dimensions among the dimensions, the second annotation result represents a comparison result of the negative evaluation scaling coefficients of any two dimensions among the dimensions, and the third annotation result represents a comparison result of the positive evaluation scaling coefficient and the negative evaluation scaling coefficient of each dimension.

15. A storage medium for storing computer executable instructions, wherein the computer executable instructions, when executed by a processor, implement the following process: Obtain sub-model scores of at least two dimensions of the target object; The obtained sub-model score is input into the trained evaluation model as described in any one of claims 1 to 7, and the total risk evaluation score of the target object is output.

Images