Scene analysis method and device, storage medium and electronic device

Abstract

The present application discloses a scenario analysis method and device, a storage medium and an electronic device, and relates to the field of artificial intelligence. The method comprises: converting a deep learning model of each target object in a target scenario into a first decision tree model, and determining a first feature of the first decision tree model, wherein the target scenario is a scenario for predicting the default risk of multiple target objects, and the first feature is a feature of a target object whose impact on the default risk is greater than a preset degree; constructing a second decision tree model based on the first feature of the first decision tree model; expressing the second decision tree model in the form of Boolean logic, and calculating the value of the target feature of each target object based on Boolean logic; analyzing the target scenario based on the value of the target feature of each target object. Through this application, the problem that the deep learning model is used in the related art to predict the default risk of customers and it is difficult to analyze the scenario of default risk is solved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence, and specifically, to a scene analysis method and device, a storage medium, and an electronic device. Background Art

[0002] In the current related technologies, in order to further improve the accuracy of business decisions, business personnel have gradually introduced deep learning models. That is, by inputting a large number of features into the neural network deep learning, training the model, and applying the model to actual business decisions such as risk default prediction and suspicious transaction screening.

[0003] However, the current related technologies have the following disadvantages:

[0004] First, there is a black box phenomenon in deep learning models. The black box refers to a phenomenon in which one does not have a complete understanding of the principles and processes of the tool used to achieve the purpose. For example, for deep learning models, we cannot fully judge the output of the model results, which input feature variables are the most important, and how the features affect the changes in the results. Therefore, the deep learning model only outputs the results, but cannot intuitively tell the user which features have the greatest impact on the results, and whether the impact of the features on the results is positive or negative, like rules or simple linear models, making it inconvenient for users to analyze the causes of the results in actual use.

[0005] Second, it is impossible to implement scenario analysis. In actual scenarios, users often need to conduct scenario analysis to simulate changes in results. For example, in the default risk scenario analysis, assuming that the customer changes from default to non-default, or from non-default to default, how will the overall risk level change. However, due to the black box phenomenon of deep learning, users cannot accurately control the output results of the model by controlling the variables of the input feature values.

[0006] Therefore, due to the black box problem of deep learning models, it is impossible to better help users make business decisions.

[0007] There is currently no effective solution to the problem that deep learning models are used in related technologies to predict customer default risks, but it is difficult to analyze default risk scenarios. Summary of the invention

[0008] The main purpose of this application is to provide a scenario analysis method and device, a storage medium and an electronic device to solve the problem in the related art that deep learning models are used to predict customer default risks and it is difficult to analyze the scenarios of default risks.

[0009] In order to achieve the above-mentioned purpose, according to one aspect of the present application, a scenario analysis method is provided. The method includes: converting the deep learning model of each target object in the target scenario into a first decision tree model, and determining the first feature of the first decision tree model, wherein the target scenario is a scenario for predicting the default risk of multiple target objects, and the first feature is a feature of the target object whose impact on the default risk is greater than a preset degree; constructing a second decision tree model based on the first feature of the first decision tree model; expressing the second decision tree model in the form of Boolean logic, and calculating the value of the target feature of each target object based on the Boolean logic, wherein the target feature is the feature after the first feature changes, and is the feature that changes the default risk prediction result; analyzing the target scenario based on the value of the target feature of each target object.

[0010] Furthermore, according to the value of the target characteristic of each target object, analyzing the target scenario includes: performing a default risk analysis on each target object according to the value of the target characteristic of each target object, and obtaining a plurality of first analysis results; according to each first analysis result, clustering the plurality of target objects by using a clustering method, and obtaining a plurality of clusters, wherein the cluster is a collection of target objects with the same characteristic changes; and analyzing the target scenario according to each cluster.

[0011] Furthermore, based on the value of the target characteristic of each target object, performing a default risk analysis on each target object includes: obtaining the value of the target characteristic of each target object and the value of the first characteristic; determining the value of the characteristic of each target object after the characteristic is changed based on the value of the target characteristic of each target object and the value of the first characteristic; performing a default risk analysis on each target object based on the value of the characteristic after the characteristic is changed.

[0012] Furthermore, analyzing the target scene based on each cluster includes: determining the value of the feature after the feature change of each cluster; obtaining a second feature based on the value of the feature after the feature change of each cluster, wherein the second feature is a feature in the target feature; and analyzing the target scene based on the second feature.

[0013] Furthermore, converting the deep learning model of each target object in the target scene into a first decision tree model includes: obtaining a third eigenvalue of each target object, wherein the third eigenvalue is a numerical value corresponding to the characteristic of each target object; perturbing the third eigenvalue a preset number of times to obtain a fourth eigenvalue; inputting the fourth eigenvalue into the deep learning model to obtain a first default risk prediction result; and performing learning and training on the first default risk prediction result to obtain the first decision tree model.

[0014] Furthermore, constructing a second decision tree model based on the first feature of the first decision tree model includes: performing perturbation processing on the eigenvalue of the first feature of the first decision tree model to obtain a fifth eigenvalue; converting the first default risk prediction result according to a preset rule to obtain a second default risk prediction result; and constructing the second decision tree model based on the fifth eigenvalue and the second default risk prediction result.

[0015] Furthermore, according to the Boolean logic, calculating the value of the target feature of each target object includes: determining the problem of the Boolean logic; solving the problem of the Boolean logic to obtain the value of the target feature of each target object.

[0016] Furthermore, before determining the first feature of the first decision tree model, the method also includes: determining the importance of each feature in the first decision tree model; and determining the first feature of each first decision tree model based on the importance of each feature and the number of first features of each first decision tree model.

[0017] Furthermore, after analyzing the target scene according to the value of the target feature of each target object, the method further includes: obtaining a second analysis result of analyzing the target scene; and determining a target strategy according to the second analysis result.

[0018] In order to achieve the above-mentioned purpose, according to another aspect of the present application, a scene analysis device is provided. The device includes: a first conversion unit, which is used to convert the deep learning model of each target object in the target scene into a first decision tree model, and determine the first feature of the first decision tree model, wherein the target scene is a scene for predicting the default risk of multiple target objects, and the first feature is a feature of the target object whose impact on the default risk is greater than a preset degree; a first construction unit, which is used to construct a second decision tree model based on the first feature of the first decision tree model; a first processing unit, which is used to express the second decision tree model in the form of Boolean logic, and calculate the value of the target feature of each target object based on the Boolean logic, wherein the target feature is the feature after the first feature changes, and is the feature that changes the default risk prediction result; a first analysis unit, which is used to analyze the target scene based on the value of the target feature of each target object.

[0019] Furthermore, the first analysis unit includes: a first analysis module, used to perform default risk analysis on each target object according to the value of the target feature of each target object, and obtain multiple first analysis results; a first clustering module, used to cluster the multiple target objects according to each first analysis result and use a clustering method to obtain multiple clusters, wherein the cluster is a collection of target objects with the same feature changes; a second analysis module, used to analyze the target scenario based on each cluster.

[0020] Furthermore, the first analysis module includes: a first acquisition submodule, used to obtain the value of the target feature and the value of the first feature of each target object; a first determination submodule, used to determine the value of the feature of each target object after the feature is changed based on the value of the target feature and the value of the first feature of each target object; and a first analysis submodule, used to perform a default risk analysis on each target object based on the value of the feature after the feature is changed.

[0021] Furthermore, the second analysis module includes: a second determination submodule, used to determine the value of the feature after the feature change of each cluster; a first processing submodule, used to obtain a second feature based on the value of the feature after the feature change of each cluster, wherein the second feature is a feature in the target feature; and a second analysis submodule, used to analyze the target scene based on the second feature.

[0022] Furthermore, the first conversion unit includes: a first acquisition module, used to obtain a third eigenvalue of each target object, wherein the third eigenvalue is a numerical value corresponding to the characteristic of each target object; a first processing module, used to perturb the third eigenvalue according to a preset number of times to obtain a fourth eigenvalue; a first input module, used to input the fourth eigenvalue into the deep learning model to obtain a first default risk prediction result; and a first training module, used to learn and train the first default risk prediction result to obtain the first decision tree model.

[0023] Furthermore, the first construction unit includes: a second processing module, used to perform perturbation processing on the eigenvalue of the first feature of the first decision tree model to obtain a fifth eigenvalue; a first conversion module, used to convert the first default risk prediction result according to a preset rule to obtain a second default risk prediction result; a first construction module, used to construct the second decision tree model based on the fifth eigenvalue and the second default risk prediction result.

[0024] Furthermore, the first processing unit includes: a first determination module, used to determine the Boolean logic problem; and a first calculation module, used to solve the Boolean logic problem to obtain the value of the target feature of each target object.

[0025] Furthermore, the device also includes: a first determination unit, used to determine the importance of each feature in the first decision tree model before determining the first feature of the first decision tree model; a second determination unit, used to determine the first feature of each first decision tree model based on the importance of each feature and the number of first features of each first decision tree model.

[0026] Furthermore, the device also includes: a first acquisition unit, which is used to acquire a second analysis result of analyzing the target scene after analyzing the target scene according to the value of the target feature of each target object; and a third determination unit, which is used to determine the target strategy based on the second analysis result.

[0027] In order to achieve the above-mentioned purpose, according to another aspect of the present application, a computer-readable storage medium is provided, wherein the storage medium stores a program, wherein the program executes any one of the scene analysis methods described above.

[0028] In order to achieve the above-mentioned purpose, according to another aspect of the present application, an electronic device is provided, which includes one or more processors and a memory, and the memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement any one of the scene analysis methods described above.

[0029] Through this application, the following steps are adopted: converting the deep learning model of each target object in the target scenario into a first decision tree model, and determining the first feature of the first decision tree model, wherein the target scenario is a scenario for predicting the default risk of multiple target objects, and the first feature is a feature of the target object whose impact on the default risk is greater than the preset degree; constructing a second decision tree model based on the first feature of the first decision tree model; expressing the second decision tree model in the form of Boolean logic, and calculating the value of the target feature of each target object based on Boolean logic, wherein the target feature is the feature after the first feature changes, and is the feature that changes the default risk prediction result; analyzing the target scenario based on the value of the target feature of each target object, solving the problem that it is difficult to analyze the scenario of default risk when using a deep learning model to predict the default risk of customers in the related technology. By converting the deep learning model of each target object in the scenario of default risk into the first feature of the first decision tree model to construct a second decision tree model, and expressing the second decision tree model in the form of Boolean logic, and then calculating the value of the target feature of each target object based on Boolean logic, and analyzing the scenario of default risk based on the value of the target feature, the accuracy of the subsequent business strategy can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0031] Figure 1 is a flowchart of a scene analysis method provided according to an embodiment of the present application;

[0032] Figure 2 is a schematic diagram of a decision tree of client A in an embodiment of the present application;

[0033] Figure 3 is a schematic diagram of a decision tree of client B in an embodiment of the present application;

[0034] Figure 4 is a schematic diagram of a decision tree of client C in an embodiment of the present application;

[0035] Figure 5 is a schematic diagram of a decision tree to be converted into Boolean logic rules in an embodiment of the present application;

[0036] Figure 6 is a flowchart of an optional scene analysis method provided according to an embodiment of the present application;

[0037] Figure 7 is a schematic diagram of a scene analysis device provided according to an embodiment of the present application;

[0038] Figure 8 is a schematic diagram of an optional scene analysis device provided according to an embodiment of the present application;

[0039] Fig. 9 is a schematic diagram of an electronic device provided according to an embodiment of the present application. DETAILED DESCRIPTION

[0040] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0041] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0042] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present application described here. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0043] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for display, data for analysis, etc.) involved in this disclosure are all information and data authorized by the user or fully authorized by all parties.

[0044] The present invention is described below in conjunction with preferred implementation steps. Figure 1 is a flowchart of a scene analysis method provided according to an embodiment of the present application, such as Figure 1 As shown, the method comprises the following steps:

[0045] Step S101, converting the deep learning model of each target object in the target scene into a first decision tree model, and determining the first feature of the first decision tree model, wherein the target scene is a scene for predicting default risk for multiple target objects, and the first feature is a feature of the target object whose impact on the default risk is greater than a preset degree.

[0046] For example, the deep learning model used in the scenario of predicting default risk for customers is converted into a decision tree model, and the more important features in the decision tree model are screened out, that is, the features that have a greater impact on customer changes in the scenario of default risk prediction.

[0047] Step S102: construct a second decision tree model based on the first feature of the first decision tree model.

[0048] For example, relatively important features in the decision tree model are input into the decision tree model, that is, the decision tree model is repeated, and the repeated decision tree model is obtained.

[0049] Step S103, expressing the second decision tree model in the form of Boolean logic, and calculating the value of the target feature of each target object based on the Boolean logic, wherein the target feature is the feature after the first feature is changed, and is the feature that causes the default risk prediction result to change.

[0050] For example, the replayed decision tree model is converted into the form of Boolean logic, and the problem of solving the reversal of the scenario results is transformed into solving a set of Boolean logic problems, and the solution of the Boolean logic problem is the feature value that can reverse the prediction result. In addition, the Boolean satisfiability problem refers to the problem of determining the satisfiability of propositional logic formulas, that is, the problem of determining whether there is an explanation that satisfies a given Boolean formula. If the result of a Boolean formula is TRUE, it is called satisfied, and vice versa. For example, the formula A and Not B. When a=TRUE and B=FALSE, A and Not B=TRUE, and the formula is satisfied. When a=TRUE and B=TRUE, A and Not B=FALSE, and the formula is not satisfied. Generally, ∧AND is represented, ∨ represents OR, Means NOT.

[0051] Step S104: analyzing the target scene according to the value of the target feature of each target object.

[0052] For example, the scenario of default risk prediction is analyzed based on the feature values ​​that can reverse the prediction results.

[0053] Through the above steps S101 to S104, a second decision tree model is constructed by converting the deep learning model of each target object in the default risk scenario into the first feature of the first decision tree model, and the second decision tree model is expressed in the form of Boolean logic. Based on the Boolean logic, the value of the target feature of each target object is calculated, and the default risk scenario is analyzed according to the value of the target feature, thereby improving the accuracy of the subsequent business strategy.

[0054] Optionally, in the scenario analysis method provided in the embodiment of the present application, converting the deep learning model of each target object in the target scenario into a first decision tree model includes: obtaining a third eigenvalue of each target object, wherein the third eigenvalue is a numerical value corresponding to the characteristic of each target object; perturbing the third eigenvalue a preset number of times to obtain a fourth eigenvalue; inputting the fourth eigenvalue into the deep learning model to obtain a first default risk prediction result; and learning and training the first default risk prediction result to obtain a first decision tree model.

[0055] For example, let the original deep learning model be f(x). And x represents a set of feature variables [x1,x2,...,x n ], f(x) represents a classifier model, and the result represents the probability of default of the customer’s credit risk.

[0056] Since the decision tree model itself has good interpretability, the decision tree model is used to approximate the deep learning model, and the decision tree model is set to g(x′).

[0057] Therefore, it is necessary to find a decision tree model g(x′) so that the results of f(x) and g(x′) are similar. Therefore, the risk default prediction of each customer is processed in the following steps, that is, each customer needs to find a decision tree model corresponding to the deep learning model through simulation.

[0058] Since each customer has a set of characteristic variables [x1, x2, ..., x N ], so the mean and variance of all features in the training samples are set to

[0059] Then, set the actual feature value of each customer to be And for x i The value of is randomly perturbed, and the basic formula of the perturbation is:

[0060]

[0061] and, Represents the normal distribution probability function. And the perturbation is M times, generally not less than 500 times, and a set of perturbed customer feature values ​​is obtained:

[0062]

[0063] ......

[0064]

[0065] Then, each Substitute f(x) into each other and get a set of prediction results:

[0066]

[0067] ......

[0068]

[0069] According to the above values, a decision tree model g(x′) is trained, and the feature input of g(x′) is required to be:

[0070]

[0071] ......

[0072]

[0073] Moreover, the corresponding output results are:

[0074]

[0075] Therefore, a decision tree model g(x′) can be trained based on the above data. Its input is consistent with the deep learning model f(x), and its output simulates the output of the deep learning model f(x). In this way, a decision tree model g(x′) with the same effect as f(x) is simulated.

[0076] In addition, by performing the above processing on each customer, a respective decision tree can be obtained. The schematic diagram of the decision tree of customer A in the embodiment of the present application is as follows: Figure 2 As shown, the schematic diagram of the decision tree of client B in the embodiment of the present application is as follows Figure 3 As shown, the schematic diagram of the decision tree of client C in the embodiment of the present application is as follows Figure 4 As shown. Moreover, Figure 2 Here, X1_1 represents the first feature of customer A, X1_2 represents the second feature of customer A, and so on. Figure 3 Here, X2_7 represents the seventh feature of customer B, X2_2 represents the second feature of customer B, and so on. Figure 4 X3_9 in represents the ninth feature of customer C, X3_3 represents the third feature of customer C, and so on. In addition, for example, if the first feature in this embodiment represents the age feature of the customer, then X1_1, X2_1, and X3_1 represent the age information of customer A, the age information of customer B, and the age information of customer C, respectively, and the same applies to the rest.

[0077] In summary, deep learning models can be simulated by decision trees with a certain degree of interpretability, so that decision trees can be used to approximately explain the results of deep learning models.

[0078] Optionally, in the scenario analysis method provided in an embodiment of the present application, before determining the first feature of the first decision tree model, the method also includes: determining the importance of each feature in the first decision tree model; determining the first feature of each first decision tree model based on the importance of each feature and the number of first features of each first decision tree model.

[0079] For example, since all features x are put into the model in the original g(x′), the decision tree will have too many nodes and it will be impossible to focus on important features. Moreover, each tree grows differently and it is impossible to extract common decision combinations. Therefore, it is necessary to find common and important features. The following method can be used to process it:

[0080] For each customer, the corresponding decision tree g i (x′), each feature will generate a feature importance value, which will be calculated in the process of generating the decision tree. The importance value of each feature is recorded as Among them, i represents the i-th customer, and n represents the feature importance value of the n-th feature variable of the i-th customer.

[0081] Moreover, for each feature variable n, we have Then for each J n Sort from largest to smallest and take the C features with the highest values ​​(the default setting is C = 20, and C ≤ N). Let these C features be z = [z1, z2, ..., z C ].

[0082] Through the above solution, the more important features in the decision tree model can be quickly and accurately screened out, thereby narrowing the scope of customer features in subsequent scenario analysis, thereby improving the timeliness of scenario analysis.

[0083] Optionally, in the scenario analysis method provided in an embodiment of the present application, constructing a second decision tree model based on the first feature of the first decision tree model includes: performing perturbation processing on the eigenvalue of the first feature of the first decision tree model to obtain a fifth eigenvalue; converting the first default risk prediction result according to a preset rule to obtain a second default risk prediction result; and constructing a second decision tree model based on the fifth eigenvalue and the second default risk prediction result.

[0084] For example, for each customer, the original Keep z features from the previous step, that is,

[0085] right Keep

[0086] right Keep ......

[0087] Keep

[0088] And convert the corresponding output results as follows:

[0089] For the default model, set the threshold to Threshold (default 0.8).

[0090] when Defined as

[0091] when Defined as

[0092] This will

[0093]

[0094] ......

[0095]

[0096] Convert to

[0097]

[0098] ......

[0099]

[0100] That is, its output becomes:

[0101]

[0102] And retrain g(z) so that each customer repeats their own decision tree g i (z).

[0103] Moreover, the input and output of the two-round decision tree are shown in Table 1:

[0104] Table 1

[0105]

[0106]

[0107] In summary, by replaying the decision tree, the prediction results of the decision tree model can be transformed, thus paving the way for the subsequent conversion of the decision tree model into the form of Boolean logic.

[0108] Optionally, in the scene analysis method provided in the embodiment of the present application, calculating the value of the target feature of each target object based on Boolean logic includes: determining a Boolean logic problem; solving the Boolean logic problem to obtain the value of the target feature of each target object.

[0109] For example, the decision tree model of each customer is converted into Boolean logic rules, that is, for each customer i, its decision tree is converted into a set of rule keys.

[0110] For example, in Boolean logic, Figure 5 The decision tree shown is converted into a rule key representation as:

[0111]

[0112] Then we set the scene. The goal of setting the scene is to find a set of Make at the same time

[0113] That is, assuming that feature z changes slightly, it becomes After that, the original prediction results were reversed.

[0114] Right now There must be At the same time z and The distance between them must be very close.

[0115] Definition hour, on the contrary

[0116]

[0117] Among them, zj Represents the true value of the customer feature; Indicates the value that has changed; MAD j represents the median of feature j; θ j The weight that represents the range of variation of feature j. The higher the weight, the smaller the range of variation allowed for feature j. It is set by the user.

[0118] For example, assuming that the customer's real feature value is z = [z1 = 1, z2 = 3, z3 = 0], for the above Boolean logic rule key, the customer's real default prediction result is 1. When z changes slightly, it is assumed to become , the customer default prediction result becomes 0, a reversal occurs.

[0119] Moreover, the goal of the solution is to reverse the original prediction result by slightly changing the customer's feature values ​​while keeping the decision tree structure unchanged.

[0120] Therefore, the problem of solving the inversion of scenario results is transformed into a problem of solving a set of Boolean logic.

[0121] Solve The main solution steps are as follows:

[0122] Input: Initialization

[0123] Output: Inverted result

[0124] process:

[0125] (1) Let δ min ←0,δ max ←1;

[0126] (2) When δ max -δ min >ε, then the following cycle is performed:

[0127] make

[0128] make

[0129] Using a SAT solver, solve And the SAT solver is a tool used to solve Boolean logic problems;

[0130] If z′ does not meet the solution conditions, it means that the original restriction requirements are too high, so let δ min ←δ, expand The range of variation of both;

[0131] If z′ is satisfied, then let δmax ←δ, increase the restriction range requirements to find a better value;

[0132] (3) Output results

[0133] In summary, through the rules of Boolean logic, the decision tree transformation is called the problem of solving Boolean logic, which simplifies the difficulty of problem analysis, so that the value of the feature that can reverse the prediction result can be obtained quickly and accurately.

[0134] Optionally, in the scenario analysis method provided in the embodiment of the present application, analyzing the target scenario according to the value of the target characteristic of each target object includes: performing a default risk analysis on each target object according to the value of the target characteristic of each target object, and obtaining multiple first analysis results; based on each first analysis result, and using a clustering method, clustering the multiple target objects to obtain multiple clusters, wherein a cluster is a collection of target objects with the same characteristic changes; and analyzing the target scenario according to each cluster.

[0135] For example, risk analysis is conducted on individual customers based on the features that have the greatest impact on changes in individual customers. Then, based on the results of risk analysis on individual customers, risk analysis is conducted on customer groups. That is, customer groups are clustered using clustering methods. Customers with the same feature changes will form a cluster, indicating that these customers have the same risk change characteristics. Based on the classified customer groups, the overall risk in the default risk scenario is analyzed.

[0136] To sum up, based on the solution results for each customer, risk cause scenario analysis can be achieved from individual customers to customer groups, from point to surface, which is more in line with the usage needs of business decision-making.

[0137] Optionally, in the scenario analysis method provided in the embodiment of the present application, performing a default risk analysis on each target object based on the value of the target feature of each target object includes: obtaining the value of the target feature of each target object and the value of the first feature; determining the value of the feature of each target object after the feature is changed based on the value of the target feature of each target object and the value of the first feature; performing a default risk analysis on each target object based on the value of the feature after the feature is changed.

[0138] For example, for each customer i, there is a corresponding set of z i and where z i is the true value, is the characteristic value that produces the reversed result after changing the basis in the previous step. Then, we conduct risk analysis on a single customer, that is, for each customer i, we can analyze the characteristics that have the greatest impact on its changes, as well as the magnitude of the changes, and provide guidance on changes in customer default predictions. After the change A customer's default status can either improve (change from 1 to 0) or worsen (change from 0 to 1).

[0139] Through the above solution, by solving decision trees and Boolean problems one by one for each customer, the obtained results can be closer to the actual situation of each customer and the solved results can be more accurate.

[0140] Optionally, in the scene analysis method provided in an embodiment of the present application, analyzing the target scene based on each cluster includes: determining the value of the feature after the feature change of each cluster; obtaining a second feature based on the value of the feature after the feature change of each cluster, wherein the second feature is a feature in the target feature; and analyzing the target scene based on the second feature.

[0141] For example, when conducting customer group risk analysis, when when when This will Convert to Then, we use clustering method to cluster the customers. Customers with the same characteristic changes will become a cluster, indicating that these customers have the same risk change characteristics.

[0142] Then conduct an overall risk analysis. And for [E1,E2,...,E C ] Sorting by high and low values, we get the features that have a greater impact on the overall risk of the customer base, which means that changes in these features will lead to changes in the default risk of the overall customer base.

[0143] Through the above solution, it is convenient to conduct an overall analysis of the default risk scenario and obtain accurate scenario analysis results.

[0144] Optionally, in the scene analysis method provided in an embodiment of the present application, after analyzing the target scene based on the value of the target feature of each target object, the method further includes: obtaining a second analysis result of analyzing the target scene; and determining the target strategy based on the second analysis result.

[0145] For example, based on the analysis results of the default risk scenarios, we can determine some characteristics that have a greater impact on the default risk outcomes, and whether these characteristics have a positive or negative impact on the outcomes. Then, based on the information from the scenario analysis results, we can analyze the causes of the outcomes and ultimately determine the subsequent business strategies.

[0146] Through the above-mentioned scheme, business strategies can be formulated accurately, which improves the accuracy of the business strategies formulated subsequently.

[0147] Figure 6 is a flowchart of an optional scene analysis method provided according to an embodiment of the present application, such as Figure 6 As shown in the figure, the process of scenario analysis specifically includes:

[0148] Step S601, proxy model conversion, converting the deep learning model into a decision tree model;

[0149] Step S602, mining the importance of features in the decision tree model;

[0150] Step S603, screening the important features and replaying the decision tree model;

[0151] Step S604, converting the decision tree model into Boolean logic rules, converting the replayed decision tree model into Boolean logic rules;

[0152] Step S605, scenario setting, finding the feature value that can reverse the prediction result;

[0153] Step S606: scenario analysis, analyzing the changes in features before and after the results are reversed, and performing scenario and decision analysis.

[0154] In summary, the scenario analysis method provided in the embodiment of the present application converts the deep learning model of each target object in the target scenario into a first decision tree model, and determines the first feature of the first decision tree model, wherein the target scenario is a scenario for predicting default risk for multiple target objects, and the first feature is a feature of the target object whose impact on the default risk is greater than a preset degree; a second decision tree model is constructed based on the first feature of the first decision tree model; the second decision tree model is expressed in the form of Boolean logic, and based on the Boolean logic, the value of the target feature of each target object is calculated, wherein the target feature is the feature after the first feature changes, and is the feature that changes the default risk prediction result; the target scenario is analyzed based on the value of the target feature of each target object, thereby solving the problem in the related art that it is difficult to analyze the scenario of default risk when using a deep learning model to predict the default risk of a customer. The second decision tree model is constructed by converting the deep learning model of each target object in the default risk scenario into the first feature of the first decision tree model, and the second decision tree model is expressed in the form of Boolean logic. Based on the Boolean logic, the value of the target feature of each target object is calculated, and the default risk scenario is analyzed according to the value of the target feature, thereby improving the accuracy of the subsequent business strategy.

[0155] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0156] The embodiment of the present application also provides a scene analysis device. It should be noted that the scene analysis device of the embodiment of the present application can be used to execute the scene analysis method provided in the embodiment of the present application. The scene analysis device provided in the embodiment of the present application is introduced below.

[0157] Figure 7 is a schematic diagram of a scene analysis device according to an embodiment of the present application. Figure 7 As shown, the device includes: a first conversion unit 701, a first construction unit 702, a first processing unit 703 and a first analysis unit 704.

[0158] Specifically, the first conversion unit 701 is used to convert the deep learning model of each target object in the target scene into a first decision tree model, and determine the first feature of the first decision tree model, wherein the target scene is a scene for predicting default risk for multiple target objects, and the first feature is a feature of the target object whose impact on the default risk is greater than a preset degree;

[0159] A first construction unit 702 is used to construct a second decision tree model according to the first feature of the first decision tree model;

[0160] The first processing unit 703 is used to express the second decision tree model in the form of Boolean logic, and calculate the value of the target feature of each target object according to the Boolean logic, wherein the target feature is the feature after the first feature is changed, and is the feature that causes the default risk prediction result to change;

[0161] The first analysis unit 704 is used to analyze the target scene according to the value of the target feature of each target object.

[0162] In summary, the scene analysis device provided by the embodiment of the present application converts the deep learning model of each target object in the target scene into a first decision tree model through the first conversion unit 701, and determines the first feature of the first decision tree model, wherein the target scene is a scene for predicting default risk for multiple target objects, and the first feature is a feature of the target object whose impact on the default risk is greater than a preset degree; the first construction unit 702 constructs a second decision tree model based on the first feature of the first decision tree model; the first processing unit 703 represents the second decision tree model in the form of Boolean logic, and calculates the value of the target feature of each target object based on Boolean logic, wherein the target feature is the feature after the first feature is changed, And it is a feature that causes the default risk prediction result to change; the first analysis unit 704 analyzes the target scenario according to the value of the target feature of each target object, which solves the problem that it is difficult to analyze the default risk scenario when using a deep learning model to predict the default risk of a customer in the related technology. The second decision tree model is constructed based on the first feature of the first decision tree model that converts the deep learning model of each target object in the default risk scenario, and the second decision tree model is expressed in the form of Boolean logic. Based on the Boolean logic, the value of the target feature of each target object is calculated, and the default risk scenario is analyzed according to the value of the target feature, thereby improving the accuracy of the subsequent business strategy.

[0163] Optionally, in the scene analysis device provided in the embodiment of the present application, the first analysis unit includes: a first analysis module, used to perform a default risk analysis on each target object according to the value of the target characteristic of each target object, and obtain multiple first analysis results; a first clustering module, used to cluster the multiple target objects according to each first analysis result and in a clustering manner, and obtain multiple clusters, wherein a cluster is a collection of target objects with the same characteristic changes; a second analysis module, used to analyze the target scene based on each cluster.

[0164] Optionally, in the scenario analysis device provided in the embodiment of the present application, the first analysis module includes: a first acquisition submodule, used to obtain the value of the target feature and the value of the first feature of each target object; a first determination submodule, used to determine the value of the feature of each target object after the feature is changed based on the value of the target feature and the value of the first feature of each target object; the first analysis submodule, used to perform a default risk analysis on each target object based on the value of the feature after the feature of each target object is changed.

[0165] Optionally, in the scene analysis device provided in the embodiment of the present application, the second analysis module includes: a second determination submodule, used to determine the value of the feature of each cluster after the feature change; a first processing submodule, used to obtain a second feature based on the value of the feature of each cluster after the feature change, wherein the second feature is a feature in the target feature; and a second analysis submodule, used to analyze the target scene based on the second feature.

[0166] Optionally, in the scene analysis device provided in the embodiment of the present application, the first conversion unit includes: a first acquisition module, used to obtain a third eigenvalue of each target object, wherein the third eigenvalue is a numerical value corresponding to the characteristic of each target object; a first processing module, used to perturb the third eigenvalue according to a preset number of times to obtain a fourth eigenvalue; a first input module, used to input the fourth eigenvalue into a deep learning model to obtain a first default risk prediction result; and a first training module, used to learn and train the first default risk prediction result to obtain a first decision tree model.

[0167] Optionally, in the scenario analysis device provided in the embodiment of the present application, the first construction unit includes: a second processing module, used to perform perturbation processing on the eigenvalue of the first feature of the first decision tree model to obtain a fifth eigenvalue; a first conversion module, used to convert the first default risk prediction result according to a preset rule to obtain a second default risk prediction result; and a first construction module, used to construct a second decision tree model based on the fifth eigenvalue and the second default risk prediction result.

[0168] Optionally, in the scene analysis device provided in an embodiment of the present application, the first processing unit includes: a first determination module, used to determine a Boolean logic problem; and a first calculation module, used to solve the Boolean logic problem to obtain the value of the target feature of each target object.

[0169] Optionally, in the scene analysis device provided in an embodiment of the present application, the device also includes: a first determination unit, used to determine the importance of each feature in the first decision tree model before determining the first feature of the first decision tree model; a second determination unit, used to determine the first feature of each first decision tree model based on the importance of each feature and the number of first features of each first decision tree model.

[0170] Optionally, in the scene analysis device provided in the embodiment of the present application, the device also includes: a first acquisition unit, used to obtain a second analysis result of analyzing the target scene after analyzing the target scene based on the value of the target feature of each target object; and a third determination unit, used to determine the target strategy based on the second analysis result.

[0171] The scene analysis device includes a processor and a memory. The above-mentioned first conversion unit 701, first construction unit 702, first processing unit 703 and first analysis unit 704 are all stored in the memory as program units, and the processor executes the above-mentioned program units stored in the memory to realize corresponding functions.

[0172] Figure 8 is a schematic diagram of an optional scene analysis device provided according to an embodiment of the present application, such as Figure 8 As shown, the device includes: a model rule conversion unit, an important rule screening unit and a scenario decision analysis unit. The model rule conversion unit is used to convert the deep learning model into a rule-based model while keeping the effect basically unchanged; because the deep model often uses a large number of features and the initial rules are very large, the important rule screening unit is used to extract important rules based on the rule-based model; the scenario decision analysis is used to convert the rule-based model into a Boolean satisfiability problem, and by setting the scenario goal, inversely deduce the rule combination that can achieve the goal.

[0173] The processor contains a kernel, which calls the corresponding program unit from the memory. One or more kernels can be set, and the accuracy of the subsequent business strategy can be improved by adjusting the kernel parameters.

[0174] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0175] An embodiment of the present invention provides a computer-readable storage medium on which a program is stored. When the program is executed by a processor, the scene analysis method is implemented.

[0176] like Fig. 9As shown, an embodiment of the present invention provides an electronic device, the device includes a processor, a memory, and a program stored in the memory and executable on the processor, and the processor implements the following steps when executing the program: converting a deep learning model of each target object in a target scene into a first decision tree model, and determining a first feature of the first decision tree model, wherein the target scene is a scene for predicting default risk for multiple target objects, and the first feature is a feature of a target object whose impact on default risk is greater than a preset degree; constructing a second decision tree model based on the first feature of the first decision tree model; expressing the second decision tree model in the form of Boolean logic, and calculating the value of a target feature of each target object based on the Boolean logic, wherein the target feature is a feature after the first feature changes, and is a feature that changes the default risk prediction result; analyzing the target scene based on the value of the target feature of each target object.

[0177] When the processor executes the program, the following steps are also implemented: according to the value of the target feature of each target object, the target scenario is analyzed, including: according to the value of the target feature of each target object, each target object is subjected to a default risk analysis to obtain a plurality of first analysis results; according to each first analysis result, the plurality of target objects are clustered by clustering to obtain a plurality of clusters, wherein the cluster is a collection of target objects with the same feature changes; according to each cluster, the target scenario is analyzed.

[0178] When the processor executes the program, the following steps are also implemented: based on the value of the target feature of each target object, a default risk analysis is performed on each target object, including: obtaining the value of the target feature of each target object and the value of the first feature; based on the value of the target feature of each target object and the value of the first feature, determining the value of the feature of each target object after the feature is changed; based on the value of the feature of each target object after the feature is changed, a default risk analysis is performed on each target object.

[0179] When the processor executes the program, the following steps are also implemented: based on each cluster, analyzing the target scene includes: determining the value of the feature after the feature of each cluster is changed; based on the value of the feature after the feature of each cluster is changed, obtaining a second feature, wherein the second feature is a feature in the target feature; based on the second feature, analyzing the target scene.

[0180] When the processor executes the program, the following steps are also implemented: converting the deep learning model of each target object in the target scene into a first decision tree model includes: obtaining a third eigenvalue of each target object, wherein the third eigenvalue is a numerical value corresponding to the characteristic of each target object; perturbing the third eigenvalue a preset number of times to obtain a fourth eigenvalue; inputting the fourth eigenvalue into the deep learning model to obtain a first default risk prediction result; and performing learning and training on the first default risk prediction result to obtain the first decision tree model.

[0181] When the processor executes the program, the following steps are also implemented: constructing a second decision tree model based on the first feature of the first decision tree model includes: perturbing the eigenvalue of the first feature of the first decision tree model to obtain a fifth eigenvalue; converting the first default risk prediction result according to a preset rule to obtain a second default risk prediction result; and constructing the second decision tree model based on the fifth eigenvalue and the second default risk prediction result.

[0182] When the processor executes the program, the following steps are also implemented: based on the Boolean logic, calculating the value of the target feature of each target object includes: determining the problem of the Boolean logic; solving the problem of the Boolean logic to obtain the value of the target feature of each target object.

[0183] When the processor executes the program, the following steps are also implemented: before determining the first feature of the first decision tree model, the method also includes: determining the importance of each feature in the first decision tree model; determining the first feature of each first decision tree model based on the importance of each feature and the number of first features of each first decision tree model.

[0184] When the processor executes the program, the following steps are also implemented: after analyzing the target scene according to the value of the target feature of each target object, the method further includes: obtaining a second analysis result of analyzing the target scene; and determining a target strategy according to the second analysis result. The device in this article can be a server, a PC, a PAD, a mobile phone, etc.

[0185] The present application also provides a computer program product, which, when executed on a data processing device, is suitable for executing a program that is initialized with the following method steps: converting a deep learning model of each target object in a target scenario into a first decision tree model, and determining a first feature of the first decision tree model, wherein the target scenario is a scenario for predicting default risk for multiple target objects, and the first feature is a feature of a target object whose impact on default risk is greater than a preset degree; constructing a second decision tree model based on the first feature of the first decision tree model; expressing the second decision tree model in the form of Boolean logic, and calculating the value of a target feature of each target object based on the Boolean logic, wherein the target feature is a feature after the first feature changes, and is a feature that changes the default risk prediction result; analyzing the target scenario based on the value of the target feature of each target object.

[0186] When executed on a data processing device, it is also suitable for executing a program that is initialized with the following method steps: analyzing the target scenario based on the value of the target feature of each target object, including: performing a default risk analysis on each target object based on the value of the target feature of each target object, to obtain multiple first analysis results; clustering the multiple target objects based on each first analysis result and using a clustering method to obtain multiple clusters, wherein the cluster is a collection of target objects with the same feature changes; analyzing the target scenario based on each cluster.

[0187] When executed on a data processing device, it is also suitable for executing a program initialized with the following method steps: based on the value of the target characteristic of each target object, performing a default risk analysis on each target object, including: obtaining the value of the target characteristic of each target object and the value of the first characteristic; based on the value of the target characteristic of each target object and the value of the first characteristic, determining the value of the characteristic after the characteristic of each target object is changed; based on the value of the characteristic after the characteristic of each target object is changed, performing a default risk analysis on each target object.

[0188] When executed on a data processing device, it is also suitable for executing a program initialized with the following method steps: analyzing the target scene based on each cluster includes: determining the value of the feature after the feature of each cluster is changed; obtaining a second feature based on the value of the feature after the feature of each cluster is changed, wherein the second feature is a feature in the target feature; and analyzing the target scene based on the second feature.

[0189] When executed on a data processing device, it is also suitable for executing a program initialized with the following method steps: converting the deep learning model of each target object in the target scene into a first decision tree model includes: obtaining a third eigenvalue of each target object, wherein the third eigenvalue is a numerical value corresponding to the characteristic of each target object; perturbing the third eigenvalue a preset number of times to obtain a fourth eigenvalue; inputting the fourth eigenvalue into the deep learning model to obtain a first default risk prediction result; and learning and training the first default risk prediction result to obtain the first decision tree model.

[0190] When executed on a data processing device, it is also suitable for executing a program initialized with the following method steps: constructing a second decision tree model based on the first feature of the first decision tree model, including: perturbing the eigenvalue of the first feature of the first decision tree model to obtain a fifth eigenvalue; converting the first default risk prediction result according to a preset rule to obtain a second default risk prediction result; constructing the second decision tree model based on the fifth eigenvalue and the second default risk prediction result.

[0191] When executed on a data processing device, it is also suitable for executing an initialized program having the following method steps: calculating the value of the target feature of each target object based on the Boolean logic, including: determining the problem of the Boolean logic; solving the problem of the Boolean logic to obtain the value of the target feature of each target object.

[0192] When executed on a data processing device, it is also suitable for executing a program initialized with the following method steps: before determining the first feature of the first decision tree model, the method also includes: determining the importance of each feature in the first decision tree model; determining the first feature of each first decision tree model based on the importance of each feature and the number of first features of each first decision tree model.

[0193] When executed on a data processing device, it is also suitable for executing a program that initializes the following method steps: after analyzing the target scene based on the value of the target feature of each target object, the method also includes: obtaining a second analysis result of analyzing the target scene; determining the target strategy based on the second analysis result.

[0194] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0195] The present application 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 the present application. 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 data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing 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.

[0196] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising 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.

[0197] These computer program instructions can also be loaded onto a computer or other programmable data processing 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 implementing 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.

[0198] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0199] The memory may include non-permanent memory 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.

[0200] 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.

[0201] 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.

[0202] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment or an embodiment in combination with software and hardware. Moreover, the present application may adopt 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.) that contain computer-usable program code.

[0203] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. A scene analysis method, characterized in that: include: Converting the deep learning model of each target object in the target scenario into a first decision tree model, and determining a first feature of the first decision tree model, wherein the target scenario is a scenario for predicting default risk for multiple target objects, and the first feature is a feature of a target object whose impact on default risk is greater than a preset degree; Constructing a second decision tree model according to the first feature of the first decision tree model; The second decision tree model is expressed in the form of Boolean logic, and according to the Boolean logic, the value of the target feature of each target object is calculated, wherein the target feature is the feature after the first feature is changed, and is the feature that changes the default risk prediction result; Analyzing the target scene according to the value of the target feature of each target object; Wherein, converting the deep learning model of each target object in the target scene into a first decision tree model includes: obtaining a third eigenvalue of each target object, wherein the third eigenvalue is a numerical value corresponding to the feature of each target object; performing perturbation processing on the third eigenvalue according to a preset number of times to obtain a fourth eigenvalue; inputting the fourth eigenvalue into the deep learning model to obtain a first default risk prediction result; performing learning training on the first default risk prediction result to obtain the first decision tree model; Constructing a second decision tree model based on the first feature of the first decision tree model includes: perturbing the eigenvalue of the first feature of the first decision tree model to obtain a fifth eigenvalue; converting the first default risk prediction result according to a preset rule to obtain a second default risk prediction result; and constructing the second decision tree model based on the fifth eigenvalue and the second default risk prediction result.

2. The method according to claim 1, characterized in that According to the value of the target feature of each target object, analyzing the target scene includes: Performing default risk analysis on each target object according to the value of the target characteristic of each target object, and obtaining a plurality of first analysis results; According to each first analysis result, the plurality of target objects are clustered by using a clustering method to obtain a plurality of clusters, wherein the cluster is a collection of target objects having the same characteristic changes; According to each cluster, the target scene is analyzed.

3. The method according to claim 2, characterized in that Based on the value of the target characteristic of each target object, the default risk analysis of each target object includes: Obtaining the value of the target feature and the value of the first feature of each target object; Determining the value of the feature after the feature change of each target object according to the value of the target feature of each target object and the value of the first feature; Conduct a default risk analysis on each target object based on the value of the characteristics after the characteristics of each target object have changed.

4. The method according to claim 2, characterized in that: According to each cluster, analyzing the target scene includes: Determine the value of the feature after the feature change of each cluster; Obtaining a second feature according to the value of the feature after the feature change of each cluster, wherein the second feature is a feature in the target feature; The target scene is analyzed according to the second feature.

5. The method according to claim 1, characterized in that According to the Boolean logic, calculating the value of the target feature of each target object includes: Determine the Boolean logic problem; The Boolean logic problem is solved to obtain the value of the target feature of each target object.

6. The method according to claim 1, characterized in that Before determining the first feature of the first decision tree model, the method further includes: Determining the importance of each feature in the first decision tree model; The first feature of each first decision tree model is determined according to the importance of each feature and the number of first features of each first decision tree model.

7. The method according to claim 1, characterized in that After analyzing the target scene according to the value of the target feature of each target object, the method further includes: Obtaining a second analysis result of analyzing the target scene; A target strategy is determined based on the second analysis result.

8. A scene analysis device, characterized in that: include: A first conversion unit, configured to convert the deep learning model of each target object in a target scenario into a first decision tree model, and determine a first feature of the first decision tree model, wherein the target scenario is a scenario for predicting default risk for multiple target objects, and the first feature is a feature of a target object whose impact on default risk is greater than a preset degree; A first construction unit, configured to construct a second decision tree model according to the first feature of the first decision tree model; a first processing unit, configured to represent the second decision tree model in the form of Boolean logic, and calculate the value of a target feature of each target object according to the Boolean logic, wherein the target feature is a feature after the first feature is changed, and is a feature that causes a default risk prediction result to change; A first analysis unit, configured to analyze the target scene according to the value of the target feature of each target object; The first conversion unit includes: a first acquisition module, used to acquire a third eigenvalue of each target object, wherein the third eigenvalue is a numerical value corresponding to the feature of each target object; a first processing module, used to perform perturbation processing on the third eigenvalue according to a preset number of times to obtain a fourth eigenvalue; a first input module, used to input the fourth eigenvalue into the deep learning model to obtain a first default risk prediction result; a first training module, used to perform learning and training on the first default risk prediction result to obtain the first decision tree model; The first construction unit includes: a second processing module, used to perform perturbation processing on the eigenvalue of the first feature of the first decision tree model to obtain a fifth eigenvalue; a first conversion module, used to convert the first default risk prediction result according to a preset rule to obtain a second default risk prediction result; a first construction module, used to construct the second decision tree model based on the fifth eigenvalue and the second default risk prediction result.

9. A computer-readable storage medium, characterized in that: The storage medium stores a program, wherein the program executes the scene analysis method described in any one of claims 1 to 7.

10. An electronic device, characterized in that: It includes one or more processors and a memory, wherein the memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the scene analysis method described in any one of claims 1 to 7.

Images