Financial data detection model training method and device, financial data detection method and device and medium

Abstract

The invention provides a financial data detection model training and financial data detection method and device and a medium, and the method comprises the steps: obtaining a financial related data set which comprises an attribute value sequence of a sample company, and the sample company comprises normal samples and abnormal samples, the attribute value sequence comprises elements of different attribute categories; determining a segmentation evaluation value of each element according to an element mutation condition and an element position distribution feature of the same attribute category and association features among elements of different attribute categories; discretizing the elements of the same attribute category through the segmentation evaluation value to obtain a discretized financial related data set; and performing model training based on the discretized financial related data set to obtain a financial data detection model. According to the invention, the calculation efficiency and generalization ability of the financial data detection model are improved, and the detection efficiency and detection accuracy of the financial data are further improved.

Description

Technical Field

[0001] The present application relates to the field of Internet technology, and in particular to a financial data detection model training and financial data detection method, device and medium. Background Art

[0002] As monitoring agencies increasingly tighten supervision and management of listed companies, the importance of anomaly detection in financial data is gradually increasing. However, due to the wide variety of listed companies, the high complexity of financial data and the wide coverage, an automated operation monitoring method is needed.

[0003] Among them, regarding automated operation monitoring, it is a method of using information technology and automation tools to monitor and manage the operation data of each enterprise, which can effectively supervise the operating efficiency and financial anomalies of each enterprise in the market.

[0004] However, existing financial data monitoring methods have problems such as low monitoring efficiency and low accuracy. Summary of the invention

[0005] The technical problem to be solved by the present application is to provide a financial data detection model training and financial data detection method, device and medium to solve the problems existing in the prior art in view of the above-mentioned deficiencies in the prior art.

[0006] In the first aspect, the present application provides a financial data detection model training method.

[0007] The methods include:

[0008] S1. Acquire a set of financial related data, wherein the set of financial related data includes a sequence of attribute values ​​of sample companies, wherein the sample companies include normal samples and abnormal samples, and the sequence of attribute values ​​includes elements of different attribute categories;

[0009] S2. Determine the segmentation evaluation value of each element according to the mutation of elements of the same attribute category, the element position distribution characteristics, and the correlation characteristics between elements of different attribute categories;

[0010] S3, discretizing the elements of the same attribute category by segmenting the evaluation values ​​to obtain a discretized financial related data set;

[0011] S4. Perform model training based on the discretized financial-related data set to obtain a financial data detection model.

[0012] In some embodiments, S2 includes:

[0013] S21, composing the elements of the same attribute category into an element sequence, and determining the first segmentation evaluation factor of each element according to the mutation difference of each element in each element sequence and the abnormal sample division;

[0014] S22, determining a second segmentation evaluation factor for each element according to the distance of each element in each element sequence relative to the first element and the last element in the element sequence to which it belongs;

[0015] S23, determining a third segmentation evaluation factor of each element according to the position distribution of each element in each attribute value sequence in the element sequence to which it belongs;

[0016] S24 . Determine a segmentation evaluation value of each element according to the first segmentation evaluation factor, the second segmentation evaluation factor, and the third segmentation evaluation factor of each element.

[0017] In some embodiments, in S21, the abnormal sample classification process of each element includes:

[0018] Mark any element as a target element, determine a first ratio of the number of abnormal samples in a first element interval before the target element to the total number of abnormal samples in the element sequence to which the target element belongs, and determine a second ratio of the number of abnormal samples in a second element interval after the target element to the total number of abnormal samples;

[0019] An abnormal classification index of the target element is determined according to a difference between the first ratio and the second ratio corresponding to the target element.

[0020] In some embodiments, in S21, the first segmentation evaluation factor is calculated by the following formula:

[0021] A1 ni =B ni ×C ni

[0022]

[0023] C ni =|α - -α + |

[0024] in:

[0025] A1 ni The first segmentation evaluation factor of the i-th element in the element sequence of the n-th attribute category;

[0026] B ni Indicates the mutation degree index of the i-th element in the element sequence of the n-th attribute category;

[0027] Cni represents the abnormal partition index of the i-th element in the element sequence of the n-th attribute category;

[0028] Δ n(i+1) The difference between the i-th element and the i+1-th element in the element sequence of the n-th attribute category;

[0029] Δ n(i-1) The difference between the i-th element and the i-1-th element in the element sequence of the n-th attribute category;

[0030] ε represents a preset constant;

[0031] α - It represents the first ratio of the number of abnormal samples in the first element interval before the i-th element in the element sequence of the n-th attribute category to the total number of abnormal samples;

[0032] α + represents the second ratio of the number of abnormal samples in the second element interval after the i-th element in the element sequence of the n-th attribute category to the total number of abnormal samples;

[0033] || represents the absolute value function.

[0034] In some embodiments, S22 includes:

[0035] Calculate the distance between each element and the first element in the element sequence to which it belongs, and obtain the first distance value;

[0036] Calculate the distance between each element and the last element in the element sequence to which it belongs, and obtain a second distance value;

[0037] For each element, a second segmentation evaluation factor is determined for the first distance value and the second distance value corresponding to the element.

[0038] In some embodiments, the second segmentation evaluation factor is calculated by the following formula:

[0039]

[0040] in:

[0041] A2 ni The second segmentation evaluation factor of the i-th element in the element sequence of the n-th attribute category;

[0042] min means to find the minimum value function;

[0043] L ni The first distance value between the ith element in the element sequence of the nth attribute category and the first element in the element sequence to which it belongs;

[0044] Lc ni The second distance value between the i-th element in the element sequence of the n-th attribute category and the last element in the element sequence to which it belongs;

[0045] exp represents an exponential function with the natural constant e as the base.

[0046] In some embodiments, S23 includes:

[0047] Determine the sorting order of each element in each attribute value sequence in the element sequence to which it belongs;

[0048] Take any element in any attribute value sequence as the base element, and take other elements in the attribute value sequence where the base element is located except the base element as comparison elements;

[0049] The average value of the difference between the sorting number corresponding to the reference element and the sorting numbers corresponding to each comparison element is calculated, and the average value of the difference is used as the third segmentation evaluation factor of the reference element.

[0050] In some embodiments, the third segmentation evaluation factor is calculated by the following formula:

[0051]

[0052] in:

[0053] A3 x represents the third segmentation evaluation factor of the benchmark element;

[0054] N represents the number of attribute categories corresponding to the sample companies;

[0055] F j Indicates the sorting number corresponding to the j-th comparison element corresponding to the reference element;

[0056] F x Indicates the sorting sequence number corresponding to the benchmark element;

[0057] || represents the absolute value function.

[0058] In some embodiments, S24 includes:

[0059] Normalizing the first segmentation evaluation factor, the second segmentation evaluation factor, and the third segmentation evaluation factor of each element to obtain each normalized result value corresponding to each element;

[0060] According to each result value after normalization corresponding to the same element, the segmentation evaluation value of the corresponding element is obtained.

[0061] In a second aspect, the present application provides a financial data detection method, the method comprising:

[0062] Get the attribute value sequence of the company to be detected;

[0063] Input the attribute value sequence of the company to be tested into the financial data testing model to obtain the financial testing results of the company to be tested;

[0064] Wherein, the financial data detection model is obtained by the financial data detection model training method according to any one of claims 1-9.

[0065] In a third aspect, the present application provides a financial data detection model training device, the device comprising:

[0066] A first acquisition module is configured to acquire a set of financial related data, wherein the set of financial related data includes a sequence of attribute values ​​of sample companies, wherein the sample companies include normal samples and abnormal samples, and the sequence of attribute values ​​includes elements of different attribute categories;

[0067] A determination module is configured to determine the segmentation evaluation value of each element according to the mutation of elements of the same attribute category, the element position distribution characteristics, and the correlation characteristics between elements of different attribute categories;

[0068] A discrete processing module, which is configured to discretize elements of the same attribute category by segmenting the evaluation values ​​to obtain a discretized financial related data set;

[0069] The model training module is configured to perform model training based on a discretized set of financial related data to obtain a financial data detection model.

[0070] In a fourth aspect, the present application provides a financial data detection device, the device comprising:

[0071] A second acquisition module is configured to acquire a sequence of attribute values ​​of the company to be detected;

[0072] A financial detection module, which is configured to input the attribute value sequence of the company to be detected into the financial data detection model to obtain the financial detection result of the company to be detected;

[0073] Wherein, the financial data detection model is obtained by the financial data detection model training method according to any one of claims 1-9.

[0074] In a fifth aspect, the present application provides a financial data processing device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to implement the above-mentioned financial data detection model training method or financial data detection method.

[0075] In a sixth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the above-mentioned financial data detection model training method or financial data detection method is implemented.

[0076] The present application provides a financial data detection model training and financial data detection method, device and medium, the method comprising: obtaining a set of financial related data, the set of financial related data comprising a sequence of attribute values ​​of a sample company, the sample company comprising a normal sample and an abnormal sample, the sequence of attribute values ​​comprising elements of different attribute categories; determining the segmentation evaluation value of each element according to the element mutation of the same attribute category, the element position distribution characteristics and the correlation characteristics between elements of different attribute categories; discretizing the elements of the same attribute category by the segmentation evaluation value to obtain a discretized set of financial related data; performing model training based on the discretized set of financial related data to obtain a financial data detection model. The present application realizes efficient and rapid discretization of a set of financial related data with continuous attributes, improves the computational efficiency and generalization ability of the financial data detection model, and further improves the detection efficiency and accuracy of financial data. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0078] Figure 1 A schematic diagram of a financial data detection model training method provided in an embodiment of the present application;

[0079] Figure 2 A schematic diagram of determining the segmentation evaluation value of each element provided in an embodiment of the present application;

[0080] Figure 3 A schematic diagram of a financial data detection method provided in an embodiment of the present application;

[0081] Figure 4 A schematic diagram of a financial data detection model training device provided in an embodiment of the present application;

[0082] Figure 5 A schematic diagram of a financial data detection device provided in an embodiment of the present application;

[0083] Figure 6 A schematic diagram of a financial data processing device provided in an embodiment of the present application.

[0084] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0085] In order to enable those skilled in the art to better understand the technical solution of the present application, the implementation mode of the present application will be further described in detail below with reference to the accompanying drawings.

[0086] It should be understood that the specific embodiments and drawings described herein are only used to explain the present application, rather than to limit the present application.

[0087] It can be understood that, in the absence of conflict, the various embodiments in the present application and the various features in the embodiments can be combined with each other.

[0088] It can be understood that, for the convenience of description, the drawings of the present application only show the parts related to the present application, while the parts unrelated to the present application are not shown in the drawings.

[0089] It can be understood that each unit and module involved in the embodiments of the present application may correspond to only one physical structure, or may be composed of multiple physical structures, or multiple units and modules may be integrated into one physical structure.

[0090] It can be understood that the terms "first", "second", etc. in the embodiments of the present application are used to distinguish different objects, or to distinguish different processing of the same object, rather than to describe a specific order of objects.

[0091] It is understandable that, in the absence of conflict, the functions and steps marked in the flowcharts and block diagrams of the present application may occur in an order different from that marked in the drawings.

[0092] It is understood that the flowcharts and block diagrams of the present application illustrate the possible architectures, functions, and operations of the systems, devices, equipment, and methods according to the various embodiments of the present application. Among them, each box in the flowchart or block diagram may represent a unit, module, program segment, code, which contains executable instructions for implementing the specified functions. Moreover, each box or combination of boxes in the block diagram and flowchart may be implemented by a hardware-based system that implements the specified functions, or by a combination of hardware and computer instructions.

[0093] It can be understood that the units and modules involved in the embodiments of the present application can be implemented by software or hardware, for example, the units and modules can be located in a processor.

[0094] Machine learning algorithms are commonly used to detect financial data fraud. Among them, the C4.5 algorithm (C4.5 decision tree) is widely used in financial data detection due to its easy understanding and interpretation, ability to process various types of data and high flexibility.

[0095] The C4.5 algorithm is a series of algorithms used in classification problems in machine learning and data mining. Its goal is supervised learning: given a data set, each tuple in it can be described by a set of attribute values, and each tuple belongs to a class in a mutually exclusive category. The goal of C4.5 is to find a mapping relationship from attribute values ​​to categories through learning, and this mapping can be used to classify new entities with unknown categories.

[0096] However, the various attributes in financial data are often numerical variables. When the number of samples is large enough, the values ​​of the same attribute in the sample basically present continuous attribute values, which causes the C4.5 algorithm to discretize the continuous attribute values. Specifically, multiple split points are inserted for a single continuous attribute, the information gain of the split points is calculated, and the discretization is performed based on the information gain. The amount of calculation when calculating the information gain increases exponentially, making the C4.5 algorithm inefficient in processing financial data with continuous attributes, which further leads to low efficiency in financial data detection.

[0097] In addition, the existing solutions have the following problems:

[0098] (1) High complexity of security management:

[0099] The encryption algorithm of the existing digital financial data management platform relies on keys to encrypt and decrypt data. Effective key management is an important part of ensuring the security of the encryption system, but the complexity and security of key management may become the target of attackers. If the key is improperly managed or expired or leaked, it may lead to data being cracked.

[0100] (2) Large computing resource requirements:

[0101] The computing resources of the existing digital financial management platform are very limited. If the encryption algorithm is more complex, a large amount of computing resources may be required to complete the security detection of financial-related data and the subsequent overall security operation of the system, especially for scenarios with high data volume and real-time processing requirements, which may affect the performance and response time of the system. It does not take advantage of the actual operation experience of users of the digital financial system.

[0102] Based on the shortcomings of the prior art, the present application provides a financial data detection method for automated operation monitoring, which determines the segmentation evaluation value of each element by analyzing the continuity characteristics of the financial-related data set, and can realize the rapid discretization processing of financial data through the segmentation evaluation value, which effectively improves the detection efficiency of financial data. When determining the segmentation evaluation value of the element, based on the financial-related data set, the element mutation of the same attribute category, the element position distribution characteristics and the association characteristics between elements of different attribute categories are analyzed, so that the numerical accuracy of the segmentation evaluation value obtained is higher; determining the segmentation evaluation value helps to avoid calculating information gain, reduces the computational overhead of logarithmic operations, improves the discretization efficiency of financial-related data, further improves the training efficiency of the financial data detection model, and improves the efficiency of financial data detection; and, when determining the segmentation evaluation value, the element position distribution characteristics are also considered, which helps to avoid the imbalance problem of the segmentation point, thereby realizing efficient and rapid discretization of continuous attribute data, which improves the computational efficiency and generalization ability of the financial data detection model, and improves the detection accuracy of financial data.

[0103] The application scenario targeted by this application is: when the traditional C4.5 algorithm processes financial data with continuity attributes, the amount of calculation required to calculate the information gain increases exponentially, making the data discretization processing efficiency of the C4.5 algorithm inefficient, which further leads to low efficiency in financial data detection.

[0104] The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0105] The present application provides a financial data detection model training and financial data detection method. The working process of the method can be implemented by electronic devices, such as computers, handheld smart terminals, etc. For the convenience of explanation, the embodiments of the present application are described with the method execution subject as a computer.

[0106] Figure 1 A schematic diagram of a financial data detection model training method provided in an embodiment of the present application, such as Figure 1 As shown, the present application provides a financial data detection model training method, the method includes S1-S4, which are as follows:

[0107] S1. Acquire a set of financial related data, wherein the set of financial related data includes a sequence of attribute values ​​of sample companies, wherein the sample companies include normal samples and abnormal samples, and the sequence of attribute values ​​includes elements of different attribute categories;

[0108] Specifically, this embodiment collects historical financial-related data of several sample companies on the automated operation monitoring platform. Since the purpose is to detect whether the company's financial data is abnormal, when obtaining the training sample data of the detection model, it is necessary to collect two types of sample companies, namely normal samples and abnormal samples. Among them, normal samples refer to sample companies that have not committed financial fraud, and abnormal samples refer to sample companies that have committed financial fraud.

[0109] Financial-related data specifically refers to continuous attribute values ​​of different attribute categories. Based on the annual reports of different sample companies under a single sector, the attribute information in the annual reports can be extracted using natural language processing to obtain a set of financial-related data. The sample companies can be listed companies, and the number of sample companies and the number of attribute categories can be set according to specific actual conditions without specific restrictions.

[0110] Among them, a single board can include classification boards such as GEM, financial board and technology board; the attribute categories collected by each sample company are the same, and each sample company can correspond to multiple attribute categories.

[0111] For example, the attribute categories corresponding to the asset class include: the proportion of accounts receivable, debt-to-asset ratio, current ratio, interest-bearing debt ratio, total asset turnover rate, asset impairment loss ratio and accounts payable ratio; the attribute categories corresponding to the cash class include: monetary funds change rate, cash ratio, free cash flow to net profit and cash sales ratio; the attribute categories corresponding to the profitability class include: total operating income multiples, non-operating income ratio, operating profit growth rate, net profit growth rate and operating income growth rate, which are all attribute values, that is, elements of each attribute category.

[0112] It should be noted that, in order to avoid the difference in the numerical distribution and data tendency of a single attribute element, this embodiment performs a forward normalization process on the elements of each attribute category in the financial-related data set, and uses the forward normalized data set as the basic data for subsequent analysis and calculation, that is, uses the forward normalized data set as the final financial-related data set. Among them, the forward normalization process refers to first performing a forward normalization process and then performing a normalization process.

[0113] S2. Determine the segmentation evaluation value of each element according to the mutation of elements of the same attribute category, the element position distribution characteristics, and the correlation characteristics between elements of different attribute categories;

[0114] In the prior art, the C4.5 algorithm has low computational efficiency when processing financial data with continuous attributes. The main reason is that in order to select the optimal segmentation threshold, the algorithm traverses the information gain value of each segmentation point, and logarithmic operations are required when calculating the information gain value, which results in a large computational overhead, making the C4.5 algorithm's financial data detection efficiency low. In addition, the C4.5 algorithm uses the optimal segmentation threshold to divide the continuous attribute into two intervals. When the difference in the interval data is small, it may cause the problem of unbalanced segmentation, which reduces the model's generalization ability and prediction accuracy.

[0115] Based on this, after obtaining the financial-related data set, this application first determines the segmentation evaluation value of each element according to the element mutation of the same attribute category, the element position distribution characteristics and the correlation characteristics between elements of different attribute categories, so as to facilitate subsequent data processing.

[0116] Figure 2 A schematic diagram of determining the segmentation evaluation value of each element provided in an embodiment of the present application, such as Figure 2 As shown, in some embodiments, S2 includes:

[0117] S21, elements of the same attribute category are grouped into element sequences, and a first segmentation evaluation factor of each element is determined according to the mutation difference of each element in each element sequence and the abnormal sample division;

[0118] S22, determining a second segmentation evaluation factor for each element according to the distance of each element in each element sequence relative to the first element and the last element in the element sequence to which it belongs;

[0119] S23, determining a third segmentation evaluation factor of each element according to the position distribution of each element in each attribute value sequence in the element sequence to which it belongs;

[0120] S24 . Determine a segmentation evaluation value of each element according to the first segmentation evaluation factor, the second segmentation evaluation factor, and the third segmentation evaluation factor of each element.

[0121] Specifically, when processing continuous attribute data, the C4.5 algorithm discretizes the continuous attribute data using segmentation points, so that normal samples and abnormal samples can be distinguished to the greatest extent by using a single or multiple segmentation points. In order to avoid the problem of low computational efficiency of the traditional segmentation point determination strategy, this embodiment requires further analysis in combination with the attribute characteristics of the financial-related data set to achieve the discretization of the continuous attribute data in the financial-related data set.

[0122] The characteristics of the elements of each attribute category when used as segmentation points may include: a large turning point in the distribution characteristics of the elements in the element sequence of the attribute category to which they belong; and the elements may be used to better distinguish normal samples from abnormal samples.

[0123] Specifically, the elements belonging to the same attribute category are first divided together to form an element sequence; in order to analyze the data distribution characteristics of the elements of the same attribute category, the elements in each element sequence need to be sorted, and the sorting order can be from small to large; according to the difference between each element in each element sequence and its two adjacent elements, the mutation degree index of each element is determined; according to the distribution of abnormal samples in the interval before and after each element in each element sequence, the abnormal division index of each element is determined; combined with the mutation degree index and the abnormal division index of each element, the first segmentation evaluation factor of each element is determined.

[0124] In some embodiments, in S21, the abnormal sample classification process of each element includes:

[0125] Mark any element as a target element, determine a first ratio of the number of abnormal samples in a first element interval before the target element to the total number of abnormal samples in the element sequence to which the target element belongs, and determine a second ratio of the number of abnormal samples in a second element interval after the target element to the total number of abnormal samples;

[0126] An abnormal classification index of the target element is determined according to a difference between the first ratio and the second ratio corresponding to the target element.

[0127] In some embodiments, in S21, the first segmentation evaluation factor is calculated by the following formula:

[0128] A1 ni =B ni ×C ni

[0129]

[0130] C ni =|α - -α + |

[0131] in:

[0132] A1 ni The first segmentation evaluation factor of the i-th element in the element sequence of the n-th attribute category;

[0133] B ni Indicates the mutation degree index of the i-th element in the element sequence of the n-th attribute category;

[0134] Cni represents the abnormal partition index of the i-th element in the element sequence of the n-th attribute category;

[0135] Δ n(i+1) The difference between the ith element and the i+1th element in the element sequence of the nth attribute category, where the difference is the absolute value of the difference between the two elements;

[0136] Δ n(i-1) The difference between the i-th element and the i-1-th element in the element sequence of the n-th attribute category;

[0137] ε represents a preset constant, which is used to prevent the denominator of the fraction from being zero. Its empirical value can be 0.01, which can be set by yourself;

[0138] α - It represents the first ratio of the number of abnormal samples in the first element interval before the i-th element in the element sequence of the n-th attribute category to the total number of abnormal samples;

[0139] α + It represents the second ratio of the number of abnormal samples in the second element interval after the i-th element in the element sequence of the n-th attribute category to the total number of abnormal samples; when determining the number of abnormal samples, the attribute values ​​corresponding to the sample companies with financial anomalies can be directly determined as abnormal data, so determining the number of abnormal samples is to determine the number of abnormal data in the element interval;

[0140] || represents the absolute value function.

[0141] In the calculation formula of the first segmentation evaluation factor, for the mutation degree index, the first element or the last element in the element sequence is generally not used as a segmentation point, so there is no need to calculate the mutation degree index of the first element and the last element in the element sequence; the mutation degree index can characterize the numerical difference between an element and its two adjacent elements. If the mutation degree index is large, it means that a large mutation occurs at the corresponding element, and the possibility of it being a segmentation point is greater. If there is an element with a small numerical difference with the previous element and a large numerical difference with the next element, it means that the mutation difference of the element is large, indicating that the element has a greater potential as a segmentation point of the element sequence to which it belongs;

[0142] For the abnormal division index, it can characterize that the proportion of abnormal samples in the two closed intervals formed by the i-th element and the first and last elements in the element sequence to which it belongs is quite different, indicating that assuming the i-th element is the segmentation point, the i-th element can better divide normal samples and abnormal samples, so the abnormal division index of the i-th element is relatively large; for the first segmentation evaluation factor, if an element changes less in the local range and the segmentation of normal samples and abnormal samples in the previous and next intervals is less obvious, the first segmentation evaluation factor of the element is smaller and more mediocre, indicating that the possibility of using the element value as a segmentation point is lower; on the contrary, if the attribute distribution at a certain element changes more and can better segment abnormal samples, the first segmentation evaluation factor of the element is smaller and the possibility of using the element value as a segmentation point is higher.

[0143] In some embodiments, S22 includes:

[0144] Calculate the distance between each element and the first element in the element sequence to which it belongs, and obtain the first distance value;

[0145] Calculate the distance between each element and the last element in the element sequence to which it belongs, and obtain a second distance value;

[0146] For each element, a second segmentation evaluation factor is determined for the first distance value and the second distance value corresponding to the element.

[0147] In this embodiment, the C4.5 algorithm uses the optimal segmentation threshold to segment the continuous element sequence into two intervals. When the data in the interval is small, it may cause an unbalanced segmentation result. Financial anomaly detection based on discretized data with poor segmentation effect will reduce the generalization ability of the model and the accuracy of the prediction results. Therefore, in order to avoid the unbalanced characteristics of the segmentation point, that is, the large deviation of the interval division, it is necessary to introduce a second segmentation evaluation factor.

[0148] In some embodiments, the second segmentation evaluation factor is calculated by the following formula:

[0149]

[0150] in:

[0151] A2 ni The second segmentation evaluation factor of the i-th element in the element sequence of the n-th attribute category;

[0152] min means to find the minimum value function;

[0153] L ni The first distance value between the ith element in the element sequence of the nth attribute category and the first element in the element sequence to which it belongs;

[0154] Lc ni The second distance value between the i-th element in the element sequence of the n-th attribute category and the last element in the element sequence to which it belongs;

[0155] exp represents an exponential function with the natural constant e as the base.

[0156] In the calculation formula of the second segmentation evaluation factor, the distance between two elements can be the Euclidean distance or the number of elements between the element intervals formed by the two elements; the second segmentation evaluation factor indicates the influence of the balance characteristics of the segmentation result on the financial detection ability of the subsequent model. The larger the selected minimum distance value, the greater the possibility that the element is in the middle of the element sequence, the higher the balance characteristics of the segmentation, and the more accurate the financial detection result; in this embodiment, exp(-min(L ni ,Lc ni )) does not have the possibility of being zero, so the special case where the denominator of the fraction is zero does not need to be considered in the calculation formula of the second segmentation evaluation factor.

[0157] In some embodiments, S23 includes:

[0158] Determine the sorting order of each element in each attribute value sequence in the element sequence to which it belongs;

[0159] Take any element in any attribute value sequence as the base element, and take other elements in the attribute value sequence where the base element is located except the base element as comparison elements;

[0160] The average value of the difference between the sorting number corresponding to the reference element and the sorting numbers corresponding to each comparison element is calculated, and the average value of the difference is used as the third segmentation evaluation factor of the reference element.

[0161] In this embodiment, the first segmentation evaluation factor only analyzes the possibility of an element being a segmentation point from the perspective of abnormal sample distribution. If the abnormal sample distribution is relatively dispersed or relatively uniform, the first segmentation evaluation factors of most elements obtained are relatively close in size. At this time, it is impossible to select an effective segmentation point only by the first segmentation evaluation factor, and further analysis is required based on the distribution characteristics of elements of different attribute categories corresponding to the same sample company.

[0162] For financial-related data sets, there are large correlation characteristics between different attribute categories corresponding to the same sample company. If a single sample company is a normal sample, the distribution patterns of the elements of different attribute categories corresponding to the sample company in the element sequence to which it belongs are relatively small; on the contrary, if a single sample company is an abnormal sample, the distribution patterns of the elements of different attribute categories corresponding to the sample company in the element sequence to which it belongs are relatively large. The large difference in distribution patterns can indicate that the attribute category distribution of the same sample company is highly disordered, and highly disordered elements have a higher potential to become segmentation points.

[0163] In some embodiments, the third segmentation evaluation factor is calculated by the following formula:

[0164]

[0165] in:

[0166] A3 x represents the third segmentation evaluation factor of the benchmark element;

[0167] N represents the number of attribute categories corresponding to the sample companies;

[0168] F j Indicates the sorting number corresponding to the j-th comparison element corresponding to the reference element;

[0169] F x Indicates the sorting sequence number corresponding to the benchmark element;

[0170] || represents the absolute value function.

[0171] In the calculation formula of the third segmentation evaluation factor, if the difference in the ranking order between the elements of a certain attribute category of a single sample company and the elements of other attribute categories is greater, it means that the correlation characteristics between the distribution of the elements of this attribute category and the elements of other attribute categories are weaker. The third segmentation evaluation factor can also characterize the distribution disorder of the elements of this attribute category. The greater the distribution disorder, the greater the possibility that the elements of this attribute category are used as segmentation points. By referring to the calculation process of the third segmentation evaluation factor of the reference element, the third segmentation evaluation factor of each element in each element sequence can be obtained.

[0172] In some embodiments, S24 includes:

[0173] Normalizing the first segmentation evaluation factor, the second segmentation evaluation factor, and the third segmentation evaluation factor of each element to obtain each normalized result value corresponding to each element;

[0174] According to each result value after normalization corresponding to the same element, the segmentation evaluation value of the corresponding element is obtained.

[0175] In this embodiment, the first segmentation evaluation factor, the second segmentation evaluation factor and the third segmentation evaluation factor of each element are all positively correlated with the segmentation evaluation value, that is, the larger the three segmentation evaluation factors are, the larger the segmentation evaluation value is.

[0176] Specifically, the first segmentation evaluation factor, the second segmentation evaluation factor and the third segmentation evaluation factor of each element are normalized to obtain the normalized result values ​​corresponding to each element; the product of the normalized result values ​​corresponding to the same element is used as the segmentation evaluation value of the corresponding element.

[0177] Optionally, the three segmentation evaluation factors of the same element may be combined by other calculation methods, for example, each result value after normalization is added together as the segmentation evaluation value of the corresponding element.

[0178] Among them, the three segmentation evaluation factors are normalized in order to unify the dimensions of the calculation process. The segmentation evaluation factors can be normalized by linear normalization.

[0179] S3, discretizing the elements of the same attribute category by segmenting the evaluation values ​​to obtain a discretized financial related data set;

[0180] S4. Perform model training based on the discretized financial-related data set to obtain a financial data detection model.

[0181] Among them, after obtaining the segmentation evaluation value of each element, a segmentation point is selected for each element sequence based on the segmentation evaluation value, and the segmentation evaluation value of the selected segmentation point is relatively large. Subsequently, the element sequence is segmented based on the selected segmentation point, and the data set after segmentation processing is used as the final input data of the model. The ratio of the training set and the test set in the input data can be set to 8:2. Among them, the ratio of the training set and the test set can be set by the implementer according to the specific actual situation, and there is no specific limitation.

[0182] Specifically, any element sequence is taken as the target element sequence, and a preset number of elements corresponding to the target segmentation evaluation values ​​are selected from all the segmentation evaluation values ​​corresponding to the target element sequence as the segmentation points of the target element sequence; the target element sequence is segmented using each segmentation point to obtain each element interval corresponding to the target element sequence; each element interval corresponding to each element sequence is obtained, and the set consisting of each element interval is used as the discretized financial-related data set; the training set in the discretized financial-related data set is substituted into the C4.5 algorithm to obtain the trained decision tree; the trained decision tree is verified and optimized using the test set to obtain the financial data detection model.

[0183] In this embodiment, for the target element sequence, the segmentation evaluation values ​​of all elements in the target element sequence are first arranged in descending order, and the number of segmentation intervals for the target element sequence is selected by a random function, denoted as W. The random interval of the random function in this embodiment is [1,7], which is a random number for selecting the random interval; it can be seen that the number of selected segmentation points is W-1, and the first W-1 elements in the target element sequence after descending order are used as segmentation points, and the target element sequence is divided into various element intervals using the segmentation points; the continuity attribute of each element sequence is discretized using the same discretization method, which can realize the discretization of the financial-related data set with continuous attributes; the discretized financial-related data set is substituted into the C4.5 algorithm, and a trained decision tree is first constructed, and then the trained decision tree is pruned using a pessimistic pruning strategy, and finally the model is trained using K-fold cross validation, and a weighted summation is performed based on the recognition accuracy of the training set and the test set, and the decision tree model with the largest sum is used as the final financial data detection model.

[0184] Figure 3 A schematic diagram of a financial data detection method provided in an embodiment of the present application, such as Figure 3 As shown, the present application provides a financial data detection method, the method comprising:

[0185] Get the attribute value sequence of the company to be detected;

[0186] Input the attribute value sequence of the company to be tested into the financial data testing model to obtain the financial testing results of the company to be tested;

[0187] Wherein, the financial data detection model is obtained through the financial data detection model training method described in the above embodiments.

[0188] Specifically, for financial data detection, we first use natural language processing to obtain the attribute information in the annual report data of the company to be detected, and then input the attribute information into the financial data detection model. The model outputs the probability of financial fraud in the corresponding company, and marks the companies with a higher probability as the financial detection results of the company to be detected. The detection results are then fed back to the audit supervision staff to facilitate further processing by the staff.

[0189] In summary, the present application provides a financial data detection model training and a financial data detection method. A financial-related data set is obtained through an automated operation monitoring platform. A first segmentation evaluation factor is obtained for the prominent features of all samples in the set under a single attribute to represent the mutation of each element under the same attribute category, which can reflect the possibility of the element as a segmentation point; and a third segmentation evaluation factor is obtained by combining the association between the elements of a certain attribute category corresponding to a single sample company and the elements of other attribute categories, reflecting the association deviation between the elements of the attribute category and the elements of other attribute categories, thereby measuring the balance and accuracy of the elements of the attribute category as segmentation points. Based on the segmentation evaluation value of each element in each element sequence, the financial-related data set can be discretized, and the discretization result is used as the input of the C4.5 algorithm to obtain a trained decision tree model; thereby, the abnormal probability is obtained for the financial data of the company to be detected using the financial data detection model to realize the detection of the financial data.

[0190] Compared with the discretization method in the traditional C4.5 algorithm, this application does not need to traverse and calculate the information gain of each segmentation point in the sample set each time, which helps to reduce the computational overhead of logarithmic operations, and obtains segmentation evaluation values ​​for the distribution characteristics of continuous attributes, and combines the distance information of the elements when calculating the segmentation evaluation value, which effectively avoids the imbalance problem of segmentation points. As a result, efficient and fast discretization processing of financial-related data sets with continuous attributes is achieved, the computational efficiency and generalization ability of the financial data detection model are improved, and the detection efficiency and accuracy of financial data are further improved.

[0191] It should be understood that, although the various steps in the flowcharts in the above-described embodiments are sequentially displayed according to the indications of the arrows, these steps are not necessarily executed sequentially in the order indicated by the arrows. Unless there is a clear description in this article, the execution of these steps is not strictly limited in order, and they can be executed in other orders. Moreover, at least a portion of the steps in the figure may include a plurality of sub-steps or a plurality of stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily to be carried out sequentially, but can be executed in turn or alternately with other steps or at least a portion of the sub-steps or stages of other steps.

[0192] Figure 4 A schematic diagram of a financial data detection model training device provided in an embodiment of the present application, such as Figure 4 As shown, the present application provides a financial data detection model training device, the device comprising:

[0193] A first acquisition module 11, which is configured to acquire a set of financial related data, wherein the set of financial related data includes a sequence of attribute values ​​of sample companies, wherein the sample companies include normal samples and abnormal samples, and the sequence of attribute values ​​includes elements of different attribute categories;

[0194] A determination module 12, which is configured to determine the segmentation evaluation value of each element according to the mutation of elements of the same attribute category, the element position distribution characteristics, and the correlation characteristics between elements of different attribute categories;

[0195] Discretization processing module 13, which is configured to discretize elements of the same attribute category by segmenting evaluation values ​​to obtain a discretized financial related data set;

[0196] The model training module 14 is configured to perform model training based on the discretized financial related data set to obtain a financial data detection model.

[0197] Regarding the limitation on the financial data detection model training device, reference may be made to the limitation on the financial data detection model training method in the above-mentioned embodiments of the present application, which will not be elaborated in this embodiment.

[0198] Figure 5 A schematic diagram of a financial data detection device provided in an embodiment of the present application, such as Figure 5 As shown, the present application provides a financial data detection device, the device comprising:

[0199] A second acquisition module 21, which is configured to acquire a sequence of attribute values ​​of a company to be detected;

[0200] A financial detection module 22, which is configured to input the attribute value sequence of the company to be detected into the financial data detection model to obtain the financial detection result of the company to be detected;

[0201] Wherein, the financial data detection model is obtained through the above-mentioned financial data detection model training method.

[0202] Regarding the limitation of the financial data detection device, reference may be made to the limitation of the financial data detection method in the above-mentioned embodiments of the present application, which will not be elaborated in this embodiment.

[0203] Figure 6 A schematic diagram of a financial data processing device provided in an embodiment of the present application, such as Figure 6 As shown, in some embodiments, the present application provides a financial data processing device, including a memory 32 and a processor 31, the memory stores a computer program, and the processor is configured to run the computer program to execute the financial data detection model training and financial data detection method in the above-mentioned embodiments of the present application.

[0204] The memory is connected to the processor, the memory may be a flash memory or a read-only memory or other memory, and the processor may be a central processing unit or a single-chip microcomputer.

[0205] In some embodiments, the present application provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the financial data detection model training and the financial data detection method in the above-mentioned embodiments of the present application are implemented.

[0206] The computer-readable storage medium includes volatile or non-volatile, removable or non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, computer program modules or other data). Computer-readable storage media include, but are not limited to, RAM (Random Access Memory), ROM (Read-Only Memory), EEPROM (Electrically Erasable Programmable read only memory), flash memory or other memory technology, CD-ROM (Compact Disc Read-Only Memory), digital versatile disk (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer.

[0207] It is to be understood that the above embodiments are merely exemplary embodiments used to illustrate the principles of the present application, but the present application is not limited thereto. For those skilled in the art, various modifications and improvements can be made without departing from the spirit and substance of the present application, and these modifications and improvements are also considered to be within the scope of protection of the present application.

Claims

1. A financial data detection model training method, characterized in that: The method comprises: S1. Acquire a set of financial related data, wherein the set of financial related data includes a sequence of attribute values ​​of sample companies, wherein the sample companies include normal samples and abnormal samples, and the sequence of attribute values ​​includes elements of different attribute categories; S2. Determine the segmentation evaluation value of each element according to the mutation of elements of the same attribute category, the element position distribution characteristics, and the correlation characteristics between elements of different attribute categories; S3, discretizing the elements of the same attribute category by segmenting the evaluation values ​​to obtain a discretized financial related data set; S4. Perform model training based on the discretized financial-related data set to obtain a financial data detection model.

2. The financial data detection model training method according to claim 1, characterized in that: S2, including: S21, elements of the same attribute category are grouped into element sequences, and a first segmentation evaluation factor of each element is determined according to the mutation difference of each element in each element sequence and the abnormal sample division; S22, determining a second segmentation evaluation factor for each element according to the distance of each element in each element sequence relative to the first element and the last element in the element sequence to which it belongs; S23, determining a third segmentation evaluation factor of each element according to the position distribution of each element in each attribute value sequence in the element sequence to which it belongs; S24 . Determine a segmentation evaluation value of each element according to the first segmentation evaluation factor, the second segmentation evaluation factor, and the third segmentation evaluation factor of each element.

3. The financial data detection model training method according to claim 1, characterized in that: In S21, the abnormal sample division process of each element includes: Mark any element as a target element, determine a first ratio of the number of abnormal samples in a first element interval before the target element to the total number of abnormal samples in the element sequence to which the target element belongs, and determine a second ratio of the number of abnormal samples in a second element interval after the target element to the total number of abnormal samples; An abnormal classification index of the target element is determined according to a difference between the first ratio and the second ratio corresponding to the target element.

4. The financial data detection model training method according to claim 3, characterized in that: In S21, the first segmentation evaluation factor is calculated by the following formula: A1 ni =B ni ×C ni C ni ＝|α - -α + | in: A1 ni The first segmentation evaluation factor of the i-th element in the element sequence of the n-th attribute category; B ni Indicates the mutation degree index of the i-th element in the element sequence of the n-th attribute category; C ni represents the abnormal partition index of the i-th element in the element sequence of the n-th attribute category; Δ n(i+1) The difference between the i-th element and the i+1-th element in the element sequence of the n-th attribute category; Δ n(i-1) The difference between the i-th element and the i-1-th element in the element sequence of the n-th attribute category; ε represents a preset constant; α - It represents the first ratio of the number of abnormal samples in the first element interval before the i-th element in the element sequence of the n-th attribute category to the total number of abnormal samples; α + represents the second ratio of the number of abnormal samples in the second element interval after the i-th element in the element sequence of the n-th attribute category to the total number of abnormal samples; || represents the absolute value function.

5. The financial data detection model training method according to claim 2, characterized in that: S22, including: Calculate the distance between each element and the first element in the element sequence to which it belongs, and obtain the first distance value; Calculate the distance between each element and the last element in the element sequence to which it belongs, and obtain a second distance value; For each element, a second segmentation evaluation factor is determined for the first distance value and the second distance value corresponding to the element.

6. The financial data detection model training method according to claim 5, characterized in that: The second segmentation evaluation factor is calculated by the following formula: in: A2 ni The second segmentation evaluation factor of the i-th element in the element sequence of the n-th attribute category; min means to find the minimum value function; L ni The first distance value between the ith element in the element sequence of the nth attribute category and the first element in the element sequence to which it belongs; Lc ni The second distance value between the i-th element in the element sequence of the n-th attribute category and the last element in the element sequence to which it belongs; exp represents an exponential function with the natural constant e as the base.

7. The financial data detection model training method according to claim 2, characterized in that: S23, including: Determine the sorting order of each element in each attribute value sequence in the element sequence to which it belongs; Take any element in any attribute value sequence as the base element, and take other elements in the attribute value sequence where the base element is located except the base element as comparison elements; The average value of the difference between the sorting number corresponding to the reference element and the sorting numbers corresponding to each comparison element is calculated, and the average value of the difference is used as the third segmentation evaluation factor of the reference element.

8. The financial data detection model training method according to claim 7, characterized in that: The third segmentation evaluation factor is calculated by the following formula: in: A3 x represents the third segmentation evaluation factor of the benchmark element; N represents the number of attribute categories corresponding to the sample companies; F j Indicates the sorting number corresponding to the j-th comparison element corresponding to the reference element; F x Indicates the sorting sequence number corresponding to the benchmark element; || represents the absolute value function.

9. The financial data detection model training method according to claim 2, characterized in that: S24, including: Normalizing the first segmentation evaluation factor, the second segmentation evaluation factor, and the third segmentation evaluation factor of each element to obtain each normalized result value corresponding to each element; According to each result value after normalization corresponding to the same element, the segmentation evaluation value of the corresponding element is obtained.

10. A financial data detection method, characterized in that: The method comprises: Get the attribute value sequence of the company to be detected; Input the attribute value sequence of the company to be tested into the financial data testing model to obtain the financial testing results of the company to be tested; Wherein, the financial data detection model is obtained by the financial data detection model training method according to any one of claims 1-9.

11. A financial data detection model training device, characterized in that: The device comprises: A first acquisition module is configured to acquire a set of financial related data, wherein the set of financial related data includes a sequence of attribute values ​​of sample companies, wherein the sample companies include normal samples and abnormal samples, and the sequence of attribute values ​​includes elements of different attribute categories; A determination module is configured to determine the segmentation evaluation value of each element according to the mutation of elements of the same attribute category, the element position distribution characteristics, and the correlation characteristics between elements of different attribute categories; A discrete processing module, which is configured to discretize elements of the same attribute category by segmenting the evaluation values ​​to obtain a discretized financial related data set; The model training module is configured to perform model training based on a discretized set of financial related data to obtain a financial data detection model.

12. A financial data detection device, characterized in that: The device comprises: A second acquisition module is configured to acquire a sequence of attribute values ​​of the company to be detected; A financial detection module, which is configured to input the attribute value sequence of the company to be detected into the financial data detection model to obtain the financial detection result of the company to be detected; Wherein, the financial data detection model is obtained by the financial data detection model training method according to any one of claims 1-9.

13. A financial data processing device, characterized in that: It comprises a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to implement the financial data detection model training method as described in any one of claims 1 to 9 or the financial data detection method as described in claim 10.

14. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by the processor, the financial data detection model training method according to any one of claims 1 to 9 or the financial data detection method according to claim 10 is implemented.

Images