A method and device for training a money laundering model and an electronic device

Abstract

The application discloses a training method and device of a money laundering model and electronic equipment, and the training method comprises the following steps: acquiring an initial labeled transaction sample set; performing machine learning model training based on the initial labeled transaction sample set according to a preselected machine learning algorithm to obtain a prediction model; acquiring an unlabeled transaction sample set, performing prediction processing on the unlabeled transaction sample set according to the prediction model, and screening at least one transaction sample to be labeled from the unlabeled transaction sample set according to a prediction result; obtaining at least one new labeled transaction sample after the at least one transaction sample to be labeled is labeled, and performing iterative update training on the prediction model according to the at least one new labeled transaction sample; and judging whether the prediction model satisfies a preset termination condition, and if yes, obtaining a money laundering model based on the prediction model, and if not, returning to the step of acquiring the unlabeled transaction sample set.

Description

[0001] This application is a divisional application of patent application No. 201910839229.7, filed on September 5, 2019, entitled "A training method, apparatus and electronic device for an anti-money laundering model". Technical Field

[0002] This invention relates to the field of machine learning model technology, and more specifically, to a training method for an anti-money laundering model, a training device for an anti-money laundering model, an electronic device, and a readable storage medium. Background Technology

[0003] Anti-money laundering refers to the prevention of money laundering activities that attempt to conceal or disguise the source and nature of proceeds and profits from crimes such as drug trafficking, organized crime, terrorism, smuggling, corruption, bribery, and crimes disrupting financial order. Anti-money laundering is a common risk control scenario and is a systematic project aimed at identifying potential money laundering activities and combating the laundering of illicit gains and profits through various channels.

[0004] With the development of machine learning and data mining technologies, machine learning models are commonly used in current technologies to identify money laundering activities. In recent years, machine learning models have achieved considerable success in anti-money laundering scenarios. However, machine learning requires all samples to be labeled by experts, and training anti-money laundering machine learning models requires a large number of samples. This leads to a significant increase in human resources and labor costs for sample labeling. Moreover, it introduces redundant samples into the training set, which is detrimental to the training of anti-money laundering models. Summary of the Invention

[0005] This invention provides a novel technical solution for training anti-money laundering models.

[0006] According to a first aspect of the present invention, a method for training an anti-money laundering model is provided, comprising:

[0007] Obtain an initial set of labeled transaction samples, where each sample in the set has been labeled as either a money laundering transaction or not a money laundering transaction;

[0008] Based on the pre-selected machine learning algorithm, a machine learning model is trained on the initial set of labeled transaction samples to obtain a prediction model;

[0009] Obtain an unlabeled transaction sample set, perform prediction processing on the unlabeled transaction sample set according to the prediction model, and select at least one transaction sample to be labeled from the unlabeled transaction sample set based on the prediction results.

[0010] After the at least one transaction sample to be labeled is labeled, at least one new labeled transaction sample is obtained, and the prediction model is iteratively updated and trained based on the at least one new labeled transaction sample;

[0011] Determine whether the prediction model meets the preset termination condition. If yes, obtain the anti-money laundering model based on the prediction model; otherwise, return to the step of obtaining the unlabeled transaction sample set.

[0012] Optionally, the step of performing prediction processing on the unlabeled transaction sample set according to the prediction model, and selecting at least one transaction sample to be labeled from the unlabeled transaction sample set based on the prediction result includes:

[0013] From the set of unlabeled transaction samples, select unlabeled transaction samples whose prediction results fall within a first set range, and use them as the transaction samples to be labeled.

[0014] Optionally, the prediction model includes a target anti-money laundering model and a screening model composed of multiple models;

[0015] The step of performing prediction processing on the unlabeled transaction sample set according to the prediction model, and selecting at least one transaction sample to be labeled from the unlabeled transaction sample set based on the prediction results includes:

[0016] The unlabeled transaction sample set is predicted based on the target anti-money laundering model to obtain the first prediction result for each unlabeled transaction sample;

[0017] From the set of unlabeled transaction samples, select unlabeled transaction samples within a first set range of the first prediction result as the candidate transaction samples to be labeled;

[0018] For each candidate transaction sample to be labeled, predictions are made according to the multiple models that make up the screening model to obtain the corresponding second prediction result;

[0019] Based on the corresponding second prediction result, at least one transaction sample to be labeled is selected from the candidate transaction samples to be labeled.

[0020] Optionally, the step of selecting the at least one transaction sample to be labeled from the candidate transaction samples to be labeled based on the corresponding second prediction result includes:

[0021] Based on the second prediction result corresponding to each candidate transaction sample to be labeled, the candidate transaction samples to be labeled are deduplicated to obtain the at least one transaction sample to be labeled.

[0022] Optionally, the step of deduplicating the candidate transaction samples to be labeled based on the second prediction result corresponding to each candidate sample to be labeled, to obtain the transaction sample to be labeled, includes:

[0023] For each candidate transaction sample to be labeled, the label value corresponding to the second prediction result within the second set range is set to the first set value; the label value corresponding to the second prediction result outside the second set range is set to the second set value.

[0024] For each candidate transaction sample to be labeled, a label value vector is obtained based on a preset order and the label value corresponding to the second prediction result of the corresponding candidate transaction sample;

[0025] The candidate transaction samples to be labeled are deduplicated based on the labeled value vector to obtain the transaction samples to be labeled.

[0026] Optionally, the termination condition includes any one or more of the following:

[0027] The number of candidate transaction samples to be labeled is less than or equal to a preset first quantity threshold;

[0028] The proportion of candidate transaction samples to be labeled selected from the unlabeled transaction sample set is less than or equal to a preset first proportion threshold.

[0029] The number of unlabeled transaction samples is less than or equal to a preset second quantity threshold;

[0030] In the unlabeled transaction sample set, the proportion of unlabeled transaction samples is less than or equal to a preset second proportion threshold.

[0031] Optionally, multiple prediction models may use different algorithms or have different hyperparameters.

[0032] Optionally, the method for obtaining the anti-money laundering model based on the prediction model includes any one of the following:

[0033] The target anti-money laundering model is used as the anti-money laundering model;

[0034] Select any one of the screening models as the anti-money laundering model;

[0035] The anti-money laundering model is obtained by combining at least two of the screening models.

[0036] The anti-money laundering model is obtained by combining the target anti-money laundering model with at least one of the screening models.

[0037] Optionally, the prediction model consists of multiple models;

[0038] The step of performing prediction processing on the unlabeled transaction sample set according to the prediction model, and selecting at least one transaction sample to be labeled from the unlabeled transaction sample set based on the prediction result includes:

[0039] The unlabeled transaction sample set is predicted based on one or at least two of the prediction models to obtain a first prediction result for each unlabeled transaction sample;

[0040] Based on the first prediction result, candidate transaction samples that meet the preset conditions are selected from the set of unlabeled transaction samples;

[0041] For each candidate transaction sample to be labeled, predictions are made according to the multiple models that make up the prediction model, and the corresponding second prediction result is obtained.

[0042] Based on the corresponding second prediction result, at least one transaction sample to be labeled is selected from the candidate transaction samples to be labeled.

[0043] Optionally, the step of predicting the unlabeled transaction sample set based on one or at least two of the prediction models to obtain a first prediction result for each unlabeled transaction sample includes:

[0044] One model is selected from the multiple models that make up the prediction model. For each unlabeled transaction sample, a prediction is made according to the selected model to obtain the corresponding first prediction result.

[0045] or,

[0046] At least two models are selected from the multiple models that make up the prediction model. For each unlabeled transaction sample, predictions are made according to the selected at least two models to obtain prediction results. The prediction results corresponding to the at least two models are then averaged to obtain a first prediction result.

[0047] Optionally, selecting the at least one transaction sample to be labeled from the candidate transaction samples based on the corresponding second prediction result includes:

[0048] For each candidate transaction sample to be labeled, the label value corresponding to the second prediction result within the second set range is set to the first set value; the label value corresponding to the second prediction result outside the second set range is set to the second set value.

[0049] For each candidate transaction sample to be labeled, a label value vector is obtained based on a preset order and the label value corresponding to the second prediction result of the corresponding candidate transaction sample;

[0050] The candidate transaction samples to be labeled are deduplicated according to the label value vector to obtain the at least one transaction sample to be labeled.

[0051] Optionally, the step of obtaining the anti-money laundering model based on the prediction model includes:

[0052] Choose any prediction model as the anti-money laundering model; or,

[0053] The anti-money laundering model is obtained by combining at least two prediction models.

[0054] Optionally, the machine learning algorithm is the random forest algorithm.

[0055] Optionally, the training method further includes:

[0056] Obtain a sample of the target transactions to be predicted;

[0057] The anti-money laundering model is used to predict the target transaction sample, and the prediction result of the target transaction sample is obtained.

[0058] Optionally, the training method further includes:

[0059] The prediction results for the target sample are displayed.

[0060] According to a second aspect of the present invention, a training apparatus for an anti-money laundering model is provided, comprising:

[0061] The initial sample acquisition module is used to acquire an initial set of labeled transaction samples, wherein each sample in the labeled sample set has been labeled as either a money laundering transaction or not a money laundering transaction;

[0062] The model initial training module is used to train a machine learning model based on the initial set of labeled transaction samples according to a pre-selected machine learning algorithm to obtain a prediction model;

[0063] The unlabeled sample screening module is used to obtain an unlabeled transaction sample set, perform prediction processing on the unlabeled transaction sample set according to the prediction model, and screen out at least one unlabeled transaction sample from the unlabeled transaction sample set according to the prediction result.

[0064] The model iterative training module is used to obtain at least one new labeled transaction sample after the at least one transaction sample to be labeled is labeled, and to iteratively update and train the prediction model based on the at least one new labeled transaction sample;

[0065] The termination condition judgment module is used to determine whether the prediction model meets the preset termination condition. If it does, an anti-money laundering model is obtained based on the prediction model; otherwise, the process returns to the step of obtaining the unlabeled transaction sample set.

[0066] Optionally, the step of performing prediction processing on the unlabeled transaction sample set according to the prediction model, and selecting at least one transaction sample to be labeled from the unlabeled transaction sample set based on the prediction result includes:

[0067] From the set of unlabeled transaction samples, select unlabeled transaction samples whose prediction results fall within a first set range, and use them as the transaction samples to be labeled.

[0068] Optionally, the prediction model includes a target anti-money laundering model and a screening model composed of multiple models;

[0069] The step of performing prediction processing on the unlabeled transaction sample set according to the prediction model, and selecting at least one transaction sample to be labeled from the unlabeled transaction sample set based on the prediction results, includes:

[0070] The unlabeled transaction sample set is predicted based on the target anti-money laundering model to obtain the first prediction result for each unlabeled transaction sample;

[0071] From the set of unlabeled transaction samples, select unlabeled transaction samples within a first set range of the first prediction result as the candidate transaction samples to be labeled;

[0072] For each candidate transaction sample to be labeled, predictions are made according to the multiple models that make up the screening model to obtain the corresponding second prediction result;

[0073] Based on the corresponding second prediction result, at least one transaction sample to be labeled is selected from the candidate transaction samples to be labeled.

[0074] Optionally, selecting the at least one transaction sample to be labeled from the candidate transaction samples based on the corresponding second prediction result includes:

[0075] Based on the second prediction result corresponding to each candidate transaction sample to be labeled, the candidate transaction samples to be labeled are deduplicated to obtain the at least one transaction sample to be labeled.

[0076] Optionally, the step of deduplicating the candidate transaction samples to be labeled based on the second prediction result corresponding to each candidate sample to be labeled, to obtain the transaction sample to be labeled, includes:

[0077] For each candidate transaction sample to be labeled, the label value corresponding to the second prediction result within the second set range is set to the first set value; the label value corresponding to the second prediction result outside the second set range is set to the second set value.

[0078] For each candidate transaction sample to be labeled, a label value vector is obtained based on a preset order and the label value corresponding to the second prediction result of the corresponding candidate transaction sample;

[0079] The candidate transaction samples to be labeled are deduplicated based on the labeled value vector to obtain the transaction samples to be labeled.

[0080] Optionally, the termination condition includes any one or more of the following:

[0081] The number of candidate transaction samples to be labeled is less than or equal to a preset first quantity threshold;

[0082] The proportion of candidate transaction samples to be labeled selected from the unlabeled transaction sample set is less than or equal to a preset first proportion threshold.

[0083] The number of unlabeled transaction samples is less than or equal to a preset second quantity threshold;

[0084] In the unlabeled transaction sample set, the proportion of unlabeled transaction samples is less than or equal to a preset second proportion threshold.

[0085] Optionally, multiple prediction models may use different algorithms or have different hyperparameters.

[0086] Optionally, the anti-money laundering model derived from the prediction model includes any one of the following:

[0087] The target anti-money laundering model is used as the anti-money laundering model;

[0088] Select any one of the screening models as the anti-money laundering model;

[0089] The anti-money laundering model is obtained by combining at least two of the screening models.

[0090] The anti-money laundering model is obtained by combining the target anti-money laundering model with at least one of the screening models.

[0091] Optionally, the prediction model consists of multiple models;

[0092] The step of performing prediction processing on the unlabeled transaction sample set according to the prediction model, and selecting at least one transaction sample to be labeled from the unlabeled transaction sample set based on the prediction result includes:

[0093] The unlabeled transaction sample set is predicted based on one or at least two of the prediction models to obtain a first prediction result for each unlabeled transaction sample;

[0094] Based on the first prediction result, candidate transaction samples that meet the preset conditions are selected from the set of unlabeled transaction samples;

[0095] For each candidate transaction sample to be labeled, predictions are made according to the multiple models that make up the prediction model, and the corresponding second prediction result is obtained.

[0096] Based on the corresponding second prediction result, at least one transaction sample to be labeled is selected from the candidate transaction samples to be labeled.

[0097] Optionally, the step of predicting the unlabeled transaction sample set based on one or at least two of the prediction models to obtain a first prediction result for each unlabeled transaction sample includes:

[0098] One model is selected from the multiple models that make up the prediction model. For each unlabeled transaction sample, a prediction is made according to the selected model to obtain the corresponding first prediction result.

[0099] or,

[0100] At least two models are selected from the multiple models that make up the prediction model. For each unlabeled transaction sample, predictions are made according to the selected at least two models to obtain prediction results. The prediction results corresponding to the at least two models are then averaged to obtain a first prediction result.

[0101] Optionally, selecting the at least one transaction sample to be labeled from the candidate transaction samples based on the corresponding second prediction result includes:

[0102] For each candidate transaction sample to be labeled, the label value corresponding to the second prediction result within the second set range is set to the first set value; the label value corresponding to the second prediction result outside the second set range is set to the second set value.

[0103] For each candidate transaction sample to be labeled, a label value vector is obtained based on a preset order and the label value corresponding to the second prediction result of the corresponding candidate transaction sample;

[0104] The candidate transaction samples to be labeled are deduplicated according to the label value vector to obtain the at least one transaction sample to be labeled.

[0105] Optionally, obtaining the anti-money laundering model based on the prediction model includes:

[0106] Choose any prediction model as the anti-money laundering model; or,

[0107] The anti-money laundering model is obtained by combining at least two prediction models.

[0108] Optionally, the machine learning algorithm is the random forest algorithm.

[0109] Optionally, the training device further includes:

[0110] A module used to obtain a sample of the target transactions to be predicted;

[0111] A module for predicting the target transaction sample based on the anti-money laundering model to obtain the prediction result of the target transaction sample.

[0112] Optionally, the training device further includes:

[0113] A module used to display the prediction results of the target sample.

[0114] According to a third aspect of the present invention, an electronic device is provided, comprising:

[0115] The training apparatus as described in the second aspect of the present invention; or...

[0116] A processor and a memory, the memory being used to store instructions for controlling the processor to execute the training method according to the first aspect of the invention.

[0117] According to a fourth aspect of the present invention, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the training method as described in the first aspect of the present invention.

[0118] In the embodiments of this invention, a predictive model is first obtained by training a machine learning algorithm based on a small initial set of labeled transaction samples using a pre-selected machine learning algorithm. Then, based on the prediction results of this model for unlabeled transaction samples, transaction samples requiring expert labeling are selected. Finally, the predictive model is iteratively trained using the newly labeled transaction samples obtained after experts have labeled the unlabeled samples, resulting in an anti-money laundering model. This reduces the number of transaction samples requiring expert labeling, making it easier for the predictive model to learn valuable information. Furthermore, by deduplicating the candidate unlabeled transaction samples, the computational load during the iterative training of the predictive model can be reduced.

[0119] Other features and advantages of the invention will become clear from the following detailed description of exemplary embodiments of the invention with reference to the accompanying drawings. Attached Figure Description

[0120] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments of the invention and, together with their description, serve to explain the principles of the invention.

[0121] Figure 1 This is a block diagram showing an example of a hardware configuration of an electronic device that can be used to implement embodiments of the present invention;

[0122] Figure 2 A flowchart illustrating the training method of the anti-money laundering model according to an embodiment of the present invention is shown;

[0123] Figure 3 A flowchart illustrating the steps of screening transaction samples to be labeled according to an embodiment of the present invention is shown;

[0124] Figure 4 A flowchart illustrating an example of a training method for an anti-money laundering model according to an embodiment of the present invention is shown.

[0125] Figure 5 A schematic block diagram of a training device for an anti-money laundering model according to an embodiment of the present invention is shown.

[0126] Figure 6 A block diagram of an electronic device according to an embodiment of the present invention is shown. Detailed Implementation

[0127] Various exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of the invention.

[0128] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the invention or its application or use.

[0129] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.

[0130] In all the examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.

[0131] It should be noted that similar labels and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be discussed further in subsequent figures.

[0132] Hereinafter, various embodiments and examples according to the present invention will be described with reference to the accompanying drawings.

[0133] <Hardware Configuration>

[0134] Figure 1 This is a block diagram illustrating the hardware configuration of an electronic device 1000 that can implement embodiments of the present invention.

[0135] Electronic device 1000 can be a portable computer, desktop computer, mobile phone, tablet computer, etc. For example... Figure 1 As shown, the electronic device 1000 may include a processor 1100, a memory 1200, an interface device 1300, a communication device 1400, a display device 1500, an input device 1600, a speaker 1700, a microphone 1800, etc. The processor 1100 may be a central processing unit (CPU), a microprocessor (MCU), etc. The memory 1200 may include, for example, ROM (Read-Only Memory), RAM (Random Access Memory), or non-volatile memory such as a hard disk. The interface device 1300 may include, for example, a USB interface, a headphone jack, etc. The communication device 1400 may be capable of wired or wireless communication, specifically including Wi-Fi communication, Bluetooth communication, 2G / 3G / 4G / 5G communication, etc. The display device 1500 may be, for example, an LCD screen, a touch screen, etc. The input device 1600 may include, for example, a touch screen, a keyboard, motion input, etc. Users can input / output voice information through the speaker 1700 and the microphone 1800.

[0136] Figure 1 The electronic devices shown are merely illustrative and in no way intended to limit the invention, its application, or use. In embodiments applied to the invention, the memory 1200 of the electronic device 1000 is used to store instructions for controlling the processor 1100 to operate and execute any of the anti-money laundering model training methods provided in the embodiments of the invention. Those skilled in the art will understand that, although... Figure 1 The electronic device 1000 is shown with multiple devices shown; however, the present invention may refer only to some of these devices. For example, the electronic device 1000 may refer only to the processor 1100 and the storage device 1200. Those skilled in the art can design instructions according to the disclosed scheme of the present invention. How the instructions control the processor to operate is well known in the art and will not be described in detail here.

[0137] <Method Implementation>

[0138] In this embodiment, a method for training an anti-money laundering model is provided. This method can be implemented by an electronic device. The electronic device can be, for example,... Figure 1 The electronic device shown is 1000.

[0139] according to Figure 2As shown, the training method for the anti-money laundering model in this embodiment may include the following steps S2100 to S2500:

[0140] Step S2100: Obtain the initial set of marked transaction samples.

[0141] Each sample in the labeled sample set has been labeled as either a money laundering transaction or not a money laundering transaction.

[0142] For example, samples that are money laundering transactions in the marked transaction sample set can be marked as 1, while samples that are not money laundering transactions can be marked as 0.

[0143] In this embodiment, each sample in the labeled transaction sample set can be labeled by an expert or by an electronic device according to pre-set rules.

[0144] Step S2200: Based on the pre-selected machine learning algorithm, train the machine learning model on the initial set of labeled transaction samples to obtain the prediction model.

[0145] In one embodiment, the pre-selected machine learning algorithm may be any one or more of the following machine learning algorithms: random forest algorithm, GBDT (Gradient Boosting Decison Tree) algorithm, XGBOOST (eXtreme Gradient Boosting) algorithm, logistic regression algorithm, neural network algorithm, etc.

[0146] Based on the labeled transaction sample set obtained in step S2100 and the pre-selected machine learning algorithm, a machine learning model can be trained to obtain at least one prediction model.

[0147] Step S2300: Obtain an unlabeled transaction sample set, perform prediction processing on the unlabeled transaction sample set according to the prediction model, and select at least one transaction sample to be labeled from the unlabeled transaction sample set based on the prediction results.

[0148] The unlabeled transaction sample set includes multiple unlabeled transaction samples, meaning that each unlabeled transaction sample has not been pre-labeled as either a money laundering transaction or not a money laundering transaction.

[0149] In the first embodiment, the prediction model can be a single model. By performing prediction processing on the unlabeled transaction sample set based on this model, a prediction result for each unlabeled transaction sample can be obtained. This prediction result can be a score between 0 and 1, representing the probability that the corresponding transaction sample is a money laundering transaction based on the prediction model.

[0150] In this embodiment, unlabeled transaction samples whose prediction results fall within a first set range can be selected from unlabeled transaction samples and used as transaction samples to be labeled.

[0151] The closer the prediction result is to 1, the greater the likelihood that the prediction model considers the corresponding transaction sample to be a money laundering transaction. Conversely, the closer the prediction result is to 0, the greater the likelihood that the prediction model considers the corresponding transaction sample not to be a money laundering transaction. In other words, the larger the absolute value of the difference between the prediction result and 0.5, the easier it is for the prediction model to determine the type of the corresponding transaction sample (i.e., whether it is a money laundering transaction or not). The larger the absolute value of the difference between the prediction result of a transaction sample and 0.5, the less labeling value it has for the prediction model. Since only transaction samples that the prediction model cannot accurately predict have labeling value, unlabeled transaction samples whose prediction results fall within a first set range can be selected as the transaction samples to be labeled.

[0152] The first set range can be pre-defined according to the application scenario or specific needs. For example, the first set range can be [0.4, 0.6]. In this case, unlabeled transaction samples with prediction results greater than or equal to 0.4 and less than or equal to 0.6 can be selected as transaction samples to be labeled.

[0153] In the second embodiment, the prediction model includes a target anti-money laundering model and a screening model consisting of multiple models.

[0154] Furthermore, the multiple prediction models employ different machine learning algorithms and / or have different hyperparameter settings, so that the multiple prediction models obtained by training the machine learning models based on the same initial set of labeled transaction samples are different.

[0155] Therefore, performing prediction processing on the unlabeled transaction sample set based on the prediction model, and selecting at least one transaction sample to be labeled from the unlabeled transaction sample set based on the prediction results, can further include, for example: Figure 3 Steps S2310 to S2340 are shown below:

[0156] Step S2310: Based on the target anti-money laundering model, predict the unlabeled transaction sample set to obtain the first prediction result for each unlabeled transaction sample.

[0157] The first prediction result can be a score between 0 and 1, used to represent the probability of predicting whether the corresponding transaction sample is a money laundering transaction based on the target anti-money laundering model.

[0158] Step S2320: Select the labeled transaction samples whose first prediction results fall within a first set range from the unlabeled transaction samples as candidate transaction samples to be labeled.

[0159] The closer the first prediction result is to 1, the greater the likelihood that the target anti-money laundering model considers the corresponding transaction sample to be a money laundering transaction. Conversely, the closer the first prediction result is to 0, the greater the likelihood that the target anti-money laundering model considers the corresponding transaction sample not to be a money laundering transaction. In other words, the larger the absolute value of the difference between the first prediction result and 0.5, the easier it is for the target anti-money laundering model to determine the type of the corresponding transaction sample (i.e., whether it is a money laundering transaction or not). The larger the absolute value of the difference between the first prediction result and 0.5, the less labeling value it has for the target anti-money laundering model. Since only transaction samples that the target anti-money laundering model cannot accurately predict have labeling value, unlabeled transaction samples whose first prediction results fall within a first set range can be selected as candidate transaction samples to be labeled.

[0160] The first set range can be pre-defined according to the application scenario or specific needs. For example, the first set range can be [0.4, 0.6]. In this case, unlabeled transaction samples with prediction results greater than or equal to 0.4 and less than or equal to 0.6 can be selected as candidate transaction samples to be labeled.

[0161] Step S2330: For each candidate transaction sample to be labeled, predictions are made according to the multiple models that make up the screening model to obtain the corresponding second prediction result.

[0162] Specifically, based on multiple models that make up the screening model, predictions can be made for each candidate transaction sample to be labeled, and the second prediction result of each model in the screening model corresponding to each candidate transaction sample to be labeled can be obtained.

[0163] For example, the prediction model includes a screening model consisting of n models. Step S2320 yields m candidate transaction samples to be labeled. The second prediction result of the j-th (j∈[1,n]) model in the screening model corresponding to the i-th (i∈[1,m]) candidate transaction sample to be labeled can be... .

[0164] Step S2340: Based on the corresponding second prediction result, at least one transaction sample to be labeled is selected from the candidate transaction samples to be labeled.

[0165] In one embodiment, the candidate transaction samples to be labeled may be deduplicated based on the second prediction result of each candidate transaction sample to be labeled, so as to obtain at least one transaction sample to be labeled.

[0166] In one embodiment, the method for deduplicating candidate transaction samples to be labeled may include the following steps S2341 to S2343:

[0167] Step S2341: For each candidate transaction sample to be labeled, set the label value corresponding to the second prediction result within the second set range to the first set value; set the label value corresponding to the second prediction result outside the second set range to the second set value.

[0168] In this embodiment, the determination of whether the candidate transaction sample to be labeled is a money laundering transaction can be made by whether the second prediction result of the candidate transaction sample to be labeled is within a second set range.

[0169] For example, if the second prediction result is within a second set range, the corresponding model predicts that the candidate transaction sample to be labeled is not a money laundering transaction; if the second prediction result exceeds the second set range, the corresponding model predicts that the candidate transaction sample to be labeled is a money laundering transaction.

[0170] The second setting range can be preset according to the application scenario or specific needs. For example, the second setting range can be [0, 0.5]. Then, the flag value of the second prediction result that is greater than or equal to 0 and less than or equal to 0.5 (i.e., within the second setting range) can be set as the first setting value, and the flag value of the second prediction result that is greater than 0.5 (i.e., outside the second setting range) can be set as the second setting value.

[0171] The first and second settings can also be specific values ​​preset according to the application scenario or specific needs, and the specific values ​​of the first and second settings can be different. For example, the first setting can be 0, and the second setting can be 1.

[0172] In one example, the number of models constituting the screening model is 3. The second prediction results obtained by predicting 4 candidate transaction samples to be labeled can be shown in Table 1 below:

[0173] Table 1

[0174]

[0175] Set the label value corresponding to the second prediction result within the second set range in Table 1 to the first set value 0; set the label value corresponding to the second prediction result outside the second set range to the second set value 1. The label values ​​for each model corresponding to the four candidate transaction samples to be labeled can be as shown in Table 2 below:

[0176] Table 2

[0177]

[0178] Step S2342: For each candidate transaction sample to be labeled, based on a preset order, a label value vector is obtained according to the label value corresponding to the second prediction result of the corresponding candidate transaction sample to be labeled.

[0179] The preset order can be the order of multiple models that make up the filtering model, and this order can be set in advance according to the application scenario or specific needs.

[0180] For example, the prediction model includes a screening model composed of n models. Step S2320 yields m candidate transaction samples to be labeled. The second prediction result of the j-th (j∈[1,n]) model in the screening model corresponds to the i-th (i∈[1,m]) candidate transaction sample to be labeled. Second prediction result The corresponding tag value can be represented as If the n models are arranged sequentially, then the label vector of the i-th candidate transaction sample can be represented as: .

[0181] In the example shown in Table 2, the label value vector of the first candidate transaction sample to be labeled can be represented as (1,0,1), the label value vector of the second candidate transaction sample to be labeled can be represented as (1,0,1), the label value vector of the third candidate transaction sample to be labeled can be represented as (1,1,1), and the label value vector of the fourth candidate transaction sample to be labeled can be represented as (1,0,1).

[0182] Step S2343: Deduplicatization is performed on the candidate transaction samples to be labeled based on the label value vector to obtain the transaction samples to be labeled.

[0183] In one embodiment, the method of deduplicating candidate transaction samples to be labeled based on the label value vector to obtain the transaction sample to be labeled can be: for each label value vector, select one from at least one corresponding candidate transaction sample to be labeled as the transaction sample to be labeled.

[0184] In the example shown in Table 2, the label value vectors of the 1st, 3rd, and 4th candidate transaction samples are all (1,0,1). Therefore, any one of these three candidate samples can be selected as the transaction sample to be labeled. Furthermore, since only the 2nd candidate transaction sample has a label value vector of (1,1,1), it can be directly used as the transaction sample to be labeled.

[0185] In the third embodiment, the prediction model consists of multiple models. Therefore, performing prediction processing on the unlabeled transaction sample set based on the prediction model, and selecting at least one transaction sample to be labeled from the unlabeled transaction sample set based on the prediction results, can further include the following steps S2350 to S2380:

[0186] Step S2350: Make predictions on the unlabeled transaction sample set based on one or at least two models of the prediction model to obtain the first prediction result for each unlabeled transaction sample.

[0187] In one embodiment, predicting an unlabeled set of transaction samples based on one or at least two models in the prediction model to obtain a first prediction result for each unlabeled transaction sample may include: selecting one model from the multiple models that make up the prediction model, and for each unlabeled transaction sample, making a prediction based on the selected model to obtain the corresponding first prediction result.

[0188] The first prediction result can be a score between 0 and 1, used to represent the probability of predicting whether the corresponding unlabeled transaction sample is a money laundering transaction based on the selected model.

[0189] In another embodiment, predicting an unlabeled set of transaction samples based on one or at least two models in the prediction model to obtain a first prediction result for each unlabeled transaction sample may include: selecting at least two models from the plurality of models that make up the prediction model; for each unlabeled transaction sample, making a prediction based on the selected at least two models respectively to obtain a prediction result, and then averaging the prediction results corresponding to the at least two models respectively to obtain a first prediction result.

[0190] The prediction result of each selected model can be a score between 0 and 1, used to represent the probability that the corresponding model predicts whether the corresponding transaction sample is a money laundering transaction. The first prediction result is the average of the prediction results of each selected model.

[0191] Step S2360: Based on the first prediction result, select candidate transaction samples that meet the preset conditions from the set of unlabeled transaction samples.

[0192] The closer the first prediction result is to 1, the greater the probability that the selected model or at least two models consider the corresponding transaction sample to be a money laundering transaction. Conversely, the closer the first prediction result is to 0, the greater the probability that the selected model or at least two models consider the corresponding transaction sample not to be a money laundering transaction. In other words, the larger the absolute value of the difference between the first prediction result and 0.5, the easier it is for the selected model or at least two models to determine the type of the corresponding transaction sample (i.e., whether it is a money laundering transaction or not). The larger the absolute value of the difference between the first prediction result of a transaction sample and 0.5, the less labeling value it has for the selected model or at least two models. Since only transaction samples that the selected model or at least two models cannot accurately predict have labeling value, unlabeled transaction samples whose first prediction results fall within a first set range can be selected as candidate transaction samples to be labeled.

[0193] The first set range can be pre-defined according to the application scenario or specific needs. For example, the first set range can be [0.4, 0.6]. In this case, unlabeled transaction samples with prediction results greater than or equal to 0.4 and less than or equal to 0.6 can be selected as candidate transaction samples to be labeled.

[0194] Step S2370: For each candidate transaction sample to be labeled, predictions are made according to the multiple models that make up the prediction model to obtain the corresponding second prediction result.

[0195] Specifically, based on multiple models that make up the prediction model, predictions can be made for each candidate transaction sample to be labeled, and the second prediction result of each model in the prediction model corresponding to each candidate transaction sample to be labeled can be obtained.

[0196] For example, the prediction model consists of k models. Step S2320 yields m candidate transaction samples to be labeled. The second prediction result of the j-th (j∈[1,k]) model in the screening model corresponding to the i-th (i∈[1,m]) candidate transaction sample to be labeled can be... .

[0197] Step S2380: Based on the corresponding second prediction result, select at least one transaction sample to be labeled from the candidate transaction samples to be labeled.

[0198] In one embodiment, selecting at least one transaction sample to be labeled from the candidate transaction samples to be labeled based on the corresponding second prediction result includes the following steps S2381 to S2383:

[0199] Step S2381: For each candidate transaction sample to be labeled, set the label value corresponding to the second prediction result within the second set range to the first set value; set the label value corresponding to the second prediction result outside the second set range to the second set value.

[0200] Step S2381 can be described with reference to the aforementioned step S2341, and will not be repeated here.

[0201] Step S2382: For each candidate transaction sample to be labeled, based on a preset order, a label value vector is obtained according to the label value corresponding to the second prediction result of the corresponding candidate transaction sample to be labeled.

[0202] Step S2382 can be described with reference to the aforementioned step S2342, and will not be repeated here.

[0203] Step S2383: Perform deduplication on the candidate transaction samples to be labeled based on the label value vector to obtain at least one transaction sample to be labeled.

[0204] Step S2383 can be described with reference to the aforementioned step S2343, and will not be repeated here.

[0205] After obtaining at least one transaction sample to be labeled according to the first, second, or third embodiment described above, proceed to step S2400.

[0206] Step S2400: After at least one transaction sample to be labeled is labeled, at least one new labeled transaction sample is obtained, and the prediction model is iteratively updated and trained based on the at least one new labeled transaction sample.

[0207] In one embodiment, the transaction samples to be labeled obtained through step S2300 may be provided to an expert who performs the labeling action, and the expert may label each transaction sample to be labeled to obtain a corresponding new labeled transaction sample.

[0208] The electronic device implementing this embodiment of the invention may, after a transaction sample to be labeled is labeled, obtain a new labeled transaction sample and iteratively update and train the prediction model based on the new labeled transaction sample.

[0209] Based on the first embodiment mentioned above, the prediction model is a model, and it can be trained by iteratively updating the model based on new labeled transaction samples.

[0210] Based on the second embodiment mentioned above, the prediction model includes a target anti-money laundering model and a screening model composed of multiple models. Then, the target anti-money laundering model and the screening model composed of multiple models can be iteratively updated and trained according to new labeled transaction samples.

[0211] Based on the third embodiment mentioned above, the prediction model consists of multiple models. Therefore, the multiple models that make up the prediction model can be iteratively updated and trained according to the new labeled transaction samples.

[0212] Step S2500: Determine whether the prediction model meets the preset termination condition. If yes, obtain the anti-money laundering model based on the prediction model; otherwise, return to the step of obtaining the unlabeled transaction sample set.

[0213] Based on the first embodiment described above, the termination conditions may include any one or more of the following:

[0214] The number of transaction samples to be labeled is less than or equal to a preset first quantity threshold;

[0215] In the unlabeled transaction sample set, the proportion of the selected transaction samples to be labeled is less than or equal to the preset first proportion threshold.

[0216] The number of unlabeled transaction samples is less than or equal to a preset second quantity threshold;

[0217] In the unlabeled transaction sample set, the proportion of unlabeled transaction samples is less than or equal to a preset second proportion threshold.

[0218] In an embodiment where the termination condition includes the number of transaction samples to be labeled being less than or equal to a preset first quantity threshold, the first quantity threshold may be set in advance according to the application scenario or specific needs. For example, the first quantity threshold may be 100. Then, when the number of transaction samples to be labeled obtained through step S2300 is less than or equal to 100, it is determined that the prediction model meets the termination condition.

[0219] In embodiments where the termination condition includes the proportion of unlabeled transaction samples selected from the unlabeled transaction sample set being less than or equal to a preset first proportion threshold, the number of unlabeled transaction samples selected in step S2300 can be determined as the number to be labeled, the number of unlabeled transaction samples in the unlabeled transaction sample set obtained in step S2300 can be determined as the unlabeled number, and the ratio of the number to be labeled to the unlabeled number can be calculated, which is the proportion of unlabeled transaction samples selected from the unlabeled transaction sample set. This first proportion threshold can be preset according to the application scenario or specific needs. For example, the first proportion threshold can be 10%. Therefore, when the proportion of unlabeled transaction samples selected from the unlabeled transaction sample set obtained in step S2300 is less than or equal to 10%, the prediction model is deemed to meet the termination condition.

[0220] In an embodiment where the termination condition includes the number of unlabeled transaction samples being less than or equal to a preset second quantity threshold, the second quantity threshold may be preset according to the application scenario or specific needs. For example, the second quantity threshold may be 100. Then, when the number of unlabeled transaction samples in the unlabeled transaction sample set obtained through step S2300 is less than or equal to 100, it is determined that the prediction model meets the termination condition.

[0221] In embodiments where the termination condition includes the proportion of unlabeled transaction samples in the unlabeled transaction sample set being less than or equal to a preset second proportion threshold, the following steps can be taken: determining the number of transaction samples to be labeled selected in step S2300 as the number to be labeled; determining the number of unlabeled transaction samples in the unlabeled transaction sample set obtained in step S2300 as the number of unlabeled samples; calculating the difference between the number of unlabeled samples and the number to be labeled; and calculating the ratio of this difference to the number of unlabeled samples, which is the proportion of unlabeled transaction samples. The second proportion threshold can be preset according to the application scenario or specific needs. For example, the second proportion threshold can be 10%. Therefore, when the proportion of unlabeled transaction samples obtained in step S2300 is less than or equal to 10%, the prediction model can be determined to meet the termination condition.

[0222] Based on the second or third embodiment described above, the termination conditions may include any one or more of the following:

[0223] The number of candidate transaction samples to be labeled is less than or equal to a preset first quantity threshold;

[0224] In the unlabeled transaction sample set, the proportion of candidate labeled transaction samples selected is less than or equal to a preset first proportion threshold.

[0225] The number of unlabeled transaction samples is less than or equal to a preset second quantity threshold;

[0226] In the unlabeled transaction sample set, the proportion of unlabeled transaction samples is less than or equal to a preset second proportion threshold.

[0227] In an embodiment where the termination condition includes the number of candidate transaction samples to be labeled being less than or equal to a preset first quantity threshold, the first quantity threshold may be preset according to the application scenario or specific needs. For example, the first quantity threshold may be 100. Then, when the number of candidate transaction samples to be labeled obtained through step S2320 or step S2360 is less than or equal to 100, the prediction model is determined to meet the termination condition.

[0228] In embodiments where the termination condition includes the proportion of candidate transaction samples to be labeled selected from the unlabeled transaction sample set being less than or equal to a preset first proportion threshold, the number of candidate transaction samples to be labeled selected through step S2320 or step S2360 can be determined as the candidate number to be labeled, and the number of unlabeled transaction samples in the unlabeled transaction sample set obtained through step S2320 or step S2360 can be determined as the unlabeled number. The ratio of the candidate number to be labeled to the unlabeled number is then calculated, which represents the proportion of candidate transaction samples to be labeled selected from the unlabeled transaction sample set. This first proportion threshold can be preset according to the application scenario or specific requirements. For example, the first proportion threshold can be 10%. Therefore, when the proportion of candidate transaction samples to be labeled selected from the unlabeled transaction sample set obtained through step S2320 or step S2360 is less than or equal to 10%, the prediction model is deemed to meet the termination condition.

[0229] In an embodiment where the termination condition includes the number of unlabeled transaction samples being less than or equal to a preset second quantity threshold, the second quantity threshold may be preset according to the application scenario or specific needs. For example, the second quantity threshold may be 100. Then, when the number of unlabeled transaction samples in the unlabeled transaction sample set obtained through step S2320 or step S2360 is less than or equal to 100, the prediction model is determined to meet the termination condition.

[0230] In embodiments where the termination condition includes the proportion of unlabeled transaction samples in the unlabeled transaction sample set being less than or equal to a preset second proportion threshold, the following steps can be taken: The number of candidate transaction samples to be labeled selected through step S2320 or step S2360 can be determined as the candidate number to be labeled; the number of unlabeled transaction samples in the unlabeled transaction sample set obtained through step S2320 or step S2360 can be determined as the unlabeled number; the difference between the unlabeled number and the candidate number to be labeled can be calculated; and the ratio of this difference to the unlabeled number is calculated, which represents the proportion of unlabeled transaction samples. This second proportion threshold can be preset according to the application scenario or specific requirements. For example, the second proportion threshold can be 10%. Therefore, when the proportion of unlabeled transaction samples obtained through step S2320 or step S2360 is less than or equal to 10%, the prediction model can be determined to meet the termination condition.

[0231] Based on the first embodiment above, the prediction model is a model. Therefore, when the prediction model meets the preset termination conditions, obtaining an anti-money laundering model based on the prediction model may include: using the prediction model as an anti-money laundering model.

[0232] Based on the second embodiment described above, the prediction model includes a target anti-money laundering model and a screening model composed of multiple models. Therefore, when the prediction model meets a preset termination condition, the anti-money laundering model obtained based on the prediction model can include any of the following:

[0233] Use the target anti-money laundering model as the anti-money laundering model;

[0234] Choose any screening model as the anti-money laundering model;

[0235] Combining at least two screening models yields an anti-money laundering model;

[0236] The anti-money laundering model is obtained by combining the target anti-money laundering model with at least one screening model.

[0237] In one embodiment, combining at least two models to obtain an anti-money laundering model may include combining the at least two models by weighted summation or weighted averaging.

[0238] Based on the third embodiment described above, the prediction model consists of multiple models. Therefore, when the prediction model meets a preset termination condition, the anti-money laundering model obtained based on the prediction model can include:

[0239] Take any one of the models that make up the prediction model as the anti-money laundering model; or,

[0240] Combining at least two models that make up a prediction model yields an anti-money laundering model.

[0241] In one embodiment, combining at least two models to obtain an anti-money laundering model may include combining the at least two models by weighted summation or weighted averaging.

[0242] If the prediction model meets the non-preset termination condition, return to the step of obtaining the unlabeled transaction sample set.

[0243] In one embodiment, the unlabeled transaction sample set obtained each time the step of obtaining the unlabeled transaction sample set is executed is equivalent to a subset of the unlabeled transaction sample set obtained in the previous execution of the step of obtaining the unlabeled transaction sample set. Here, "previous" as used in this specification refers to any time preceding the current execution.

[0244] In the embodiments of this invention, a predictive model is first obtained by training a machine learning algorithm based on a small initial set of labeled transaction samples using a pre-selected machine learning algorithm. Then, based on the prediction results of this model for unlabeled transaction samples, transaction samples requiring expert labeling are selected. Finally, the predictive model is iteratively trained using the newly labeled transaction samples obtained after experts have labeled the unlabeled samples, resulting in an anti-money laundering model. This reduces the number of transaction samples requiring expert labeling, making it easier for the predictive model to learn valuable information. Furthermore, by deduplicating the candidate unlabeled transaction samples, the computational load during the iterative training of the predictive model can be reduced.

[0245] After obtaining the anti-money laundering model, the method may further include:

[0246] Obtain a sample of the target transactions to be predicted;

[0247] The anti-money laundering model is used to predict the target transaction sample, and the prediction result of the target transaction sample is obtained.

[0248] In one embodiment, the method may further include: displaying the prediction results of the target transaction sample so that supervisors can perform corresponding control measures on the target transaction sample based on the prediction results. For example, the prediction results may be used to determine whether the target transaction sample is a money laundering transaction, and if it is a money laundering transaction, to punish the person who performed the transaction or to revoke the corresponding transaction.

[0249] <Example 1>

[0250] The following will combine Figure 4 The training method for the anti-money laundering model provided in this embodiment will be further explained. For example... Figure 4 As shown, the method includes:

[0251] Step S4001: Obtain the initial set of marked transaction samples.

[0252] Step S4002: Based on the pre-selected machine learning algorithm, the machine learning model is trained on the initial set of labeled transaction samples to obtain the target anti-money laundering model and the screening model composed of multiple models.

[0253] Step S4003: Obtain the unlabeled transaction sample set.

[0254] Step S4004: Based on the target anti-money laundering model, predict the unlabeled transaction sample set to obtain the first prediction result for each unlabeled transaction sample.

[0255] Step S4005: Select the labeled transaction samples whose first prediction results fall within a first set range from the unlabeled transaction samples as candidate transaction samples to be labeled.

[0256] Step S4006: For each candidate transaction sample to be labeled, predictions are made according to the multiple models that make up the screening model to obtain the corresponding second prediction result.

[0257] Step S4007: For each candidate transaction sample to be labeled, the label value corresponding to the second prediction result within the second set range is set to the first set value; the label value corresponding to the second prediction result outside the second set range is set to the second set value.

[0258] Step S4008: For each candidate transaction sample to be labeled, based on a preset order, a label value vector is obtained according to the label value corresponding to the second prediction result of the corresponding candidate transaction sample to be labeled.

[0259] Step S4009: Deduplication is performed on the candidate transaction samples to be labeled based on the label value vector to obtain the transaction samples to be labeled.

[0260] Step S4010: After at least one transaction sample to be labeled is labeled, at least one new labeled transaction sample is obtained, and the target anti-money laundering model and the screening model are iteratively updated and trained based on the at least one new labeled transaction sample.

[0261] Step S4011: Determine whether the prediction model meets the preset termination conditions.

[0262] If the termination condition is met, proceed to step S4012; if the termination condition is not met, return to step S4003.

[0263] Step S4012: Use the target anti-money laundering model as the anti-money laundering model.

[0264] <Device Embodiment>

[0265] In this embodiment, a training device 5000 for an anti-money laundering model is provided, such as... Figure 5 As shown, it includes an initial sample acquisition module 5100, a model initial training module 5200, a sample screening module 5300, a model iterative training module 5400, and a termination condition judgment module 5500.

[0266] The aforementioned initial sample acquisition module 5100 is used to acquire an initial set of labeled transaction samples, wherein each sample in the labeled sample set has been labeled as either a money laundering transaction or not a money laundering transaction.

[0267] The aforementioned model initial training module 5200 is used to train a machine learning model based on an initial set of labeled transaction samples according to a pre-selected machine learning algorithm, so as to obtain a prediction model.

[0268] The aforementioned unlabeled sample screening module 5300 is used to obtain an unlabeled transaction sample set, perform prediction processing on the unlabeled transaction sample set according to the prediction model, and screen out at least one unlabeled transaction sample from the unlabeled transaction sample set based on the prediction results.

[0269] The aforementioned model iterative training module 5400 is used to obtain at least one new labeled transaction sample after at least one transaction sample to be labeled is labeled, and to iteratively update and train the prediction model based on at least one new labeled transaction sample.

[0270] The aforementioned termination condition judgment module 5500 is used to determine whether the prediction model meets the preset termination condition. If it does, an anti-money laundering model is obtained based on the prediction model. Otherwise, the module controls the unlabeled sample screening module 5300 to obtain the unlabeled transaction sample set, perform prediction processing on the unlabeled transaction sample set according to the prediction model, and screen out at least one unlabeled transaction sample from the unlabeled transaction sample set based on the prediction result.

[0271] In one embodiment, performing prediction processing on an unlabeled transaction sample set according to a prediction model, and selecting at least one transaction sample to be labeled from the unlabeled transaction sample set based on the prediction results, includes:

[0272] The unlabeled transaction samples whose prediction results fall within a first set range are selected from the unlabeled transaction sample set and used as the transaction samples to be labeled.

[0273] In one embodiment, the predictive model includes a target anti-money laundering model and a screening model consisting of multiple models;

[0274] The process involves predicting the unlabeled transaction sample set using a prediction model, and then selecting at least one transaction sample to be labeled from the unlabeled transaction sample set based on the prediction results.

[0275] Based on the target anti-money laundering model, predictions are made on the unlabeled transaction sample set to obtain the first prediction result for each unlabeled transaction sample;

[0276] The unlabeled transaction samples within a first set range of the first prediction result are selected from the unlabeled transaction sample set and used as candidate transaction samples to be labeled.

[0277] For each candidate transaction sample to be labeled, predictions are made according to the multiple models that make up the screening model to obtain the corresponding second prediction result;

[0278] Based on the corresponding second prediction result, at least one transaction sample to be labeled is selected from the candidate transaction samples to be labeled.

[0279] In one embodiment, selecting at least one transaction sample to be labeled from the candidate transaction samples based on the corresponding second prediction result includes:

[0280] Based on the second prediction result corresponding to each candidate transaction sample to be labeled, the candidate transaction samples to be labeled are deduplicated to obtain at least one transaction sample to be labeled.

[0281] In one embodiment, based on the second prediction result corresponding to each candidate sample to be labeled, the candidate transaction samples to be labeled are deduplicated to obtain the following transaction samples to be labeled:

[0282] For each candidate transaction sample to be labeled, the label value corresponding to the second prediction result within the second set range is set to the first set value; the label value corresponding to the second prediction result outside the second set range is set to the second set value.

[0283] For each candidate transaction sample to be labeled, a label value vector is obtained based on a preset order and the label value corresponding to the second prediction result of the corresponding candidate transaction sample;

[0284] The candidate transaction samples to be labeled are deduplicated based on the label value vector to obtain the transaction samples to be labeled.

[0285] In one embodiment, the termination condition includes any one or more of the following:

[0286] The number of candidate transaction samples to be labeled is less than or equal to a preset first quantity threshold;

[0287] In the unlabeled transaction sample set, the proportion of the selected candidate transaction samples to be labeled is less than or equal to the preset first proportion threshold.

[0288] The number of unlabeled transaction samples is less than or equal to a preset second quantity threshold;

[0289] In the unlabeled transaction sample set, the proportion of unlabeled transaction samples is less than or equal to a preset second proportion threshold.

[0290] In one embodiment, multiple prediction models employ different algorithms or have different hyperparameters.

[0291] In one embodiment, the anti-money laundering model derived from the prediction model includes any one of the following:

[0292] Use the target anti-money laundering model as the anti-money laundering model;

[0293] Choose any screening model as the anti-money laundering model;

[0294] Combining at least two screening models yields an anti-money laundering model;

[0295] The anti-money laundering model is obtained by combining the target anti-money laundering model with at least one screening model.

[0296] In one embodiment, the prediction model consists of multiple models;

[0297] The process involves predicting the unlabeled transaction sample set using a prediction model, and then selecting at least one transaction sample to be labeled from the unlabeled transaction sample set based on the prediction results.

[0298] Based on one or at least two prediction models, a first prediction result is obtained for each unlabeled transaction sample set.

[0299] Based on the first prediction result, candidate transaction samples that meet the preset conditions are selected from the set of unlabeled transaction samples.

[0300] For each candidate transaction sample to be labeled, predictions are made separately according to the multiple models that make up the prediction model to obtain the corresponding second prediction result;

[0301] Based on the corresponding second prediction result, at least one transaction sample to be labeled is selected from the candidate transaction samples to be labeled.

[0302] In one embodiment, predicting the unlabeled transaction sample set based on one or at least two prediction models to obtain a first prediction result for each unlabeled transaction sample includes:

[0303] One model is selected from the multiple models that make up the prediction model. For each unlabeled transaction sample, a prediction is made based on the selected model to obtain the corresponding first prediction result.

[0304] or,

[0305] Select at least two models from the multiple models that make up the prediction model. For each unlabeled transaction sample, make a prediction based on the selected at least two models to obtain the prediction result. Then, average the prediction results corresponding to the at least two models to obtain the first prediction result.

[0306] In one embodiment, selecting at least one transaction sample to be labeled from the candidate transaction samples based on the corresponding second prediction result includes:

[0307] For each candidate transaction sample to be labeled, the label value corresponding to the second prediction result within the second set range is set to the first set value; the label value corresponding to the second prediction result outside the second set range is set to the second set value.

[0308] For each candidate transaction sample to be labeled, a label value vector is obtained based on a preset order and the label value corresponding to the second prediction result of the corresponding candidate transaction sample;

[0309] The candidate transaction samples to be labeled are deduplicated based on the label value vector to obtain at least one transaction sample to be labeled.

[0310] In one embodiment, deriving an anti-money laundering model based on a predictive model includes:

[0311] Choose any prediction model as an anti-money laundering model; or,

[0312] Combining at least two prediction models yields an anti-money laundering model.

[0313] In one embodiment, the machine learning algorithm is the random forest algorithm.

[0314] In one embodiment, the training device 5000 further includes:

[0315] A module used to obtain a sample of the target transactions to be predicted;

[0316] This module is used to predict target transaction samples based on anti-money laundering models and obtain the prediction results of the target transaction samples.

[0317] In one embodiment, the training device 5000 further includes:

[0318] A module used to display the prediction results for the target sample.

[0319] Those skilled in the art will understand that the anti-money laundering model training device 5000 can be implemented in various ways. For example, the anti-money laundering model training device 5000 can be implemented by configuring a processor with instructions. For example, the instructions can be stored in ROM, and when the device is started, the instructions can be read from ROM into a programmable device to implement the anti-money laundering model training device 5000. For example, the anti-money laundering model training device 5000 can be embedded in a dedicated device (e.g., an ASIC). The anti-money laundering model training device 5000 can be divided into independent units, or they can be combined together. The anti-money laundering model training device 5000 can be implemented by one of the various implementation methods described above, or by a combination of two or more of the various implementation methods described above.

[0320] In this embodiment, the anti-money laundering model training device 5000 can have various implementation forms. For example, the anti-money laundering model training device 5000 can be any functional module running in a software product or application that provides transaction services, or peripheral embeddings, plug-ins, patches, etc. of these software products or applications, or it can be the software products or applications themselves.

[0321] <Electronic Devices>

[0322] In this embodiment, an electronic device 6000 is also provided. The electronic device 6000 may be... Figure 1 The electronic device shown is 1000.

[0323] In one aspect, the electronic device 6000 may include the aforementioned anti-money laundering model training device 5000, used to implement the anti-money laundering model training method of any embodiment of the present invention.

[0324] On the other hand, such as Figure 6 As shown, the electronic device 6000 may further include a processor 6100 and a memory 6200, the memory 6200 being used to store executable instructions; the processor 6100 being used to control the electronic device 6000 to execute the training method of the anti-money laundering model according to any embodiment of the present invention, according to the instructions.

[0325] In this embodiment, the electronic device 6000 can be a mobile phone, tablet computer, handheld computer, desktop computer, laptop computer, workstation, game console, or other devices.

[0326] Computer-readable storage media

[0327] In this embodiment, a computer-readable storage medium is also provided, on which a computer program is stored. When the computer program is executed by a processor, it implements the training method of the anti-money laundering model as in any embodiment of the present invention.

[0328] This invention can be an apparatus, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of the invention.

[0329] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination of the foregoing. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0330] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0331] The computer program instructions used to perform the operations of this invention may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing state information from the computer-readable program instructions. This electronic circuitry can execute the computer-readable program instructions to implement various aspects of the invention.

[0332] Various aspects of the present invention are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0333] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that, when executed by the processor of the computer or other programmable data processing apparatus, they create means for implementing the functions / actions specified in one or more blocks of the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner; thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing aspects of the functions / actions specified in one or more blocks of the flowchart and / or block diagram.

[0334] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0335] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions. It will be known to those skilled in the art that implementation in hardware, implementation in software, and implementation using a combination of software and hardware are equivalent.

[0336] The various embodiments of the present invention have been described above. These descriptions are exemplary and not exhaustive, and are not limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or technical improvements to the embodiments in the market, or to enable others skilled in the art to understand the embodiments disclosed herein. The scope of the invention is defined by the appended claims.

Claims

1. A training method for an anti-money laundering model, characterized in that, include: Obtain an initial set of labeled transaction samples, where each sample in the set has been labeled as either a money laundering transaction or not a money laundering transaction; Based on the pre-selected machine learning algorithm, a machine learning model is trained on the initial set of labeled transaction samples to obtain a prediction model; Obtain an unlabeled transaction sample set, perform prediction processing on the unlabeled transaction sample set according to the prediction model, and select at least one transaction sample to be labeled from the unlabeled transaction sample set based on the prediction results. After the at least one transaction sample to be labeled is labeled, at least one new labeled transaction sample is obtained, and the prediction model is iteratively updated and trained based on the at least one new labeled transaction sample; Determine whether the prediction model meets the preset termination condition. If yes, obtain the anti-money laundering model based on the prediction model; otherwise, return to the step of obtaining the unlabeled transaction sample set. The prediction model consists of multiple models; The step of performing prediction processing on the unlabeled transaction sample set according to the prediction model, and selecting at least one transaction sample to be labeled from the unlabeled transaction sample set based on the prediction result includes: The unlabeled transaction sample set is predicted based on one or at least two of the prediction models to obtain a first prediction result for each unlabeled transaction sample; Based on the first prediction result, candidate transaction samples that meet the preset conditions are selected from the set of unlabeled transaction samples; For each candidate transaction sample to be labeled, predictions are made according to the multiple models that make up the prediction model, to obtain the second prediction result of each model in the prediction model corresponding to each candidate transaction sample to be labeled; Based on the corresponding second prediction result, at least one transaction sample to be labeled is selected from the candidate transaction samples to be labeled; Wherein, the first prediction result is a score between 0 and 1, and the preset condition is that the absolute value of the difference between the first prediction result and 0.5 is less than or equal to a set threshold. The termination conditions include any one or more of the following: The number of candidate transaction samples to be labeled is less than or equal to a preset first quantity threshold; The proportion of candidate transaction samples to be labeled in the unlabeled transaction sample set is less than or equal to a preset first proportion threshold. The number of unlabeled transaction samples is less than or equal to a preset second quantity threshold; In the unlabeled transaction sample set, the proportion of unlabeled transaction samples is less than or equal to a preset second proportion threshold.

2. The training method according to claim 1, characterized in that, The step of predicting the unlabeled transaction sample set based on one or at least two of the prediction models to obtain a first prediction result for each unlabeled transaction sample includes: One model is selected from the multiple models that make up the prediction model. For each unlabeled transaction sample, a prediction is made according to the selected model to obtain the corresponding first prediction result. or, At least two models are selected from the multiple models that make up the prediction model. For each unlabeled transaction sample, predictions are made according to the selected at least two models to obtain prediction results. The prediction results corresponding to the at least two models are then averaged to obtain a first prediction result.

3. The training method according to claim 1, characterized in that, The step of selecting at least one transaction sample to be labeled from the candidate transaction samples to be labeled based on the corresponding second prediction result includes: For each candidate transaction sample to be labeled, the label value corresponding to the second prediction result within the second set range is set to the first set value; the label value corresponding to the second prediction result outside the second set range is set to the second set value. The result of the corresponding model predicting whether the candidate transaction sample to be labeled is a money laundering transaction is determined by whether the second prediction result of the candidate transaction sample is within the second set range. For each candidate transaction sample to be labeled, based on the preset order of the multiple models that make up the prediction model, the label values ​​corresponding to the second prediction results of the corresponding candidate transaction sample are sorted to obtain a label value vector. The candidate transaction samples to be labeled are deduplicated according to the label value vector to obtain the at least one transaction sample to be labeled.

4. The training method according to claim 1, characterized in that, The steps for obtaining the anti-money laundering model based on the prediction model include: Choose any prediction model as the anti-money laundering model; or, The anti-money laundering model is obtained by combining at least two prediction models.

5. The training method according to claim 1, characterized in that, The machine learning algorithm mentioned is the random forest algorithm.

6. A prediction method for an anti-money laundering model, characterized in that, include: Obtain a sample of the target transactions to be predicted; The anti-money laundering model obtained by the training method as described in any one of claims 1 to 5 is used to predict the target transaction sample, thereby obtaining the prediction result of the target transaction sample.

7. The prediction method according to claim 6, characterized in that, The prediction method further includes: The prediction results for the target transaction sample are displayed.

8. A training device for an anti-money laundering model, characterized in that, include: The initial sample acquisition module is used to acquire an initial set of labeled transaction samples, wherein each sample in the labeled sample set has been labeled as either a money laundering transaction or not a money laundering transaction; The model initial training module is used to train a machine learning model based on the initial set of labeled transaction samples according to a pre-selected machine learning algorithm to obtain a prediction model; The unlabeled sample screening module is used to obtain an unlabeled transaction sample set, perform prediction processing on the unlabeled transaction sample set according to the prediction model, and screen out at least one unlabeled transaction sample from the unlabeled transaction sample set according to the prediction result. The model iterative training module is used to obtain at least one new labeled transaction sample after the at least one transaction sample to be labeled is labeled, and to iteratively update and train the prediction model based on the at least one new labeled transaction sample; The termination condition judgment module is used to determine whether the prediction model meets the preset termination condition. If it does, an anti-money laundering model is obtained based on the prediction model; otherwise, the process returns to the step of obtaining the unlabeled transaction sample set. The prediction model consists of multiple models; The step of performing prediction processing on the unlabeled transaction sample set according to the prediction model, and selecting at least one transaction sample to be labeled from the unlabeled transaction sample set based on the prediction result includes: The unlabeled transaction sample set is predicted based on one or at least two of the prediction models to obtain a first prediction result for each unlabeled transaction sample; Based on the first prediction result, candidate transaction samples that meet the preset conditions are selected from the set of unlabeled transaction samples; For each candidate transaction sample to be labeled, predictions are made according to the multiple models that make up the prediction model, to obtain the second prediction result of each model in the prediction model corresponding to each candidate transaction sample to be labeled; Based on the corresponding second prediction result, at least one transaction sample to be labeled is selected from the candidate transaction samples to be labeled. Wherein, the first prediction result is a score between 0 and 1, and the preset condition is that the absolute value of the difference between the first prediction result and 0.5 is less than or equal to a set threshold. The termination conditions include any one or more of the following: The number of candidate transaction samples to be labeled is less than or equal to a preset first quantity threshold; The proportion of candidate transaction samples to be labeled in the unlabeled transaction sample set is less than or equal to a preset first proportion threshold. The number of unlabeled transaction samples is less than or equal to a preset second quantity threshold; In the unlabeled transaction sample set, the proportion of unlabeled transaction samples is less than or equal to a preset second proportion threshold.

9. The training device according to claim 8, characterized in that, The step of predicting the unlabeled transaction sample set based on one or at least two of the prediction models to obtain a first prediction result for each unlabeled transaction sample includes: One model is selected from the multiple models that make up the prediction model. For each unlabeled transaction sample, a prediction is made according to the selected model to obtain the corresponding first prediction result. or, At least two models are selected from the multiple models that make up the prediction model. For each unlabeled transaction sample, predictions are made according to the selected at least two models to obtain prediction results. The prediction results corresponding to the at least two models are then averaged to obtain a first prediction result.

10. The training device according to claim 8, characterized in that, The step of selecting at least one transaction sample to be labeled from the candidate transaction samples to be labeled based on the corresponding second prediction result includes: For each candidate transaction sample to be labeled, the label value corresponding to the second prediction result within the second set range is set to the first set value; the label value corresponding to the second prediction result outside the second set range is set to the second set value. The result of the corresponding model predicting whether the candidate transaction sample to be labeled is a money laundering transaction is determined by whether the second prediction result of the candidate transaction sample is within the second set range. For each candidate transaction sample to be labeled, based on the preset order of the multiple models that make up the prediction model, the label values ​​corresponding to the second prediction results of the corresponding candidate transaction sample are sorted to obtain a label value vector. The candidate transaction samples to be labeled are deduplicated according to the label value vector to obtain the at least one transaction sample to be labeled.

11. The training device according to claim 8, characterized in that, The anti-money laundering model derived from the prediction model includes: Choose any prediction model as the anti-money laundering model; or, The anti-money laundering model is obtained by combining at least two prediction models.

12. The training device according to claim 8, characterized in that, The machine learning algorithm mentioned is the random forest algorithm.

13. A predictive device for an anti-money laundering model, characterized in that, include: A module used to obtain a sample of the target transactions to be predicted; A module for predicting the target transaction sample based on the anti-money laundering model obtained by the training device as described in any one of claims 8 to 12, and obtaining the prediction result of the target transaction sample.

14. The prediction device according to claim 13, characterized in that, The prediction device further includes: A module used to display the prediction results of the target transaction sample.

15. An electronic device comprising: The apparatus as described in any one of claims 8-14; or, A processor and a memory, the memory being used to store instructions for controlling the processor to perform the method according to any one of claims 1 to 7.

16. A computer-readable storage medium having a computer program stored thereon, the computer program implementing the method as described in any one of claims 1 to 7 when executed by a processor.

Images