Fraud risk prediction method and device, equipment and storage medium

Abstract

The invention provides a fraud risk prediction method and device, equipment and a storage medium. The method comprises the steps of obtaining a claim knowledge base; constructing a feature layer according to the claim knowledge base; determining m tasks, and generating a target sub-prediction model corresponding to each task; constructing n expert layers, and performing deep learning on each target sub-prediction model by adopting each expert layer to obtain a weight of each expert layer corresponding to each target sub-prediction model; constructing a self-attention layer; according to the feature layer, the target sub-prediction model corresponding to each task, the n expert layers and the self-attention layer, a fraud risk multi-task prediction model is established, modeling is carried out for different tasks at the same time, representation information is shared among related tasks, fraud risk prediction corresponding to different tasks is achieved, the generalization performance of fraud risk prediction is improved, and the fraud risk prediction efficiency is improved. And the fraud risk can be comprehensively and accurately identified.

Description

Technical Field

[0001] The present application relates to the field of financial anti-fraud technology, and specifically to a fraud risk prediction method, device, equipment and storage medium. Background Art

[0002] With the rapid development of Internet finance and payment methods, fraudulent activities using online channels of commercial banks have become frequent, affecting financial security. Financial anti-fraud is an important area of ​​the financial industry. Accurately identifying and predicting fraud risks in financial activities is an important guarantee for financial security.

[0003] In traditional technology, anti-fraud algorithms are mainly based on machine learning methods to optimize a specific indicator. In order to achieve this goal, a single model or integrated algorithm is trained to complete a single classification prediction task. In essence, it is a single-task learning algorithm. Then, during the training process, the performance of the model is improved by fine-tuning the parameters.

[0004] However, the anti-fraud algorithms in existing technologies have poor generalization performance and are unable to comprehensively and accurately identify fraud risks. Summary of the invention

[0005] The present application provides a fraud risk prediction method, device, equipment and storage medium, thereby solving the problem that the anti-fraud algorithm in the prior art has poor generalization performance and cannot comprehensively and accurately identify fraud risks.

[0006] In a first aspect, the present application provides a fraud risk prediction method, comprising:

[0007] Access to the claims knowledge base;

[0008] Building a feature layer based on the claims knowledge base;

[0009] Determine m tasks, and generate a target sub-prediction model corresponding to each task, wherein m is a positive integer greater than 1;

[0010] Constructing n expert layers, and using each expert layer to perform deep learning on each target sub-prediction model to obtain a weight of each expert layer corresponding to each target sub-prediction model, wherein n is a positive integer greater than 1;

[0011] Constructing a self-attention layer, wherein the self-attention layer is used to adjust the weights of each expert layer corresponding to each target sub-prediction model;

[0012] A fraud risk multi-task prediction model is established based on the feature layer, the target sub-prediction models corresponding to each task, the n expert layers and the self-attention layer, wherein the fraud risk multi-task prediction model is used to predict the fraud risk corresponding to each task of the case.

[0013] Here, the present application provides a fraud risk prediction method based on multi-task learning, and establishes a feature layer shared by multiple tasks based on the claims knowledge base to capture the complex relationships and patterns between features and reduce the risk of overfitting. In order to achieve multi-task prediction, respective target sub-prediction models are generated according to multiple tasks. The target sub-prediction model can predict the fraud risk of the corresponding task, and then multiple expert layers are constructed to score the target sub-prediction model. In combination with the self-attention mechanism, a self-attention layer is constructed to perform weighted self-learning on the expert's scores. Different expert scoring weights can be learned for different tasks. Through the above-mentioned feature layer, the target sub-prediction model corresponding to each task, the expert layer and the self-attention layer, a multi-task prediction model for fraud risk can be obtained. Modeling is performed simultaneously for different tasks, and representation information is shared between related tasks to achieve fraud risk prediction corresponding to different tasks, thereby improving the generalization performance of fraud risk prediction and being able to comprehensively and accurately identify fraud risks.

[0014] Optionally, constructing a feature layer according to the claims knowledge base includes:

[0015] The claims knowledge base is subjected to data extraction processing to obtain detailed information of multiple cases, wherein the claims knowledge base includes case information of multiple cases; the detailed information of the multiple cases is subjected to data preprocessing to obtain multiple preprocessed detailed information; the multiple preprocessed detailed information is subjected to feature conversion processing to obtain multiple vector information; and a feature layer is constructed based on the multiple vector information.

[0016] This application extracts detailed information of multiple cases from the claims knowledge base. The multiple cases here can correspond to different tasks. Through data preprocessing and feature conversion processing of the extracted detailed information, each original feature of the case can be mapped to a dense vector space, which helps to capture the complex relationships and patterns between features. The feature layer is shared by all tasks, which promotes better generalization performance of the model on the original task and improves the accuracy of fraud risk prediction.

[0017] Optionally, the detailed information includes at least one of case type, claim amount, policyholder information, accident description, number of accidents and policyholder customer level.

[0018] Among them, this application can extract a variety of detailed information of the case, specifically, it can be one or more of the case type, claim amount, policyholder information, accident description, number of accidents and policyholder customer level. Through the precise extraction of the above information, the detailed information related to fraud risk can be accurately screened out. The extraction of multiple detailed information is also more suitable for multi-task learning, which further improves the accuracy of fraud risk prediction.

[0019] Optionally, generating a target sub-prediction model corresponding to each task includes:

[0020] Obtain training data samples corresponding to each task and the original sub-prediction model corresponding to each task, wherein the training data samples include case samples and task labels corresponding to the case samples; according to the training data samples corresponding to each task, train the corresponding original sub-prediction model to obtain the target sub-prediction model corresponding to each task.

[0021] Among them, the present application models different tasks separately. Specifically, for each task, the corresponding training data samples and the original sub-prediction model are obtained. By using the training data samples to train the original sub-prediction model, a target sub-prediction model focusing on fraud risk prediction of the corresponding task can be obtained, which can accurately realize the fraud risk prediction of the corresponding task. The establishment of multiple target sub-prediction models realizes fraud risk prediction of multiple tasks corresponding to different types of cases, promotes the generalization performance of fraud risk prediction, and can accurately and comprehensively predict fraud risks.

[0022] Optionally, the constructing of n expert layers includes:

[0023] Determine n deep learning objectives and parameter sharing strategies; perform expert layer construction processing according to the n deep learning objectives and the parameter sharing strategies to obtain n expert layers.

[0024] Here, certain parameters can be shared between the expert layers of the present application, which enhances the generalization ability of the model and further improves the accuracy of fraud risk prediction.

[0025] Optionally, the tasks include high claim case identification and fraud case identification.

[0026] Among them, the present application can simultaneously identify high-payout cases and fraud cases, which increases the generalization capability of fraud risk prediction and can accurately and comprehensively predict fraud risks.

[0027] Optionally, after establishing the fraud risk multi-task prediction model according to the feature layer, the target sub-prediction models corresponding to the tasks, the n expert layers and the self-attention layer, the method further includes:

[0028] Obtain case information of the case to be predicted; input the case information of the case to be predicted into the fraud risk multi-task prediction model to determine the fraud risk prediction results of the case to be predicted corresponding to each task through the output results of the fraud risk multi-task prediction model.

[0029] Here, the present application can perform multi-task fraud risk prediction on the predicted cases based on the constructed multi-task fraud risk prediction model, thereby achieving a comprehensive and accurate prediction of the fraud risks of financial cases and improving the security of financial cases.

[0030] In a second aspect, the present application provides a fraud risk prediction device, comprising:

[0031] An acquisition module is used to acquire a claims knowledge base;

[0032] A first processing module, configured to construct a feature layer according to the claims knowledge base;

[0033] The second processing module is used to determine m tasks and generate a target sub-prediction model corresponding to each task, wherein m is a positive integer greater than 1;

[0034] The third processing module is used to construct n expert layers, and use each expert layer to perform deep learning on each target sub-prediction model to obtain a weight of each expert layer corresponding to each target sub-prediction model, wherein n is a positive integer greater than 1;

[0035] A fourth processing module is used to construct a self-attention layer, wherein the self-attention layer is used to adjust the weights of each expert layer corresponding to each target sub-prediction model;

[0036] A module is established to establish a fraud risk multi-task prediction model based on the feature layer, the target sub-prediction models corresponding to each task, the n expert layers and the self-attention layer, wherein the fraud risk multi-task prediction model is used to predict the fraud risk corresponding to each task of the case.

[0037] Optionally, the first processing module is specifically configured to:

[0038] Performing data extraction processing on the claims knowledge base to obtain detailed information of multiple cases, wherein the claims knowledge base includes case information of multiple cases;

[0039] Performing data preprocessing on the detailed information of the multiple cases to obtain multiple preprocessed detailed information;

[0040] Performing feature conversion processing on the plurality of preprocessed detailed information to obtain a plurality of vector information;

[0041] A feature layer is constructed according to the multiple vector information.

[0042] Optionally, the detailed information includes at least one of case type, claim amount, policyholder information, accident description, number of accidents and policyholder customer level.

[0043] Optionally, the second processing module is specifically configured to:

[0044] Obtaining training data samples corresponding to each task and original sub-prediction models corresponding to each task, wherein the training data samples include case samples and task labels corresponding to the case samples;

[0045] According to the training data samples corresponding to each task, the corresponding original sub-prediction model is trained to obtain the target sub-prediction model corresponding to each task.

[0046] Optionally, the third processing module is specifically used to:

[0047] Determine n deep learning objectives and parameter sharing strategies;

[0048] An expert layer construction process is performed according to the n deep learning objectives and the parameter sharing strategy to obtain n expert layers.

[0049] Optionally, the tasks include high claim case identification and fraud case identification.

[0050] Optionally, after the establishment module is used to establish the fraud risk multi-task prediction model according to the feature layer, the target sub-prediction model corresponding to each task, the n expert layers and the self-attention layer, the apparatus further includes a prediction module for:

[0051] Obtain case information of the case to be predicted;

[0052] The case information of the case to be predicted is input into the fraud risk multi-task prediction model, so as to determine the fraud risk prediction results of the case to be predicted corresponding to each task through the output results of the fraud risk multi-task prediction model.

[0053] In a third aspect, the present application provides a fraud risk prediction device, comprising: at least one processor and a memory;

[0054] The memory stores computer-executable instructions;

[0055] The at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor performs the fraud risk prediction method as described in the first aspect and various possible designs of the first aspect.

[0056] In a fourth aspect, the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer execution instructions. When a processor executes the computer execution instructions, the fraud risk prediction method as described in the first aspect and various possible designs of the first aspect is implemented.

[0057] The fraud risk prediction method, device, equipment and storage medium provided by the present application, wherein the method establishes a feature layer shared by multiple tasks based on a claims knowledge base to capture the complex relationships and patterns between features and reduce the risk of overfitting. In order to achieve multi-task prediction, respective target sub-prediction models are generated according to multiple tasks, and the target sub-prediction models can predict the fraud risks of corresponding tasks. Then, multiple expert layers for scoring the target sub-prediction models are constructed, and a self-attention mechanism is combined to construct a self-attention layer to perform weighted self-learning on the expert scores. Different expert scoring weights can be learned for different tasks. Through the above-mentioned feature layer, the target sub-prediction models corresponding to each task, the expert layer and the self-attention layer, a multi-task prediction model for fraud risk can be obtained. Modeling is performed simultaneously for different tasks, and representation information is shared between related tasks to achieve fraud risk prediction corresponding to different tasks, thereby improving the generalization performance of fraud risk prediction and being able to comprehensively and accurately identify fraud risks. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0059] Figure 1 A schematic diagram of a fraud risk prediction system architecture provided in an embodiment of the present application;

[0060] Figure 2 A flowchart of a fraud risk prediction method provided in an embodiment of the present application;

[0061] Figure 3 A flowchart of another fraud risk prediction method provided in an embodiment of the present application;

[0062] Figure 4 A schematic diagram of the structure of a self-attention layer provided in an embodiment of the present application;

[0063] Figure 5 A schematic diagram of the structure of a fraud risk multi-task prediction model provided in an embodiment of the present application;

[0064] Figure 6 A schematic diagram of the structure of a fraud risk prediction device provided in an embodiment of the present application;

[0065] Figure 7 A schematic diagram of the structure of a fraud risk prediction device provided in an embodiment of the present application.

[0066] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0067] Here, exemplary embodiments will be described in detail, and examples thereof are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. Instead, they are only examples of devices and methods consistent with some aspects of the present application.

[0068] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0069] Traditional anti-fraud algorithms are mainly based on machine learning methods to optimize a specific indicator, such as fraud case identification. In order to achieve this goal, a single model or integrated algorithm is trained to complete a single classification prediction task. In essence, it is a single-task learning algorithm, and then the performance of the model is improved by fine-tuning the parameters during the training process. Although this may achieve the best results for a certain indicator, it may ignore some potential information. Specifically, this information is the supervised labeled data of other tasks related to anti-fraud risk identification.

[0070] In order to solve the above problems, the embodiments of the present application provide a fraud risk prediction method, device, equipment and medium, which simultaneously models multiple tasks, shares the bottom feature layer between different tasks, shares representation information between related tasks, reduces the risk of overfitting, and promotes better generalization performance of the model on the original task.

[0071] Optional, Figure 1 A schematic diagram of a fraud risk prediction system architecture provided in an embodiment of the present application. Figure 1 In the above architecture, the above architecture includes at least one of a data acquisition device 101, a processing device 102 and a display device 103.

[0072] It is understandable that the structure illustrated in the embodiment of the present application does not constitute a specific limitation on the fraud risk prediction system architecture. In other feasible implementations of the present application, the above architecture may include more or fewer components than shown in the figure, or combine certain components, or split certain components, or arrange the components differently, which can be determined according to the actual application scenario and is not limited here. Figure 1 The components shown may be implemented in hardware, software, or a combination of software and hardware.

[0073] In a specific implementation process, the data acquisition device 101 may include an input / output interface and may also include a communication interface. The data acquisition device 101 may be connected to the processing device via the input / output interface or the communication interface.

[0074] The processing device 102 can establish a feature layer shared by multiple tasks based on the claims knowledge base to capture the complex relationships and patterns between features and reduce the risk of overfitting. In order to achieve multi-task prediction, based on multiple tasks, respective target sub-prediction models are generated. The target sub-prediction models can predict the fraud risks of the corresponding tasks, and then multiple expert layers are constructed to score the target sub-prediction models. In combination with the self-attention mechanism, a self-attention layer is constructed to perform weighted self-learning on the expert scores. Different expert scoring weights can be learned for different tasks. Through the above-mentioned feature layer, the target sub-prediction models corresponding to each task, the expert layer and the self-attention layer, a multi-task prediction model for fraud risk is obtained.

[0075] The display device 103 may also be a touch display screen or a screen of a terminal device, which is used to receive user instructions while displaying the above content to achieve interaction with the user.

[0076] It should be understood that the above-mentioned processing device can be implemented by a processor reading instructions in a memory and executing the instructions, or it can be implemented by a chip circuit.

[0077] In addition, the network architecture and business scenarios described in the embodiments of the present application are intended to more clearly illustrate the technical solutions of the embodiments of the present application, and do not constitute a limitation on the technical solutions provided in the embodiments of the present application. Ordinary technicians in this field can know that with the evolution of network architecture and the emergence of new business scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0078] The technical solution of the present application is described in detail below in conjunction with specific embodiments:

[0079] Optionally, Figure 2 A flowchart of a fraud risk prediction method provided in an embodiment of the present application is shown below. The execution subject of the embodiment of the present application may be Figure 1 The specific execution subject can be determined according to the actual application scenario. Figure 2 As shown, the method comprises the following steps:

[0080] S201: Obtain the claims knowledge base.

[0081] S202: Construct a feature layer based on the claims knowledge base.

[0082] Optionally, a feature layer is constructed based on the claims knowledge base, including:

[0083] Data extraction is performed on the claims knowledge base to obtain detailed information of multiple cases, wherein the claims knowledge base includes case information of multiple cases; data preprocessing is performed on the detailed information of multiple cases to obtain multiple preprocessed detailed information; feature conversion is performed on the multiple preprocessed detailed information to obtain multiple vector information; and a feature layer is constructed based on the multiple vector information.

[0084] In one possible implementation, important case details are extracted based on the claims knowledge base to construct an embedding layer. The specific method is as follows:

[0085] Data preprocessing: First, based on the claims knowledge base, extract important case details, such as case type, claim amount, policyholder information, accident description, etc. This information needs to be preprocessed, including missing value processing, outlier detection, text data cleaning and standardization, etc. It can also include the number of accidents or policyholder customer level.

[0086] Feature Engineering: Convert non-numeric features (such as case type, accident description) into numeric ones. You can use pre-trained models such as Word Embedding (Word Embedding), such as Word to Vector (Word2Vec), Global Vectors for Word Representation (GloVe), or Bidirectional Encoder Representations from Transformers (BERT) to convert text information into vector form. For numeric features, normalization or standardization can be performed.

[0087] Build feature layer (Embedding Layer): Use the Embedding layer to map each feature into a dense vector space, which helps capture complex relationships and patterns between features.

[0088] The embodiment of the present application extracts detailed information of multiple cases from the claims knowledge base, where the multiple cases can correspond to different tasks. Through data preprocessing and feature conversion processing of the extracted detailed information, each original feature of the case can be mapped to a dense vector space, which helps to capture the complex relationships and patterns between features. The feature layer is shared by all tasks, which promotes better generalization performance of the model on the original task and improves the accuracy of fraud risk prediction.

[0089] Optionally, the detailed information includes at least one of case type, claim amount, policyholder information, accident description, number of accidents and policyholder customer level.

[0090] Among them, the embodiment of the present application can extract a variety of detailed information of the case, specifically, it can be one or more of the case type, claim amount, policyholder information, accident description, number of accidents and policyholder customer level. Through the precise extraction of the above information, the detailed information related to the fraud risk can be accurately screened out. The extraction of multiple detailed information is also more suitable for multi-task learning, which further improves the accuracy of fraud risk prediction.

[0091] S203: Determine m tasks, and generate a target sub-prediction model corresponding to each task.

[0092] Wherein, m is a positive integer greater than 1.

[0093] Here, m can be determined according to actual conditions, and the present application embodiment does not impose any specific restrictions on this. Figure 1 The data collection device 101 receives m tasks input by a user.

[0094] Optionally, generating a target sub-prediction model corresponding to each task includes:

[0095] Obtain training data samples corresponding to each task and the original sub-prediction model corresponding to each task, wherein the training data samples include case samples and task labels corresponding to the case samples; according to the training data samples corresponding to each task, train the corresponding original sub-prediction model to obtain the target sub-prediction model corresponding to each task.

[0096] Optionally, the tasks include high claim case identification and fraud case identification.

[0097] Among them, the embodiment of the present application can simultaneously identify high-payout cases and fraud cases, increase the generalization ability of fraud risk prediction, and can accurately and comprehensively predict fraud risks.

[0098] Optionally, tasks may also correspond to different types of fraud cases. For example, multiple types of fraud cases corresponding to different tasks may be divided according to the amount of the fraud cases, or multiple types of fraud cases may be divided according to one or more of the data such as the case type, claim amount, policyholder information, accident description, number of accidents, and policyholder customer level to which the fraud cases belong. This enables accurate prediction of different fraud cases and improves the accuracy of fraud risk prediction.

[0099] In a possible implementation, taking m=2 as an example, the embodiment of the present application constructs 0-1 samples for high compensation cases and fraud cases respectively to achieve modeling of task A (taskA) and task B (taskB).

[0100] Task definition:

[0101] Task A: Identification of high-payout cases, the goal is to predict whether a claim case falls within the high-payout range.

[0102] Task B: Fraud case identification, which can also be fraud case identification, aims to predict whether a claim case involves fraud.

[0103] Sample construction:

[0104] From the historical claims data, we construct a 0-1 sample based on the claim amount and fraud mark. High claim cases and fraud cases are marked as 1, and others are marked as 0.

[0105] Model training:

[0106] Use algorithms such as Deep Neural Network (DNN), Random Forest or eXtreme Gradient Boosting (XGBoost) to model TaskA and TaskB respectively.

[0107] Among them, the embodiments of the present application respectively model different tasks. Specifically, for each task, the corresponding training data samples and the original sub-prediction model are obtained. By using the training data samples to train the original sub-prediction model, a target sub-prediction model focusing on fraud risk prediction of the corresponding task can be obtained, which can accurately realize the fraud risk prediction of the corresponding task. The establishment of multiple target sub-prediction models realizes fraud risk prediction of multiple tasks corresponding to different types of cases, promotes the generalization performance of fraud risk prediction, and can accurately and comprehensively predict fraud risks.

[0108] S204: construct n expert layers, and use each expert layer to perform deep learning on each target sub-prediction model to obtain the weight of each expert layer corresponding to each target sub-prediction model.

[0109] Wherein, n is a positive integer greater than 1.

[0110] Here, n can be determined according to actual conditions, and the embodiments of the present application do not impose any specific limitation on this.

[0111] S205: Construct self-attention layer.

[0112] Among them, the self-attention layer is used to adjust the weights of each expert layer corresponding to each target sub-prediction model.

[0113] S206: Establish a fraud risk multi-task prediction model based on the feature layer, the target sub-prediction model corresponding to each task, n expert layers and the self-attention layer.

[0114] Among them, the fraud risk multi-task prediction model is used to predict the fraud risk corresponding to each task of the case.

[0115] Here, an embodiment of the present application provides a fraud risk prediction method based on multi-task learning. A feature layer shared by multiple tasks is established based on a claims knowledge base to capture the complex relationships and patterns between features and reduce the risk of overfitting. In order to achieve multi-task prediction, respective target sub-prediction models are generated according to multiple tasks. The target sub-prediction model can predict the fraud risk of the corresponding task, and then multiple expert layers are constructed to score the target sub-prediction model. In combination with the self-attention mechanism, a self-attention layer is constructed to perform weighted self-learning on the expert's scores. Different expert scoring weights can be learned for different tasks. Through the above-mentioned feature layer, the target sub-prediction model corresponding to each task, the expert layer and the self-attention layer, a multi-task prediction model for fraud risk can be obtained. Modeling is performed simultaneously for different tasks, and representation information is shared between related tasks to achieve fraud risk prediction corresponding to different tasks, thereby improving the generalization performance of fraud risk prediction and being able to comprehensively and accurately identify fraud risks.

[0116] Optionally, after establishing the fraud risk multi-task prediction model according to the feature layer, the target sub-prediction model corresponding to each task, the n expert layers and the self-attention layer, it also includes:

[0117] Obtain case information of the case to be predicted; input the case information of the case to be predicted into the fraud risk multi-task prediction model to determine the fraud risk prediction results of the case to be predicted corresponding to each task through the output results of the fraud risk multi-task prediction model.

[0118] Here, the embodiment of the present application can perform multi-task fraud risk prediction on the predicted case based on the constructed multi-task fraud risk prediction model, thereby achieving a comprehensive and accurate prediction of the fraud risk of financial cases and improving the security of financial cases.

[0119] Optionally, Figure 3 A flowchart of another fraud risk prediction method provided in an embodiment of the present application is shown in FIG. Figure 3 As shown, the method includes:

[0120] S301: Acquire a claims knowledge base.

[0121] S302: Construct a feature layer based on the claims knowledge base.

[0122] S303: Determine m tasks and generate a target sub-prediction model corresponding to each task.

[0123] Wherein, m is a positive integer greater than 1.

[0124] Among them, the implementation method of steps S301-S303 is the same as the implementation method of steps S201-S203, and will not be repeated here.

[0125] S304: Determine n deep learning objectives and parameter sharing strategies.

[0126] Optionally, you can Figure 1 The data acquisition device 101 receives n deep learning objectives and parameter sharing strategies input by the user.

[0127] S305: Perform expert layer construction processing according to n deep learning objectives and parameter sharing strategies to obtain n expert layers.

[0128] S306: Use each expert layer to perform deep learning on each target sub-prediction model to obtain the weight of each expert layer corresponding to each target sub-prediction model.

[0129] Wherein, n is a positive integer greater than 1.

[0130] S307: Construct a self-attention layer.

[0131] Among them, the self-attention layer is used to adjust the weights of each expert layer corresponding to each target sub-prediction model.

[0132] Optionally, Figure 4 A schematic diagram of the structure of a self-attention layer provided in an embodiment of the present application is shown in FIG. Figure 4 As shown in the figure, the self-attention layer converts each element of the input sequence (for example, the embedded representation of each word) into three vectors, which are query (Q), key (K), and value (V). The attention score of the self-attention layer represents the correlation or matching degree between the query vector and the key vector. The larger the value, the higher the correlation between the two. The output of this layer is obtained by combining the attention score.

[0133] In a possible implementation, taking n=2 as an example, the specific implementation of the expert layer and the self-attention layer is described. The expert layer is a parameter sharing layer constructed by three experts. A neural network is used to simulate that three experts score the prediction results of the two tasks respectively.

[0134] Expert Build:

[0135] Design three expert layers, each of which performs deep learning on a certain aspect of the claims case. For example, expert layer 1 may focus on the semantic analysis of the case description, expert layer 2 may focus on the credibility of the policyholder information, and expert layer 3 may focus on the rationality of the claim amount.

[0136] Parameter sharing: Certain parameters can be shared between expert layers to enhance the generalization ability of the model.

[0137] The self-attention layer is used to self-learn the weights of the experts’ scores. Different expert score weights can be learned for different tasks, and it is determined which experts are more important for each task. Finally, prediction outputs are made for different tasks. Therefore, the model takes into account the relevance and differences of tasks.

[0138] Self-Attention Mechanism:

[0139] A self-attention layer is added above the expert layer. This layer learns the importance of different experts to different tasks by calculating the weights output by the expert layer. This enables the model to adaptively adjust the weights of experts for different tasks, thereby making more accurate predictions for TaskA and TaskB.

[0140] Prediction output:

[0141] Based on the weighted output of the expert layer, predictions are made for Task A and Task B respectively. For high-payout cases and fraud cases, the model can give more accurate judgments.

[0142] S308: Establish a fraud risk multi-task prediction model based on the feature layer, the target sub-prediction model corresponding to each task, n expert layers and the self-attention layer.

[0143] Among them, the fraud risk multi-task prediction model is used to predict the fraud risk corresponding to each task of the case.

[0144] Optionally, Figure 5 A schematic diagram of the structure of a fraud risk multi-task prediction model provided in an embodiment of the present application is shown in FIG. Figure 5 As shown in the figure, the fraud risk multi-task prediction model inputs feature embedding (feature embs) and feature combination (idcross), the input is connected to the connection layer (concat), the connection layer is connected to the three expert layers, and is connected to task A and task B through the self-attention mechanism layer. Task A and task B in the figure refer to the corresponding target sub-prediction models. Finally, the accurate prediction results of task A and task B can be output.

[0145] Here, certain parameters can be shared among the expert layers in the embodiments of the present application, enhancing the generalization ability of the model and further improving the accuracy of fraud risk prediction.

[0146] Figure 6 FIG. is a schematic structural diagram of a fraud risk prediction device provided by an embodiment of the present application. As Figure 6 shown, the device in the embodiment of the present application includes: an acquisition module 601, a first processing module 602, a second processing module 603, a third processing module 604, a fourth processing module 605, and an establishment module 606. The fraud risk prediction device here may be the above-mentioned processing device, the processor itself, or a chip or integrated circuit that implements the functions of the processor. It should be noted here that the division of the acquisition module 601, the first processing module 602, the second processing module 603, the third processing module 604, the fourth processing module 605, and the establishment module 606 is only a logical function division, and physically the two may be integrated or independent.

[0147] Among them,

[0148] The acquisition module is used to acquire a claims knowledge base;

[0149] The first processing module is used to construct a feature layer according to the claims knowledge base;

[0150] The second processing module is used to determine m tasks and generate target sub-prediction models corresponding to each task, where m is a positive integer greater than 1;

[0151] The third processing module is used to construct n expert layers and perform deep learning on each target sub-prediction model using each expert layer to obtain the weights of each expert layer corresponding to each target sub-prediction model, where n is a positive integer greater than 1;

[0152] The fourth processing module is used to construct a self-attention layer, where the self-attention layer is used to adjust the weights of each expert layer corresponding to each target sub-prediction model;

[0153] The establishment module is used to establish a fraud risk multi-task prediction model according to the feature layer, the target sub-prediction models corresponding to each task, the n expert layers, and the self-attention layer, where the fraud risk multi-task prediction model is used to perform fraud risk prediction corresponding to each task on a case.

[0154] Optionally, the first processing module is specifically used for:

[0155] Perform data extraction processing on the claims knowledge base to obtain the detailed information of multiple cases, where the claims knowledge base includes the case information of multiple cases;

[0156] Performing data preprocessing on detailed information of multiple cases to obtain multiple preprocessed detailed information;

[0157] Performing feature conversion processing on a plurality of pre-processed detailed information to obtain a plurality of vector information;

[0158] Construct a feature layer based on multiple vector information.

[0159] Optionally, the detailed information includes at least one of case type, claim amount, policyholder information, accident description, number of accidents and policyholder customer level.

[0160] Optionally, the second processing module is specifically used for:

[0161] Obtaining training data samples corresponding to each task and the original sub-prediction model corresponding to each task, wherein the training data samples include case samples and task labels corresponding to the case samples;

[0162] According to the training data samples corresponding to each task, the corresponding original sub-prediction model is trained to obtain the target sub-prediction model corresponding to each task.

[0163] Optionally, the third processing module is specifically used for:

[0164] Determine n deep learning objectives and parameter sharing strategies;

[0165] The expert layer construction process is performed according to n deep learning objectives and parameter sharing strategies to obtain n expert layers.

[0166] Optionally, tasks include high claim case identification and fraud case identification.

[0167] Optionally, after the establishment module is used to establish the fraud risk multi-task prediction model according to the feature layer, the target sub-prediction model corresponding to each task, the n expert layers and the self-attention layer, the above-mentioned device also includes a prediction module for:

[0168] Obtain case information of the case to be predicted;

[0169] The case information of the case to be predicted is input into the fraud risk multi-task prediction model to determine the fraud risk prediction results of the case to be predicted corresponding to each task through the output results of the fraud risk multi-task prediction model.

[0170] refer to Figure 7, which shows a schematic diagram of the structure of a fraud risk prediction device 700 suitable for implementing the embodiment of the present disclosure, and the fraud risk prediction device 700 can be a terminal device or a server. The terminal device can include but is not limited to mobile terminals such as mobile phones, laptops, digital broadcast receivers, personal digital assistants (PDAs), tablet computers (Portable Android Devices, PADs), portable multimedia players (PMPs), vehicle terminals (such as vehicle navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 7 The fraud risk prediction device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0171] like Figure 7 As shown, the fraud risk prediction device 700 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 701, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 702 or a program loaded from a storage device 708 to a random access memory (RAM) 703. Various programs and data required for the operation of the fraud risk prediction device 700 are also stored in the RAM 703. The processing device 701, the ROM 702, and the RAM 703 are connected to each other via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0172] Typically, the following devices may be connected to the I / O interface 705: input devices 706 including, for example, a touch screen, a touchpad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; output devices 707 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 708 including, for example, a magnetic tape, a hard disk, etc.; and communication devices 709. The communication device 709 may allow the fraud risk prediction device 700 to communicate with other devices wirelessly or by wire to exchange data. Although Figure 7 The fraud risk prediction apparatus 700 having various devices is shown, but it should be understood that it is not required to implement or possess all the devices shown. More or fewer devices may be implemented or possessed instead.

[0173] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device 709, or installed from a storage device 708, or installed from a ROM 702. When the computer program is executed by the processing device 701, the above-mentioned functions defined in the method of the embodiment of the present disclosure are executed.

[0174] It should be noted that the computer-readable medium disclosed above may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in combination with an instruction execution system, device or device. In the present disclosure, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, in which a computer-readable program code is carried. This propagated data signal may take a variety of forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above. The computer readable signal medium may also be any computer readable medium other than a computer readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0175] The computer-readable medium may be included in the fraud risk prediction device; or it may exist independently without being assembled into the fraud risk prediction device.

[0176] The computer-readable medium carries one or more programs. When the one or more programs are executed by the fraud risk prediction device, the fraud risk prediction device executes the method shown in the above embodiment.

[0177] Computer program code for performing the operations of the present disclosure may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as an independent software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of 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).

[0178] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some implementations as replacements, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0179] The units involved in the embodiments described in the present disclosure may be implemented by software or hardware. The name of a unit does not limit the unit itself in some cases. For example, the first acquisition unit may also be described as a "unit for acquiring at least two Internet Protocol addresses".

[0180] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), and the like.

[0181] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0182] The fraud risk prediction device of the embodiment of the present application can be used to execute the technical solutions in the above-mentioned method embodiments of the present application. Its implementation principles and technical effects are similar and will not be repeated here.

[0183] An embodiment of the present application also provides a computer-readable storage medium, which stores computer-executable instructions. When the computer-executable instructions are executed by a processor, they are used to implement any of the above-mentioned fraud risk prediction methods.

[0184] An embodiment of the present application also provides a computer program product, including a computer program, which, when executed by a processor, is used to implement any of the above-mentioned fraud risk prediction methods.

[0185] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0186] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0187] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the invention disclosed herein. The present application is intended to cover any modification, use or adaptation of the present application, which follows the general principles of the present application and includes common knowledge or customary techniques in the art that are not disclosed in the present application. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present application are indicated by the following claims.

[0188] It should be understood that the present application is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.

Claims

1. A fraud risk prediction method, characterized in that: include: Access to the claims knowledge base; Building a feature layer based on the claims knowledge base; Determine m tasks, and generate a target sub-prediction model corresponding to each task, wherein m is a positive integer greater than 1; Constructing n expert layers, and using each expert layer to perform deep learning on each target sub-prediction model to obtain a weight of each expert layer corresponding to each target sub-prediction model, wherein n is a positive integer greater than 1; Constructing a self-attention layer, wherein the self-attention layer is used to adjust the weights of each expert layer corresponding to each target sub-prediction model; A fraud risk multi-task prediction model is established based on the feature layer, the target sub-prediction models corresponding to each task, the n expert layers and the self-attention layer, wherein the fraud risk multi-task prediction model is used to predict the fraud risk corresponding to each task of the case.

2. The method according to claim 1, characterized in that The feature layer is constructed according to the claims knowledge base, including: Performing data extraction processing on the claims knowledge base to obtain detailed information of multiple cases, wherein the claims knowledge base includes case information of multiple cases; Performing data preprocessing on the detailed information of the multiple cases to obtain multiple preprocessed detailed information; Performing feature conversion processing on the plurality of preprocessed detailed information to obtain a plurality of vector information; A feature layer is constructed according to the multiple vector information.

3. The method according to claim 2, characterized in that The detailed information includes at least one of case type, claim amount, policyholder information, accident description, number of accidents and policyholder customer level.

4. The method according to claim 1, characterized in that: The generating of the target sub-prediction model corresponding to each task includes: Obtaining training data samples corresponding to each task and original sub-prediction models corresponding to each task, wherein the training data samples include case samples and task labels corresponding to the case samples; According to the training data samples corresponding to each task, the corresponding original sub-prediction model is trained to obtain the target sub-prediction model corresponding to each task.

5. The method according to any one of claims 1 to 4, characterized in that: The construction of n expert layers includes: Determine n deep learning objectives and parameter sharing strategies; An expert layer construction process is performed according to the n deep learning objectives and the parameter sharing strategy to obtain n expert layers.

6. The method according to any one of claims 1 to 4, characterized in that: The tasks include high claim case identification and fraud case identification.

7. The method according to any one of claims 1 to 4, characterized in that: After establishing the fraud risk multi-task prediction model according to the feature layer, the target sub-prediction models corresponding to the tasks, the n expert layers and the self-attention layer, the method further includes: Obtain case information of the case to be predicted; The case information of the case to be predicted is input into the fraud risk multi-task prediction model, so as to determine the fraud risk prediction results of the case to be predicted corresponding to each task through the output results of the fraud risk multi-task prediction model.

8. A fraud risk prediction device, characterized in that: include: An acquisition module is used to acquire a claims knowledge base; A first processing module, configured to construct a feature layer according to the claims knowledge base; The second processing module is used to determine m tasks and generate a target sub-prediction model corresponding to each task, wherein m is a positive integer greater than 1; The third processing module is used to construct n expert layers, and use each expert layer to perform deep learning on each target sub-prediction model to obtain a weight of each expert layer corresponding to each target sub-prediction model, wherein n is a positive integer greater than 1; A fourth processing module is used to construct a self-attention layer, wherein the self-attention layer is used to adjust the weights of each expert layer corresponding to each target sub-prediction model; A module is established to establish a fraud risk multi-task prediction model based on the feature layer, the target sub-prediction models corresponding to each task, the n expert layers and the self-attention layer, wherein the fraud risk multi-task prediction model is used to predict the fraud risk corresponding to each task of the case.

9. A fraud risk prediction device, characterized in that: include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 7 when executed by a processor.

Images