Multi-data source fusion-based risk assessment

Abstract

Disclosed in the embodiments of the present disclosure are a multi-data source fusion-based risk assessment method and a related apparatus. The multi-data source fusion-based risk assessment method comprises: acquiring a target time series data set corresponding to a target user, the target time series data set comprising first target time series data corresponding to an internal risk assessment model and second target time series data corresponding to each external data source; inputting the target time series data set into a risk assessment model to obtain a target time series fusion feature set and target mask features corresponding to the target time series data set; and, on the basis of the target time series fusion feature set and the target mask features, outputting a target risk assessment result, the target mask features comprising a mask value corresponding to each external data source, and the risk assessment model being obtained by performing training on the basis of time series data sets having known risk assessment results.

Description

Risk assessment based on multi-data source fusion TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of computers, and particularly relates to risk assessment based on multi-data source fusion. BACKGROUND

[0002] In the real scene of industry models, the data richness and quality determine the quality of the model results, so timely updating the model training data source is the key to establishing a user interaction model. At the same time, with the rapid development of various technologies, the data generated by humans every day is growing exponentially. In order to ensure the performance of the model and meet the requirements of society, the model needs to frequently cope with the situation of adding and cutting off data sources.

[0003] In the related data fusion scene, the data source is often fused by means of logistic regression or tree model. Specifically, every time a data source is added or reduced, the original old model needs to be abandoned, and the new model needs to be trained on the full data. SUMMARY

[0004] The embodiments of the present disclosure provide a risk assessment method based on multi-data source fusion and related devices, which can solve the problem of multi-data source fusion by only adjusting the input of the model, without the need for previous feature engineering, model structure modification and other work, so as to effectively realize the input of variable-length data source quantity, improve the efficiency and flexibility of risk assessment based on multi-data source fusion, and further eliminate the missing and cut-off data source information by means of masking, avoid the noise generated by the missing or cut-off data source, and improve the performance and accuracy of the risk assessment model based on multi-data source fusion. The above technical solutions are as follows.

[0005] In a first aspect, the embodiments of the present disclosure provide a risk assessment method based on multi-data source fusion, comprising: obtaining a target time series data set corresponding to a target user; the target time series data set comprises first target time series data corresponding to an internal risk assessment model and second target time series data corresponding to each external data source; inputting the target time series data set into a risk assessment model to obtain a target time series fusion feature set and a target mask feature corresponding to the target time series data set, and outputting a target risk assessment result corresponding to the target user based on the target time series fusion feature set and the target mask feature; wherein the target mask feature comprises a mask value corresponding to each external data source; and the risk assessment model is trained based on a time series data set of known risk assessment results.

[0006] In a possible implementation, the target time sequence fusion feature set includes the first target time sequence fusion feature corresponding to the first target time sequence data and the second target time sequence fusion feature corresponding to each of the second target time sequence data; and the outputting of the target risk assessment result corresponding to the target user based on the target time sequence fusion feature set and the target mask feature includes: performing feature fusion on the second target time sequence fusion feature corresponding to each of the second target time sequence data based on the target mask feature by using a self-attention mechanism to obtain a target data source fusion feature; and outputting the target risk assessment result corresponding to the target user based on the first target time sequence fusion feature and the target data source fusion feature.

[0007] In a possible implementation, the risk assessment model includes a time sequence fusion layer, a mask mechanism layer, a transformer layer, and a residual connection layer; the time sequence fusion layer is configured to map and aggregate the first target time sequence data and each of the second target time sequence data in a time dimension to obtain the first target time sequence fusion feature corresponding to the first target time sequence data and the second target time sequence fusion feature corresponding to each of the second target time sequence data; the mask mechanism layer is configured to perform mask processing on each of the second target time sequence data to obtain the target mask feature; the transformer layer is configured to calculate, by using a self-attention mechanism, target attention of the second target time sequence fusion feature corresponding to each of the second target time sequence data in the target time sequence fusion feature set to the first target time sequence fusion feature, and perform feature fusion on each of the second target time sequence fusion feature based on the target mask feature and the target attention corresponding to each of the second target time sequence fusion feature to obtain a target data source fusion feature; and the residual connection layer is configured to output the target risk assessment result corresponding to the target user based on the target fusion feature obtained by weighted fusion of the first target data source feature and the target data source fusion feature.

[0008] In a possible implementation, the transformer layer is specifically configured to map the first target time sequence fusion feature and each of the second target time sequence fusion feature in a data source dimension to obtain the first target data source feature corresponding to the first target time sequence fusion feature and the second target data source feature corresponding to each of the second target time sequence fusion feature, calculate target attention of each of the second target data source feature to the first target data source feature by using a self-attention mechanism, and perform feature fusion on each of the second target data source feature based on the target mask feature and the target attention corresponding to each of the second target data source feature to obtain a target data source fusion feature.

[0009] In a possible implementation, the mask mechanism layer is specifically configured to: set the mask value corresponding to the second target time series data as 0 when the second target time series data is empty, and set the mask value corresponding to the second target time series data as 1 when the second target time series data is not empty; and / or set the mask value corresponding to the second target time series data of a non-target external data source as 0 when the external data source is the non-target external data source, and set the mask value corresponding to the second target time series data of a target external data source as 1 when the external data source is the target external data source; wherein the non-target external data source is used to represent an external data source that does not need to participate in the risk assessment of the target user, and the target external data source is used to represent an external data source that needs to participate in the risk assessment of the target user.

[0010] In a possible implementation, the risk assessment model further includes a target constraint layer, which is configured to calculate a loss result corresponding to the risk assessment model, and train the risk assessment model based on the target loss result. The loss result is determined based on a first loss and / or a second loss. The first loss is determined based on an actual risk assessment result and a target risk assessment result output by the risk assessment model. The second loss is determined based on the target risk assessment result output by the risk assessment model and an internal model risk assessment result output by an internal risk assessment model.

[0011] In a possible implementation, the method further includes: obtaining an actual risk assessment result corresponding to the target user; calculating a target loss result corresponding to the risk assessment model based on the actual risk assessment result corresponding to the target user, an internal model risk assessment result, and a target risk assessment result; setting a first parameter of the risk assessment model as a fixed parameter, and updating a second parameter of the risk assessment model based on the target loss result; wherein the second parameter is closer to an output layer of the risk assessment model than the first parameter.

[0012] In a possible implementation, the first target time series data and the second target time series data each correspond to a same target time period.

[0013] In a second aspect, the embodiments of the present disclosure provide a risk assessment device based on multi-data source fusion. The device comprises: a first acquisition module configured to acquire a target time series data set corresponding to a target user; the target time series data set comprises first target time series data corresponding to an internal risk assessment model and second target time series data corresponding to each external data source; a risk assessment module configured to input the target time series data set into a risk assessment model, obtain a target time series fusion feature set and a target mask feature corresponding to the target time series data set, and output a target risk assessment result corresponding to the target user based on the target time series fusion feature set and the target mask feature; wherein the target mask feature comprises a mask value corresponding to each external data source; and the risk assessment model is trained based on a time series data set of known risk assessment results.

[0014] In a possible implementation, the target time series fusion feature set comprises first target time series fusion features corresponding to the first target time series data and second target time series fusion features corresponding to each second target time series data; and the risk assessment module comprises: a feature fusion unit configured to fuse the second target time series fusion features corresponding to each second target time series data based on the target mask feature using a self-attention mechanism to obtain target data source fusion features; and an output unit configured to output the target risk assessment result corresponding to the target user based on the first target time series fusion features and the target data source fusion features.

[0015] In a possible implementation, the risk assessment model comprises a time series fusion layer, a mask mechanism layer, a transformer layer, and a residual connection layer; the time series fusion layer is configured to map and aggregate the first target time series data and each second target time series data in the time dimension respectively to obtain the first target time series fusion features corresponding to the first target time series data and the second target time series fusion features corresponding to each second target time series data; the mask mechanism layer is configured to perform mask processing on each second target time series data to obtain the target mask feature; the transformer layer is configured to calculate target attention of each second target time series fusion feature corresponding to each second target time series data in the target time series fusion feature set to the first target time series fusion feature using a self-attention mechanism, and fuse each second target time series fusion feature based on the target mask feature and the target attention corresponding to each second target time series fusion feature to obtain target data source fusion features; and the residual connection layer is configured to output the target risk assessment result corresponding to the target user based on the target fusion features obtained by weighted fusion of the first target data source features and the target data source fusion features.

[0016] In a possible implementation, the transformer layer is specifically configured to: map the first target time sequence fusion feature and each of the second target time sequence fusion features in a data source dimension respectively to obtain a first target data source feature corresponding to the first target time sequence fusion feature and a second target data source feature corresponding to each of the second target time sequence fusion features, and calculate target attention of each of the second target data source features to the first target data source feature by using a self-attention mechanism, and perform feature fusion on each of the second target data source features based on the target mask feature and the target attention corresponding to each of the second target data source features to obtain a target data source fusion feature.

[0017] In a possible implementation, the mask mechanism layer is specifically configured to: set the mask value corresponding to the second target time sequence data to 0 in a case where the second target time sequence data is empty, and set the mask value corresponding to the second target time sequence data to 1 in a case where the second target time sequence data is not empty; and / or set the mask value of the second target time sequence data corresponding to the non-target external data source to 0 in a case where the external data source is a non-target external data source, and set the mask value of the second target time sequence data corresponding to the target external data source to 1 in a case where the external data source is a target external data source; wherein the non-target external data source is used to represent an external data source that does not need to participate in the risk assessment of the target user; and the target external data source is used to represent an external data source that needs to participate in the risk assessment of the target user.

[0018] In a possible implementation, the risk assessment model further includes a target constraint layer, the target constraint layer is configured to calculate a loss result corresponding to the risk assessment model, and train the risk assessment model based on the target loss result; the loss result is determined based on a first loss and / or a second loss, the first loss is determined based on an actual risk assessment result and a target risk assessment result output by the risk assessment model, and the second loss is determined based on the target risk assessment result output by the risk assessment model and an internal model risk assessment result output by an internal risk assessment model.

[0019] In a possible implementation, the risk assessment device based on multi-data source fusion further includes: a second acquisition module, configured to acquire an actual risk assessment result corresponding to the target user; a loss calculation module, configured to calculate a target loss result corresponding to the risk assessment model based on the actual risk assessment result corresponding to the target user, the internal model risk assessment result, and the target risk assessment result; and a parameter updating module, configured to set a first parameter of the risk assessment model as a fixed parameter, and update a second parameter of the risk assessment model based on the target loss result, wherein the second parameter is closer to an output layer of the risk assessment model than the first parameter.

[0020] In a possible implementation, the target time period corresponding to each of the first target time sequence data and the second target time sequence data is the same.

[0021] In a third aspect, an electronic device is provided, including: a processor and a memory; the processor is connected with the memory; the memory is configured to store executable program code; the processor runs a program corresponding to the executable program code by reading the executable program code stored in the memory, so as to execute the method provided in the first aspect or any possible implementation manner of the first aspect of the present disclosure.

[0022] In a fourth aspect, a computer storage medium is provided, which stores a plurality of instructions, and the instructions are adapted to be loaded by a processor and execute the method provided in the first aspect or any possible implementation manner of the first aspect of the present disclosure.

[0023] In a fifth aspect, a computer program product containing instructions is provided, when the computer program product runs on a computer or a processor, the computer or the processor executes the method provided in the first aspect or any possible implementation manner of the first aspect of the present disclosure.

[0024] In an aspect, more data information can be absorbed in the time dimension through the timing input, and the problem of unstable user risk assessment caused by data loss can be avoided. In another aspect, the features of the first target time series data and the features of each second target time series data after time series fusion are fused based on the mask values corresponding to each external data source. Through the mask method, the fusion problem of multiple data sources can be solved by only adjusting the input of the model. Without the need for previous feature engineering, model structure modification, and other work, the input of variable-length data source quantity can be effectively realized, the efficiency and flexibility of risk assessment based on multi-data source fusion are improved, the missing and interrupted data source information is excluded, the noise caused by data source loss or interruption is avoided, and the performance and accuracy of the risk assessment model based on multi-data source fusion for risk assessment are improved. In another aspect, the first target time series data corresponding to the internal risk assessment model is allowed to participate in risk assessment, which can improve the voice of the internal model evaluation result corresponding to the internal risk assessment model in the risk assessment model to some extent, improve the consistency of the target risk assessment result of the target user before and after the risk assessment using external data sources, and avoid the poor consistency of the model evaluation result before and after the risk assessment using external data sources or different external data sources, which leads to a huge difference in the risk assessment result of the user before and after the risk assessment, and causes poor user experience of the user. BRIEF DESCRIPTION OF DRAWINGS

[0025] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure, the drawings required to be used in the embodiments will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present disclosure, and other drawings can be obtained by those skilled in the art without creative labor.

[0026] FIG. 1 is an architecture schematic diagram of a risk assessment system based on multi-data source fusion according to an example embodiment of the present disclosure;

[0027] FIG. 2 is a flow schematic diagram of a risk assessment method based on multi-data source fusion according to an example embodiment of the present disclosure;

[0028] FIG. 3 is an architecture schematic diagram of a risk assessment model according to an example embodiment of the present disclosure;

[0029] FIG. 4 is a flow schematic diagram of another risk assessment method based on multi-data source fusion according to an example embodiment of the present disclosure;

[0030] FIG. 5 is a structure schematic diagram of a risk assessment device based on multi-data source fusion according to an example embodiment of the present disclosure;

[0031] FIG. 6 is a structural schematic diagram of an electronic device according to an example embodiment of the present disclosure. DETAILED DESCRIPTION

[0032] The technical solutions in the example embodiments of the present disclosure will be described clearly and completely below with reference to the drawings in the example embodiments of the present disclosure.

[0033] The terms "first", "second", "third", and the like in the present disclosure and claims and the above drawings are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device including a series of steps or units is not limited to the listed steps or units, but can optionally include steps or units not listed or can optionally include other steps or units inherent to the process, method, product, or device.

[0034] It should be noted that the information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data for analysis, stored data, displayed data, etc.), and signals involved in the example embodiments of the present disclosure are all authorized by the user or fully authorized by all parties, and the collection, use, and processing of related data need to comply with relevant laws, regulations, and standards in relevant countries and regions. For example, the target time series data set and the actual risk assessment result involved in the present disclosure are obtained under sufficient authorization.

[0035] Next, please refer to FIG. 1, which is an architecture schematic diagram of a risk assessment system based on multi-data source fusion according to an example embodiment of the present disclosure. As shown in FIG. 1, the risk assessment system based on multi-data source fusion includes a terminal 110 and a server 120. The terminal 110 can include one or more terminals corresponding to users. A user version of software can be installed in the terminal 110, which is used to implement functions such as providing credit services and risk assessment services for users.

[0036] Optionally, the terminal 110 can be installed with a target application and a plurality of non-target applications in cooperation with the target application, but is not limited to this. The target application is an application that needs to be risk assessed, which can be but is not limited to built-in with an internal risk assessment model, used to perform risk assessment on a target user based on target internal data of the target user in the target application, and obtain an internal model risk assessment result. The non-target application is an application that can provide external data of the target user for the risk assessment process corresponding to the target application, and one non-target application can correspond to one external data source involved in the risk assessment process corresponding to the target application.

[0037] Optionally, the terminal 110 can establish a data relationship with the network, and establish a data connection relationship with the server 120 through the network, for example, send the target time series data set corresponding to the target user to the server 120, etc. Any one of the terminals 110 can be, but is not limited to, a mobile phone, a tablet computer, a notebook computer, etc. installed with a user version software.

[0038] The server 120 can be a server capable of providing risk assessment based on multi-data source fusion. The server 120 can obtain first target time series data, second target time series data, etc. from a plurality of terminals 110 or non-target application corresponding servers through a network. The server 120 can also obtain a target time series data set corresponding to a target user, the target time series data set including first target time series data corresponding to an internal risk assessment model and second target time series data corresponding to each external data source; and input the above target time series data set into the risk assessment model to obtain a target time series fusion feature set and a target mask feature corresponding to the above target time series data set, and output a target risk assessment result corresponding to the target user based on the above target time series fusion feature set and the target mask feature. The above target mask feature includes a mask value corresponding to each of the above external data sources; and the above risk assessment model is trained based on a time series data set of a known risk assessment result.

[0039] It can be understood that the server 120 can be, but is not limited to, a hardware server, a virtual server, a cloud server, etc.

[0040] The above risk assessment based on multi-data source fusion is not limited to being executed by the server 120, but can also be executed by any one of the terminals 110. The embodiments of the present disclosure do not make specific limitations on this, and the following embodiments are all described by taking the risk assessment based on multi-data source fusion executed by the terminal 110 as an example.

[0041] The network can be a medium providing a communication link between the server 120 and any one of the terminals 110, or can be an Internet including network devices and transmission media, but is not limited to this. The transmission medium can be a wired link, for example, but not limited to, a coaxial cable, an optical fiber, and a digital subscriber line (DSL), etc., or a wireless link, for example, but not limited to, wireless fidelity (WIFI), Bluetooth, and a mobile device network, etc.

[0042] It can be understood that the number of terminals 110 and servers 120 in the risk assessment system based on multi-data source fusion shown in FIG. 1 is only an example, and in a specific implementation, the risk assessment system based on multi-data source fusion can include any number of terminals and servers, and the embodiments of the present disclosure do not make specific limitations thereon. For example, but not limited to, the terminal 110 can be a terminal cluster composed of multiple terminals, and the server 120 can be a server cluster composed of multiple servers.

[0043] Next, in combination with FIG. 1, the risk assessment method based on multi-data source fusion provided by the embodiments of the present disclosure is introduced. For details, please refer to FIG. 2, which is a flowchart of a risk assessment method based on multi-data source fusion provided by an exemplary embodiment of the present disclosure. As shown in FIG. 2, the risk assessment method based on multi-data source fusion includes the following steps.

[0044] S202, obtaining a target time series data set corresponding to a target user, the target time series data set including first target time series data corresponding to an internal risk assessment model and second target time series data corresponding to each external data source.

[0045] Specifically, the first target time series data can include but is not limited to a plurality of internal model risk assessment results arranged in time sequence; the internal model risk assessment result can be obtained by the internal risk assessment model based on the internal data of the target user, that is, the terminal or server corresponding to the target application collects the internal data (such as but not limited to internal behavior data or content financial data, etc.) of the target user in the target application according to a preset frequency or in real time, and inputs the internal data into the internal risk assessment model to output the internal model risk assessment result corresponding to the target user. The internal risk assessment model can be trained based on the internal data of a plurality of users with known risk assessment results, and can adopt a tree model structure. The internal risk assessment model can be set in the terminal or server corresponding to the target application that needs to perform user risk assessment. The second target time series data can include but is not limited to a plurality of external data corresponding to the external data source arranged in time sequence. The external data source can be a non-target application having a cooperative relationship with the target application or other institutions except the target institution corresponding to the target application. The external data can be but not limited to behavior data or financial data of the target user in the non-target application or other institutions, and other data related to risk assessment. The server for risk assessment can receive the first target time series data provided by the terminal or server corresponding to the target application, and receive the second target time series data provided by each external data source.

[0046] It can be understood that, in order to ensure the effectiveness and accuracy of the risk assessment, the target time period corresponding to each of the first target time series data and the second target time series data should be the same, for example but not limited to, both being time series data composed of data in the previous year or the previous month sorted in chronological order.

[0047] In S204, the target time series data set is input into the risk assessment model to obtain a target time series fusion feature set and a target mask feature corresponding to the target time series data set, and output a target risk assessment result corresponding to the target user based on the target time series fusion feature set and the target mask feature. The target mask feature includes a mask value corresponding to each external data source.

[0048] Specifically, the target risk assessment result can include but is not limited to a target risk score corresponding to the target user or whether the target user belongs to a risk user. The target risk score can include but is not limited to a probability of the target user being overdue after credit or a probability of the target user being currently at risk of fraud, etc. The specific implementation can be determined according to the actual risk assessment scenario, and the embodiments of the present disclosure are not limited thereto. The risk assessment model is trained based on a time series data set of known risk assessment results. The time series data set includes first time series data and second time series data corresponding to each external data source.

[0049] Optionally, the target time series fusion feature set includes first target time series fusion features corresponding to the first target time series data and second target time series fusion features corresponding to each second target time series data. The implementation process of the risk assessment model outputting a target risk assessment result corresponding to the target user based on the target time series fusion feature set and the target mask feature can include but is not limited to: first, based on the target mask feature, using a self-attention mechanism to fuse the second target time series fusion features corresponding to each second target time series data, for example but not limited to, using the self-attention mechanism to calculate the target attention of the second target time series fusion features corresponding to each second target time series data to the first target time series fusion features, and based on the target mask feature and the target attention of each second target time series fusion feature, fusing the second target time series fusion features to obtain target data source fusion features; then, based on the first target time series fusion features and the target data source fusion features, outputting a target risk assessment result corresponding to the target user.

[0050] In the embodiments of the present disclosure, the risk assessment model utilizes the self-attention mechanism to perform feature fusion on the first target time sequence fusion feature corresponding to the first target time sequence data and the second target time sequence fusion feature corresponding to each second target time sequence data in the target time sequence fusion feature set, to obtain a target data source fusion feature, so that the fusion problem of multiple data sources can be solved by adjusting the input of the model only, without the need for feature engineering, model structure modification and other work in advance, so as to effectively realize the input of variable-length data source quantity and improve the efficiency and flexibility of risk assessment based on multi-data source fusion.

[0051] It can be understood that the number of external data sources corresponding to the input in the risk assessment model training process and the number of external data sources corresponding to the input in the risk assessment process can be the same or different, which is not limited in the embodiments of the present disclosure. That is, in the embodiments of the present disclosure, even if the number of external data sources corresponding to the input in the risk assessment process is different from that in the model training process, the second target time sequence data of the external data source lacking in the input can be masked by the mask feature directly, without the need for retraining the model or modifying the model structure, so as to flexibly and efficiently realize risk assessment of different input quantities or types of external data sources.

[0052] In the embodiments of the present disclosure, on the one hand, time sequence input can absorb more data information in the time dimension, avoiding the problem of unstable user risk assessment caused by data loss; on the other hand, the features of the first target time sequence data and each second target time sequence data in the target time sequence fusion feature set are fused based on the mask values corresponding to each external data source, so that the fusion problem of multiple data sources can be solved by adjusting the input of the model only, without the need for feature engineering, model structure modification and other work in advance, so as to effectively realize the input of variable-length data source quantity, improve the efficiency and flexibility of risk assessment based on multi-data source fusion, and eliminate the missing and interrupted data source information, avoiding the noise caused by data source loss or interruption, and improving the performance and accuracy of the risk assessment model based on multi-data source fusion; on the other hand, by allowing the first target time sequence data corresponding to the internal risk assessment model to participate in risk assessment, the voice of the internal model evaluation result corresponding to the internal risk assessment model in the risk assessment model can be improved to some extent, the consistency of the target risk assessment result of the target user before and after the risk assessment based on the external data source is improved, and the problem of poor user experience caused by the large difference between the risk assessment results of the user before and after the risk assessment based on the external data source or different external data sources is avoided.

[0053] In some possible embodiments, as shown in FIG. 3, the above risk assessment model can include, but is not limited to, a time fusion layer, a mask mechanism layer, a transformer layer, and a residual connection layer. Among them: the time fusion layer is configured to respectively map and aggregate the first target time series data and each of the second target time series data in the time dimension to obtain first target time fusion features corresponding to the first target time series data and second target time fusion features corresponding to each of the second target time series data. Since each kind of user data has more or less decay in the time dimension, that is, the older the data, the less gain it brings. Therefore, the time fusion layer can more accurately convert each data source (internal data source and each external data source) from a time series input into a time fusion feature through a learnable time weight parameter (i.e., mapping and aggregation in the time dimension).

[0054] The mask mechanism layer is configured to perform mask processing on each of the second target time series data to obtain target mask features, so as to mask the data source information that the model does not want to see through the mask mechanism, and avoid data noise caused by data source missing or flow interruption.

[0055] Optionally, the implementation process of performing mask processing on each of the second target time series data to obtain target mask features can include, but is not limited to: in a case where the second target time series data corresponding to a certain external data source is empty, it indicates that there is no data of the target user at the external data source or the target user's external data at the external data source is not authorized to be used, and then the mask value corresponding to the second target time series data can be set to 0 to inform the risk assessment model that the second target time series data corresponding to the external data source will not participate in the risk assessment, and the fusion weight corresponding to the second target time series data can be directly set to 0 to avoid noise caused by data source missing; and in a case where the second target time series data is not empty, it indicates that the target user's external data at the external data source has been authorized to be taken this time, and then the mask value corresponding to the second target time series data of the external data can be set to 1 to indicate that the second target time series data corresponding to the external data source needs to participate in the risk assessment. And / or, in a case where the external data source is a non-target external data source, it indicates that the external data source does not need to participate in the risk assessment of the target user, and then the mask value of the second target time series data corresponding to the non-target external data source can be directly set to 0; in a case where the external data source is a target external data source, it indicates that the external data source needs to participate in the risk assessment of the target user, and then the mask value of the second target time series data corresponding to the target external data source can be set to 1. The non-target external data source and the target external data source can be set and updated according to actual risk assessment requirements, and the embodiments of the present disclosure do not limit this.

[0056] The transformer layer is configured to calculate target attention of each second target time series fusion feature corresponding to the second target time series data to the first target time series fusion feature by using a self-attention mechanism, and perform feature fusion on each second target time series fusion feature based on the target mask feature and the target attention corresponding to each second target time series fusion feature, to obtain a target data source fusion feature. That is, after the first target time series data and the second target time series data of each external data source are input into the time series fusion layer and the mask mechanism layer, the data are input into the transformer layer. The multi-head self-attention mechanism can be used to calculate the attention of each data source to other data sources, and the fusion weight probability is obtained by using the softmax method, so as to realize the fusion of different data sources.

[0057] In the embodiment of the present disclosure, the multi-head attention mechanism in the transformer can well solve the problem of changes in the number of data sources. Similarly, in NLP, a sentence of different lengths can be input, and here each data source can be regarded as a word in NLP. The mutual weight between data sources is dynamically calculated by using the self-attention method, and the number of data source inputs that the model can accept does not need to be displayed or set in advance. When the data source is interrupted or added, the model can realize small sample parameter fine-tuning without any modification, and the corresponding cost is low.

[0058] Optionally, the transformer layer can be but is not limited to being specifically used for: mapping the first target time series fusion feature and each second target time series fusion feature in the data source dimension respectively, to obtain a first target data source feature corresponding to the first target time series fusion feature and a second target data source feature corresponding to each second target time series fusion feature, so that the target time series data corresponding to the data source dimension of each data source (internal data source and each external data source) can be more accurately obtained by using the learnable data source weight parameter (i.e., mapping in the data source dimension), and the model understandable representation (target data source feature) of the data source dimension, and the target attention of each second target data source feature to the first target data source feature is calculated by using a self-attention mechanism, and each second target data source feature is fused based on the target mask feature and the target attention corresponding to each second target data source feature, to obtain a target data source fusion feature, for example but not limited to: the fusion weight of the second target data source feature with the mask value of 0 is set to 0, the fusion weight of the second target data source feature with the mask value of 1 is set to 0, and the target data source fusion feature is obtained by weighting and fusing the fusion weight probability calculated according to the corresponding target attention.

[0059] The residual connection layer is configured to output a target risk assessment result corresponding to the target user based on the first target data source feature and the target data source fusion feature.

[0060] Optionally, next, please continue to refer to FIG. 3. As shown in FIG. 3, the risk assessment model can further include, but is not limited to, a target constraint layer. The target constraint layer is configured to calculate a target loss result corresponding to the risk assessment model, and train the risk assessment model based on the target loss result. The target loss result is determined based on a first loss and / or a second loss. The first loss (i.e., a classification loss) can be determined based on an actual risk assessment result and a target risk assessment result output by the risk assessment model, and is used to represent a gap between the actual risk assessment result corresponding to the target user and the target risk assessment result output by the model. The second loss (i.e., a consistency loss) can be determined based on the target risk assessment result output by the risk assessment model and an internal model risk assessment result output by the internal risk assessment model. That is, when the number of external data sources input into the risk assessment model changes and it is desired to improve the accuracy of the model or to train the risk assessment model, the risk assessment model can be constrained by multiple losses to make the risk assessment model as close as possible to the internal model assessment result before the external data sources are fused, while maintaining the performance of the risk assessment model, thereby improving the stability of the risk assessment.

[0061] Next, please refer to FIG. 4, which is a flowchart of another risk assessment method based on multi-data source fusion provided by an example embodiment of the present disclosure. As shown in FIG. 4, the risk assessment method based on multi-data source fusion includes the following steps.

[0062] S402, obtaining a target time series data set corresponding to a target user, the target time series data set including a first target time series data corresponding to an internal risk assessment model and a second target time series data corresponding to each external data source.

[0063] Specifically, S402 is consistent with S202, which will not be repeated here.

[0064] S404, input the target time series data set into the risk assessment model to obtain a target time series fusion feature set and a target mask feature corresponding to the target time series data set, and output a target risk assessment result corresponding to the target user based on the target time series fusion feature set and the target mask feature, the target mask feature including a mask value corresponding to each external data source.

[0065] Specifically, S404 is consistent with S204, which will not be repeated here.

[0066] S406, obtain an actual risk assessment result corresponding to the target user.

[0067] Specifically, if the external data source input into the risk assessment model changes this time, or subsequent more accurate risk assessment is desired, after obtaining the target risk assessment result corresponding to the target user, the actual risk assessment result corresponding to the target user can be collected, such as but not limited to whether the actual situation is overdue or the actual situation is fraud, etc., so as to update the risk assessment model according to the actual risk assessment result of the target user in subsequent, and improve the model precision.

[0068] S408, based on the actual risk assessment result corresponding to the target user, the internal model risk assessment result and the target risk assessment result, calculate a target loss result corresponding to the risk assessment model.

[0069] Specifically, the above S408 can be but not limited to executed by the target constraint layer of the risk assessment model. That is, a first loss can be calculated based on the actual risk assessment result corresponding to the target user and the target risk assessment result output by the model using a preset classification constraint function, and a second loss function can be calculated based on the actual risk assessment result corresponding to the target user and the internal model risk assessment result using a preset consistency constraint function. Then calculate the sum of the first loss and the second loss to obtain the target loss result corresponding to the risk assessment model.

[0070] S410, set the first parameter of the risk assessment model as a fixed parameter, and update the second parameter of the risk assessment model based on the target loss result.

[0071] Specifically, after obtaining the target loss result corresponding to the risk assessment model, in order to reduce the model training and updating cost, the first parameter of the risk assessment model can be but not limited to set as a fixed parameter, that is, the first parameter remains unchanged, and then the second parameter of the risk assessment model is updated based on the target loss result, that is, only the second parameter in the model is updated and trained. The second parameter is closer to the output layer of the risk assessment model than the first parameter, and the first parameter and the second parameter can be pre-set according to actual training requirements, which is not limited in the present embodiment.

[0072] It can be understood that the more the first parameters, i.e., the fixed parameters, are set, the lower the cost of model updating is, and the less time is needed for updating; the less the first parameters, i.e., the fixed parameters, are set, the higher the accuracy of the updated model is, and the better the updating effect is.

[0073] Next, please refer to FIG. 5, which is a structural schematic diagram of a risk assessment device based on multi-data source fusion provided by an exemplary embodiment of the present disclosure. As shown in FIG. 5, the risk assessment device based on multi-data source fusion 500 includes: a first acquisition module 510, configured to acquire a target time series data set corresponding to a target user; the target time series data set includes first target time series data corresponding to an internal risk assessment model and second target time series data corresponding to each external data source; a risk assessment module 520, configured to input the target time series data set into a risk assessment model to obtain a target time series fusion feature set and a target mask feature corresponding to the target time series data set, and output a target risk assessment result corresponding to the target user based on the target time series fusion feature set and the target mask feature; wherein the target mask feature includes a mask value corresponding to each external data source; the risk assessment model is trained based on a time series data set of known risk assessment results.

[0074] In a possible implementation, the target time series fusion feature set includes first target time series fusion features corresponding to the first target time series data and second target time series fusion features corresponding to each second target time series data; the risk assessment module 520 includes: a feature fusion unit, configured to perform feature fusion on the second target time series fusion features corresponding to each second target time series data based on the target mask feature by using a self-attention mechanism to obtain target data source fusion features; and an output unit, configured to output the target risk assessment result corresponding to the target user based on the first target time series fusion features and the target data source fusion features.

[0075] In a possible implementation, the risk assessment model comprises a time sequence fusion layer, a mask mechanism layer, a transformer layer, and a residual connection layer; the time sequence fusion layer is configured to map and aggregate the first target time sequence data and each of the second target time sequence data in a time dimension, to obtain first target time sequence fusion features corresponding to the first target time sequence data and second target time sequence fusion features corresponding to each of the second target time sequence data; the mask mechanism layer is configured to perform mask processing on each of the second target time sequence data, to obtain target mask features; the transformer layer is configured to calculate, by using a self-attention mechanism, target attention of each of the second target time sequence fusion features on the first target time sequence fusion features, and perform feature fusion on each of the second target time sequence fusion features based on the target mask features and the target attention of each of the second target time sequence fusion features, to obtain target data source fusion features; and the residual connection layer is configured to output a target risk assessment result corresponding to the target user based on a target fusion feature obtained by weighted fusion of the first target data source features and the target data source fusion features.

[0076] In a possible implementation, the transformer layer is specifically configured to map the first target time sequence fusion features and each of the second target time sequence fusion features in a data source dimension, to obtain first target data source features corresponding to the first target time sequence fusion features and second target data source features corresponding to each of the second target time sequence fusion features, calculate target attention of each of the second target data source features on the first target data source features by using a self-attention mechanism, and perform feature fusion on each of the second target data source features based on the target mask features and the target attention of each of the second target data source features, to obtain target data source fusion features.

[0077] In a possible implementation, the mask mechanism layer is specifically configured to set a mask value corresponding to the second target time sequence data to 0 in a case where the second target time sequence data is empty, and set the mask value corresponding to the second target time sequence data to 1 in a case where the second target time sequence data is not empty; and / or set a mask value of the second target time sequence data corresponding to a non-target external data source to 0 in a case where the external data source is the non-target external data source, and set a mask value of the second target time sequence data corresponding to a target external data source to 1 in a case where the external data source is the target external data source; wherein the non-target external data source is used to represent an external data source that does not need to participate in risk assessment of the target user; and the target external data source is used to represent an external data source that needs to participate in risk assessment of the target user.

[0078] In a possible implementation, the risk assessment model further includes: a target constraint layer; the target constraint layer is configured to calculate a target loss result corresponding to the risk assessment model, and train the risk assessment model based on the target loss result; the target loss result is determined based on a first loss and / or a second loss, the first loss is determined based on an actual risk assessment result and a target risk assessment result output by the risk assessment model, and the second loss is determined based on the target risk assessment result output by the risk assessment model and an internal model risk assessment result output by an internal risk assessment model.

[0079] In a possible implementation, the risk assessment device 500 based on multi-data-source fusion further includes: a second acquisition module, configured to acquire an actual risk assessment result corresponding to the target user; a loss calculation module, configured to calculate a target loss result corresponding to the risk assessment model based on the actual risk assessment result corresponding to the target user, an internal model risk assessment result, and a target risk assessment result; and a parameter updating module, configured to set a first parameter of the risk assessment model as a fixed parameter, and update a second parameter of the risk assessment model based on the target loss result; the second parameter is closer to an output layer of the risk assessment model than the first parameter.

[0080] In a possible implementation, the first target time sequence data and the second target time sequence data each correspond to a same target time period.

[0081] The division of the modules in the risk assessment device based on multi-data-source fusion is merely for example, and in other embodiments, the risk assessment device based on multi-data-source fusion can be divided into different modules as needed to complete all or part of the functions of the risk assessment device based on multi-data-source fusion. The implementation of each module in the risk assessment device based on multi-data-source fusion provided in the embodiments of the present disclosure can be in the form of a computer program. The computer program can run on a terminal or a server. The program modules constituted by the computer program can be stored on the memory of the terminal or the server. When the computer program is executed by the processor, all or part of the steps of the risk assessment method based on multi-data-source fusion described in the embodiments of the present disclosure are implemented.

[0082] Next, please refer to FIG. 6, which is a structural schematic diagram of an electronic device provided in an example embodiment of the present disclosure. As shown in FIG. 6, the electronic device 600 can include at least one processor 610, at least one communication bus 620, a user interface 630, at least one network interface 640, and a memory 650. The communication bus 620 can be used to realize the connection and communication of the above-mentioned components.

[0083] The user interface 630 can include a display and a camera, and can further include a standard wired interface, a wireless interface, and the like.

[0084] The network interface 640 can include a Bluetooth module, a near field communication (NFC) module, a wireless fidelity (Wi-Fi) module, and the like.

[0085] The processor 610 can include one or more processing cores. The processor 610 connects various parts in the electronic device 600 through various interfaces and lines, and performs various functions of the electronic device 600 and processes data by running or executing instructions, programs, code sets, or instruction sets stored in the memory 650, and calling data stored in the memory 650. Alternatively, the processor 610 can be implemented in at least one of a digital signal processing (DSP), a field-programmable gate array (FPGA), and a programmable logic array (PLA). The processor 610 can be integrated with a combination of one or more of a central processing unit (CPU), a graphics processing unit (GPU), and a modem. The CPU is mainly used to process an operating system, a user interface, and an application program. The GPU is used to render and draw content to be displayed on the display. The modem is used to process wireless communication. It can be understood that the modem can also not be integrated into the processor 610, but can be implemented by a separate chip.

[0086] The memory 650 can include a random access memory (RAM) and can also include a read-only memory (ROM). Optionally, the memory 650 includes a non-transitory computer-readable medium. The memory 650 can be used to store instructions, programs, codes, code sets, or instruction sets. The memory 650 can include a program storage area and a data storage area, where the program storage area can store instructions for implementing an operating system, instructions for at least one function (such as a obtaining function, a risk assessment function, a masking function, etc.), instructions for implementing the various method embodiments described above, and the like; the data storage area can store data related to the various method embodiments described above, and the like. The memory 650 can also optionally be at least one storage device located remotely from the aforementioned processor 610. As shown in FIG. 6, the memory 650, as a computer storage medium, can include an operating system, a network communication module, a user interface module, and program instructions.

[0087] In some possible embodiments, the electronic device 600 described above can be the aforementioned risk assessment device based on multi-data source fusion, and the processor 610 can be configured to invoke program instructions stored in the memory 650 and specifically perform the following operations: obtaining a target time series data set corresponding to a target user; the target time series data set includes first target time series data corresponding to an internal risk assessment model and second target time series data corresponding to each external data source; inputting the target time series data set into a risk assessment model to obtain a target time series fusion feature set corresponding to the target time series data set and a target masking feature, and outputting a target risk assessment result corresponding to the target user based on the target time series fusion feature set and the target masking feature; wherein the target masking feature includes a masking value corresponding to each external data source; the risk assessment model is trained based on a time series data set of known risk assessment results.

[0088] In some possible embodiments, the target time series fusion feature set includes first target time series fusion features corresponding to the first target time series data and second target time series fusion features corresponding to each second target time series data; when the processor 610 performs the operation of outputting the target risk assessment result corresponding to the target user based on the target time series fusion feature set and the target masking feature, it is specifically configured to perform: using a self-attention mechanism to perform feature fusion on the second target time series fusion features corresponding to each second target time series data based on the target masking feature to obtain a target data source fusion feature; and outputting the target risk assessment result corresponding to the target user based on the first target time series fusion features and the target data source fusion feature.

[0089] In some possible embodiments, the risk assessment model comprises a time sequence fusion layer, a mask mechanism layer, a transformer layer, and a residual connection layer. The time sequence fusion layer is configured to map and aggregate the first target time sequence data and each of the second target time sequence data in a time dimension, to obtain first target time sequence fusion features corresponding to the first target time sequence data and second target time sequence fusion features corresponding to each of the second target time sequence data. The mask mechanism layer is configured to perform mask processing on each of the second target time sequence data, to obtain target mask features. The transformer layer is configured to calculate, by using a self-attention mechanism, target attention of each of the second target time sequence fusion features on the first target time sequence fusion features, and perform feature fusion on each of the second target time sequence fusion features based on the target mask features and the target attention of each of the second target time sequence fusion features, to obtain target data source fusion features. The residual connection layer is configured to output a target risk assessment result corresponding to the target user based on a weighted fusion of the first target data source features and the target data source fusion features.

[0090] In some possible embodiments, the transformer layer is specifically configured to map the first target time sequence fusion features and each of the second target time sequence fusion features in a data source dimension, to obtain first target data source features corresponding to the first target time sequence fusion features and second target data source features corresponding to each of the second target time sequence fusion features, and calculate, by using a self-attention mechanism, target attention of each of the second target data source features on the first target data source features, and perform feature fusion on each of the second target data source features based on the target mask features and the target attention of each of the second target data source features, to obtain target data source fusion features.

[0091] In some possible embodiments, the mask mechanism layer is specifically configured to perform the following operations: in a case where the second target time sequence data is empty, setting a mask value corresponding to the second target time sequence data to 0, and in a case where the second target time sequence data is not empty, setting the mask value corresponding to the second target time sequence data to 1; and / or in a case where the external data source is a non-target external data source, setting a mask value of the second target time sequence data corresponding to the non-target external data source to 0, and in a case where the external data source is a target external data source, setting a mask value of the second target time sequence data corresponding to the target external data source to 1; wherein the non-target external data source is used to represent an external data source that does not need to participate in risk assessment of the target user, and the target external data source is used to represent an external data source that needs to participate in risk assessment of the target user.

[0092] In some possible embodiments, the risk assessment model further includes: a target constraint layer; the target constraint layer is configured to calculate a loss result corresponding to the risk assessment model, and train the risk assessment model based on the target loss result; the loss result is determined based on a first loss and / or a second loss, the first loss is determined based on an actual risk assessment result and a target risk assessment result output by the risk assessment model, and the second loss is determined based on the target risk assessment result output by the risk assessment model and an internal model risk assessment result output by an internal risk assessment model.

[0093] In some possible embodiments, the processor 610 is further configured to perform: obtaining an actual risk assessment result corresponding to the target user; calculating a target loss result corresponding to the risk assessment model based on the actual risk assessment result corresponding to the target user, an internal model risk assessment result, and a target risk assessment result; setting a first parameter of the risk assessment model as a fixed parameter, and updating a second parameter of the risk assessment model based on the target loss result; and the second parameter is closer to an output layer of the risk assessment model than the first parameter.

[0094] In some possible embodiments, the first target time sequence data and the second target time sequence data each correspond to a same target time period.

[0095] The embodiments of the present disclosure further provide a computer readable storage medium, which stores instructions. When the instructions are run on a computer or a processor, the computer or the processor performs one or more steps in the above embodiments. When the constituent modules of the above risk assessment device based on multi-data source fusion are realized in the form of software function units and sold or used as independent products, the constituent modules can be stored in the above computer readable storage medium.

[0096] In the above embodiments, all or part of the methods can be implemented by software, hardware, firmware, or any combination thereof. When implemented by software, all or part of the methods can be implemented in the form of a computer program product. The above computer program product includes one or more computer instructions. When the above computer program instructions are loaded and executed on a computer, all or part of the above processes or functions according to the embodiments of the present disclosure are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted by the computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center through a wired (such as a coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.) manner. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. that includes one or more available media sets. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a magnetic tape), an optical medium (for example, a digital versatile disc (DVD)), or a semiconductor medium (for example, a solid state disk (SSD)), etc.

[0097] A person of ordinary skill in the art can understand that all or part of the above-mentioned method embodiments can be implemented by a computer program to instruct the relevant hardware, and the program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the above-mentioned embodiments of the method. The storage medium includes ROM, RAM, magnetic or optical disks, and various program code storage media. In the case of no conflict, the technical features in the embodiments and the embodiments can be combined arbitrarily.

[0098] The above-described embodiments are merely described as preferred embodiments of the present disclosure, and do not limit the scope of the present disclosure. Without departing from the design spirit of the present disclosure, various modifications and improvements of the technical solutions of the present disclosure made by a person of ordinary skill in the art shall fall within the protection scope determined by the claims.

[0099] The above describes particular embodiments of the present disclosure. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims and the specification can be performed in a different order and still accomplish the desired result. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve the desired results. In certain implementations, multitasking and parallel processing can be advantageous.

Claims

1. A risk assessment method based on multi-data source fusion, the method comprising: obtaining a target time series data set corresponding to a target user; the target time series data set comprises first target time series data corresponding to an internal risk assessment model and second target time series data corresponding to each external data source; inputting the target time series data set into a risk assessment model to obtain a target time series fusion feature set corresponding to the target time series data set and a target mask feature, and outputting a target risk assessment result corresponding to the target user based on the target time series fusion feature set and the target mask feature; wherein the target mask feature comprises a mask value corresponding to each external data source; the risk assessment model is trained based on a time series data set of known risk assessment results.

2. The method of claim 1, wherein the target time series fusion feature set comprises first target time series fusion features corresponding to the first target time series data and second target time series fusion features corresponding to each second target time series data; the outputting of the target risk assessment result corresponding to the target user based on the target time series fusion feature set and the target mask feature comprises: using a self-attention mechanism to fuse the second target time series fusion features corresponding to each second target time series data based on the target mask feature to obtain target data source fusion features; outputting the target risk assessment result corresponding to the target user based on the first target time series fusion features and the target data source fusion features.

3. The method of claim 2, wherein the risk assessment model comprises a time series fusion layer, a mask mechanism layer, a transformer layer, and a residual connection layer; the time series fusion layer is configured to map and aggregate the first target time series data and each second target time series data in the time dimension to obtain the first target time series fusion features corresponding to the first target time series data and the second target time series fusion features corresponding to each second target time series data; the mask mechanism layer is configured to mask each second target time series data to obtain the target mask feature; the transformer layer is configured to calculate target attention of the second target time series fusion features corresponding to each second target time series data in the target time series fusion feature set to the first target time series fusion features using a self-attention mechanism, and fuse each second target time series fusion feature based on the target mask feature and the target attention corresponding to each second target time series fusion feature to obtain target data source fusion features; the residual connection layer is configured to output the target risk assessment result corresponding to the target user based on the target fusion features obtained by weighted fusion of the first target data source features and the target data source fusion features.

4. The method of claim 3, wherein the transformer layer is specifically configured to: map the first target time-series fusion feature and each of the second target time-series fusion features in a data source dimension to obtain a first target data source feature corresponding to the first target time-series fusion feature and a second target data source feature corresponding to each of the second target time-series fusion features, and calculate a target attention of each of the second target data source features to the first target data source feature using a self-attention mechanism, and perform feature fusion on each of the second target data source features based on the target mask feature and the target attention corresponding to each of the second target data source features to obtain a target data source fusion feature.

5. The method of claim 3, wherein the mask mechanism layer is specifically configured to: set a mask value corresponding to the second target time-series data as 0 in a case that the second target time-series data is empty, and set the mask value corresponding to the second target time-series data as 1 in a case that the second target time-series data is not empty; and / or set a mask value corresponding to the second target time-series data of a non-target external data source as 0 in a case that the external data source is the non-target external data source, and set the mask value corresponding to the second target time-series data of a target external data source as 1 in a case that the external data source is the target external data source; wherein the non-target external data source is used to represent an external data source that does not need to participate in the risk assessment of the target user, and the target external data source is used to represent an external data source that needs to participate in the risk assessment of the target user. a target constraint layer; the target constraint layer is configured to calculate a target loss result corresponding to the risk assessment model, and train the risk assessment model based on the target loss result; the target loss result is determined based on a first loss and / or a second loss, the first loss is determined based on an actual risk assessment result and a target risk assessment result output by the risk assessment model, and the second loss is determined based on the target risk assessment result output by the risk assessment model and an internal model risk assessment result output by an internal risk assessment model.

7. The method of claim 1, further comprising: obtaining an actual risk assessment result corresponding to the target user; calculating a target loss result corresponding to the risk assessment model based on the actual risk assessment result corresponding to the target user, an internal model risk assessment result, and a target risk assessment result; setting a first parameter of the risk assessment model as a fixed parameter, and updating a second parameter of the risk assessment model based on the target loss result; wherein the second parameter is closer to an output layer of the risk assessment model than the first parameter. wherein 8. The method of any one of claims 1-7, wherein a target time period corresponding to each of the first target time-series data and the second target time-series data is the same.

6. The method of claim 3, the risk assessment model further comprising:

9. A risk assessment device based on multi-data source fusion, the device comprising: a first obtaining module configured to obtain a target time-series data set corresponding to a target user; ​ ​ ​ ​ ​ ​ ​ ​ ​ The target time series data set includes first target time series data corresponding to an internal risk assessment model and second target time series data corresponding to each external data source; The risk assessment module is configured to input the target time series data set into a risk assessment model to obtain a target time series fusion feature set and a target mask feature corresponding to the target time series data set, and output a target risk assessment result corresponding to the target user based on the target time series fusion feature set and the target mask feature. The target mask feature includes a mask value corresponding to each external data source. The risk assessment model is trained based on a time series data set of known risk assessment results.

10. An electronic device comprising: A processor and a memory; The processor is connected to the memory; The memory is configured to store executable program codes; The processor runs a program corresponding to the executable program codes by reading the executable program codes stored in the memory, so as to execute the method according to any one of claims 1-8.

11. A computer storage medium storing a plurality of instructions, the instructions being adapted to be loaded and executed by a processor to perform the method steps according to any one of claims 1-8.

12. A computer program product containing instructions, which, when the computer program product is run on a computer or a processor, cause the computer or the processor to execute the method according to any one of claims 1-8.

Images