Business processing method, device, equipment and program product

Abstract

The invention discloses a business processing method and device, equipment and a program product, and relates to the technical field of artificial intelligence. The method comprises the following steps: acquiring service processing associated data, service processing party data, service processing constraint data and service feedback data of a to-be-processed service, and generating a current multi-dimensional feature vector; a weight calculation routing network is adopted to analyze the current multi-dimensional feature vector, and the current weight of each target service model is obtained; respectively analyzing the current multi-dimensional feature vector by adopting each target service model to obtain a current service detection result of each target service model on the to-be-processed service; and according to the current weight of each target service model, fusing each current service detection result to obtain a target fusion result, and according to the target fusion result, generating a current service processing instruction. According to the technical scheme of the embodiment of the invention, the service processing efficiency and the applicability of the service processing method are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, and particularly relates to a business processing method, device, equipment and program product. BACKGROUND

[0002] Automatic business can use computer programs to execute business orders according to preset rules. Automatic business can quickly process a large amount of data and make corresponding business decisions. With the development of the market and the progress of data analysis technology, automatic business has become an indispensable part of the market.

[0003] Although automatic business can improve market efficiency and reduce business costs to a certain extent, traditional automatic business still has the problem of limited data processing capability, which is difficult to process and analyze massive real-time data, has decision delay or low decision efficiency, and has the problem of poor generalization ability, poor application. SUMMARY

[0004] The present application provides a business processing method, device, equipment and program product, which improves the efficiency of business processing and the application of business processing method.

[0005] According to an aspect of the present application, a business processing method is provided, the method comprising:

[0006] Obtaining business processing associated data, business processing party data, business processing constraint data and business feedback data of a to-be-processed business, and generating a current multi-dimensional feature vector based on the business processing associated data, the business processing party data, the business processing constraint data and the business feedback data;

[0007] Analyzing the current multi-dimensional feature vector using a weight calculation routing network to obtain a current weight of each target business model;

[0008] Using each target business model to analyze the current multi-dimensional feature vector respectively to obtain a current business detection result of each target business model on the to-be-processed business;

[0009] Fusing each current business detection result according to the current weight of each target business model to obtain a target fusion result, and generating a current business processing instruction according to the target fusion result.

[0010] According to another aspect of the present application, a business processing device is provided, the device comprising:

[0011] The data acquisition module is configured to acquire service processing associated data, service processing party data, service processing constraint data and service feedback data of a to-be-processed service, and generate a current multi-dimensional feature vector based on the service processing associated data, the service processing party data, the service processing constraint data and the service feedback data.

[0012] The current weight calculation module is configured to analyze the current multi-dimensional feature vector by using a weight calculation routing network to obtain a current weight of each target service model.

[0013] The current service detection module is configured to analyze the current multi-dimensional feature vector by using each target service model to obtain a current service detection result of each target service model for the to-be-processed service.

[0014] The current service processing module is configured to fuse each current service detection result according to the current weight of each target service model to obtain a target fusion result, and generate a current service processing instruction according to the target fusion result.

[0015] According to another aspect of the present application, an electronic device is provided, which comprises:

[0016] at least one processor; and

[0017] a memory connected with the at least one processor in communication; wherein

[0018] the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to execute the service processing method according to any one of the embodiments of the present application.

[0019] According to another aspect of the present application, a computer readable storage medium is provided, which stores computer instructions for enabling a processor to implement the service processing method according to any one of the embodiments of the present application when executed by the processor.

[0020] According to another aspect of the present application, a computer program product is provided, which comprises a computer program for implementing the service processing method according to any one of the embodiments of the present application when executed by a processor.

[0021] The technical scheme of the embodiment of the application obtains service processing associated data, service processing party data, service processing constraint data and service feedback data of a to-be-processed service, generates a current multi-dimensional feature vector based on the service processing associated data, the service processing party data, the service processing constraint data and the service feedback data, adopts multi-source data, improves the accuracy of service processing, analyzes the current multi-dimensional feature vector by using a weight calculation routing network, obtains a current weight of each target service model, adopts each target service model to analyze the current multi-dimensional feature vector respectively, obtains a current service detection result of each target service model on the to-be-processed service, fuses each current service detection result according to the current weight of each target service model, obtains a target fusion result, and generates a current service processing instruction according to the target fusion result. The weight calculation routing network and each target service model are introduced, multi-dimensional detection and multi-dimensional fusion are performed on the service processing associated data, the service processing party data, the service processing constraint data and the service feedback data of the to-be-processed service, a service processing instruction of the to-be-processed service is obtained, the service processing associated data, the service processing party data, the service processing constraint data and the service feedback data of the to-be-processed service are comprehensively considered, and the efficiency of service processing and the applicability of the service processing method are improved.

[0022] It should be understood that the content described in this part is not intended to identify key or important features of the embodiments of the application, nor is it intended to limit the scope of the application. Other features of the application will become apparent from the following description. BRIEF DESCRIPTION OF DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0024] Figure 1 is a flowchart of a service processing method provided by the first embodiment of the application;

[0025] Figure 2 is a flowchart of a service processing method provided by the second embodiment of the application;

[0026] Figure 3 is a structural schematic diagram of a service processing device provided by the third embodiment of the application;

[0027] Figure 4 is a structural schematic diagram of an electronic device for implementing the service processing method of the embodiment of the application. DETAILED DESCRIPTION

[0028] In the following, the technical solutions in the embodiments of the present application will be described clearly and completely with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all the other embodiments obtained by a person of ordinary skill in the art without creative work should belong to the protection scope of the present application.

[0029] It should be noted that the terms "first", "second" and the like in the description and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in other than the order illustrated or described herein. In addition, the terms "comprise" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device that includes a list of steps or units does not necessarily limit to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to such processes, methods, products or devices.

[0030] Embodiment one

[0031] Figure 1 A flowchart of a service processing method provided by the embodiment one of the present application. The embodiment of the present application can be applicable to the case of processing resource transfer service. The method can be executed by a service processing device, which can be realized in the form of hardware and / or software, and can be configured in an electronic device carrying service processing function, such as a client or a server.

[0032] Referring to Figure 1 The service processing method shown in the figure comprises:

[0033] S101, obtaining service processing association data, service processing party data, service processing constraint data and service feedback data of a to-be-processed service, and generating a current multi-dimensional feature vector based on the service processing association data, the service processing party data, the service processing constraint data and the service feedback data.

[0034] The to-be-processed service can be a service to be executed by a computer program according to a preset rule. For example, the to-be-processed service can be a to-be-processed transaction. That is, the to-be-processed service can be a transaction order to be executed by a computer program according to a preset rule.

[0035] The business processing associated data can be processing data associated with the to-be-processed business. The business processing associated data can represent the to-be-processed business from the dimension of business execution. For example, the business processing associated data can include real-time resource transfer amount corresponding to the to-be-processed business, business execution record, and other business information associated with the to-be-processed business.

[0036] The business processing party can be a party that processes the to-be-processed business. The business processing party data can be used to represent the processing capability and processing standard of the business processing party for the to-be-processed business. For example, the business processing party data can include business processing data of the business processing party in a preset time period, business analysis data of the business processing party, and business evaluation standard of the business processing party. The business processing data of the business processing party in the first preset time period can be used to represent the business processing situation of the business processing party in the preset time period. The first preset time period can be a preset analysis time of the business processing party. For example, the first preset time period can include 1 year, 6 months, 3 months, and 1 month. The business analysis data of the business processing party can be used to represent the analysis result of the business processing party for the business. For example, the business analysis data of the business processing party can be an industry analysis report. The business evaluation standard of the business processing party can be used to represent the evaluation dimension referred to by the business processing party for evaluating the to-be-processed business. Optionally, the business evaluation standard of the business processing party can include at least one business evaluation index. For example, the business evaluation standard of the business processing party can be an evaluation index system file.

[0037] The business constraint data can be a constraint condition of the to-be-processed business during processing. Optionally, different to-be-processed businesses have corresponding industry standard files and regulatory files. Correspondingly, the business constraint data can include the industry standard file and the regulatory file corresponding to the to-be-processed business. The industry standard file corresponding to the to-be-processed business can be an industry standard file required to be referred to when processing the to-be-processed business. The industry standard file of the to-be-processed file can be formulated by an industry standard institution. The industry standard file corresponding to the to-be-processed business can represent the processing limitation condition of the to-be-processed business from the dimension of the industry standard institution and reflect the inclination degree of the industry standard institution to the to-be-processed business. The regulatory file of the to-be-processed file can be a file of a regulatory agency required to be referred to when processing the to-be-processed business. The regulatory file of the to-be-processed file can be formulated by a regulatory agency. The regulatory file corresponding to the to-be-processed business can represent the processing limitation condition of the to-be-processed business from the dimension of the regulatory agency and reflect the inclination degree of the regulatory agency to the to-be-processed business.

[0038] The business feedback data can be feedback data of a business user on the to-be-processed business. The business feedback data can be used to represent the use tendency of the business user on the to-be-processed business. Optionally, under the premise of obtaining the data acquisition authorization, the business feedback data can be sourced from a private data collection platform or a public data collection platform. For example, the private data collection platform of the business feedback data can be the device. For example, the business feedback data can include platform text stream and real-time news, etc.

[0039] The current multi-dimensional feature vector can be used to represent the comprehensive result of the business processing association data, the business processing party data, the business processing constraint data and the business feedback data of the to-be-processed business.

[0040] Specifically, under the premise of obtaining the authorization, the application programming interface (API) between the device and other platforms can be used to obtain the business processing association data, the business processing party data, the business processing constraint data and the business feedback data of the to-be-processed business. The business processing association data, the business processing party data, the business processing constraint data and the business feedback data of the to-be-processed business can be vectorized to generate the current multi-dimensional feature vector. Optionally, a distributed data warehouse can be established. The business processing association data, the business processing party data, the business processing constraint data and the business feedback data of the to-be-processed business can be classified and stored according to the data type. For example, the real-time business processing association data can use a columnar storage structure, and the corresponding timestamp accuracy can be up to the microsecond level. A full-text index database can be established for unstructured data. A relational database can be constructed for the data with an association relationship to establish the cross-checking relationship between the data.

[0041] Optionally, time synchronization processing of multi-source data can be performed on the business processing associated data of the to-be-processed business, so as to realize millisecond timestamp alignment of the business processing associated data of the to-be-processed business. Optionally, the business processing associated data of the to-be-processed business after timestamp alignment can be vectorized to obtain a business processing associated vector. Optionally, a data standardization converter can be used to unify the data dimensions of the data of each business processing party. Optionally, the data of each business processing party after data dimension unification can be vectorized to obtain a business processing direction vector. Optionally, a natural language processing technology can be used to perform entity recognition and event classification on the business processing constraint data to obtain an analysis result of the business processing constraint data. The analysis result of the business processing constraint data includes the business processing party, the business and the business constraint information corresponding to the business processing constraint data. Optionally, the analysis result of the business processing constraint data can be vectorized to obtain a business processing constraint vector. Optionally, a sentiment dictionary can be used to analyze the business feedback data to obtain information such as text data, a text publisher, a data propagation range and a data propagation speed. Optionally, the information such as the text data, the text publisher, the data propagation range and the data propagation speed can be vectorized to obtain a business feedback vector. Optionally, the business processing associated vector, the business processing direction vector, the business processing constraint vector and the business feedback vector can be integrated to obtain a current multi-dimensional feature vector.

[0042] S102, using a weight calculation routing network to analyze the current multi-dimensional feature vector to obtain a current weight of each target business model.

[0043] The weight calculation routing network can be used to determine the current weight of each target business model corresponding to the current multi-dimensional feature vector. Illustratively, the weight calculation routing network can include an input layer, a hidden layer and an output layer. The input layer of the weight calculation routing network can be used to receive the current multi-dimensional feature vector. The hidden layer of the weight calculation routing network can adopt a 3-layer MLP (Multilayer Perceptron) structure, the activation function can be a Gaussian error linear unit, and a attention mechanism module can be embedded. The output layer of the weight calculation routing network can apply a noise Top-k gating mechanism to generate the current weight of each target business model. The value of k can be dynamically adjusted according to the number of target business models. The current weight can be used to weight the output result of each target business model. The current weight can be used to represent the degree of association or the degree of influence between the target business model and the to-be-processed business. The target business model can be used to analyze the current multi-dimensional feature vector from different dimensions.

[0044] In an optional embodiment of the present application, each target business model includes a time series prediction model, a multivariate regression model, a text classification model and a heterogeneous graph learning model.

[0045] The time series prediction model can predict the multi-dimensional feature vector of the next time period adjacent to the current time period based on the current multi-dimensional feature vector of the current time period. Optionally, the time series prediction model can also predict the business processing association vector of the next time period adjacent to the current time period based on the business processing association vector in the current multi-dimensional feature vector of the current time period.

[0046] The multivariate regression model can be used to generate evaluation results of the current multi-dimensional feature vector in each corresponding data dimension. Optionally, the multivariate regression model can also be used to determine the evaluation results of the business processing party in each data dimension according to the business processing direction vector in the current multi-dimensional feature vector.

[0047] The text classification model can detect the degree of inclination to the to-be-processed business based on the current multi-dimensional feature vector. Optionally, the text classification model can also detect the degree of inclination of the industry standard formulating agency or the regulatory agency to the to-be-processed business (or the degree of inclination of the business user to the to-be-processed business) according to the business processing constraint vector (or the business feedback vector) in the current multi-dimensional feature vector.

[0048] The heterogeneous graph learning model can determine the corresponding association graph based on the current multi-dimensional feature vector. Optionally, the heterogeneous graph learning model can also determine the association graph corresponding to the business processing direction vector based on the business processing direction vector in the current multi-dimensional feature vector.

[0049] The scheme can further improve the business processing efficiency by selecting typical target business models and specificizing each target business model into a time series prediction model, a multivariate regression model, a text classification model, and a heterogeneous graph learning model.

[0050] Specifically, the current multi-dimensional feature vector can be input into the weight calculation routing network for analysis to obtain the current weights of each target business model.

[0051] S103, each target business model is used to analyze the current multi-dimensional feature vector respectively to obtain the current business detection result of each target business model to the to-be-processed business.

[0052] The current business detection result can be the detection result of the target business model to the current multi-dimensional feature vector of the to-be-processed business. The current business detection result can be used to detect the current multi-dimensional feature vector from a single dimension.

[0053] Specifically, the current multi-dimensional feature vector can be input into each target business model for analysis to obtain the current business detection result of each target business model to the to-be-processed business.

[0054] Optionally, the service processing association vector in the current multi-dimensional feature vector can be input into the time series prediction model to obtain a current service detection result of the to-be-processed service by the time series prediction model; the service processing direction vector in the current multi-dimensional feature vector can be input into the multivariate regression model to obtain a current service detection result of the to-be-processed service by the multivariate regression model; the service processing constraint vector (or the service feedback vector) in the current multi-dimensional feature vector can be input into the text classification model to obtain a current service detection result of the to-be-processed service by the text classification model; and the service processing direction vector in the current multi-dimensional feature vector can be input into the heterogeneous graph learning model to obtain a current service detection result of the to-be-processed service by the heterogeneous graph learning model.

[0055] In an optional embodiment of the present application, each target service model is used to analyze the current multi-dimensional feature vector to obtain a current service detection result of the to-be-processed service by each target service model, including: obtaining a service type of the to-be-processed service, and determining a target number according to the service type of the to-be-processed service; screening a target number of target service models in front of the order from each target service model according to the order of the current weight, updating each target service model; and using the updated each target service model to analyze the current multi-dimensional feature vector to obtain a current service detection result of the to-be-processed service by each target service model.

[0056] The service type can be used to represent the processing scene of the to-be-processed service. The processing requirements of to-be-processed services of different service types are different. The target number can be the number of target service detection models for detecting the to-be-processed service. For example, the service type can include a high-frequency service processing scene and other service processing scenes. Compared with the other service processing scenes, the high-frequency service processing scene has higher requirements for the efficiency of service processing. Correspondingly, the target number corresponding to the high-frequency service processing scene is lower than the target number corresponding to the other service processing scenes. For example, the target number corresponding to the high-frequency service processing scene can be 2; and the target number corresponding to the other service processing scenes can be 3.

[0057] Specifically, the service type of the to-be-processed service can be obtained. The target number corresponding to the service type can be determined based on the service type of the to-be-processed service. The current weight can be sorted. The target service model corresponding to the current weight in front of the order of the target number can be screened from each target service model to update each target service model for processing the current multi-dimensional feature vector of the to-be-processed service. The updated each target service model can be used to analyze the current multi-dimensional feature vector to obtain a current service detection result of the to-be-processed service by each target service model.

[0058] The scheme introduces the service type of the to-be-processed service, filters the target service models, selects target service models with higher influence degree on the to-be-processed service, detects the current multi-dimensional feature vector of the to-be-processed service based on the filtered target service models, and improves the detection efficiency and accuracy of the to-be-processed service while taking into account the correlation between the to-be-processed service and the target service models.

[0059] In S104, the current service detection results are fused according to the current weights of the target service models to obtain a target fusion result, and a current service processing instruction is generated according to the target fusion result.

[0060] The target fusion result can be a fusion result of the current service detection results. The current service detection result can represent the detection of the to-be-processed service from a single dimension. The target fusion result can represent the detection of the to-be-processed service from a comprehensive dimension. The target fusion result obtained by fusing the current service detection results with the current weights not only realizes the fusion of the current service detection results, but also takes into account the influence degree of each target service model on the to-be-processed service, thereby improving the accuracy of the target fusion result. The current service processing instruction can be used to indicate how to process the to-be-processed service. There is a corresponding relationship between the service processing instruction and the fusion result. Based on the fusion result, the corresponding service processing instruction can be determined. For example, the current service processing instruction can include executing the to-be-processed service and prohibiting the execution of the to-be-processed service.

[0061] Specifically, the target fusion result can be obtained by weighting and summing the current service detection results based on the current weights of the target service models. The corresponding current service processing instruction can be determined according to the target fusion result.

[0062] Optionally, the current service detection results can be standardized and calibrated first to eliminate the dimensional differences between the current service detection results, and then the current service detection results can be fused according to the current weights of the target service models to obtain the target fusion result.

[0063] The technical scheme of the embodiment of the present application obtains the service processing associated data, the service processing party data, the service processing constraint data and the service feedback data of the to-be-processed service, generates the current multi-dimensional feature vector based on the service processing associated data, the service processing party data, the service processing constraint data and the service feedback data, adopts multi-source data, improves the accuracy of service processing, analyzes the current multi-dimensional feature vector by adopting the weight calculation routing network, obtains the current weight of each target service model, adopts each target service model to analyze the current multi-dimensional feature vector respectively, obtains the current service detection result of each target service model to the to-be-processed service, fuses each current service detection result according to the current weight of each target service model, obtains the target fusion result, and generates the current service processing instruction according to the target fusion result. The weight calculation routing network and each target service model are introduced, the service processing associated data, the service processing party data, the service processing constraint data and the service feedback data of the to-be-processed service are subjected to multi-dimensional detection and multi-dimensional fusion, the service processing instruction of the to-be-processed service is obtained, the service processing associated data, the service processing party data, the service processing constraint data and the service feedback data of the to-be-processed service are comprehensively considered, and the efficiency of service processing and the applicability of the service processing method are improved.

[0064] Embodiment two

[0065] Figure 2A flowchart of a service processing method provided for the second embodiment of the present application. The second embodiment of the present application further increases, before the step of "obtaining service processing associated data, service processing party data, service processing constraint data and service feedback data of the to-be-processed service", the steps of "obtaining actual service processing instructions of the historical service and service processing associated data, service processing party data, service processing constraint data and service feedback data of the historical service, and generating a historical multi-dimensional feature vector based on the service processing associated data, service processing party data, service processing constraint data and service feedback data of the historical service"; using an initial routing network to analyze the historical multi-dimensional feature vector to obtain historical weights of each initial service model; using each initial service model to analyze the historical multi-dimensional feature vector respectively to obtain historical service detection results of the historical service of each initial service model; fusing each historical service detection result according to the historical weights of each initial service model to obtain a historical fusion result, and generating a historical service processing instruction according to the historical fusion result; calculating an overall loss of the initial routing network and each initial service model according to the historical service processing instruction and the actual service processing instruction; and adjusting the initial routing network according to the overall loss, returning to execute the step of using the initial routing network to analyze the historical multi-dimensional feature vector to obtain the historical weights of each initial service model until the overall loss of the initial routing network and each initial service model converges to obtain a weight calculation routing network. The training process of the weight calculation routing network and each target service model is introduced, the weight calculation routing network is trained first, and then each target service model is trained, which further improves the processing efficiency and accuracy of the to-be-processed service. It should be noted that the parts not described in detail in the second embodiment of the present application can be referred to the descriptions of other embodiments.

[0066] Referring to Figure 2 The service processing method shown in the figure comprises:

[0067] S201, obtaining actual service processing instructions of the historical service and service processing associated data, service processing party data, service processing constraint data and service feedback data of the historical service, and generating a historical multi-dimensional feature vector based on the service processing associated data, service processing party data, service processing constraint data and service feedback data of the historical service.

[0068] The historical business can be a business that has been executed according to preset rules by using a computer program. For example, the historical business can be an automatic transaction, i.e., a transaction order that has been executed according to preset rules by using a computer program. The actual business processing instruction can be an actual business processing instruction of the historical business. The historical multi-dimensional feature vector can be used to represent a comprehensive result of the business processing association data, the business processing party data, the business processing constraint data, and the business feedback data of the historical business.

[0069] Specifically, under the premise of authorization, an application programming interface (API) between the device and other platforms can be used to obtain the actual business processing instruction of the historical business and the business processing association data, the business processing party data, the business processing constraint data, and the business feedback data of the historical business. The business processing association data, the business processing party data, the business processing constraint data, and the business feedback data of the historical business can be vectorized to generate the historical multi-dimensional feature vector.

[0070] In an optional embodiment of the present application, before obtaining the actual business processing instruction of the historical business and the business processing association data, the business processing party data, the business processing constraint data, and the business feedback data of the historical business, the method further includes applying a regularization constraint to each target business model.

[0071] The regularization constraint can be used to prevent overfitting of a single target business model. For example, the regularization constraint can be an L2 regularization constraint.

[0072] The present solution can prevent overfitting of each target business model by applying a regularization constraint to each target business model.

[0073] S202, using an initial routing network, analyzing the historical multi-dimensional feature vector to obtain historical weights of each initial business model.

[0074] The initial routing network can be an untrained routing network. For example, the initial routing network can include an input layer, a hidden layer, and an output layer. The initial business model can be an untrained target business model. The initial business model can be used to analyze the historical multi-dimensional feature vector from different dimensions. The historical weights can be used to weight the output results of each initial business model. The historical weights can be used to represent the degree of association or the degree of influence between the initial business model and the to-be-processed business.

[0075] Specifically, the historical multi-dimensional feature vector can be input into the initial routing network to analyze the historical multi-dimensional feature vector to obtain the historical weights of each initial business model.

[0076] S203, analyze the historical multi-dimensional feature vectors respectively by using the initial service models to obtain historical service detection results of the historical service by the initial service models.

[0077] The historical service detection result can be a detection result of the historical multi-dimensional feature vector of the historical service by the initial service model. The historical service detection result can be used to detect the historical multi-dimensional feature vector from a single dimension.

[0078] Specifically, the historical multi-dimensional feature vector can be input into the initial service models, and the historical multi-dimensional feature vector can be analyzed respectively by the initial service models to obtain the historical service detection results of the historical service by the initial service models.

[0079] S204, fuse the historical service detection results according to the historical weights of the initial service models to obtain a historical fusion result, and generate a historical service processing instruction according to the historical fusion result.

[0080] The historical fusion result can be a fusion result of the historical service detection results. The historical service detection result can represent the detection of the historical service from a single dimension. The historical fusion result can represent the detection of the historical service from a comprehensive dimension. The historical service processing instruction can be a service processing instruction of the historical service predicted based on the historical fusion result. For example, the historical service processing instruction can include executing the historical service and prohibiting the execution of the historical service.

[0081] Specifically, the historical fusion result can be obtained by weighted summation of the historical service detection results based on the historical weights of the initial service models. The corresponding historical service processing instruction can be determined according to the historical fusion result.

[0082] S205, calculate the overall loss of the initial routing network and the initial service models according to the historical service processing instruction and the actual service processing instruction.

[0083] The historical service processing instruction can be a service processing instruction of the historical service predicted based on the historical fusion result. The actual service processing instruction can be an actual service processing instruction of the historical service. For example, the actual service processing instruction can include executing the historical service and prohibiting the execution of the historical service.

[0084] The overall loss can be used to represent the deviation degree between the historical service processing instruction predicted based on the initial routing network and the initial service models and the actual service processing instruction.

[0085] Specifically, the overall loss of the initial routing network and the initial service models can be determined according to the difference between the historical service processing instruction and the actual service processing instruction.

[0086] S206, according to the overall loss, the initial routing network is adjusted, and the step of returning to execute the initial routing network is adopted to analyze the historical multi-dimensional feature vector, and the historical weight of each initial service model is obtained, until the overall loss of the initial routing network and each initial service model converges, and the weight calculation routing network is obtained.

[0087] S207, the initial service model is adjusted, and the step of returning to execute each initial service model is adopted to analyze the historical multi-dimensional feature vector, and the historical service detection result of each initial service model to the historical service is obtained, until the overall loss of the weight calculation routing network and each initial service model converges, and each target service model is obtained.

[0088] S208, the service processing associated data, the service processing party data, the service processing constraint data and the service feedback data of the to-be-processed service are obtained, and the current multi-dimensional feature vector is generated based on the service processing associated data, the service processing party data, the service processing constraint data and the service feedback data.

[0089] S209, the weight calculation routing network is adopted to analyze the current multi-dimensional feature vector, and the current weight of each target service model is obtained.

[0090] S210, each target service model is adopted to analyze the current multi-dimensional feature vector, and the current service detection result of each target service model to the to-be-processed service is obtained.

[0091] S211, according to the current weight of each target service model, the current service detection result is fused to obtain a target fusion result, and the current service processing instruction is generated according to the target fusion result.

[0092] The technical scheme of the embodiment of the application, by calculating the overall loss of the initial routing network and each initial service model according to the historical service processing instruction and the actual service processing instruction before processing the service, adjusting the initial routing network according to the overall loss, returning to execute the initial routing network, analyzing the historical multi-dimensional feature vector, obtaining the historical weight of each initial service model, until the overall loss of the initial routing network and each initial service model converges, obtaining the weight calculation routing network, adjusting each initial service model, returning to execute each initial service model, analyzing the historical multi-dimensional feature vector, obtaining the historical service detection result of each initial service model to the historical service, until the overall loss of the weight calculation routing network and each initial service model converges, obtaining each target service model, introducing the training process of the weight calculation routing network and each target service model, and further improving the processing efficiency and accuracy of the to-be-processed service by training the weight calculation routing network first and then training each target service model.

[0093] In an optional embodiment of the present application, after obtaining each target business model, the method further comprises: performing static verification on the weight calculation routing network and each target business model to obtain a static verification result; performing dynamic verification on the weight calculation routing network and each target business model to obtain a dynamic verification result; and determining a target optimization mode of the weight calculation routing network and each target business model according to the static verification result and / or the dynamic verification result.

[0094] The static verification result and the dynamic verification result can be used to verify the verification accuracy of the weight calculation routing network and each target business model.

[0095] Specifically, a K-fold cross-validation (preferably, K≥5) and an independent test set verification (preferably, a hold-out ratio≥30%) double-track parallel mechanism can be used to perform static verification on the weight calculation routing network and each target business model in a preset time period, to obtain an average prediction accuracy of the weight calculation routing network and each target business model in the preset time period, i.e., to obtain the static verification result. A time rolling window verification mechanism can be established, a configurable time window parameter (such as a window length and a rolling step) can be set, and a forward chain test can be used to perform dynamic verification on the weight calculation routing network and each target business model, to obtain the dynamic verification result. When the static verification results of the weight calculation routing network and each target business model in three continuous preset time periods exceed a tolerance range (such as ε=5%), a hyperparameter evolutionary algorithm (HEA) can be used to perform nonlinear search optimization on the weight calculation routing network and each target business model. When the dynamic verification result of the weight calculation routing network and each target business model exceeds a preset threshold (such as δ=15%), a model structure reconstruction process can be triggered, and weights can be calculated by the weight calculation routing network to reassign the target business model.

[0096] The present scheme further improves the accuracy of the weight calculation routing network and each target business model by increasing dynamic verification and static verification of the weight calculation routing network and each target business model, determining an optimization mode of the weight calculation routing network and each target business model based on the static verification result and the dynamic verification result, and thereby improving the accuracy of business processing.

[0097] In an optional embodiment of the present application, after obtaining each target business model, the method further comprises: detecting a stability index, a risk control index, and a generalization difference index of the weight calculation routing network and each target business model; and adjusting parameters of the weight calculation routing network and each target business model according to the stability index, the risk control index, and / or the generalization difference index.

[0098] The stability index can be used to characterize the stability of the weight calculation routing network and each target business model. The risk control index can be used to characterize the risk control capability of the weight calculation routing network and each target business model. The generalization difference index characterizes the generalization of the weight calculation routing network and each target business model.

[0099] Specifically, the stability index, the risk control index, and the generalization difference index of the weight calculation routing network and each target business model are detected. When the stability index is less than or equal to a preset stability threshold, when the risk control index is less than or equal to a preset risk control threshold, and / or when the generalization difference index is less than or equal to a preset generalization difference threshold, the weight calculation routing network and each target business model can be parameter adjusted.

[0100] The scheme further improves the accuracy of the weight calculation routing network and each target business model by increasing the key index verification and parameter adjustment of the weight calculation routing network and each target business model, thereby improving the accuracy of business processing.

[0101] On the basis of the above-mentioned embodiments, the application further provides a preferred embodiment. The business processing method comprises:

[0102] 1. Data collection and processing.

[0103] Step one, multi-source data acquisition and classified storage.

[0104] Specifically, cross-platform historical transaction data acquisition can be performed through a standardized API interface. The historical transaction data includes historical transaction business processing associated data, business processing party data, business processing constraint data, and business feedback data. The business processing associated data includes real-time prices, transaction-by-transaction records, and order book depth information. The business processing party data includes business processing party regular financial reports, industry analysis reports, and valuation index systems. The business processing constraint data includes policy documents and regulatory agency announcement texts. The business feedback data includes social media platform text streams and news media real-time news.

[0105] Optionally, a distributed data warehouse can be established for classified storage according to data types. For example, the business processing associated data can be stored in a columnar storage structure, and the timestamp accuracy of the data can reach the microsecond level. A full-text index database can be established for unstructured text data. A relational database can be constructed for financial data to establish inter-index reconciliation relationships.

[0106] Step two, data preprocessing.

[0107] Specifically, an order book dynamic reconstruction algorithm can be used to perform time synchronization processing of multi-source data on the business processing associated data of the to-be-processed business, so as to realize millisecond-level timestamp alignment of the business processing associated data of the to-be-processed business. A data standardization converter can be used to unify the data dimensions of the data of each business processing party. A natural language processing technology can be used to perform entity recognition and event classification on the business processing constraint data, to obtain an analysis result of the business processing constraint data. The analysis result of the business processing constraint data includes the business processing party, the business, and the business constraint information corresponding to the business processing constraint data. A sentiment dictionary can be used to analyze the business feedback data, to obtain information such as text data, a text publisher, a data propagation range, and a data propagation speed.

[0108] Step four, multi-dimensional feature engineering construction.

[0109] Specifically, the business processing associated vector, the business processing direction vector, the business processing constraint vector, and the business feedback vector can be integrated to obtain a current multi-dimensional feature vector.

[0110] Step five, feature optimization screening.

[0111] Specifically, the influence degree of a feature on a to-be-processed transaction can be verified in different ways. If it is detected that the influence degree of the feature on the to-be-processed transaction is less than or equal to a preset influence degree threshold, the corresponding feature is deleted.

[0112] Illustratively, feature effectiveness verification can be performed. For example, mutual information method can be used to evaluate the nonlinear correlation between the feature and the yield rate; causal test can be applied to confirm the predictive lead of the feature.

[0113] Illustratively, feature redundancy elimination can be implemented. For example, a variance inflation factor detection module can be constructed to eliminate multicollinearity features; for another example, feature clustering analysis can be deployed to merge high-similarity feature clusters.

[0114] Illustratively, a dynamic feature optimization mechanism can be established. For example, a contribution value evaluation system based on the quarterly gradient boosting method can be developed, a feature stability monitoring module can be constructed to track the effective period of the feature, and a feature combination optimizer can be designed to mine the interactive feature enhancement effect.

[0115] 2. Constructing each target business model.

[0116] Step one, constructing an intelligent transaction analysis system.

[0117] The system architecture configuration can include a centralized routing scheduling module, a plurality of expert sub-model (i.e., target business model) libraries, and a strategy fusion module, and asynchronous communication between the modules is realized through a distributed message queue.

[0118] The expert sub-models can include a time series prediction model, a multivariate regression model, a text classification model, and a heterogeneous graph learning model.

[0119] The time series prediction model can perform trend prediction on the business processing association vector. That is, the next time period business processing association vector can be predicted based on the current time period business processing association vector. The multivariate regression model can be used to determine the evaluation results of the business processing party in each data dimension according to the business processing direction amount in the current multi-dimensional feature vector. The text classification model can be used to detect the tendency of the industry standard formulating organization or the regulatory organization to the to-be-processed business (or the tendency of the business user to the to-be-processed business) according to the business processing constraint vector (or the business feedback vector) in the current multi-dimensional feature vector. The heterogeneous graph learning model can determine the association graph corresponding to the business processing direction amount based on the business processing direction amount in the current multi-dimensional feature vector.

[0120] Step two, weight calculation routing network construction and training.

[0121] The structure of the weight calculation routing network can include an input layer, a hidden layer, and an output layer. The input layer of the weight calculation routing network can be used to receive the current multi-dimensional feature vector. The hidden layer of the weight calculation routing network can adopt a 3-layer MLP (Multilayer Perceptron) structure, the activation function can be a Gaussian error linear unit, and an attention mechanism module can be embedded. The output layer of the weight calculation routing network can apply a noise Top-k gating mechanism to generate the current weights of each target business model. The value of k can be dynamically adjusted according to the number of target business models.

[0122] Specifically, the joint training mechanism can be used to train the weight calculation routing network and each target business model. For example, the parameters of the target business model can be first frozen to minimize the expert allocation cross-entropy loss, and the weight calculation routing network can be trained alone. The parameters of the target business model can be unfrozen, and the multi-task loss function of the weight calculation routing network and the target business model can be jointly optimized. L2 constraint can be applied to the balance of each target business model to prevent specific experts from being overloaded.

[0123] The specific steps of the model training in which the parameters of the target business model are unfrozen and the multi-task loss function of the weight calculation routing network and the target business model is jointly optimized are as follows:

[0124] The target business model and the data input and output of the weighted routing network. The input of each target business model is the historical multidimensional feature vector X after feature processing. The historical multidimensional feature vector may include business processing association vectors (such as price trend features), business processing direction quantities (such as financial indicator features), business feedback vectors (such as sentiment scores and news trigger tags), etc. The output of the target business model is the historical detection result of the target business model on the input historical multidimensional feature vector X. For example, historical detection results may include the probability of future price increases or decreases, event impact scores, and asset scores.

[0125] The input of the weight calculation routing network is also the historical multidimensional feature vector X, and its output is the historical weight vector [w1,w2,...,w n ], the historical weight vector represents the contribution or call probability of each target business model under the input of the historical multidimensional feature vector X. The historical weight vector is used to guide the combination or activation of each target business model.

[0126] This solution can use an end-to-end training approach to optimize the weight calculation routing network and each target service model. The training process includes the following steps:

[0127] (1) Sample construction: The training sample set can be constructed using labeled historical transaction data. The input data is the historical multi-dimensional feature vector X, and the output is the actual fusion result y (including the real labels of future price increase or decrease probability, event impact score, and asset score).

[0128] (2) Forward propagation: Each historical multidimensional feature vector X generates a historical weight w through the weight calculation routing network, and is passed to all or part of the target business model to obtain the historical detection results output by each target business model.

[0129] (3) Output fusion: The historical detection results output by the target business model are analyzed based on the historical weights w. Perform weighted combination to obtain historical fusion results

[0130] (4) Loss calculation: based on historical fusion results Calculate the overall loss function with the actual fusion result y The overall loss is applied simultaneously to each target service model and the weight calculation routing network.

[0131] (5) Back propagation and parameter update: The back propagation algorithm is used to perform gradient updates on the parameters of each target business model and the parameters of the weight calculation routing network to optimize the model structure.

[0132] Step three, model performance verification and generalization ability guarantee.

[0133] Model performance verification is performed in the following ways:

[0134] Optionally, a hierarchical verification system can be used, including:

[0135] (a) Static verification module: Through K-fold cross-validation (K≥5) and independent test set verification (Hold-out Ratio≥30%), calculate the average prediction accuracy of the model in historical data (MAPE≤8%), and obtain the static verification result. If the static verification result (Δ_metric) of the static verification module exceeds the tolerance range (ε=5%) for three consecutive periods, start the hyperparameter evolution algorithm for nonlinear search optimization.

[0136] (b) Dynamic verification module: Establish a time rolling window verification mechanism, set configurable time window parameters (window length L∈[60,250] trading days, rolling step S∈[5,20] trading days), perform forward chain test, and obtain dynamic verification result. When the static verification result (R_decay) of the dynamic verification module exceeds the preset threshold (δ=15%), trigger the model structure reconstruction process, and optimize the combination of each target business model through weight calculation routing network weight redistribution.

[0137] Optionally, three types of core indicators can be monitored simultaneously.

[0138] (1) Yield stability indicator: Calculate the annual yield volatility (σ≤15%) and Sharpe ratio (Sharpe Ratio≥1.2).

[0139] (2) Risk control indicator: Evaluate the maximum drawdown rate (Max Drawdown≤20%) and drawdown recovery period (T_recovery≤60 trading days).

[0140] (3) Generalization difference indicator: Quantify the KL divergence (D_KL≤0.15) and feature distribution offset (Δ_dist≤10%) between the training set and the validation set.

[0141] When the stability indicator is less than or equal to the preset stability threshold, when the risk control indicator is less than or equal to the preset risk control threshold, and / or when the generalization difference indicator is less than or equal to the preset generalization difference threshold, the weight calculation routing network and each target business model can be adjusted.

[0142] 3. Dynamic target business model allocation and transaction execution.

[0143] Deploy the trained target business models and weight calculation routing network to the real trading system to build a prediction-execution-feedback closed-loop architecture. Through the API interface, real-time business processing associated data, business processing party data, business processing constraint data and business feedback data of the to-be-processed transaction are received and input synchronously to the target business models and weight calculation routing network for transaction instruction prediction to generate the current transaction instruction.

[0144] Step one, feature preprocessing.

[0145] Specifically, the business processing associated data, business processing party data, business processing constraint data and business feedback data of the to-be-processed transaction can be received in real time to generate a 32-dimensional current multi-dimensional feature vector.

[0146] Step two, routing decision process.

[0147] Specifically, the current multi-dimensional feature vector can be input into the weight calculation routing network to obtain the current weight of each target business model through the gating mechanism.

[0148] Optionally, if the business type of the to-be-processed transaction is a high-frequency transaction scenario, the Top-2 expert is enabled (i.e., the target number of target business models is 2); if the business type of the to-be-processed transaction is an event-driven scenario (i.e., other transaction scenarios), the Top-3 expert is enabled (i.e., the target number of target business models is 3).

[0149] Optionally, the inference interface of the activated target business model is called in parallel through a thread pool.

[0150] Step three, result fusion.

[0151] Specifically, the current detection results output by each target business model can be standardized and calibrated to eliminate dimensional differences. Each current detection result can be weighted and fused according to the current weight, and the specific formula is:

[0152] Final Output =Σ(Softmax(w i )*Expert i (x));

[0153] Wherein, Final Output is the target fusion result; w i is the current weight corresponding to the target business model; Expert i (x) is the current detection result of the target business model.

[0154] Step four, current transaction instruction generation and execution.

[0155] Specifically, according to the target fusion result, the corresponding current transaction signal (including can trade and prohibit trade) is generated.

[0156] 5. Transaction execution.

[0157] Step one, execute optimization.

[0158] Obtain the order book data associated with the transaction, and dynamically adjust the order placement rhythm and order splitting strategy based on the order book data.

[0159] Step two, audit tracking.

[0160] Specifically, the complete decision link and the current weight of each target business model can be recorded.

[0161] Optionally, a visual backtracking interface can be provided to support regulatory compliance review.

[0162] The scheme fuses multiple target business models with professional capabilities, models and predicts market trends, fundamental information, sentiment signals and event-driven factors in multiple dimensions, and each target business model is activated on demand under the dispatch of the weight calculation routing network, realizing accurate response to different trading scenarios; In real-time operation, market data and transaction feedback are continuously monitored, and factors such as price deviation, strategy yield fluctuation and expert performance change are combined to dynamically adjust the routing mechanism and expert weight, so that after inputting a large amount of real-time financial data to the model, an accurate trading instruction can be given, realizing real-time iteration and optimization of the trading instruction; In addition, through this mechanism, the weight calculation routing network and each target business model can not only switch the strategy focus in time according to the market structure change (such as from trend trading to event-driven trading), but also trigger strategy fine-tuning or replacement when the model output deviates from the expectation, which can significantly improve the adaptability and accuracy of the trading instruction and reduce the consumption of manpower; Through real-time analysis of market dynamics and model output deviation, the transaction risk is identified and responded in advance; Before executing the trading instruction, the system combines high-frequency market fluctuations, major news events, and emotional mutation signals to make comprehensive judgments and identify potential market risk areas in advance; In the process of transaction execution, the actual yield trend and the deviation from the expectation are dynamically monitored, and stop loss, reduce position or strategy switching operations are triggered to avoid expanding losses in extreme market conditions; At the same time, the system introduces a dynamic drawdown control mechanism and an expert model failure detection mechanism to ensure that risk exposure can be actively reduced in various abnormal situations, thereby significantly improving the risk avoidance ability of the trading strategy and enhancing the fund stability and security of the overall account.

[0163] Embodiment three

[0164] Figure 3A structural schematic diagram of a service processing device provided for embodiment three of the present application. The embodiment of the present application can be applied to the case of processing resource transfer services. The device can execute a service processing method. The device can be realized in the form of hardware and / or software. The device can be configured in an electronic device that bears service processing functions, such as a client or a server.

[0165] Referring to Figure 3 The service processing device shown includes a current data acquisition module 301, a current weight calculation module 302, a current service detection module 303, and a current service processing module 304. The data acquisition module 301 is configured to acquire service processing associated data, service processing party data, service processing constraint data, and service feedback data of a to-be-processed service, and generate a current multi-dimensional feature vector based on the service processing associated data, the service processing party data, the service processing constraint data, and the service feedback data. The current weight calculation module 302 is configured to analyze the current multi-dimensional feature vector by using a weight calculation routing network to obtain a current weight of each target service model. The current service detection module 303 is configured to analyze the current multi-dimensional feature vector by using each target service model to obtain a current service detection result of the to-be-processed service by each target service model. The current service processing module 304 is configured to fuse each current service detection result according to the current weight of each target service model to obtain a target fusion result, and generate a current service processing instruction according to the target fusion result.

[0166] The technical scheme of the embodiment of the present application acquires service processing associated data, service processing party data, service processing constraint data, and service feedback data of a to-be-processed service, and generates a current multi-dimensional feature vector based on the service processing associated data, the service processing party data, the service processing constraint data, and the service feedback data. The multi-source data is used to improve the accuracy of service processing. The current multi-dimensional feature vector is analyzed by using a weight calculation routing network to obtain a current weight of each target service model. Each target service model is used to analyze the current multi-dimensional feature vector to obtain a current service detection result of the to-be-processed service by each target service model. Each current service detection result is fused according to the current weight of each target service model to obtain a target fusion result. A current service processing instruction is generated according to the target fusion result. The weight calculation routing network and each target service model are introduced. The service processing associated data, the service processing party data, the service processing constraint data, and the service feedback data of the to-be-processed service are subjected to multi-dimensional detection and multi-dimensional fusion to obtain the service processing instruction of the to-be-processed service. The service processing associated data, the service processing party data, the service processing constraint data, and the service feedback data of the to-be-processed service are comprehensively considered to improve the efficiency of service processing and the applicability of the service processing method.

[0167] In an optional embodiment of the present application, the current service detection module 303 comprises: a target quantity calculation unit configured to obtain the service type of the to-be-processed service, and determine a target quantity according to the service type of the to-be-processed service; a target service model updating unit configured to update each of the target service models according to the ranking of the current weights, and filter the target quantity of target service models ranked in the front; and a target service detection unit configured to analyze the current multi-dimensional feature vector by using the updated target service models respectively, and obtain the current service detection result of the to-be-processed service by each of the target service models.

[0168] In an optional embodiment of the present application, the device further comprises: a historical data acquisition module configured to acquire the actual fusion result of historical services and the service processing associated data, the service processing party data, the service processing constraint data and the service feedback data of the historical services before the service processing associated data, the service processing party data, the service processing constraint data and the service feedback data of the to-be-processed service are acquired, and generate a historical multi-dimensional feature vector based on the service processing associated data, the service processing party data, the service processing constraint data and the service feedback data of the historical services; a historical weight calculation module configured to analyze the historical multi-dimensional feature vector by using the initial routing network, and obtain the historical weight of each initial service model; a historical service detection module configured to analyze the historical multi-dimensional feature vector by using each of the initial service models respectively, and obtain the historical service detection result of the historical services by each of the initial service models; a historical fusion result generation module configured to fuse each of the historical service detection results according to the historical weight of each of the initial service models, and obtain a historical fusion result; a whole loss calculation module configured to calculate the whole loss of the initial routing network and each of the initial service models according to the historical fusion result and the actual fusion result; a weight calculation routing network parameter adjustment module configured to adjust the parameters of the initial routing network according to the whole loss, and return to execute the step of analyzing the historical multi-dimensional feature vector by using the initial routing network to obtain the historical weight of each initial service model until the whole loss of the initial routing network and each of the initial service models converges, and obtain a weight calculation routing network; and a target service model parameter adjustment module configured to adjust the parameters of each of the initial service models, and return to execute the step of analyzing the historical multi-dimensional feature vector by using each of the initial service models respectively to obtain the historical service detection result of the historical services by each of the initial service models until the whole loss of the weight calculation routing network and each of the initial service models converges, and obtain each target service model.

[0169] In an optional embodiment of the present application, the device further comprises a regularization constraint module configured to apply a regularization constraint to each target business model before the actual fusion result of the historical business and the business processing association data, the business processing party data, the business processing constraint data and the business feedback data of the historical business are obtained.

[0170] In an optional embodiment of the present application, the device further comprises a static verification module configured to perform static verification on the weight calculation routing network and each target business model after the target business models are obtained, to obtain a static verification result; a dynamic verification module configured to perform dynamic verification on the weight calculation routing network and each target business model, to obtain a dynamic verification result; and a first optimization module configured to determine a target optimization mode of the weight calculation routing network and each target business model according to the static verification result and / or the dynamic verification result.

[0171] In an optional embodiment of the present application, the device further comprises a key indicator detection module configured to detect a stability indicator, a risk control indicator and a generalization difference indicator of the weight calculation routing network and each target business model after the target business models are obtained; and a second optimization module configured to perform parameter adjustment on the weight calculation routing network and each target business model according to the stability indicator, the risk control indicator and / or the generalization difference indicator.

[0172] In an optional embodiment of the present application, each target business model comprises a time series prediction model, a multivariate regression model, a text classification model and a heterogeneous graph learning model.

[0173] The business processing device provided by the embodiments of the present application can execute the business processing method provided by any of the embodiments of the present application, and has the corresponding function modules and beneficial effects of the execution method.

[0174] In the technical solution of the embodiments of the present application, the collected information is information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the related data comply with relevant laws, regulations and standards of relevant countries and regions, necessary security measures are taken, public order and good customs are not violated, and corresponding operation portals are provided for the user to select authorization or refusal.

[0175] Embodiment Four

[0176] According to the embodiments of the present application, the present application further provides an electronic device, a readable storage medium and a computer program product.

[0177] Figure 4A structural diagram of an electronic device 400 that can be used to implement embodiments of the present application is shown. The electronic device is intended to represent various forms of digital computers, such as laptops, desktops, tablets, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular telephones, smartphones, wearable devices (e.g., headsets, glasses, watches, etc.), and other similar computing devices. The components shown here, their connections and relationships, and their functions, are meant to be examples only, and are not meant to limit implementations of the present application described and / or claimed in this document.

[0178] As shown in Figure 4 The electronic device 400 includes at least one processor 401, and memory, such as read-only memory (ROM) 402, random access memory (RAM) 403, etc., communicatively connected to the at least one processor 401, where the memory stores computer programs executable by the at least one processor. The processor 401 can perform various appropriate actions and processes according to the computer programs stored in the read-only memory (ROM) 402 or loaded into the random access memory (RAM) 403 from the storage unit 408. In the RAM 403, various programs and data required for the operation of the electronic device 400 can also be stored. The processor 401, the ROM 402, and the RAM 403 are connected to each other through a bus 404. An input / output (I / O) interface 405 is also connected to the bus 404.

[0179] Various components in the electronic device 400 are connected to the I / O interface 405, including an input unit 406, such as a keyboard, a mouse, etc., an output unit 407, such as various types of displays, speakers, etc., a storage unit 408, such as a magnetic disk, an optical disk, etc., and a communication unit 409, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 409 allows the electronic device 400 to exchange information / data with other devices through a computer network, such as the Internet, and / or various telecommunication networks.

[0180] The processor 401 can be various general and / or special purpose processing components with processing and computing capabilities. Some examples of the processor 401 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 401 performs various methods and processes described above, such as the business processing method.

[0181] In some embodiments, the business processing method can be implemented as a computer program tangibly embodied in a computer readable storage medium, e.g., storage unit 408. In some embodiments, parts or all of the computer program can be loaded and / or installed onto electronic device 400 via, e.g., ROM 402 and / or communication unit 409. When the computer program is loaded onto RAM 403 and executed by processor 401, one or more steps of the business processing method described above can be performed. Alternatively, in other embodiments, processor 401 can be configured to perform the business processing method by other means, e.g., with the aid of firmware.

[0182] Various implementations of the systems and techniques described above can be realized in digital electronic circuitry, integrated circuitry, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a system on a chip (SOC), a complex programmable logic device (CPLD), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include implementation in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which can be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device.

[0183] Computer programs used to implement the methods of the application can be written in any combination of one or more programming languages. These computer programs can be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the computer program, when executed by the processor of the machine, implements the functions / acts specified in the flowcharts and / or block diagrams. The computer program can be executed entirely on a machine, partially on a machine, partially on a machine as part of a standalone software package, or entirely on a remote machine or server.

[0184] In the context of the present application, a computer-readable storage medium can be a tangible medium that can contain or store a computer program for use by or in connection with an instruction execution system, apparatus, or device. A computer-readable storage medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium can be a machine-readable signal medium. More specific examples of a machine-readable storage medium will include one or more lines of a program of instructions in a transitory signal, a portable computer diskette, 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 disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0185] To provide for interaction with a user, the systems and techniques described here can be implemented on an electronic device having a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the electronic device. Other kinds of devices can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form, including acoustic, speech, or tactile input.

[0186] The systems and techniques described here can be implemented in a computing system that includes a back end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front end component (e.g., a user computer having a graphical user interface or a Web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back end, middleware, or front end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0187] The computing system can include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system, to solve the defects of large management difficulty and weak business scalability in traditional physical host and VPS (Virtual Private Server) service.

[0188] It should be understood that the various forms of flow shown above can be reordered, added to, or have steps deleted. For example, the steps described in the present application can be performed in parallel, in series, or in a different order, as long as the desired results of the technical solutions of the present application can be achieved, which are not limited herein.

[0189] The above detailed description does not constitute a limitation on the protection scope of the present application. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent replacements, and improvements made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A business processing method, characterized in that: The method comprises: Acquire business processing association data, business processing party data, business processing constraint data, and business feedback data of the business to be processed, and generate a current multidimensional feature vector based on the business processing association data, the business processing party data, the business processing constraint data, and the business feedback data; Using a weight calculation routing network, analyzing the current multi-dimensional feature vector to obtain the current weight of each target service model; Using each of the target service models, respectively analyzing the current multi-dimensional feature vector, to obtain a current service detection result of each of the target service models on the service to be processed; According to the current weight of each target business model, the current business detection results are fused to obtain a target fusion result, and a current business processing instruction is generated according to the target fusion result.

2. The method according to claim 1, characterized in that The adopting of each of the target service models to analyze the current multi-dimensional feature vector respectively to obtain the current service detection result of each of the target service models on the service to be processed includes: Obtaining the business type of the business to be processed, and determining a target quantity according to the business type of the business to be processed; According to the ranking of the current weights, the target business models of the target number ranked first are selected from the target business models, and the target business models are updated; The updated target service models are used to analyze the current multi-dimensional feature vectors respectively to obtain the current service detection results of the target service models on the service to be processed.

3. The method according to claim 1, characterized in that Before obtaining the business processing related data, business processing party data, business processing constraint data and business feedback data of the business to be processed, the method further includes: Acquire actual business processing instructions of historical businesses, as well as business processing associated data, business processing party data, business processing constraint data, and business feedback data of the historical businesses, and generate a historical multidimensional feature vector based on the business processing associated data, the business processing party data, the business processing constraint data, and the business feedback data of the historical businesses; Using an initial routing network, analyzing the historical multi-dimensional feature vectors to obtain historical weights of each initial service model; Using each of the initial business models, respectively analyzing the historical multi-dimensional feature vectors, to obtain historical business detection results of each of the initial business models on the historical business; fusing the historical service detection results according to the historical weights of the initial service models to obtain a historical fusion result, and generating a historical service processing instruction according to the historical fusion result; Calculating the overall loss of the initial routing network and each of the initial service models according to the historical service processing instructions and the actual service processing instructions; Adjusting parameters of the initial routing network according to the overall loss, returning to the step of adopting the initial routing network and analyzing the historical multidimensional feature vectors to obtain historical weights of each initial business model, until the overall loss of the initial routing network and each initial business model converges to obtain a weighted routing network; Adjust the parameters of each of the initial business models, return to execute the steps of using each of the initial business models, analyze the historical multidimensional feature vectors respectively, and obtain the historical business detection results of each of the initial business models on the historical business, until the overall loss of the weight calculation routing network and each of the initial business models converges, and each target business model is obtained.

4. The method according to claim 3, characterized in that Before obtaining the actual business processing instructions of the historical business and the business processing associated data, business processing party data, business processing constraint data and business feedback data of the historical business, the following is also included: Regularization constraints are imposed on each of the target business models.

5. The method according to claim 3, characterized in that After obtaining each target business model, the method further includes: Performing static verification on the weight calculation routing network and each of the target service models to obtain a static verification result; Dynamically verifying the weight calculation routing network and each of the target service models to obtain a dynamic verification result; According to the static verification result and / or the dynamic verification result, a target optimization mode of the weight calculation routing network and each of the target service models is determined.

6. The method according to claim 3, characterized in that After obtaining each target business model, the method further includes: Detecting the weight calculation routing network and each of the target business models to determine stability indicators, risk control indicators, and generalization difference indicators; According to the stability index, the risk control index and / or the generalization difference index, the weight calculation routing network and each of the target business models are adjusted.

7. The method according to claim 1, characterized in that Each of the target business models includes a time series prediction model, a multivariate regression model, a text classification model and a heterogeneous graph learning model.

8. A business processing device, characterized in that: The device comprises: a current data acquisition module, configured to acquire business processing association data, business processing party data, business processing constraint data, and business feedback data of the business to be processed, and generate a current multidimensional feature vector based on the business processing association data, the business processing party data, the business processing constraint data, and the business feedback data; A current weight calculation module, configured to analyze the current multi-dimensional feature vector using a weight calculation routing network to obtain a current weight of each target service model; A current service detection module, configured to analyze the current multi-dimensional feature vector using each target service model to obtain a current service detection result of each target service model on the service to be processed; The current business processing module is used to fuse the current business detection results according to the current weights of the target business models to obtain a target fusion result, and generate a current business processing instruction according to the target fusion result.

9. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor. The computer program is executed by the at least one processor so that the at least one processor can execute the business processing method according to any one of claims 1 to 7.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the computer program implements the service processing method according to any one of claims 1 to 7.

Images