CN112711578B

CN112711578B - Big data denoising method for cloud computing service and cloud computing financial server

Info

Publication number: CN112711578B
Application number: CN202011602094.1A
Authority: CN
Inventors: 陈静
Original assignee: Shenzhen Panoramic Network Co ltd
Current assignee: Shenzhen Panoramic Network Co.,Ltd.
Priority date: 2020-12-30
Filing date: 2020-12-30
Publication date: 2021-09-21
Anticipated expiration: 2040-12-30
Also published as: CN112711578A

Abstract

The embodiment of the application provides a big data denoising method for cloud computing service and a cloud computing financial server, which can fully utilize information pushing configuration information of a plurality of information pushing services mapped by big data of the service to be denoised, and can predict a label to be denoised in a targeted manner by combining different information pushing services in an actual application process, so that the information pushing configuration information of each information pushing service can be used for mutual denoising complementation in a subsequent denoising process, and the denoising accuracy is greatly improved.

Description

Big data denoising method for cloud computing service and cloud computing financial server

Technical Field

The application relates to the technical field of big data, in particular to a big data denoising method for cloud computing service and a cloud computing financial server.

Background

The development and progress of big data technology provides people with new tools, namely, a method for recognizing problems and analyzing the problems from a wider field of view, more dimensions and more omnidirection. However, some organizations have mastered a certain amount of customer information data, and have neglected research on data analysis tools and methodologies in order to master large data. In financial transactions, this can affect their identification and prevention of risk, and cause the accumulation and spread of risk.

Although, to date, there is no uniform and authoritative definition of big data. But an important feature of big data is that it should include both structural data and information that appears as non-structural data when generated. And the small data mainly refers to traditional two-dimensional structural data. From a processing perspective, large data may increase dramatically with the amount of data, where the data noise may increase rapidly. Sometimes, the data noise increases faster than the amount of data. Therefore, in the field of big data, the cost of mining, screening and cleaning the big data is obviously higher than that of small data.

Based on this, how to perform effective data denoising processing on big data so as to ensure the accuracy of the subsequent big data mining process and ensure the high-quality operation of business services is a technical problem to be solved urgently in the field.

Disclosure of Invention

In order to overcome at least the above defects in the prior art, the present application aims to provide a big data denoising method for cloud computing services and a cloud computing financial server, which can make full use of information push configuration information of a plurality of information push services mapped by big data of services to be denoised, and perform targeted denoising label prediction by combining different information push services in the practical application process, so that the information push configuration information of each information push service can be utilized for mutual denoising complementation in the subsequent denoising process, and the denoising accuracy rate is greatly improved.

In a first aspect, the present application provides a big data denoising method for cloud computing services, which is applied to a cloud computing financial server, the cloud computing financial server is in communication connection with a plurality of information service terminals, the cloud computing financial server is implemented according to a cloud computing platform, and the method includes:

acquiring service big data to be denoised, and acquiring information push configuration information of a plurality of information push services mapped by the service big data, wherein the service big data is a service data set collected based on a cloud computing service;

analyzing the information push configuration information into a corresponding push element set, and inputting the push element set into a corresponding decision unit in a trained big data denoising decision model; each decision unit at least comprises a decision model, and the decision model of each decision unit processes a push element set corresponding to the information push service;

predicting according to big data denoising decision characteristics output by a plurality of decision units through a prediction module in the big data denoising decision model, and outputting a big data denoising label to which the business big data belongs;

and carrying out big data denoising on the business big data according to the big data denoising label to which the business big data belongs.

In a possible design example of the first aspect, the obtaining information push configuration information of a plurality of information push services to which the traffic big data is mapped includes:

acquiring service calling nodes corresponding to a plurality of information push services;

determining corresponding information push service from the service big data according to the service calling node;

and acquiring the information push configuration information from the determined plurality of information push services.

In a possible design example of the first aspect, the parsing the information push configuration information into a corresponding push element set includes:

splitting the information pushing configuration information according to the intention requirement;

with the intention demand elements as units, carrying out intention chain construction on the intention demand elements obtained by splitting the intention demand according to the incidence relation in the information push configuration information to obtain a candidate push element set;

when the information pushing configuration information is nonstandard configuration information, the candidate pushing element set is regulated to a pushing element set with a preset intention demand element quantity, and the obtained pushing element set after regulation corresponds to the information pushing configuration information;

and when the information pushing configuration information is standard configuration information, directly taking the candidate pushing element set as a pushing element set corresponding to the information pushing configuration information.

In a possible design example of the first aspect, the step of outputting, by a prediction module in the big data denoising decision model, a big data denoising tag to which the business big data belongs according to prediction performed by a big data denoising decision feature output by a plurality of decision units includes:

fusing big data denoising decision characteristics output by the decision units to obtain fused big data denoising decision characteristics;

predicting the fused big data denoising decision characteristics into denoising relevance parameters corresponding to each preset big data denoising label through a prediction module in the big data denoising decision model;

selecting the largest denoising correlation parameter from the predicted denoising correlation parameters;

and outputting the preset big data denoising label corresponding to the maximum denoising correlation parameter as a big data denoising label belonging to the business big data.

In a possible design example of the first aspect, each decision unit includes a preset input condition, a corresponding push element set of each information push service includes a push activation tag, and the step of inputting the push element set into a corresponding decision unit in a trained big data denoising decision model includes:

reading a push activation tag of the set of push elements;

when the read push activation tag accords with the input condition of the corresponding decision unit, inputting the push element set to the corresponding decision unit, otherwise, prompting that the push element set does not meet the input condition; or

Each decision unit comprises a preset condition of the quantity of the intention demand elements, and the step of inputting the push element set into the corresponding decision unit in the trained big data denoising decision model comprises the following steps:

and determining the quantity of the elements required by the intention of the push element set, inputting the push element set to the corresponding decision unit when the determined quantity of the elements required by the intention meets the quantity condition of the elements required by the intention of the corresponding decision unit, and otherwise, prompting that the push element set does not meet the quantity condition of the elements required by the intention.

In one possible design example of the first aspect, the big data denoising decision model is obtained by training:

acquiring candidate service big data, and determining standard configuration information mapped by the candidate service big data;

respectively matching preset necessary subscription conditions with the standard configuration information of each candidate service big data, and taking the corresponding candidate service big data as a service big data sample when the matching is successful;

acquiring a preset big data denoising label corresponding to a successfully matched necessary subscription condition, and marking the preset big data denoising label as a big data denoising label corresponding to the business big data sample;

acquiring information push configuration information of a plurality of information push services mapped by the business big data sample, analyzing the information push configuration information into a corresponding push element set, and inputting the push element set into a corresponding decision unit in a big data denoising decision model; each decision unit at least comprises a decision model, and the decision model of each decision unit processes a push element set corresponding to the information push service;

fusing big data denoising decision characteristics output by a plurality of decision units to obtain fused big data denoising decision characteristics, and predicting the fused big data denoising decision characteristics into denoising correlation parameters corresponding to each preset big data denoising label through a prediction module in the big data denoising decision model;

selecting the maximum denoising relevance parameter from the predicted denoising relevance parameters, outputting a preset big data denoising label corresponding to the maximum denoising relevance parameter as to-be-determined prediction information, adjusting the model parameter of the big data denoising decision model according to the to-be-determined prediction information and the loss function value of the big data denoising label, and continuing training until the training is finished when the training stopping condition is met, thereby obtaining the big data denoising decision model.

In a possible design example of the first aspect, the step of denoising the big data of the service according to the big data denoising tag to which the big data of the service belongs includes:

acquiring service data to be denoised in at least one service data area of the big data denoising label corresponding to the service big data, acquiring noise service characteristic data of the service data area, and respectively acquiring a global denoising operation rule and an initial block denoising operation rule of the service data area based on a software service denoising mode and a non-software service denoising mode according to the noise service characteristic data;

performing denoising label supplement processing on the initial block denoising operation rule to obtain a target block denoising operation rule;

performing rule splicing on the global denoising operation rule and the target block denoising operation rule respectively based on a software service denoising mode and a non-software service denoising mode to obtain target software service denoising rule configuration information and target non-software service denoising rule configuration information;

and updating a denoising model according to the target software service denoising rule configuration information and the target non-software service denoising rule configuration information to obtain a target denoising model, and denoising the service data to be denoised by the target denoising model.

In a possible design example of the first aspect, the step of performing denoising label supplementary processing on the initial block denoising operation rule to obtain a target block denoising operation rule includes:

acquiring de-noising label distribution of the initial block de-noising operation rule;

matching target denoising label distribution with an incidence relation with the denoising label distribution from a preconfigured denoising label distribution preset set;

and supplementing the denoising operation rule matched with the target denoising label distribution to the initial block denoising operation rule according to the target denoising label distribution to obtain the target block denoising operation rule.

In a possible design example of the first aspect, the business data region is a privacy authorized data region, the target block denoising operation rule includes a key denoising operation rule corresponding to a key denoising node of a privacy authorized data element, and the step of performing rule concatenation on the global denoising operation rule and the target block denoising operation rule based on a software service denoising mode and a non-software service denoising mode respectively to obtain target software service denoising rule configuration information and target non-software service denoising rule configuration information includes: respectively carrying out rule attribute unification on supplementary operation rule partitions of each key denoising operation rule to obtain a unified key denoising operation rule with the same content as the global denoising operation rule template, combining the unified key denoising operation rules to obtain a member denoising operation rule of a privacy authorized data element, and carrying out rule splicing on the global denoising operation rule and the member denoising operation rule of the privacy authorized data element based on a software service denoising mode and a non-software service denoising mode respectively to obtain target software service denoising rule configuration information and target non-software service denoising rule configuration information;

or, the global denoising operation rule includes a denoising operation rule of a global software service denoising mode and a denoising operation rule of a global non-software service denoising mode, the target block denoising operation rule includes a denoising operation rule of a block software service denoising mode and a denoising operation rule of a unit non-software service denoising mode, and the step of performing rule splicing on the global denoising operation rule and the target block denoising operation rule based on the software service denoising mode and the non-software service denoising mode respectively to obtain target software service denoising rule configuration information and target non-software service denoising rule configuration information includes: performing rule splicing on a denoising operation rule of the global software service denoising mode and a denoising operation rule of the blocking software service denoising mode, configuring the denoising operation rule of the rule splicing to integrate a global software service denoising mode feature and a blocking software service denoising mode feature to obtain target software service denoising rule configuration information, performing rule splicing on the denoising operation rule of the global non-software service denoising mode and the denoising operation rule of the unit non-software service denoising mode in each denoising enabling flow, and configuring the denoising operation rule of each denoising enabling flow rule splicing to integrate the global non-software service denoising mode feature and the unit non-software service denoising mode feature to obtain the target non-software service denoising rule configuration information;

wherein, the global denoising operation rule and the target block denoising operation rule both correspond to at least one denoising enabling process, the denoising operation rule of the global software service denoising mode and the denoising operation rule of the block software service denoising mode are regularly spliced, and the denoising operation rule of the regular splicing is configured to integrate the global software service denoising mode feature and the block software service denoising mode feature to obtain the target software service denoising rule configuration information, including: and carrying out rule splicing on the denoising operation rule of the global software service denoising mode and the denoising operation rule of the blocking software service denoising mode in each denoising enabling flow, and configuring the denoising operation rule spliced by each denoising enabling flow rule so as to integrate the global software service denoising mode characteristic and the blocking software service denoising mode characteristic, thereby obtaining the target software service denoising rule configuration information.

For example, in a possible design example of the first aspect, the step of obtaining the noisy traffic feature data of the traffic data region includes:

carrying out data item-by-data item denoising feature extraction on the service data region;

and obtaining software service denoising mode characteristic information and non-software service denoising mode characteristic information of the service data region according to the denoising characteristic extraction result of the data item by data item, and taking the information as the noise service characteristic data.

For example, in a possible design example of the first aspect, the step of obtaining, according to the noise service characteristic data, a global denoising operation rule and an initial block denoising operation rule of the service data region based on a software service denoising mode and a non-software service denoising mode includes:

carrying out denoising indexing on the service data region by a global denoising indexing model according to the noise service characteristic data to obtain a global denoising operation rule;

denoising and indexing the service data region by a unit denoising and indexing model according to the noise service characteristic data to obtain the initial block denoising operation rule;

the unit denoising index model comprises a key denoising node denoising index model of privacy authorized data elements;

the unit denoising index model denoises and indexes the service data region according to the noise service characteristic data to obtain the initial block denoising operation rule, and the method comprises the following steps:

and denoising and indexing the service data region according to the noise service characteristic data by using the key denoising node denoising index model of the privacy authorized data element, and determining the obtained key denoising node denoising operation rule as the initial block denoising operation rule.

For example, in a possible design example of the first aspect, the step of performing denoising model updating processing according to the target software service denoising rule configuration information and the target non-software service denoising rule configuration information to obtain a target denoising model includes:

mapping the target software service denoising rule configuration information and the target non-software service denoising rule configuration information into target denoising rule configuration information according to each denoising node combination, acquiring at least two denoising nodes included in the target denoising rule configuration information, and acquiring denoising feature information of each denoising node in the at least two denoising nodes;

determining global weighted denoising feature information aiming at the target denoising rule configuration information based on the denoising feature information of each denoising node;

determining distributed denoising feature information of the target denoising rule configuration information on a denoising distribution label based on denoising feature label components of the denoising feature information of each denoising node on the denoising distribution label;

and determining a deep learning denoising model of the target denoising rule configuration information aiming at denoising calling operation associated with the denoising distribution label according to the global weighted denoising feature information and the distributed denoising feature information, performing denoising model updating processing based on the deep learning denoising model to obtain a target denoising model, and denoising the service data to be denoised by the target denoising model.

For example, in a possible design example of the first aspect, the determining global weighted denoising feature information for the target denoising rule configuration information based on the denoising feature information of each denoising node includes:

acquiring a label feature vector of a structured denoising distribution label and a label feature vector of an unstructured denoising distribution label, wherein the denoising feature information of each denoising node respectively comprises the label feature vector of the structured denoising distribution label and the label feature vector of the unstructured denoising distribution label;

determining the distribution of the unit label characteristic vectors corresponding to each denoising node based on the label characteristic vectors of the structured denoising distribution labels and the label characteristic vectors of the unstructured denoising distribution labels corresponding to each denoising node;

and determining the global weighted denoising feature information according to the unit label feature vector distribution corresponding to each denoising node and the node number of the at least two denoising nodes.

For instance, in one possible design example of the first aspect, the denoising distribution label comprises a structured denoising distribution label;

denoising feature label components of the denoising feature information of each denoising node on the denoising distribution label comprise label feature vectors of the structured denoising distribution label;

the determining distributed denoising feature information of the target denoising rule configuration information on the denoising distribution label based on the denoising feature label component of the denoising feature information of each denoising node on the denoising distribution label comprises:

acquiring label feature vectors of the structured denoising distribution labels of the denoising feature information of each denoising node on the structured denoising distribution labels respectively;

determining a structured denoising distribution label value corresponding to each denoising node according to the label feature vector of the structured denoising distribution label of each denoising node on the structured denoising distribution label;

determining the distributed denoising feature information according to the structured denoising distribution label value corresponding to each denoising node and the number of the at least two denoising nodes;

the de-noising distribution label comprises an unstructured de-noising distribution label;

denoising feature label components of the denoising feature information of each denoising node on the denoising distribution label comprise label feature vectors of the unstructured denoising distribution label;

acquiring label feature vectors of the unstructured denoising distribution labels of the denoising feature information of each denoising node on the unstructured denoising distribution labels respectively;

determining an unstructured denoising distribution label value corresponding to each denoising node according to the label feature vector of the unstructured denoising distribution label of each denoising node on the unstructured denoising distribution label;

and determining the distributed denoising characteristic information according to the unstructured denoising distribution label value corresponding to each denoising node and the node number of the at least two denoising nodes.

In a second aspect, an embodiment of the present application further provides a big data denoising device for cloud computing service, which is applied to a cloud computing financial server, where the cloud computing financial server is in communication connection with a plurality of information service terminals, and the cloud computing financial server is implemented according to a cloud computing platform, and the device includes:

the device comprises an acquisition module, a denoising module and a denoising module, wherein the acquisition module is used for acquiring service big data to be denoised and acquiring information pushing configuration information of a plurality of information pushing services mapped by the service big data, and the service big data is a service data set collected based on cloud computing service;

the input module is used for analyzing the information push configuration information into a corresponding push element set and inputting the push element set into a corresponding decision unit in a trained big data denoising decision model; each decision unit at least comprises a decision model, and the decision model of each decision unit processes a push element set corresponding to the information push service;

the prediction module is used for predicting according to the big data denoising decision characteristics output by the decision units through the prediction module in the big data denoising decision model and outputting a big data denoising label to which the business big data belongs;

and the denoising module is used for denoising the big data of the business according to the big data denoising label to which the big data of the business belongs.

In a third aspect, an embodiment of the present application further provides a big data denoising system for cloud computing services, where the big data denoising system for cloud computing services includes a cloud computing financial server and a plurality of information service terminals in communication connection with the cloud computing financial server;

the cloud computing financial server is configured to:

In a fourth aspect, the present invention further provides a cloud computing financial server, where the cloud computing financial server includes a processor, a machine-readable storage medium, and a network interface, where the machine-readable storage medium, the network interface, and the processor are connected through a bus system, the network interface is configured to be communicatively connected to at least one information service terminal, the machine-readable storage medium is configured to store a program, an instruction, or code, and the processor is configured to execute the program, the instruction, or code in the machine-readable storage medium to perform the big data denoising method for cloud computing service in the first aspect or any one of the possible design examples in the first aspect.

In a fifth aspect, an embodiment of the present application provides a computer-readable storage medium, where instructions are stored in the computer-readable storage medium, and when the instructions are executed, the computer is caused to execute the big data denoising method for cloud computing services in the first aspect or any one of the possible design examples of the first aspect.

According to any one of the aspects, the information pushing configuration information of a plurality of information pushing services mapped by the service big data to be denoised can be fully utilized, and the label prediction of the targeted denoising is carried out by combining different information pushing services in the practical application process, so that the information pushing configuration information of each information pushing service can be utilized for mutual denoising complementation in the subsequent denoising process, and the denoising accuracy is greatly improved.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that need to be called in the embodiments are briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present application and therefore should not be considered as limiting the scope, and for those skilled in the art, other related drawings can be obtained according to the drawings without inventive efforts.

Fig. 1 is a schematic view of an application scenario of a big data denoising system for cloud computing service according to an embodiment of the present application;

fig. 2 is a schematic flowchart of a big data denoising method for cloud computing services according to an embodiment of the present disclosure;

fig. 3 is a functional module schematic diagram of a big data denoising device for cloud computing service according to an embodiment of the present disclosure;

fig. 4 is a schematic block diagram of structural components of a cloud computing financial server for implementing the big data denoising method for cloud computing services according to the embodiment of the present application.

Detailed Description

The present application will now be described in detail with reference to the drawings, and the specific operations in the method embodiments may also be applied to the apparatus embodiments or the system embodiments.

Fig. 1 is an interaction diagram of a big data denoising system 10 for cloud computing services according to an embodiment of the present application. The big data denoising system 10 for the cloud computing business may include a cloud computing financial server 100 and an information service terminal 200 communicatively connected with the cloud computing financial server 100. The big data denoising system 10 for cloud computing services shown in fig. 1 is only one possible example, and in other possible embodiments, the big data denoising system 10 for cloud computing services may also include only at least some of the components shown in fig. 1 or may also include other components.

According to the invention concept of the technical solution provided by the present application, the cloud computing financial server 100 provided by the present application can be applied to scenes such as smart medical, smart city management, smart industrial internet, general service monitoring management, and the like, in which a big data technology or a cloud computing technology can be applied, and for example, the cloud computing financial server can also be applied to, but not limited to, new energy vehicle system management, smart cloud office, cloud platform data processing, cloud game data processing, cloud live broadcast processing, cloud vehicle management platform, block chain financial micro-service link platform, and the like.

In this embodiment, the cloud computing financial server 100 and the information service terminal 200 in the big data denoising system 10 for the cloud computing service may cooperatively perform the big data denoising method for the cloud computing service described in the following method embodiment, and specific steps of the cloud computing financial server 100 and the information service terminal 200 may refer to the detailed description of the following method embodiment.

In order to solve the technical problem in the foregoing background art, fig. 2 is a schematic flow chart of a big data denoising method for cloud computing services according to an embodiment of the present application, where the big data denoising method for cloud computing services according to the present embodiment may be executed by the cloud computing financial server 100 shown in fig. 1, and the details of the big data denoising method for cloud computing services are described below.

Step S110, obtaining service big data to be denoised, and obtaining information push configuration information of a plurality of information push services mapped by the service big data.

In this embodiment, the service big data may refer to a set of service operation data, and the service operation data may be initiation operation of the software application service, browsing operation of the software application service, interactive operation of the software application service, change operation of related setting information in the software application service, and the like, but is not limited thereto.

The information push service is configured by a set of connected intention requirement elements with information push significance in the intention requirement elements, and can be used for controlling a data source for information push. For example, the information push configuration information of the information push service is configuration information of the intention requirement element under different push rules, such as data source index configuration information, push frequency configuration information, and the like.

In one possible design example, the business big data includes business topics and business topic contents. The business theme contents under different business themes belong to different information push services. The cloud computing financial server 100 may start scanning from the intention requirement element name of the business big data, attribute the currently scanned business topic content to the business topic content under the previously scanned business topic, and may correspondingly obtain information push configuration information of a plurality of information push services by obtaining a plurality of business topics in the business big data.

In one possible design example, the cloud computing financial server 100 may obtain a service invocation node (e.g., a service invocation node a of an order information push service for an order payment page) corresponding to each of the plurality of information push services, determine, according to the service invocation node, a corresponding information push service from the business big data (e.g., the order information push service for the service invocation node a), and obtain information push configuration information from the determined plurality of information push services (e.g., obtain corresponding information push configuration information from the order information push service).

Step S120, analyzing the information pushing configuration information into a corresponding pushing element set, and inputting the pushing element set into a corresponding decision unit in the trained big data denoising decision model. Each decision unit at least comprises a decision model, and the decision model of each decision unit processes a push element set corresponding to the information push service.

Wherein the pushed element set is an intention requirement element set having an order of arrangement. For example, after obtaining the information push configuration information of the plurality of information push services mapped by the business big data, the cloud computing financial server 100 may perform intent requirement splitting on the information push configuration information in an intent requirement splitting manner. After the cloud computing financial server 100 splits the information push configuration information according to the intention requirement, the intention requirement elements obtained by splitting the intention requirement are subjected to intention chain construction, and a push element set corresponding to the information push configuration information is obtained.

The intention requirement splitting method includes multiple ways, for example, an intention requirement splitting algorithm based on string matching, an intention requirement splitting algorithm based on semantic analysis, or an intention requirement splitting algorithm based on statistics, and the like. The intent requirement splitting algorithm based on character string matching is a forward maximum matching algorithm, a reverse maximum matching algorithm, a minimum segmentation algorithm or a bidirectional maximum matching algorithm.

In one possible design example, the cloud computing financial server 100 may split the intention requirement into intention requirement elements, construct an intention chain for the intention requirement elements according to an association relationship in the information push configuration information, and obtain a push element set corresponding to the information push configuration information. In one possible design example, the cloud computing financial server 100 may split the intention requirement into the intention requirement elements, perform intention chain construction on the intention requirement elements in a random sequence manner, and obtain a push element set corresponding to the information push configuration information.

The decision model may include a convolutional layer, in the convolutional layer of the decision model, a plurality of feature maps exist, each feature map includes a plurality of neurons, and all neurons of the same feature map share one convolutional kernel. The convolution kernel is the weight of the corresponding neuron, and represents a feature. The convolution kernel is generally initialized in the form of a random decimal matrix, a reasonable convolution kernel is obtained through learning in the training process of the network, and the convolution layer can reduce the connection among layers in the neural network and reduce the risk of overfitting. In this embodiment, the convolution layer may have one layer or a plurality of layers.

The big data denoising decision model can have a plurality of groups of decision units, and a plurality of groups of data can be input. The data input from each group of decision units are processed by a separate decision model, and finally, the output of different decision units is fused together by the prediction module to be used as the input of the prediction module.

In the big data denoising decision model adopted in this embodiment, the feature matrix output by the front layer may be mapped to data corresponding to each preset big data denoising label, so that the big data denoising labels to which the multiple groups of push element sets input through the regression layer belong are output.

For example, the cloud computing financial server 100 may obtain a decision unit corresponding to an information push service to which information push configuration information corresponding to a push element set belongs, and then input the push element set into a corresponding decision unit in a trained big data denoising decision model. Each decision unit at least comprises a decision model, and the decision model of each decision unit processes a push element set corresponding to the information push service.

In one possible design example, the cloud computing financial server 100 may preset a corresponding relationship between an input pushed element set and a decision unit when training a big data denoising decision model. For example, an identifier corresponding to a corresponding information push service is added to a push element set, and then different decision units in a big data denoising decision model are set to input only one push element set corresponding to the identifier. Therefore, the training algorithm of the corresponding decision unit can be ensured to correctly train the corresponding data in the training process of the big data denoising decision model. When the push element set is input into a decision unit in the trained big data denoising decision model, the push element set is input into the corresponding decision unit according to the preset corresponding relation between the input push element set and the decision unit.

And step S130, forecasting according to the big data denoising decision characteristics output by the plurality of decision units through a forecasting module in the big data denoising decision model, and outputting a big data denoising label to which the business big data belongs.

For example, the cloud computing financial server 100 may fuse the big data denoising decision features output by the multiple decision units to obtain a fused big data denoising decision feature, use the fused big data denoising decision feature as an input of a prediction module in a trained big data denoising decision model, and output a big data denoising tag to which the business big data belongs through the prediction module.

In a possible design example, the cloud computing financial server 100 may output, through a prediction module in a trained big data denoising decision model, a denoising relevance parameter of each preset big data denoising tag to which the business big data belongs, and predict the business big data to be denoised to a big data denoising tag corresponding to a maximum denoising relevance parameter.

And step S140, carrying out big data denoising on the business big data according to the big data denoising label to which the business big data belongs.

Based on the steps, the information push configuration information of a plurality of information push services mapped by the service big data to be denoised is analyzed into corresponding push element sets, the push element sets are respectively input into decision units corresponding to the information push services to which the push element sets belong in a trained big data denoising decision model, and the decision model of each decision unit processes one push element set corresponding to the information push service, so that a plurality of groups of push element sets of the service big data to be denoised can be subjected to convolution processing. And predicting according to the big data denoising decision characteristics output by the plurality of decision units through a prediction module in the big data denoising decision model, and outputting a big data denoising label to which the business big data belongs. Therefore, information push configuration information of a plurality of information push services mapped by service big data to be denoised can be fully utilized, and the label prediction of the targeted denoising is carried out by combining different information push services in the practical application process, so that the information push configuration information of each information push service can be utilized to realize mutual denoising complementation in the subsequent denoising process, and the denoising accuracy is greatly improved.

In a possible design example, for step S120, the information push configuration information may be split into intention demand elements, and the intention demand elements obtained by splitting the intention demand elements are subjected to intention chain construction according to the association relationship in the information push configuration information by taking the intention demand elements as units, so as to obtain a push element set corresponding to the information push configuration information.

For example, the cloud computing financial server 100 may push configuration information for the acquired information, and perform intent requirement splitting using an intent requirement splitting algorithm. After the information push configuration information is split according to the intention requirement, the intention requirement elements obtained by splitting the intention requirement are subjected to intention chain construction according to the incidence relation in the information push configuration information by taking the intention requirement elements as units, and a push element set corresponding to the information push configuration information is obtained.

In the above embodiment, the information push configuration information is split according to the intention requirement, and then the intention requirement elements obtained by splitting the intention requirement are constructed according to the association relationship in the information push configuration information by using the intention requirement elements as units, so as to obtain the push element set corresponding to the information push configuration information.

In a possible design example, taking an intention requirement element as a unit, performing intention chain construction on the intention requirement element obtained by splitting an intention requirement according to an incidence relation in information push configuration information, and obtaining a push element set corresponding to the information push configuration information includes: and taking the intention demand elements as units, performing intention chain construction on the intention demand elements obtained by splitting the intention demand according to the incidence relation in the information push configuration information to obtain a candidate push element set. When the information pushing configuration information is nonstandard configuration information, regulating the candidate pushing element set into a pushing element set with a preset intention demand element quantity, wherein the regulated pushing element set corresponds to the information pushing configuration information; and when the information pushing configuration information is standard configuration information, directly taking the candidate pushing element set as a pushing element set corresponding to the information pushing configuration information.

It is worth mentioning that the non-standard configuration information is information push configuration information presented in a non-standard configuration structure. The standard configuration information is information push configuration information presented by a standard configuration structure, for example, after the cloud computing financial server 100 splits the information push configuration information according to an intention requirement, the intention requirement elements obtained by splitting the intention requirement are subjected to intention chain construction according to the intention requirement elements in the association relationship in the information push configuration information by taking the intention requirement elements as units, so as to obtain a candidate push element set.

In one possible design example, the information push configuration information is non-standard configuration information, such as that in an information push service, the information push configuration information is composed of at least one non-standard configuration structure. In practical situations, when the information push configuration information is a non-standard configuration structure, some information push configuration information has more intended demand elements, some information push configuration information has less intended demand elements, and even the number of contents is different greatly. At this time, after the cloud computing financial server 100 performs intent demand splitting on the information push configuration information of which the information push configuration information is a non-standard configuration structure to obtain a candidate push element set, the number of intent demand elements of the candidate push element set is counted, and when the number of intent demand elements of the candidate push element set is greater than the number of preset intent demand elements, only the intent demand elements of the preset intent demand elements in the candidate push element set are taken to form a push element set corresponding to the information push configuration information. When the number of the intention demand elements of the candidate push element set is smaller than the preset number of the intention demand elements, supplementing the intention demand elements behind the candidate push element set, so that the number of the intention demand elements of the candidate push element set after supplementing the intention demand elements is the preset number of the intention demand elements.

For example, the preset number is 200. And when the number of the intention demand elements of the candidate push element set is more than 200, only the top 200 intention demand elements are taken as the push element set corresponding to the information push configuration information. Or when the number of the intention demand elements of the candidate push element set is greater than 200, randomly extracting 200 intention demand elements from the candidate push element set to form a new push element set as the push element set corresponding to the information push configuration information. When the number of the intention requirement elements of the candidate push element set is less than 200, supplementing the intention requirement elements behind the candidate push element set, so that the number of the intention requirement elements of the candidate push element set after supplementing the intention requirement elements is 200.

In a possible design example, the information push configuration information is standard configuration information, and the candidate push element set may be directly used as the push element set corresponding to the information push configuration information. The standard configuration information is a set of intention requirement elements with a specific format, for example, the information pushing configuration information in the intention requirement element label block is a series of intention requirement element sets.

In the above embodiment, due to the difference between the presentation structure of the information push configuration information and the content quantity of the information push configuration information, the information push configuration information of different presentation structures is split according to the intention requirement to obtain the candidate push element set, and the push element set corresponding to the information push configuration information is obtained in different manners, so that the influence generated when the presentation structure of the information push configuration information, the quantity difference of the information push configuration information, and the like predict the big service data to be denoised can be avoided, and the prediction accuracy of the big service data is further improved.

In one possible design example, step S130 may be implemented by the following exemplary substeps, described in detail below.

And the substep S131, fusing the big data denoising decision characteristics output by the decision units to obtain a fused big data denoising decision characteristic.

For example, after the cloud computing financial server 100 inputs a plurality of groups of pushed element sets into decision units in a trained big data denoising decision model, each decision unit outputs a big data denoising decision feature corresponding to the input pushed element set. The output big data denoising decision feature is a feature big data denoising decision feature of a pushing element set output after a convolution layer in a trained big data denoising decision model convolves the big data denoising decision feature of the intention demand element. And fusing the big data denoising decision characteristics output by the plurality of decision units according to the sequence of outputting the big data denoising decision characteristics to obtain fused big data denoising decision characteristics.

And a substep S132, predicting the merged big data denoising decision characteristics into denoising relevance parameters corresponding to each preset big data denoising label through a prediction module in the big data denoising decision model.

For example, the cloud computing financial server 100 may use the merged big data denoising decision feature as an input of a prediction module, and perform dimension reduction on the merged big data denoising decision feature through the prediction module in the trained big data denoising decision model, and then map the merged big data denoising decision feature into a denoising correlation parameter corresponding to each preset big data denoising label.

For example, if the fused big data denoising decision feature is a 60-dimensional big data denoising decision feature and the number of the preset big data denoising tags is 10, the 60-dimensional fused big data denoising decision feature can be mapped to a denoising correlation parameter corresponding to each preset big data denoising tag through a prediction module in the big data denoising decision model after being subjected to dimension reduction, that is, the denoising correlation parameter is mapped to a 10-dimensional big data denoising decision feature. And the data of each dimension corresponds to the service big data and belongs to the denoising relevance parameter of the preset big data denoising label.

In the substep S133, the largest denoising correlation parameter is selected from the predicted denoising correlation parameters.

For example, the cloud computing financial server 100 may determine the maximum denoising correlation parameter in the denoising correlation parameters by predicting the service big data to be denoised to the denoising correlation parameter of each preset big data denoising tag and comparing the denoising correlation parameters one by one.

And a substep S134, outputting the preset big data denoising label corresponding to the maximum denoising relevance parameter as a big data denoising label belonging to the business big data.

For example, the cloud computing financial server 100 may predict the service big data to be denoised to a preset big data denoising tag corresponding to the maximum denoising relevance parameter.

In the above embodiment, the big data denoising decision characteristics output by the plurality of decision units are fused to obtain the fused big data denoising decision characteristics, the fused big data denoising decision characteristics are predicted to be denoising correlation parameters corresponding to each preset big data denoising label through a prediction module in a big data denoising decision model, then the preset big data denoising label corresponding to the maximum denoising correlation parameter is output to be the big data denoising label belonging to the business big data, and the information push configuration information in the business big data to be denoised can be fully utilized, so that the information push configuration information of each information push service can be utilized for mutual denoising complementation in the subsequent denoising process, and the denoising accuracy is greatly improved.

In a possible design example, each decision unit may further include a preset input condition, the push element set corresponding to each information push service includes a push activation tag, and for step S120, the push activation tag of the push element set may be specifically read, and when the read push activation tag meets the input condition of the corresponding decision unit, the push element set is input to the corresponding decision unit, otherwise, the push element set is prompted not to meet the input condition.

Wherein the preset input condition is a condition preset by the cloud computing financial server 100 and allowing the push element set to be input. Each decision unit contains preset input conditions, and specifically, each decision unit only allows the input of a push element set containing a specific push activation tag. The push activation ticket is a specific label that can be used to distinguish between different push types. The push activation tag may specifically be at least one of a letter, a symbol, an image, and a chinese character. In this embodiment, the push activation tag of the push element set may be used to uniquely identify a corresponding information push service, such as a service name of the information push service.

In one possible design example, after the cloud computing financial server 100 parses the information push configuration information into a corresponding push element set, a push activation tag may be inserted at the head of the push element set, where the inserted push activation tag is used to uniquely identify a corresponding information push service, and specifically may be a number, a chinese character, or a letter, such as "1", "intention requirement element description", or "a".

For example, before inputting the push element set into the decision unit, the cloud computing financial server 100 may read a push activation tag of the push element set to be input, and determine whether the read push activation tag meets an input condition of the corresponding decision unit. And when the read push activation tag accords with the input condition of the corresponding decision unit, inputting the push element set into the corresponding decision unit, otherwise, prompting that the push element set does not meet the input condition.

For example, assuming that the input condition of one current decision unit is to allow only the push element set including the "a 1" push activation tag to be input, when the push activation tag of the push element set extracted by the cloud computing financial server 100 is "a 1", the cloud computing financial server 100 may input the push element set to the corresponding decision unit. When the push activation tag of the push element set extracted by the cloud computing financial server 100 is not "a 1", such as "a 2", the cloud computing financial server 100 prompts that the push element set does not satisfy the input condition.

In the above embodiment, according to the preset input condition included in each decision unit, only the push element set corresponding to the push activation tag meeting the input condition is controlled to be input, so that it can be ensured that the push element set input to the decision unit is correct, the influence on the applicability of the big data denoising decision model due to the incorrect input of the push element set is avoided, and the prediction accuracy of the business big data is improved.

In a possible design example, each decision unit includes a preset condition of the number of the intention requirement elements, and for step S120, the number of the intention requirement elements of the push element set may be specifically determined, and when the determined number of the intention requirement elements meets the condition of the number of the intention requirement elements of the corresponding decision unit, the push element set is input to the corresponding decision unit, otherwise, the push element set is prompted not to meet the condition of the number of the intention requirement elements.

The preset condition of the quantity of the intended demand elements is a condition which is required to be met by the quantity of the intended demand elements of the push element set of the input decision unit, which is preset by the cloud computing financial server 100. Each decision unit contains a preset condition of the number of the intended demand elements, and specifically, each decision unit only allows the input of a push element set with the number of the intended demand elements being greater than, less than or equal to a preset number, or the decision unit only allows the input of a push element set with the number of the intended demand elements being within a preset range.

For example, before the cloud computing financial server 100 inputs the pushed element set into the decision unit, the quantity of the intention requirement elements of the pushed element set may be counted, and whether the counted quantity of the intention requirement elements meets the condition of the quantity of the intention requirement elements of the corresponding decision unit is determined. And when the counted number of the elements with the intention requirement meets the number condition of the elements with the intention requirement of the corresponding decision unit, inputting the push element set to the corresponding decision unit, otherwise, prompting that the push element set does not meet the number condition of the elements with the intention requirement.

In one possible design example, the cloud computing financial server 100 sets that each decision unit only allows the push element set with a specific intent requirement element number to be input, when the information push configuration information is analyzed into the corresponding push element set, the push element sets corresponding to different information push services are respectively normalized into the push element sets with the specific intent requirement element number corresponding to the corresponding decision units, and then the push element sets are respectively input into the corresponding decision units in the trained big data denoising decision model.

In the above embodiment, according to the respective preset conditions of the quantity of the intention demand elements included in each decision unit, only the push element set meeting the conditions of the quantity of the intention demand elements can be controlled to be input, so that the push element set input to the decision unit can be ensured to be correct, and the influence on the applicability of the big data denoising decision model due to the incorrect input of the push element set is avoided.

On the basis of the above description, in one possible design example, the big data denoising decision model is obtained by training through the following steps, which are described in detail below.

Step S101, acquiring candidate service big data, and determining standard configuration information mapped by the candidate service big data.

And step S102, respectively matching preset necessary subscription conditions with the standard configuration information of each candidate service big data, and when the matching is successful, taking the corresponding candidate service big data as a service big data sample.

Step S103, acquiring a preset big data denoising label corresponding to the successfully matched necessary subscription condition, and marking the preset big data denoising label as a big data denoising label corresponding to the business big data sample.

Step S104, obtaining information pushing configuration information of a plurality of information pushing services mapped by the business big data samples, analyzing the information pushing configuration information into corresponding pushing element sets, and inputting the pushing element sets into corresponding decision units in the big data denoising decision model. Each decision unit at least comprises a decision model, and the decision model of each decision unit processes a push element set corresponding to the information push service.

And S105, fusing the big data denoising decision characteristics output by the decision units to obtain fused big data denoising decision characteristics, and predicting the fused big data denoising decision characteristics into denoising correlation parameters corresponding to each preset big data denoising label through a prediction module in the big data denoising decision model.

And S106, selecting the maximum denoising relevance parameter from the predicted denoising relevance parameters, outputting a preset big data denoising label corresponding to the maximum denoising relevance parameter as to-be-determined prediction information, adjusting the model parameter of a big data denoising decision model according to the to-be-determined prediction information and the loss function value of the big data denoising label, continuing training until the training stop condition is met, and obtaining the big data denoising decision model.

In this embodiment, the pushed element set may be input into a corresponding decision unit in the big data denoising decision model. Each decision unit at least comprises a decision model, and the decision model of each decision unit processes a push element set corresponding to the information push service.

For example, the cloud computing financial server 100 may obtain a decision unit corresponding to an information push service to which information push configuration information corresponding to a push element set belongs, and then input a push element set corresponding to a business big data sample into a corresponding decision unit in a big data denoising decision model respectively. Each decision unit at least comprises a decision model, and the decision model of each decision unit processes a push element set corresponding to the information push service.

The information to be predicted is a prediction result output by the big data denoising decision model after a business big data sample is input to the big data denoising decision model in the training process.

For example, the cloud computing financial server 100 may fuse the big data denoising decision features output by the multiple decision units to obtain a fused big data denoising decision feature. And the fused big data denoising decision characteristic is used as the input of a prediction module in a big data denoising decision model, a big data denoising label to which a business big data sample belongs is output through the prediction module, and the big data denoising label output in the model training process is used as the to-be-determined prediction information.

In a possible design example, in the training process of the big data denoising decision model, the cloud computing financial server 100 may output, through a prediction module in the big data denoising decision model, a denoising correlation parameter of each preset big data denoising tag to which a business big data sample belongs, predict the business big data sample to a big data denoising tag corresponding to the maximum denoising correlation parameter, and use the big data denoising tag corresponding to the big data denoising tag as the information to be predicted.

Wherein, the training stopping condition is a condition for ending the big data denoising decision model training. The training stopping condition can be that the preset iteration times are reached, or the prediction performance index of the big data denoising decision model after the model parameters are adjusted reaches the preset index. And adjusting the model parameters of the big data denoising decision model, namely adjusting the model parameters of the big data denoising decision model.

For example, the cloud computing financial server 100 may compare the to-be-predicted information with the loss function value of the preset big data denoising tag, so as to adjust the model parameters of the big data denoising decision model in the direction of reducing the loss function value. And if the training stopping condition is not met after the model parameters are adjusted, returning to the step S104 to continue training until the training stopping condition is met, and ending the training.

In one possible design example, the loss function value of the to-be-predicted information and the preset big data denoising label can be measured by a cost function. The cost function is a function of the model parameters and can measure a loss function value between the undetermined prediction information of the big data denoising decision model and a preset big data denoising label. The cloud computing financial server 100 may end training when the value of the cost function is smaller than a preset value, so as to obtain a big data denoising decision model for predicting business big data. Functions such as cross entropy or mean square error may be selected as the cost function.

In this way, by inputting the push element sets corresponding to the information push configuration information of the plurality of information push services mapped by the business big data sample into the decision units corresponding to the information push services to which the push element sets belong in the big data denoising decision model respectively, the decision model of each decision unit processes one push element set corresponding to one information push service, and the convolution processing can be performed on a plurality of groups of push element sets of the business big data sample. And then, model parameters are adjusted through the undetermined prediction information output by the prediction module and the loss function value of the corresponding big data denoising label, so that a big data denoising decision model is trained. Therefore, the big data denoising decision model is trained through the information push configuration information of the plurality of information push services in the business big data sample, so that the trained big data denoising decision model can predict the big data denoising label corresponding to the business big data to be denoised.

The necessary subscription condition is that when the candidate service big data simultaneously satisfies a plurality of conditions, the preset big data denoising label can be marked as the big data denoising label corresponding to the candidate service big data. The necessary subscription condition is a sufficient unnecessary condition of a big data denoising label corresponding to a certain candidate service big data belonging to a certain necessary subscription condition.

For example, for the "a" big data denoising tag, a plurality of necessary subscription conditions may be set, for example, when three or more of "a 1", "a 2", "A3", "a 4" and "a 5" must be simultaneously included in the information push configuration information of the information push service corresponding to the candidate service big data, the candidate service big data is labeled as the "a" big data denoising tag.

For example, the cloud computing financial server 100 may preset a plurality of necessary subscription conditions, and then automatically match the standard configuration information of each candidate business big data through the plurality of necessary subscription conditions.

For example, when one of the preset necessary subscription conditions is matched with the standard configuration information of the candidate service big data, it is determined that the matching is successful, and the cloud computing financial server 100 takes the candidate service big data successfully matched as a service big data sample.

For example, when the preset necessary subscription conditions are respectively matched with the standard configuration information of each candidate service big data, the cloud computing financial server 100 may record the candidate service big data successfully matched and the preset big data denoising tag corresponding to the corresponding necessary subscription conditions, and obtain the preset big data denoising tag corresponding to the necessary subscription conditions.

For example, the cloud computing financial server 100 may label a preset big data denoising label corresponding to a necessary subscription condition for successfully matching the business big data sample as a big data denoising label of the business big data sample.

In the above embodiment, the cloud computing financial server 100 obtains the service big data sample and the corresponding big data denoising label by respectively matching the preset necessary subscription conditions with the standard configuration information of each candidate service big data, so that the service big data sample and the corresponding big data denoising label are automatically matched with the candidate service big data through the plurality of necessary subscription conditions, and the efficiency of labeling the candidate service big data is improved on the premise of ensuring the accuracy of the obtained service big data sample and the corresponding big data denoising label.

For example, in one possible design example, the model training method for business big data prediction further includes a step of retransferring the business big data samples, and obtaining the number of the business big data samples corresponding to the same big data denoising label.

In one possible design example, when the preset necessary subscription condition is respectively matched with the standard configuration information of each candidate service big data, the cloud computing financial server 100 respectively counts the number of the candidate service big data of each big data denoising label, which is successfully matched, through a counter.

In one possible design example, the cloud computing financial server 100 may count the number of business big data samples corresponding to the same big data denoising label by scanning all the business big data samples. And when the number is larger than the preset number, performing descending transfer on the service big data samples corresponding to the same big data denoising label to obtain the preset number of service big data samples.

In the step of drop transfer, a part of preset number of large traffic data samples are reserved in a sampling mode. Sampling is to extract a part of the service big data samples from all the service big data samples corresponding to the same big data denoising label. Sampling means, such as simple random sampling, systematic sampling or hierarchical sampling, etc.

In one possible design example, when the number of the business big data samples corresponding to the same big data denoising label is greater than a preset number, the cloud computing financial server 100 may sample the business big data samples corresponding to the big data denoising label and extract the preset number of the business big data samples.

In one possible design example, when the number of the business big data samples corresponding to the same big data denoising tag is greater than a preset number, the cloud computing financial server 100 may sample the business big data samples corresponding to the big data denoising tag, extract the number of the business big data samples corresponding to the big data denoising tag, and delete the extracted business big data samples from the business big data samples corresponding to the same big data denoising tag by a number different from the preset number, to obtain a preset number of business big data samples.

And when the number is smaller than the preset number, the service big data samples corresponding to the same big data denoising label are subjected to over-transfer to obtain the preset number of service big data samples.

Wherein, the over-transition is to copy part of the extracted service big data sample in a sampling mode. In one possible design example, the cloud computing financial server 100 may sample the business big data samples corresponding to the same big data denoising label, and extract the business big data samples of a number that is a difference between a preset number and the number of the business big data samples corresponding to the big data denoising label. And copying the extracted large traffic data sample. And taking the original service big data sample corresponding to the same big data denoising label and the copied service big data sample as a preset number of service big data samples.

In one possible design example, the cloud computing financial server 100 may repeatedly sample the business big data samples corresponding to the same big data denoising label, and repeatedly sample the business big data samples from the business big data samples corresponding to the same big data denoising label until the number of the business big data samples reaches a preset number.

In the above embodiment, in order to avoid the influence on the big data denoising decision model training in the model training process caused by the unbalanced number of the business big data samples corresponding to different big data denoising labels, the business big data samples are retransferred. Therefore, the number of the service big data samples corresponding to the same big data denoising label is controlled to be the preset number, the model training effect and efficiency of the big data denoising decision model can be improved, and the trained big data denoising decision model can accurately predict the service big data.

In the above embodiment, due to the presentation structure of the information push configuration information and the loss function value of the content quantity of the information push configuration information, the information push configuration information with different presentation structures is split according to the intention requirement to obtain the candidate push element set, and the push element set corresponding to the information push configuration information is obtained in different manners. When the model training is carried out on the big data denoising decision model through the obtained pushing element set, the influence on the model training caused by the presentation structure of the information pushing configuration information or the quantity difference of the information pushing configuration information and the like can be avoided, and the prediction accuracy of the big data denoising decision model for business big data prediction can be improved.

In a possible design example, the big data denoising decision characteristics output by the decision units are fused to obtain a fused big data denoising decision characteristic. Predicting the fused big data denoising decision characteristics into denoising relevance parameters corresponding to each preset big data denoising label through a prediction module in a big data denoising decision model, selecting the maximum denoising relevance parameter from the predicted denoising relevance parameters, and outputting the preset big data denoising label corresponding to the maximum denoising relevance parameter as to-be-determined prediction information.

For example, after the cloud computing financial server 100 inputs a plurality of groups of pushed element sets of a business big data sample into decision units in a big data denoising decision model, each decision unit outputs a big data denoising decision feature corresponding to the input pushed element set. And fusing the big data denoising decision characteristics output by the plurality of decision units according to the sequence of outputting the big data denoising decision characteristics to obtain fused big data denoising decision characteristics. The cloud computing financial server 100 takes the fused big data denoising decision-making characteristics as the input of a prediction module, and after dimension reduction is performed on the fused big data denoising decision-making characteristics through the prediction module in the big data denoising decision-making model, the fused big data denoising decision-making characteristics are mapped into denoising correlation parameters corresponding to each preset big data denoising label. The cloud computing financial server 100 predicts the service big data samples to the denoising relevance parameters of each preset big data denoising label for one-by-one comparison, determines the maximum denoising relevance parameter in the denoising relevance parameters, and outputs the preset big data denoising label corresponding to the maximum denoising relevance parameter as undetermined prediction information of the big data denoising label belonging to the service big data sample.

In the above embodiment, the big data denoising decision characteristics output by the multiple decision units are fused to obtain a fused big data denoising decision characteristic, the fused big data denoising decision characteristic is predicted to be a denoising correlation parameter corresponding to each preset big data denoising label through a prediction module in a big data denoising decision model, and then the preset big data denoising label corresponding to the largest denoising correlation parameter is output to be undetermined prediction information of the big data denoising label belonging to the business big data sample. The information push configuration information in the business big data sample can be fully utilized, so that the information push configuration information of each information push service can be mutually verified and supplemented, and the prediction accuracy of a big data denoising decision model for business big data prediction is improved.

In one possible design example, further to step S140, this may be achieved by the following exemplary substeps, described in detail below.

Step S141, acquiring service data to be denoised, including at least one service data area, sent by a denoising service, acquiring noise service characteristic data of the service data area, and respectively acquiring a global denoising operation rule and an initial block denoising operation rule of the service data area based on a software service denoising mode and a non-software service denoising mode according to the noise service characteristic data.

In this embodiment, a plurality of cloud computing-based denoising services are run in the cloud computing financial server 100, and the denoising services are used for denoising and reporting abnormal behavior information.

The global denoising operation rule can be a denoising operation rule used for describing global information of a service data area, and the global denoising operation rule can represent the global denoising information, pay attention to the global denoising property, and have strong noise. The block denoising operation rule may be a denoising operation rule for describing unit information of the service data region, and may be a denoising operation rule corresponding to at least one unit process, a rule attribute of the block denoising operation rule may be less than that of the global denoising operation rule, and a region concerned by the block denoising operation rule is smaller, so as to generate more denoising details.

In a possible implementation manner, taking the privacy authorized data area as an example, the global denoising operation rule may be a global denoising operation rule of the privacy authorized data element that represents the condition of the global privacy authorized data element, the global denoising operation rule of the privacy authorized data element includes information of the fuzzy global privacy authorized data element, the blocking denoising operation rule may be a key denoising node denoising operation rule that represents the service access unit, the service reading unit, and the service writing unit, and the key denoising operation rule includes more specific unit area detail information.

In a possible implementation manner, the global denoising operation rule based on the software service denoising mode may be a denoising operation rule of the global software service denoising mode, and the global denoising operation rule based on the non-software service denoising mode may be a denoising operation rule of the global non-software service denoising mode. The initial block de-noising operation rule based on the software service de-noising mode can be a de-noising operation rule of the initial block software service de-noising mode, and the initial block de-noising operation rule based on the non-software service de-noising mode can be a de-noising operation rule of the initial unit non-software service de-noising mode.

In a possible implementation manner, the denoising operation rule of the global software service denoising mode and the denoising operation rule of the initial blocking software service denoising mode of the service data region can be obtained according to the software service denoising mode characteristic information, and the denoising operation rule of the global non-software service denoising mode and the denoising operation rule of the initial unit non-software service denoising mode of the service data region can be obtained according to the non-software service denoising mode characteristic information.

And S142, performing denoising label supplement processing on the initial block denoising operation rule to obtain a target block denoising operation rule.

In this embodiment, the denoising label supplementation processing is performed on the initial block denoising operation rule, and the denoising obtained after the denoising label supplementation processing is used as the target block denoising operation rule. For example, the denoising operation rule of the denoising mode of the initial blocking software service and the denoising operation rule of the denoising mode of the non-software service of the initial unit may be subjected to denoising label supplementary processing, so as to obtain the denoising operation rule of the denoising mode of the target blocking software service and the denoising operation rule of the denoising mode of the non-software service of the target unit as the target blocking denoising operation rule.

The supplementary processing of the de-noising label can refer to supplementary processing of a supplementary de-noising label set in the de-noising process. The complementary processing may be similar de-noised tag addition processing that performs de-noised tags. The denoising label can refer to denoising label description information generated in a malicious information denoising process, however, in many cases, many denoising labels have other similar associated denoising labels, so that a subsequent denoising model is updated by performing supplementary processing on a supplementary denoising label set in the denoising process, and more data information of denoising label dimensions can be increased.

And S143, performing rule splicing on the global denoising operation rule and the target block denoising operation rule respectively based on the software service denoising mode and the non-software service denoising mode to obtain target software service denoising rule configuration information and target non-software service denoising rule configuration information.

The target software service denoising rule configuration information is software service denoising rule configuration information obtained by integrating a global software service denoising mode feature and a blocking software service denoising mode feature, and the target non-software service denoising rule configuration information is non-software service denoising rule configuration information obtained by integrating a global non-software service denoising mode feature and a unit non-software service denoising mode feature. In addition, the target software service denoising rule configuration information and the target non-software service denoising rule configuration information can be both strategy node configuration sets.

In this embodiment, the global denoising operation rule and the target blocking denoising operation rule are regularly spliced, so that target denoising information including global features and unit features can be obtained, where the target denoising information includes target software service denoising rule configuration information and target non-software service denoising rule configuration information.

In a possible implementation manner, the global denoising operation rule and the target block denoising operation rule are regularly spliced based on the software service denoising mode to obtain target software service denoising rule configuration information, and the global denoising operation rule and the target block denoising operation rule are regularly spliced based on the non-software service denoising mode to obtain target non-software service denoising rule configuration information.

In a possible implementation manner, denoising label supplementation processing can be performed on the global denoising operation rule, and the global denoising operation rule after denoising label supplementation processing and the target block denoising operation rule are subjected to rule splicing to obtain corresponding target software service denoising rule configuration information and target non-software service denoising rule configuration information.

And S144, updating the denoising model according to the target software service denoising rule configuration information and the target non-software service denoising rule configuration information to obtain a target denoising model, and denoising the service data to be denoised by the target denoising model.

In a possible implementation manner, the template content of the block denoising operation rule is often less than the global denoising operation rule, and if the global denoising operation rule and the block denoising operation rule are to be regularly spliced, the template contents of the two are required to be the same, and then the operation rule segments on the same template node are fused to obtain denoising information fused with the global feature and the unit feature. Based on the above, the rule attribute of the block denoising operation rule needs to be unified, so that the contents of the block denoising operation rule and the global denoising operation rule template are consistent.

Based on the above steps, the embodiment respectively obtains the global denoising operation rule and the initial block denoising operation rule based on the software service denoising mode and the non-software service denoising mode in the service data region according to the noise service characteristic data, and performs denoising label supplement processing on the initial block denoising operation rule to obtain the target block denoising operation rule, and regularly splices the global denoising operation rule and the target block denoising operation rule based on the software service denoising mode and the non-software service denoising mode respectively to obtain the target software service denoising rule configuration information and the target non-software service denoising rule configuration information supplemented by the denoising label, so that the target denoising model obtained by updating according to the target software service denoising rule configuration information and the target non-software service denoising rule configuration information can supplement more data information of the denoising label dimension, and improving the subsequent denoising effect.

In a possible implementation manner, for step S142, in the process of performing denoising label supplementation processing on the initial block denoising operation rule to obtain the target block denoising operation rule, the following exemplary sub-steps may be implemented, which are described in detail below.

In the substep S1421, denoising label distribution of the initial block denoising operation rule is obtained.

In the substep S1422, a target denoising label distribution having an association relationship with the denoising label distribution is matched from a preconfigured denoising label distribution preset set.

In this embodiment, the existence of the association relationship may refer to existence of a hierarchical relationship or a parallel relationship, for example, for the denoising tag a, the target denoising tag having the association relationship with the denoising tag a may refer to other denoising tags a2, denoising tags A3 and the like at a level above the denoising tag a1, the denoising tag a1 may have a parallel relationship with the denoising tag a2 and the denoising tag A3, or the denoising tag a at a level above the denoising tag a1 may also be the target denoising tag having the association relationship with the denoising tag a.

And a substep S1423 of supplementing the denoising operation rule matched with the target denoising label distribution to the initial block denoising operation rule according to the target denoising label distribution to obtain a target block denoising operation rule.

In this embodiment, the denoising operation rule matched with the target denoising tag distribution corresponding to the denoising service may be supplemented to the initial block denoising operation rule to obtain the target block denoising operation rule.

In one possible implementation manner, the service data region may be a privacy authorized data region, and the target block denoising operation rule may include a key denoising operation rule corresponding to a key denoising node of a privacy authorized data element. In this way, in step S143, the rule attribute unification may be performed on the supplementary operation rule partitions of the key denoising operation rules, respectively, to obtain a unified key denoising operation rule having the same content as the global denoising operation rule template, the unified denoising operation rules are merged to obtain a member denoising operation rule of the privacy authorized data element, and the member denoising operation rules of the global denoising operation rule and the privacy authorized data element are regularly spliced based on the software service denoising mode and the non-software service denoising mode, respectively, to obtain target software service denoising rule configuration information and target non-software service denoising rule configuration information.

For another example, in another possible implementation, the global denoising operation rule includes a denoising operation rule of a global software service denoising mode and a denoising operation rule of a global non-software service denoising mode, and the target block denoising operation rule includes a denoising operation rule of a block software service denoising mode and a denoising operation rule of a unit non-software service denoising mode. Thus, in step S143, the denoising operation rule of the global software service denoising mode and the denoising operation rule of the blocking software service denoising mode may be regularly spliced, the denoising operation rule of the regular splicing is configured to integrate the global software service denoising mode feature and the blocking software service denoising mode feature to obtain the target software service denoising rule configuration information, the denoising operation rule of the global non-software service denoising mode and the denoising operation rule of the unit non-software service denoising mode are regularly spliced in each denoising enabling flow, and the denoising operation rule of each denoising enabling flow rule splicing is configured to integrate the global non-software service denoising mode feature and the unit non-software service denoising mode feature to obtain the target non-software service denoising rule configuration information.

The global denoising operation rule and the target block denoising operation rule both can correspond to at least one denoising enabling process, so that the denoising operation rule of the global software service denoising mode and the denoising operation rule of the block software service denoising mode can be regularly spliced in each denoising enabling process, and the denoising operation rule spliced by each denoising enabling process rule is configured to integrate the global software service denoising mode characteristic and the block software service denoising mode characteristic, so that target software service denoising rule configuration information is obtained.

In a possible implementation manner, in the step S141, in the process of obtaining noise service characteristic data of the service data region, data-item-by-data-item denoising feature extraction may be performed on the service data region, and then software service denoising mode feature information and non-software service denoising mode feature information of the service data region are obtained according to a result of the data-item-by-data-item denoising feature extraction, and are used as the noise service characteristic data.

In a possible implementation manner, still referring to step S141, in the process of respectively obtaining a global denoising operation rule and an initial block denoising operation rule of a service data region based on a software service denoising mode and a non-software service denoising mode according to noise service characteristic data, a global denoising index model may perform denoising indexing on the service data region according to the noise service characteristic data to obtain a global denoising operation rule, and a unit denoising index model performs denoising indexing on the service data region according to the noise service characteristic data to obtain an initial block denoising operation rule.

The unit denoising index model can comprise a key denoising node denoising index model of the privacy authorization data element. The denoising index is to capture a denoising operation rule corresponding to a key denoising node from an original denoising record data log, and specifically, a data positioning and indexing scheme in the prior art may be referred to, which is not limited here.

In this way, in the process of denoising and indexing the service data region by the unit denoising index model according to the noise service characteristic data to obtain the initial block denoising operation rule, the key denoising node denoising index model of the privacy authorization data element can denoise and index the service data region according to the noise service characteristic data, and the obtained key denoising node denoising operation rule is determined as the initial block denoising operation rule.

In a possible implementation manner, for step S144, in the process of performing denoising model updating processing according to the target software service denoising rule configuration information and the target non-software service denoising rule configuration information to obtain the target denoising model, the following exemplary sub-steps may be implemented, which are described in detail below.

And a substep S1441, mapping the target software service denoising rule configuration information and the target non-software service denoising rule configuration information into target denoising rule configuration information according to each denoising node combination, acquiring at least two denoising nodes included in the target denoising rule configuration information, and acquiring denoising feature information of each denoising node in the at least two denoising nodes.

And a substep S1442, determining global weighted denoising feature information aiming at the target denoising rule configuration information based on the denoising feature information of each denoising node.

And a substep S1443, determining distributed denoising characteristic information of the target denoising rule configuration information on the denoising distribution label based on the denoising characteristic label component of the denoising characteristic information of each denoising node on the denoising distribution label.

And a substep S1444 of determining a deep learning denoising model of the target denoising rule configuration information aiming at the denoising calling operation associated with the denoising distribution label according to the global weighted denoising feature information and the distributed denoising feature information, performing denoising model updating processing based on the deep learning denoising model to obtain a target denoising model, and denoising the service data to be denoised by the target denoising model.

Therefore, the deep learning denoising model related to the target denoising rule configuration information can be determined according to the relation between the distributed denoising feature information and the global weighting denoising feature information of the target denoising rule configuration information, so that the denoising model is updated, the iterative updating of the denoising cooperation rule among multiple safe denoising systems is conveniently carried out on the denoising model, and the denoising effect is improved.

In one possible implementation, for sub-step S1442, this may be achieved by the following exemplary embodiments:

(1) and acquiring label characteristic vectors of structured denoising distribution labels and label characteristic vectors of unstructured denoising distribution labels, which are respectively included in denoising characteristic information of each denoising node.

(2) And determining the distribution of the unit label characteristic vectors corresponding to each denoising node based on the label characteristic vectors of the structured denoising distribution labels and the label characteristic vectors of the unstructured denoising distribution labels corresponding to each denoising node.

(3) And determining global weighted denoising feature information according to the unit label feature vector distribution corresponding to each denoising node and the node number of at least two denoising nodes.

In one possible implementation, the denoising distribution label may include a structured denoising distribution label, and the denoising feature label component of the denoising feature information of each denoising node on the denoising distribution label includes a label feature vector of the structured denoising distribution label.

Thus, for sub-step S1443, this may be achieved by the following exemplary embodiments:

(1) and acquiring label feature vectors of the structured denoising distribution labels of the denoising feature information of each denoising node on the structured denoising distribution labels respectively.

(2) And determining a structured denoising distribution label value corresponding to each denoising node according to the label feature vector of the structured denoising distribution label of each denoising node on the structured denoising distribution label.

(3) And determining distributed denoising characteristic information according to the structured denoising distribution label value corresponding to each denoising node and the node number of at least two denoising nodes.

In one possible implementation, the denoising distribution label may further include an unstructured denoising distribution label, and the denoising feature label component of the denoising feature information of each denoising node on the denoising distribution label includes a label feature vector of the unstructured denoising distribution label.

(4) and acquiring label feature vectors of the unstructured denoising distribution labels of the denoising feature information of each denoising node on the unstructured denoising distribution labels respectively.

(5) And determining the unstructured denoising distribution label value corresponding to each denoising node according to the label feature vector of the unstructured denoising distribution label of each denoising node on the unstructured denoising distribution label.

(6) And determining distributed denoising characteristic information according to the unstructured denoising distribution label value corresponding to each denoising node and the node number of at least two denoising nodes.

Fig. 3 is a schematic functional module diagram of a big data denoising device 300 for cloud computing services according to an embodiment of the present disclosure, and in this embodiment, functional modules of the big data denoising device 300 for cloud computing services may be divided according to the method embodiment executed by the cloud computing financial server 100, that is, the following functional modules corresponding to the big data denoising device 300 for cloud computing services may be used to execute the method embodiments executed by the cloud computing financial server 100. The big data denoising apparatus 300 for cloud computing services may include an obtaining module 310, an input module 320, a prediction module 330, and a denoising module 340, and the functions of the functional modules of the big data denoising apparatus 300 for cloud computing services are described in detail below.

The obtaining module 310 is configured to obtain service big data to be denoised, and obtain information push configuration information of a plurality of information push services mapped by the service big data, where the service big data is a service data set collected based on a cloud computing service. The obtaining module 310 may be configured to perform the step S110, and the detailed implementation of the obtaining module 310 may refer to the detailed description of the step S110.

The input module 320 is configured to analyze the information push configuration information into a corresponding push element set, and input the push element set into a corresponding decision unit in the trained big data denoising decision model. Each decision unit at least comprises a decision model, and the decision model of each decision unit processes a push element set corresponding to the information push service. The input module 320 may be configured to perform the step S120, and the detailed implementation of the input module 320 may refer to the detailed description of the step S120

And the prediction module 330 is configured to perform prediction according to the big data denoising decision characteristics output by the multiple decision units through a prediction module in the big data denoising decision model, and output a big data denoising tag to which the business big data belongs. The prediction module 330 may be configured to perform the step S130, and the detailed implementation of the prediction module 330 may refer to the detailed description of the step S130.

And the denoising module 340 is configured to perform big data denoising on the business big data according to the big data denoising tag to which the business big data belongs. The denoising module 340 may be configured to perform the step S140, and the detailed implementation of the denoising module 340 may refer to the detailed description of the step S140.

It should be noted that the division of each module of the above apparatus is only a logical division, and the actual implementation may be wholly or partially integrated into one physical business state object, or may be physically separated. And these modules may all be implemented in software invoked by a processing element. Or may be implemented entirely in hardware. And part of the modules can be realized in the form of calling software by the processing element, and part of the modules can be realized in the form of hardware. For example, the obtaining module 310 may be a processing element separately set up, or may be implemented by being integrated into a chip of the apparatus, or may be stored in a memory of the apparatus in the form of program code, and the processing element of the apparatus calls and executes the functions of the obtaining module 310. Other modules are implemented similarly. In addition, all or part of the modules can be integrated together or can be independently realized. The processing element described herein may be an integrated circuit having signal processing capabilities. In implementation, each step of the above method or each module above may be implemented by an integrated logic circuit of hardware in a processor element or an instruction in the form of software.

Fig. 4 illustrates a hardware structure diagram of a cloud computing financial server 100 for implementing the big data denoising method for cloud computing business, according to an embodiment of the present disclosure, and as shown in fig. 4, the cloud computing financial server 100 may include a processor 110, a machine-readable storage medium 120, a bus 130, and a transceiver 140.

In a specific implementation process, at least one processor 110 executes computer-executable instructions stored in a machine-readable storage medium 120 (for example, an obtaining module 310, an input module 320, a prediction module 330, and a denoising module 340 included in a big data denoising apparatus 300 for cloud computing services shown in fig. 3), so that the processor 110 may execute a big data denoising method for cloud computing services according to the above method embodiment, where the processor 110, the machine-readable storage medium 120, and the transceiver 140 are connected by a bus 130, and the processor 110 may be configured to control transceiving actions of the transceiver 140, so as to perform data transceiving with the aforementioned information service terminal 200.

For a specific implementation process of the processor 110, reference may be made to the above-mentioned method embodiments executed by the cloud computing financial server 100, which implement principles and technical effects similar to each other, and details of this embodiment are not described herein again.

In the embodiment shown in fig. 4, it should be understood that the Processor may be a Central Processing Unit (CPU), other general purpose Processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), etc. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like. The steps of a method disclosed in connection with the present invention may be embodied directly in a hardware processor, or in a combination of the hardware and software modules within the processor.

The machine-readable storage medium 120 may comprise high-speed RAM memory and may also include non-volatile storage NVM, such as at least one disk memory.

The bus 130 may be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended ISA (EISA) bus, among others. The bus 130 may be divided into an address bus, a data bus, a control bus, and the like. For ease of illustration, the buses in the figures of the present application are not limited to only one bus or one type of bus.

In addition, a readable storage medium is provided, and computer execution instructions are stored in the readable storage medium, and when a processor executes the computer execution instructions, the big data denoising method for cloud computing service is implemented as above.

Finally, it should be understood that the examples in this specification are only intended to illustrate the principles of the examples in this specification. Other variations are also possible within the scope of this description. Thus, by way of example, and not limitation, alternative configurations of the embodiments of the specification can be considered consistent with the teachings of the specification. Accordingly, the embodiments of the present description are not limited to only those embodiments explicitly described and depicted herein.

Claims

1. A big data denoising method for cloud computing business is applied to a cloud computing financial server, the cloud computing financial server is in communication connection with a plurality of information service terminals, the cloud computing financial server is realized according to a cloud computing platform, and the method comprises the following steps:

2. The big data denoising method for cloud computing service according to claim 1, wherein the step of obtaining information push configuration information of a plurality of information push services to which the service big data is mapped comprises:

3. The big data denoising method for cloud computing service according to claim 1, wherein the step of parsing the information push configuration information into a corresponding push element set comprises:

4. The big data denoising method for cloud computing services according to any one of claims 1 to 3, wherein the step of outputting a big data denoising label to which the big data of the service belongs by predicting according to the big data denoising decision characteristics output by the plurality of decision units through a prediction module in the big data denoising decision model comprises:

5. The big data denoising method for cloud computing services according to claim 1, wherein each decision unit includes a preset input condition, a corresponding push element set of each information push service includes a push activation tag, and the step of inputting the push element set into a corresponding decision unit in a trained big data denoising decision model includes:

reading a push activation tag of the set of push elements;

6. The big data denoising method for cloud computing services according to claim 1, wherein the big data denoising decision model is obtained by training in the following way:

7. The big data denoising method for the cloud computing service as claimed in claim 1, wherein the step of denoising the big data of the service according to the big data denoising tag to which the big data of the service belongs comprises:

8. The big data denoising method for cloud computing service according to claim 7, wherein the step of performing denoising label supplementary processing on the initial block denoising operation rule to obtain a target block denoising operation rule comprises:

9. The big data denoising method for cloud computing services according to claim 7, wherein the service data region is a privacy authorized data region, the target blocking denoising operation rule comprises a key denoising operation rule corresponding to a key denoising node of a privacy authorized data element, and the step of regularly splicing the global denoising operation rule and the target blocking denoising operation rule based on a software service denoising mode and a non-software service denoising mode respectively to obtain target software service denoising rule configuration information and target non-software service denoising rule configuration information comprises: respectively carrying out rule attribute unification on supplementary operation rule partitions of each key denoising operation rule to obtain a unified key denoising operation rule with the same content as the global denoising operation rule template, combining the unified key denoising operation rules to obtain a member denoising operation rule of a privacy authorized data element, and carrying out rule splicing on the global denoising operation rule and the member denoising operation rule of the privacy authorized data element based on a software service denoising mode and a non-software service denoising mode respectively to obtain target software service denoising rule configuration information and target non-software service denoising rule configuration information;

10. A cloud computing financial server, characterized in that the cloud computing financial server comprises a processor, a machine-readable storage medium, and a network interface, the machine-readable storage medium, the network interface, and the processor are connected through a bus system, the network interface is used for being connected with at least one information service terminal in a communication manner, the machine-readable storage medium is used for storing programs, instructions, or codes, and the processor is used for executing the programs, instructions, or codes in the machine-readable storage medium to execute the big data denoising method for cloud computing business of any one of claims 1 to 9.