CN114647790A

CN114647790A - Big data mining method and cloud AI (Artificial Intelligence) service system applied to behavior intention analysis

Info

Publication number: CN114647790A
Application number: CN202210336753.4A
Authority: CN
Inventors: 刘中申
Original assignee: Individual
Current assignee: Individual
Priority date: 2022-04-01
Filing date: 2022-04-01
Publication date: 2022-06-21

Abstract

The embodiment of the application provides a big data mining method and a cloud AI service system applied to behavior intention analysis, wherein big data of user behaviors are analyzed through a target behavior intention mining network trained in advance, corresponding behavior intention mining characteristics are output, behavior intention thermal labels corresponding to target users are determined according to the behavior intention mining characteristics of different target users, pushing grouping distribution is carried out on the target users based on the behavior intention thermal labels corresponding to the target users, target user clusters under each target pushing grouping are determined, and information pushing is carried out on business service pages corresponding to the target user clusters under each target pushing grouping. In this way, the present application can improve the directionality of information push content by further mining the thermal labels for pushing reference bases after mining behavioral intentions, thereby pushing information after grouping users.

Description

Big data mining method and cloud AI (Artificial Intelligence) service system applied to behavior intention analysis

Technical Field

The application relates to the technical field of big data and AI, in particular to a big data mining method and a cloud AI service system applied to behavior intention analysis.

Background

With the development of cloud computing and big data technology, various cloud services are developed at the cloud end to meet the internet requirements of users, and the cloud service development method becomes the core competitiveness of more and more internet service providers. In the process of using the cloud services, a large amount of user behavior data can be generated by users, and under the condition of obtaining the authorization permission of the users, the internet service providers can analyze the user behavior data so as to further mine preference intentions of different users, and further provide more matched content services for the users. In the related art, the directionality of the information push content needs to be improved for the conventional information push manner of the user.

Disclosure of Invention

In order to overcome at least the above defects in the prior art, an object of the present application is to provide a big data mining method and a cloud AI service system applied to behavior intention analysis.

In a first aspect, the present application provides a big data mining method applied to behavior intention analysis, which is applied to a cloud AI service system, and the method includes:

acquiring user behavior big data of a target user to be mined, analyzing the user behavior big data according to a pre-trained target behavior intention mining network, and outputting corresponding behavior intention mining characteristics;

determining behavior intention thermal labels corresponding to the target users according to the behavior intention mining characteristics of different target users;

pushing and grouping distribution is carried out on each target user based on behavior intention thermal labels corresponding to each target user so as to determine a target user cluster under each target pushing group, wherein each target pushing group corresponds to one pushing rule, and the pushing rule corresponds to the latest pushing strategy of the corresponding behavior intention thermal label;

and carrying out information push on the business service page corresponding to the target user cluster under each target push group.

In some concepts, the step of obtaining the user behavior big data to be mined, analyzing the user behavior big data according to a pre-trained target behavior intention mining network, and outputting corresponding behavior intention mining characteristics includes:

acquiring user behavior training data for updating a network of a first behavior intention mining network, and taking training data member clusters in the user behavior training data as candidate training data member clusters;

outputting member cluster variables of the candidate training data member clusters according to the mining of the first behavior intention mining network, loading the member cluster variables of the candidate training data member clusters to a training effectiveness decision network in mapping connection with the first behavior intention mining network, calling the training effectiveness decision network to perform training effectiveness decision on the candidate training data member clusters based on the network updating round number of the first behavior intention mining network and effective value evaluation configuration information set by the training effectiveness decision network, and generating training effectiveness decision information of the candidate training data member clusters;

performing behavior intention characteristic decision on member cluster variables of the candidate training data member cluster to generate behavior intention characteristic variables of the candidate training data member cluster, performing combined training cost value evaluation on the user behavior training data based on the behavior intention characteristic variables of the candidate training data member cluster and the training effectiveness decision information, and outputting a combined training cost value of the user behavior training data;

updating the network weight information of the first behavior intention mining network based on the combined training cost value, generating a second behavior intention mining network for performing behavior intention characteristic analysis on target user behavior data according to the first behavior intention mining network after the network weight information is updated, and performing behavior intention characteristic analysis on user behavior big data by using the second behavior intention mining network as the target behavior intention mining network.

For example, in some concepts, the method further comprises:

acquiring full-dimensional unit behavior data, and configuring a negative association behavior data area corresponding to each unit behavior data cluster in the full-dimensional unit behavior data;

configuring a multi-dimensional training data member cluster for network updating of a first behavior intention mining network according to each unit behavior data cluster in the full-dimensional unit behavior data and a negative association behavior data area corresponding to each unit behavior data cluster;

and performing training unit distribution on training data member clusters in the multi-dimensional training data member clusters based on the total negative correlation statistics in the negative correlation behavior data area to generate user behavior training data corresponding to the full-dimensional unit behavior data.

For example, in some concepts, the acquiring full-dimensional unit behavior data and configuring a negative association behavior data region corresponding to each unit behavior data cluster in the full-dimensional unit behavior data includes:

acquiring reference user behavior big data for splitting unit behavior data, splitting the reference user behavior big data to generate a first number of unit behavior data clusters corresponding to the reference user behavior big data, clustering the first number of unit behavior data clusters according to behavior category information of the first number of unit behavior data clusters to generate a second number of clustering units corresponding to the full-dimensional unit behavior data, wherein the first number of unit behavior data clusters is used as the full-dimensional unit behavior data; one clustering unit corresponds to one behavior category information;

acquiring a unit behavior data cluster i from the full-dimension unit behavior data, and respectively taking a third number of clustering units extracted from the clustering units corresponding to the unit behavior data cluster i as target clustering units in the second number of clustering units; wherein i is a positive integer no greater than a first number; the number of the target clustering units is a third number;

extracting third quantity and fourth quantity of unit behavior data from unit association behavior data areas corresponding to a third quantity of target clustering units to serve as first unit behavior data, configuring a first type of negative association behavior data area corresponding to a unit behavior data cluster i according to the extracted third quantity and fourth quantity of first unit behavior data, extracting third quantity and fourth quantity of unit behavior data from a global association behavior data area corresponding to a second quantity of clustering units to serve as second unit behavior data, and configuring a second type of negative association behavior data area corresponding to the unit behavior data cluster i according to the extracted third quantity and fourth quantity of second unit behavior data; the first unit behavior data in the first type negative correlation behavior data area and the unit behavior data in the unit behavior data cluster i have the same behavior category information; the unit behavior data in the second type negative correlation behavior data area and the unit behavior data in the unit behavior data cluster i have different behavior category information;

and taking the first type negative correlation behavior data area and the second type negative correlation behavior data area as negative correlation behavior data areas of the unit behavior data cluster i.

For example, in some embodiments, the clustering the first number of cell behavior data clusters according to the behavior category information of the first number of cell behavior data clusters to generate the second number of clustering units corresponding to the full-dimensional cell behavior data includes:

loading the first quantity of unit behavior data clusters contained in the full-dimension unit behavior data to a behavior type decision model, and deciding the behavior type information of each unit behavior data cluster in the first quantity of unit behavior data clusters by the behavior type decision model; the behavior type information of each unit behavior data cluster is determined when the behavior type quantity of the behavior type information of the unit behavior data in the corresponding unit behavior data cluster reaches a preset quantity; the preset number is determined by the number of the unit behavior data in the corresponding unit behavior data cluster;

and in the first number of unit behavior data clusters, based on the behavior category information of each unit behavior data cluster, allocating the unit behavior data clusters with the same behavior category information to the same clustering unit, and outputting a second number of clustering units corresponding to the full-dimension unit behavior data.

For example, in some concepts, the full-dimensional unit behavior data includes a first number of unit behavior data clusters, and one unit behavior data cluster includes a fourth number of unit behavior data; the negative associated behavior data areas corresponding to the unit behavior data clusters respectively comprise a first type negative associated behavior data area and a second type negative associated behavior data area;

configuring a multidimensional training data member cluster for network updating of a first behavior intention mining network according to each unit behavior data cluster in the full-dimensional unit behavior data and a negative association behavior data area corresponding to each unit behavior data cluster, wherein the multidimensional training data member cluster comprises:

acquiring a target unit behavior data cluster from each unit behavior data cluster in the full-dimension unit behavior data;

selecting one unit behavior data from the fourth quantity of unit behavior data of the target unit behavior data cluster as anchor unit behavior data of the target unit behavior data cluster, taking unit behavior data except the anchor unit behavior data from the fourth quantity of unit behavior data of the target unit behavior data cluster as positive unit behavior data of the target unit behavior data cluster, and outputting a positive unit behavior data combination of the target unit behavior data cluster according to the anchor unit behavior data of the target unit behavior data cluster and the positive unit behavior data of the target unit behavior data cluster; the number of positive unit behavior data combinations of the target unit behavior data cluster is a fourth number (a fourth number-1) 1/2;

extracting 2 x a fifth number of unit behavior data from a negative association behavior data area corresponding to the target unit behavior data cluster as negative unit behavior data of the target unit behavior data cluster; the negative unit behavior data of the target unit behavior data cluster comprises a fifth number of first negative unit behavior data and a fifth number of second negative unit behavior data; the fifth quantity of first negative unit behavior data is extracted from the first type of negative association behavior data area corresponding to the target unit behavior data cluster; the fifth quantity of second negative unit behavior data is extracted from a second type of negative associated behavior data region corresponding to the target unit behavior data cluster;

configuring a training data member cluster of the target unit behavior data cluster based on the positive unit behavior data combination of the target unit behavior data cluster and the negative unit behavior data of the target unit behavior data cluster, and generating the training data member cluster of each unit behavior data cluster until each unit behavior data cluster in the full-dimensional unit behavior data is used as the target unit behavior data cluster; the number of the training data member clusters of each unit behavior data cluster is fourth number (fourth number-1) fifth number;

and taking the training data member cluster of each unit behavior data cluster as a multi-dimensional training data member cluster for updating the network of the first behavior intention mining network.

For example, in some concepts, the first type of negative association behavior data region corresponding to the target unit behavior data cluster includes a third number × a fourth number of first unit behavior data; the second type negative association behavior data area corresponding to the target unit behavior data cluster comprises a third quantity and a fourth quantity of second unit behavior data;

the method further comprises the following steps:

wandering to select one unit behavior data from the fourth number of unit behavior data of the target unit behavior data cluster as positive unit behavior data of the target unit behavior data cluster, and regarding the unit behavior data except the wandered positive unit behavior data from the fourth number of unit behavior data of the target unit behavior data cluster as anchor unit behavior data of the target unit behavior data cluster;

searching first unit behavior data which is the same as the anchor unit behavior data of the target unit behavior data cluster in the third quantity-fourth quantity of first unit behavior data, and taking the searched first unit behavior data which is the same as the anchor unit behavior data of the target unit behavior data cluster as collaborative behavior extraction data;

taking first unit behavior data other than the cooperative behavior extraction data as target behavior extraction data in the third number-fourth number of first unit behavior data; the number of data of the target behavior extraction data is (third number + fourth number-1);

determining a first feature cost value between the collaborative behavior extraction data and (third quantity, fourth quantity and-1) target behavior extraction data, taking the target behavior extraction data corresponding to the determined minimum first feature cost value as local first negative unit behavior data of the collaborative behavior extraction data, determining a second feature cost value between the collaborative behavior extraction data and third quantity, fourth quantity and second unit behavior data, and taking the second unit behavior data corresponding to the determined minimum second feature cost value as global first negative unit behavior data of the collaborative behavior extraction data;

and updating the first negative unit behavior data of the target unit behavior data cluster in the negative unit behavior data of the target unit behavior data cluster according to the local first negative unit behavior data of the cooperative behavior extraction data and the global first negative unit behavior data of the cooperative behavior extraction data.

In a second aspect, an embodiment of the present application further provides a cloud AI service system, where the cloud AI service system includes at least one processor; and a memory communicatively coupled to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to execute to implement the big data mining method applied to behavioral intent analysis of the first aspect above.

Based on the above aspects, the user behavior big data is analyzed through a target behavior intention mining network finished by pre-training, corresponding behavior intention mining characteristics are output, behavior intention thermodynamic labels corresponding to all target users are determined according to the behavior intention mining characteristics of different target users, pushing grouping distribution is carried out on all target users based on the behavior intention thermodynamic labels corresponding to all target users, target user clusters under each target pushing grouping are determined, and information pushing is carried out on business service pages corresponding to the target user clusters under each target pushing grouping. In this way, the present application can improve the directionality of information push content by further mining the thermal labels for pushing reference bases after mining behavioral intentions, thereby pushing information after grouping users.

Drawings

Fig. 1 is a schematic architecture diagram of a big data mining system applied to behavior intention analysis according to an embodiment of the present application;

FIG. 2 is a schematic flow chart of a big data mining method applied to behavior intention analysis according to an embodiment of the present disclosure;

fig. 3 is a schematic block diagram of a structure of a cloud AI service system for implementing the big data mining method applied to behavior intention analysis according to the embodiment of the present disclosure.

Detailed Description

Fig. 1 is an architecture diagram of a big data mining system 10 applied to behavior intention analysis according to an embodiment of the present application. The big data mining system 10 applied to behavior intention analysis may include a cloud AI service system 100 and a cloud AI service system 200 communicatively connected to the cloud AI service system 100. The big data mining system 10 applied to behavior intention analysis shown in fig. 1 is only one possible example, and in other possible embodiments, the big data mining system 10 applied to behavior intention analysis may also include only at least some of the components shown in fig. 1 or may also include other components.

In this embodiment, the cloud AI service system 100 and the cloud AI service system 200 in the big data mining system 10 applied to behavior intention analysis may execute the big data mining method applied to behavior intention analysis according to the following method embodiments, and the detailed description of the following method embodiments may be referred to for the specific execution steps of the cloud AI service system 100 and the cloud AI service system 200.

Step S110, acquiring user behavior big data of a target user to be mined, analyzing the user behavior big data according to a target behavior intention mining network trained in advance, and outputting corresponding behavior intention mining characteristics.

In this embodiment, the user behavior big data may be a set of behavior data (such as attention behavior data, click behavior data, sharing behavior data, and the like) generated by any target user during the use process of the online service, and the behavior data may reflect a behavior intention (such as an interest point, a preference point, an inflection point, and the like) of the target user, so as to facilitate subsequent reference basis collection for information pushing.

And step S120, determining behavior intention thermal labels corresponding to the target users according to the behavior intention mining characteristics of different target users.

In this embodiment, the behavior intention mining feature may represent a feature point set formed by behavior intentions, and in order to improve information analysis efficiency and accuracy, the occurrence frequency of feature points formed by each behavior intention may be counted, and then, feature points formed by behavior intentions greater than a set frequency may be determined as behavior intention thermal labels corresponding to respective target users.

Step S130, pushing grouping distribution is carried out on each target user based on the behavior intention heat label corresponding to each target user, so that a target user cluster under each target pushing grouping is determined.

Each target pushing group corresponds to one pushing rule, and the pushing rule corresponds to the latest pushing strategy of the corresponding behavior intention thermal label. In this embodiment, for different behavior intention thermal tags, an operator may flexibly designate a latest push policy, which may be changed in real time and may be set by the operator, for example, the latest push policy may configure different content tags of push content and corresponding push strengths.

Step S140, performing information push on the service page corresponding to the target user cluster under each target push group.

In this embodiment, after determining the target user cluster in each target push group, information push may be performed on the service page corresponding to the target user cluster in each target push group.

Based on the steps, the user behavior big data are analyzed through a pre-trained target behavior intention mining network, corresponding behavior intention mining characteristics are output, behavior intention thermal labels corresponding to all target users are determined according to the behavior intention mining characteristics of different target users, pushing grouping distribution is carried out on all target users based on the behavior intention thermal labels corresponding to all target users, target user clusters under each target pushing group are determined, and information pushing is carried out on business service pages corresponding to the target user clusters under each target pushing group. In this way, the present application can improve the directionality of information push content by further mining the thermal labels for pushing reference bases after mining behavioral intentions, thereby pushing information after grouping users.

Wherein, step S110 of the above embodiment may be further implemented by the following exemplary steps.

Step S101, obtaining user behavior training data for updating the network of the first behavior intention mining network, and taking training data member clusters in the user behavior training data as candidate training data member clusters.

Before the step S101, the cloud AI service system (e.g., a distributed server system) may also perform the following steps in advance: for example, when the distributed server system obtains the full-dimensional unit behavior data, a negative association behavior data area corresponding to each unit behavior data cluster in the full-dimensional unit behavior data may be configured; the distributed server system can configure a multi-dimensional training data member cluster for network updating of the first behavior intention mining network according to each unit behavior data cluster in the full-dimensional unit behavior data and a negative association behavior data area corresponding to each unit behavior data cluster; the distributed server system can perform training unit distribution on training data member clusters in the multi-dimensional training data member clusters based on total negative correlation statistics in the negative correlation behavior data area to generate user behavior training data corresponding to full-dimensional unit behavior data.

The number of batches of user behavior training data acquired by the cloud AI service system (e.g., a distributed server system) for network updating the first behavior intention mining network may be T1. Then, for the T1 pieces of user behavior training data, the candidate training data members in the T1 pieces of user behavior training data may be collectively referred to as multi-dimensional candidate training data members, so that the multi-dimensional candidate training data members corresponding to the T1 pieces of user behavior training data may be subjected to behavior intention feature analysis according to the first behavior intention mining network and the training effectiveness decision network, so as to perform P (e.g., P = 20) round network convergence optimization on the T1 pieces of user behavior training data.

When the T1 user behavior training data are acquired in the embodiment of the present application, all training data member clusters in the T1 user behavior training data may be collectively referred to as multidimensional training data member clusters, and then the training data member clusters in the multidimensional training data member clusters may be collectively used as candidate training data member clusters, so that the following step S102 may be further performed subsequently.

The T1 pieces of user behavior training data are obtained by the cloud AI service system (e.g., a distributed server system) performing training unit allocation on training data member clusters in the multidimensional training data member clusters based on the total negative association statistics in the configured negative association behavior data area. In some embodiments, the total negative correlation statistic in each negative correlation behavior data region may be QX (here, QX =2 × third number × fourth number), and the number of training data member clusters in each user behavior training data may also be QX (i.e., 2 × third number × fourth number).

Since a training data member cluster needs to include three candidate training data members (i.e., one anchor unit behavior data, one positive unit behavior data, and one negative unit behavior data). Therefore, for each user behavior training data, the number of candidate training data members corresponding to the training data member cluster in each user behavior training data may be T2, where T2 may be 3 × QX =3 × 2 × third number × fourth number =6 × third number × fourth number. Accordingly, for the T1 pieces of user behavior training data, the number of all training data member clusters configured for the web update of the first behavior intention mining network (i.e., T3, the number of multi-dimensional training data member clusters described above), where T3= T1 × QX, and the number of multi-dimensional candidate training data members corresponding to the multi-dimensional training data member clusters T4= T1 × T2= T1 × 3 × QX =3 × T3.

The distributed server system can further predict behavior category information of the unit behavior data in each unit behavior data cluster according to the multi-behavior category decision model, and further can generate behavior category information of the corresponding unit behavior data cluster based on the behavior category information of the unit behavior data in each unit behavior data cluster.

In some embodiments, the distributed server system may intelligently predict behavior category information of unit behavior data in each unit behavior data cluster according to the multi-behavior category decision model, which means that when a large number of multi-dimensional training data member clusters used for network updating of the first behavior intention mining network are obtained in the embodiments of the present application, it is not necessary to manually label behavior category information of each training data member cluster in the multi-dimensional training data member clusters, and it is not necessary to manually label behavior category information of candidate training data members in each training data member cluster.

Optionally, in this embodiment of the application, when a new added unit behavior data cluster exists, the new added unit behavior data cluster and the first number of unit behavior data clusters are used together as new full-dimensional unit behavior data, so as to obtain a new multidimensional training data member cluster for performing network update on the first behavior intention mining network according to the new full-dimensional unit behavior data update.

The distributed server system can extract 2 × fifth quantity of unit behavior data from the negative association behavior data area corresponding to the target unit behavior data cluster as the negative unit behavior data of the target unit behavior data cluster; for example, the negative unit behavior data of the target unit behavior data cluster includes a fifth number of first negative unit behavior data and a fifth number of second negative unit behavior data; wherein a fifth number of the first negative unit behavior data are extracted from the first type of negative associated behavior data region corresponding to the target unit behavior data cluster; a fifth quantity of second negative unit behavior data is extracted from the second type negative association behavior data region corresponding to the target unit behavior data cluster; the distributed server system can configure and obtain a training data member cluster of a target unit behavior data cluster based on a positive unit behavior data combination of the target unit behavior data cluster and negative unit behavior data of the target unit behavior data cluster, and can obtain the training data member cluster of each unit behavior data cluster until each unit behavior data cluster in full-dimensional unit behavior data is used as the target unit behavior data cluster; the number of the training data member clusters of each unit behavior data cluster is fourth number (fourth number-1) fifth number; the distributed server system may use the training data member cluster of each unit behavior data cluster as a multidimensional training data member cluster for performing network update on the first behavior intention mining network, so that the following step S102 may be further performed after the distribution of the training units is performed.

Therefore, in the process of mining the first negative unit behavior data, the distributed server system may randomly extract the unit behavior data in the same clustering unit as the first negative unit behavior data of the anchor unit behavior data, may further output (third quantity, fourth quantity, 1) target behavior extraction data in a third quantity, fourth quantity, of the first unit behavior data included in the first negative unit behavior data region corresponding to the unit behavior data cluster when determining the positive unit behavior data in a certain unit behavior data cluster according to the wandering, and may further find the target behavior extraction data having the minimum feature cost value with the anchor unit behavior data as the local first negative unit behavior data (i.e. a first negative unit behavior data) according to a manner of comparing the feature cost values, and in the third quantity-fourth quantity of second unit behavior data contained in the second negative unit behavior data area corresponding to the unit behavior data cluster, according to a manner of comparing feature cost values, the second unit behavior data having the smallest feature cost value with the anchor unit behavior data is found as global first negative unit behavior data (i.e. another first negative unit behavior data), so that according to the found global first negative unit behavior data and local first negative unit behavior data, the second unit behavior data is combined with the fourth quantity (fourth quantity-1) = 1/2=6 positive unit behavior data of the corresponding unit behavior data cluster, and finally, a hard training data member cluster for performing network updating on the first behavior intention mining network is formed.

Step S102, mining and outputting member cluster variables of candidate training data member clusters according to a first behavior intention mining network, loading the member cluster variables of the candidate training data member clusters to a training effectiveness decision network in mapping connection with the first behavior intention mining network, and performing training effectiveness decision on the candidate training data member clusters by the training effectiveness decision network based on the network updating turns of the first behavior intention mining network and effective value evaluation configuration information set by the training effectiveness decision network to generate training effectiveness decision information of the candidate training data member clusters;

for example, the cloud AI service system (e.g., a distributed server system) may load the candidate training data member clusters to a first behavioral intention mining network, perform variable coding on the candidate training data member clusters by the first behavioral intention mining network, and take member cluster variables of the candidate training data member clusters mined by the first behavioral intention mining network as first updated member cluster variables; the cloud AI service system (for example, a distributed server system) can load a first update member cluster variable to a training effectiveness decision network in mapping relation with a first behavior intention mining network when the number of network update rounds of the first behavior intention mining network reaches an initial update round number, and the training effectiveness decision network outputs first effective value evaluation configuration information set by the training effectiveness decision network based on a first decision data sequence corresponding to user behavior training data; the cloud AI service system (for example, a distributed server system) can determine a first combined training cost value of the user behavior training data according to the first decision data sequence, update network weight information of the first behavior intention mining network according to the first combined training cost value, use the first behavior intention mining network after weight update as a fuzzy behavior intention mining network, and update the number of network updating rounds of the first behavior intention mining network; the cloud AI service system (for example, a distributed server system) can take member cluster variables of candidate training data member clusters mined by the fuzzy behavior intention mining network as second updating member cluster variables, load the second updating member cluster variables to the training effectiveness decision network when the updated network updating rounds reach the preset updating rounds, and perform training effectiveness decision on the candidate training data member clusters by the training effectiveness decision network based on the configuration information evaluated by the first effective values set by the training effectiveness decision network to generate training effectiveness decision information of the candidate training data member clusters.

In a network convergence optimization stage, that is, in a process that the distributed server system performs network training on the first behavior intention mining network according to a large number of candidate training data member clusters obtained in the data preparation stage, the initial behavior variables of the candidate training data member clusters (that is, deep convolution feature vectors of the candidate training data member clusters are obtained by extraction) can be obtained by extraction according to a behavior variable extraction unit of the first behavior intention mining network, and then the extracted initial behavior variables can be loaded to a coding unit of the first behavior intention mining network, and the initial behavior variables are coded by the coding unit, at this time, the distributed server system can use a coding vector set obtained by coding as a first updated member cluster variable mined by the first behavior intention mining network.

Step S103, performing behavior intention characteristic decision on member cluster variables of candidate training data member clusters to generate behavior intention characteristic variables of the candidate training data member clusters, performing combined training cost value evaluation on user behavior training data based on the behavior intention characteristic variables of the candidate training data member clusters and training effectiveness decision information, and outputting combined training cost values of the user behavior training data;

for example, the distributed server system may perform behavior intention feature decision on the member cluster variables of the candidate training data member clusters to generate behavior intention feature variables of the candidate training data member clusters, perform frequent item variable extraction on the behavior intention feature variables of the candidate training data member clusters to generate frequent item variables of the candidate training data member clusters; the distributed server system can mine the network updating round number of the network based on the first action intention, and output a first influence factor corresponding to the first decision data sequence, a second influence factor corresponding to the second decision data sequence and a third influence factor corresponding to the third decision data sequence; the distributed server system can output a first training convergence index of a training data member cluster corresponding to the first decision data sequence based on first decision output information of the training data member cluster corresponding to the first decision data sequence, output a second training convergence index of the training data member cluster corresponding to the second decision data sequence based on second decision output information of the training data member cluster corresponding to the second decision data sequence, and output a third training convergence index of the training data member cluster corresponding to the third decision data sequence based on third decision output information of the training data member cluster corresponding to the third decision data sequence; the distributed server system can output a cluster training convergence index of the user behavior training data when the network updates the number of rounds based on a weighted value of the first training convergence index and the first influence factor, a weighted value of the second training convergence index and the second influence factor, and a weighted value of the third training convergence index and the third influence factor; the distributed server system can determine the intention evaluation value of a candidate training data member in the candidate training data member cluster on a target intention evaluation dimension according to the behavior intention characteristic variable of the candidate training data member cluster, determine the intention evaluation variable of the candidate training data member on the target intention evaluation dimension according to the frequent item variable of the candidate training data member cluster, and output the intention evaluation cost value of the candidate training data member in the network updating round number according to the intention evaluation value of the candidate training data member on the target intention evaluation dimension and the intention evaluation variable of the candidate training data member on the target intention evaluation dimension; the distributed server system can evaluate the cost value based on the cluster training convergence index of the user behavior training data in the network updating round number and the intention of candidate training data members in the network updating round number, and generate the combined training cost value of the user behavior training data.

And step S104, updating the network weight information of the first behavior intention mining network based on the combined training cost value, generating a second behavior intention mining network for performing behavior intention characteristic analysis on target user behavior data according to the first behavior intention mining network after updating the network weight information, and performing behavior intention characteristic analysis on user behavior big data by using the second behavior intention mining network as the target behavior intention mining network based on the second behavior intention mining network.

In some embodiments, the cloud AI service system obtains user behavior training data for network updating of the first behavior intention mining network, and may use training data member clusters in the user behavior training data as candidate training data member clusters; the cloud AI service system can mine and output member cluster variables of candidate training data member clusters according to the first action intention mining network, load the member cluster variables of the candidate training data member clusters to a training effectiveness decision network in mapping connection with the first action intention mining network, and perform training effectiveness decision on the candidate training data member clusters by the training effectiveness decision network based on the network updating round number of the first action intention mining network and effective value evaluation configuration information set by the training effectiveness decision network to generate training effectiveness decision information of the candidate training data member clusters; the cloud AI service system can perform behavior intention characteristic decision on member cluster variables of candidate training data member clusters to generate behavior intention characteristic variables of the candidate training data member clusters, perform combined training cost value evaluation on user behavior training data based on the behavior intention characteristic variables of the candidate training data member clusters and training effectiveness decision information, and output a combined training cost value of the user behavior training data; the cloud AI service system can update the network weight information of the first behavior intention mining network based on the combined training cost value, generate a second behavior intention mining network for performing behavior intention characteristic analysis on target user behavior data according to the first behavior intention mining network after updating the network weight information, and perform behavior intention characteristic analysis on user behavior big data by using the second behavior intention mining network as the target behavior intention mining network. In some embodiments, when the cloud AI service system is used for training user behavior training data of an initial behavior intention mining network, training data member clusters in the user behavior training data may be obtained, so that network convergence optimization may be performed on the first behavior intention mining network pair according to the training data member clusters obtained in batch, and in the process of optimizing the first behavior intention mining network by network convergence, real-time decision accuracy judgment (i.e., training effectiveness decision) may be performed on the training data member clusters, so as to perform combined training cost value evaluation on the user behavior training data based on training effectiveness decision information and behavior intention characteristic variables, so as to reversely update network weight information according to the combined training cost value obtained by learning, so that the network may be mined according to the first behavior intention after updating the network weight information, and generating a second behavior intention mining network for performing behavior intention characteristic analysis on the target user behavior data, and performing behavior intention characteristic analysis on the user behavior big data based on the second behavior intention mining network as the target behavior intention mining network. In other words, according to the embodiment of the application, in the process of optimizing the first behavior intention mining network through network convergence, real-time decision precision judgment (namely training effectiveness decision) is performed on training data member clusters, so that the reliability of network convergence optimization is effectively improved during combined training, and further the precision of subsequent big data mining is improved, so that behavior intention mining characteristics of target users obtained through mining are more accurate, the push grouping distribution of the target users is further more reasonable, and the directionality of information push contents is finally improved.

The method may comprise at least the following steps S201-S210:

step S201, when full-dimensional unit behavior data is obtained, configuring a negative association behavior data area corresponding to each unit behavior data cluster in the full-dimensional unit behavior data;

for example, the cloud AI service system (e.g., a distributed server system) may obtain reference user behavior big data for splitting the unit behavior data, split the unit behavior data for the reference user behavior big data, output a first number of unit behavior data clusters corresponding to the reference user behavior big data, use the first number of unit behavior data clusters as full-dimensional unit behavior data, group the first number of unit behavior data clusters according to behavior category information of the first number of unit behavior data clusters, and generate a second number of grouped units corresponding to the full-dimensional unit behavior data; one clustering unit corresponds to one behavior category information; the cloud AI service system (for example, a distributed server system) can acquire a unit behavior data cluster i from the full-dimensional unit behavior data, and in a second number of clustering units, a third number of clustering units extracted from the clustering units where the unit behavior data cluster i is located are used as target clustering units; wherein i is a positive integer no greater than a first number; the number of the target clustering units is a third number; the cloud AI service system (e.g., a distributed server system) may extract a third quantity of unit behavior data from the unit associated behavior data regions corresponding to the third quantity of target clustering units as first unit behavior data, configure a first type of negative associated behavior data region corresponding to the unit behavior data cluster i according to the extracted third quantity of the first unit behavior data, extract a third quantity of the fourth quantity of the unit behavior data from the global associated behavior data regions corresponding to the second quantity of clustering units as second unit behavior data, and configure a second type of negative associated behavior data region corresponding to the unit behavior data cluster i according to the extracted third quantity of the second unit behavior data; the first unit behavior data in the first type negative correlation behavior data area and the unit behavior data in the unit behavior data cluster i have the same behavior type information; the unit behavior data in the second type negative correlation behavior data area and the unit behavior data in the unit behavior data cluster i have different behavior category information; the cloud AI service system (e.g., a distributed server system) may use the first type negative association behavior data area and the second type negative association behavior data area as negative association behavior data areas of the unit behavior data cluster i.

The first number of unit behavior data clusters are clustered according to the behavior category information of the first number of unit behavior data clusters, and a second number of clustering units corresponding to the full-dimensional unit behavior data are generated, for example: loading a first number of unit behavior data clusters contained in the full-dimensional unit behavior data to a behavior category decision model, and deciding the behavior category information of each unit behavior data cluster in the first number of unit behavior data clusters by the behavior category decision model; the behavior type information of each unit behavior data cluster is determined when the behavior type quantity of the behavior type information of the unit behavior data in the corresponding unit behavior data cluster reaches a preset quantity; the preset number is determined by the number of the unit behavior data corresponding to the unit behavior data in the unit behavior data cluster. And in the first number of unit behavior data clusters, based on the behavior type information of each unit behavior data cluster, allocating the unit behavior data clusters with the same behavior type information to the same clustering unit, and outputting a second number of clustering units corresponding to the full-dimensional unit behavior data.

Step S202, configuring a multi-dimensional training data member cluster for network updating of the first behavior intention mining network according to each unit behavior data cluster in the full-dimensional unit behavior data and a negative association behavior data area corresponding to each unit behavior data cluster;

for example, the full-dimensional unit behavior data includes a first number of unit behavior data clusters, and one unit behavior data cluster contains a fourth number of unit behavior data; the negative associated behavior data areas corresponding to the unit behavior data clusters respectively comprise a first type negative associated behavior data area and a second type negative associated behavior data area; at this time, the distributed server system may obtain a target unit behavior data cluster from each unit behavior data cluster in the full-dimensional unit behavior data; the distributed server system can select one unit behavior data from the fourth quantity of unit behavior data of the target unit behavior data cluster as anchor unit behavior data of the target unit behavior data cluster, takes the unit behavior data except the anchor unit behavior data as positive unit behavior data of the target unit behavior data cluster from the fourth quantity of unit behavior data of the target unit behavior data cluster, and outputs a positive unit behavior data combination of the target unit behavior data cluster according to the anchor unit behavior data of the target unit behavior data cluster and the positive unit behavior data of the target unit behavior data cluster; the number of positive unit behavior data combinations of the target unit behavior data cluster is a fourth number (a fourth number-1) 1/2; the distributed server system can extract 2 × fifth unit behavior data from the negative association behavior data area corresponding to the target unit behavior data cluster as the negative unit behavior data of the target unit behavior data cluster; the negative unit behavior data of the target unit behavior data cluster comprises a fifth number of first negative unit behavior data and a fifth number of second negative unit behavior data; a fifth quantity of first negative unit behavior data is extracted from the first type negative association behavior data area corresponding to the target unit behavior data cluster; a fifth quantity of second negative unit behavior data is extracted from the second type negative association behavior data region corresponding to the target unit behavior data cluster; the distributed server system can configure and obtain a training data member cluster of the target unit behavior data cluster based on the positive unit behavior data combination of the target unit behavior data cluster and the negative unit behavior data of the target unit behavior data cluster, and generates the training data member cluster of each unit behavior data cluster until each unit behavior data cluster in the full-dimensional unit behavior data is taken as the target unit behavior data cluster; the number of the training data member clusters of each unit behavior data cluster is fourth number (fourth number-1) fifth number; the distributed server system can use the training data member cluster of each unit behavior data cluster as a multi-dimensional training data member cluster for updating the network of the first behavior intention mining network.

The first type of negative association behavior data area corresponding to the target unit behavior data cluster comprises a third quantity and a fourth quantity of first unit behavior data; and the second type negative association behavior data area corresponding to the target unit behavior data cluster comprises a third quantity and a fourth quantity of second unit behavior data. On this basis, in the embodiment of the present application, one unit behavior data may be wandered and selected from among the fourth number of unit behavior data of the target unit behavior data cluster as positive unit behavior data of the target unit behavior data cluster, and unit behavior data other than the wandered positive unit behavior data may be taken as anchor unit behavior data of the target unit behavior data cluster from among the fourth number of unit behavior data of the target unit behavior data cluster. Then, searching first unit behavior data which is the same as the anchor unit behavior data of the target unit behavior data cluster in the third quantity-fourth quantity of first unit behavior data, taking the searched first unit behavior data which is the same as the anchor unit behavior data of the target unit behavior data cluster as collaborative behavior extraction data,

taking first unit behavior data other than the cooperative behavior extraction data as target behavior extraction data in the third number-fourth number of first unit behavior data; the number of data of the target behavior extraction data is (third number + fourth number-1). Then, a first feature cost value between the collaborative behavior extraction data and (a third quantity x a fourth quantity-1) target behavior extraction data is determined, the target behavior extraction data corresponding to the determined minimum first feature cost value is used as local first negative unit behavior data of the collaborative behavior extraction data, a second feature cost value between the collaborative behavior extraction data and a third quantity x a fourth quantity of second unit behavior data is determined, and the second unit behavior data corresponding to the determined minimum second feature cost value is used as global first negative unit behavior data of the collaborative behavior extraction data. Therefore, the first negative unit behavior data of the target unit behavior data cluster can be updated in the negative unit behavior data of the target unit behavior data cluster according to the local first negative unit behavior data of the cooperative behavior extraction data and the global first negative unit behavior data of the cooperative behavior extraction data.

Step S203, based on the total negative correlation statistics in the negative correlation behavior data area, training unit distribution is carried out on the training data member clusters in the multi-dimensional training data member clusters, and user behavior training data corresponding to the full-dimensional unit behavior data are generated.

Step S204, acquiring user behavior training data for updating the network of the first behavior intention mining network, and taking a training data member cluster in the user behavior training data as a candidate training data member cluster;

step S205, mining and outputting member cluster variables of the candidate training data member clusters according to the first behavior intention mining network, loading the member cluster variables of the candidate training data member clusters to a training effectiveness decision network in mapping connection with the first behavior intention mining network, and performing training effectiveness decision on the candidate training data member clusters by the training effectiveness decision network based on the network updating turns of the first behavior intention mining network and effective value evaluation configuration information set by the training effectiveness decision network to generate training effectiveness decision information of the candidate training data member clusters.

For example, the candidate training data member clusters may be loaded to the first behavioral intent mining network, variable encoded by the first behavioral intent mining network, and member cluster variables of the candidate training data member clusters mined by the first behavioral intent mining network as first updated member cluster variables. Then, when the number of network updating rounds of the first behavior intention mining network reaches the initial number of updating rounds, loading the first updating member cluster variable to a training effectiveness decision network in mapping connection with the first behavior intention mining network, and calling the training effectiveness decision network to output first effective value evaluation configuration information set by the training effectiveness decision network based on a first decision data sequence corresponding to the user behavior training data. Then, determining a first combined training cost value of the user behavior training data according to the first decision data sequence, updating network weight information of the first behavior intention mining network according to the first combined training cost value, taking the first behavior intention mining network after weight updating as a fuzzy behavior intention mining network, and updating the network updating round number of the first behavior intention mining network. Then, the member cluster variables of the candidate training data member clusters mined by the fuzzy behavior intention mining network are used as second updating member cluster variables, when the updated network updating round number reaches a preset updating round number, the second updating member cluster variables are loaded to the training effectiveness decision network, the training effectiveness decision network is called to evaluate configuration information based on a first effective value set by the training effectiveness decision network, training effectiveness decisions are carried out on the candidate training data member clusters, and training effectiveness decision information of the candidate training data member clusters is generated.

For example, loading the candidate training data member clusters to the first behavioral intent mining network, variable encoding the candidate training data member clusters by the first behavioral intent mining network, taking member cluster variables of the candidate training data member clusters mined by the first behavioral intent mining network as first updated member cluster variables, comprises: loading the candidate training data member cluster to a behavior variable extraction unit of the first behavior intention mining network, and extracting the initial behavior variable of the candidate training data member cluster by the behavior variable extraction unit; loading the extracted initial behavior variables to a coding unit of the first behavior intention mining network, and coding the initial behavior variables by the coding unit; and taking the coding vector set obtained by coding as a first updating member cluster variable mined by the first behavior intention mining network.

For example, when the number of network update rounds of the first behavior intention mining network reaches the initial number of update rounds, loading the first update member cluster variable to a training effectiveness decision network in mapping relation with the first behavior intention mining network, and calling the training effectiveness decision network to output first effective value evaluation configuration information set by the training effectiveness decision network based on a first decision data sequence corresponding to the user behavior training data, the method includes: when the number of network updating rounds of the first behavior intention mining network reaches the initial number of updating rounds, loading the first updating member cluster variable to a training effectiveness decision network in mapping connection with the first behavior intention mining network, and calling the training effectiveness decision network to take a decision data sequence corresponding to the candidate training data member cluster as a first decision data sequence corresponding to the user behavior training data; updating and recording the initial effective value evaluation index of the negative unit behavior data in the candidate training data member cluster in the first decision data sequence; obtaining a confidence threshold of the training effectiveness decision network during the initial updating round number, outputting the initial confidence threshold of the training effectiveness decision network during the initial updating round number according to an initial effective value evaluation index recorded by updating and the confidence threshold during the initial updating round number, and taking effective value evaluation configuration information set by the training effectiveness decision network as first effective value evaluation configuration information according to the initial confidence threshold.

For example, in the first decision data sequence, updating an initial valid value evaluation index recording negative unit behavior data in the candidate training data member cluster, including: in the first decision data sequence, outputting the total cluster number of the candidate training data member clusters participating in updating, acquiring anchor unit behavior data of a first target training data member cluster and negative unit behavior data of the first target training data member cluster from the candidate training data member clusters, and taking a unit behavior data cluster in which the anchor unit behavior data of the first target training data member cluster is located as a target unit behavior data cluster; acquiring a sixth quantity of unit behavior data in the target unit behavior data cluster, determining a behavior coding variable of each unit behavior data in the sixth quantity of unit behavior data according to the first updating member cluster variable, and taking an average behavior coding variable determined by the behavior coding variable of each unit behavior data as a cluster center of the target unit behavior data cluster; in the sixth amount of unit behavior data in the target unit behavior data cluster, using the searched unit behavior data having the minimum feature cost value with the negative unit behavior data in the first target training data member cluster as first target unit behavior data, using the feature cost value between the negative unit behavior data in the first target training data member cluster and the first target unit behavior data as a first target cost value, and using the feature cost value between the negative unit behavior data in the first target training data member cluster and the cluster center of the target unit behavior data cluster as a second target cost value; taking the ratio of the first target cost value to the second target cost value as an evaluation index of an effective value to be updated of the negative unit behavior data of the first target training data member cluster; and outputting an initial effective value evaluation index of the negative unit behavior data in the candidate training data member cluster according to the total cluster number and the effective value evaluation index to be updated of the negative unit behavior data of the first target training data member cluster.

For example, the method includes taking a member cluster variable of the candidate training data member cluster mined by the fuzzy behavior intention mining network as a second updated member cluster variable, loading the second updated member cluster variable to the training effectiveness decision network when the updated network update round number reaches a preset update round number, calling the training effectiveness decision network to evaluate configuration information based on a first effective value set by the training effectiveness decision network, performing training effectiveness decision on the candidate training data member cluster, and generating training effectiveness decision information of the candidate training data member cluster.

(1) Taking the member cluster variable of the candidate training data member cluster mined by the fuzzy behavior intention mining network as a second updating member cluster variable, loading the second updating member cluster variable to the training effectiveness decision network when the updated network updating round number reaches a preset updating round number, calling the training effectiveness decision network to obtain the anchor unit behavior data of a second target training data member cluster and the negative unit behavior data of the second target training data member cluster from the candidate training data member cluster, and taking the unit behavior data cluster in which the anchor unit behavior data of the second target training data member cluster is located as a unit behavior data cluster to be learned;

(2) acquiring a seventh amount of unit behavior data in the unit behavior data cluster to be learned, determining a behavior coding variable of each unit behavior data in the seventh amount of unit behavior data according to the second updating member cluster variable, and taking an average behavior coding variable determined by the behavior coding variable of each unit behavior data as a cluster center of the unit behavior data cluster to be learned;

(3) outputting a cost average value corresponding to the unit behavior data cluster to be learned based on the characteristic cost values between the seventh number of unit behavior data in the unit behavior data cluster to be learned and the cluster center of the unit behavior data cluster to be learned respectively, and taking the cost average value as the intra-cluster cost value of the unit behavior data cluster to be learned;

(4) in the seventh amount of unit behavior data in the unit behavior data cluster to be learned, using the searched unit behavior data having the minimum feature cost value with the negative unit behavior data in the second target training data member cluster as second target unit behavior data, using the feature cost value between the negative unit behavior data in the second target training data member cluster and the second target unit behavior data as a third target cost value, and using the feature cost value between the negative unit behavior data in the second target training data member cluster and the cluster center of the unit behavior data cluster to be learned as a fourth target cost value;

(5) taking the ratio of the third target cost value to the fourth target cost value as a cost coefficient between the negative unit behavior data of the second target training data member cluster and the unit behavior data cluster to be learned;

(6) and according to the cost coefficient, the intra-cluster cost value and first effective value evaluation configuration information set by the training effectiveness decision network, performing training effectiveness decision on the second target training data member cluster to generate training effectiveness decision information of the second target training data member cluster, and generating the training effectiveness decision information of each training data member cluster in the candidate training data member cluster until each training data member cluster in the candidate training data member cluster is taken as the second target training data member cluster.

For example, the first decision data sequence corresponding to the user behavior training data includes a first decision data sequence, a second decision data sequence, and a third decision data sequence; the decision accuracy of the first decision data sequence is higher than the decision accuracy of the second decision data sequence, and the decision accuracy of the second decision data sequence is higher than the decision accuracy of the third decision data sequence.

Based on the above description, this embodiment may output, in each of the candidate training data member clusters, a training data member cluster for loading to the first decision data sequence, a training data member cluster for loading to the second decision data sequence, and a training data member cluster for loading to the third decision data sequence based on training validity decision information of each of the candidate training data member clusters, reject, in the first decision data sequence corresponding to the user behavior training data, the training data member cluster corresponding to the third decision data sequence, use, as a target decision data sequence, the first decision data sequence from which the training data member cluster corresponding to the third decision data sequence is rejected, and update the training validity decision network according to the training data member cluster in the target decision data sequence The first valid value set by the network evaluates the configuration information.

Step S206, performing behavior intention characteristic decision on the member cluster variables of the candidate training data member clusters to generate behavior intention characteristic variables of the candidate training data member clusters, performing combined training cost value evaluation on the user behavior training data based on the behavior intention characteristic variables of the candidate training data member clusters and training effectiveness decision information, and outputting combined training cost values of the user behavior training data.

The training effectiveness decision information of the candidate training data member cluster comprises first decision output information of a training data member cluster corresponding to a first decision data sequence, second decision output information of a training data member cluster corresponding to a second decision data sequence and third decision output information of a training data member cluster corresponding to a third decision data sequence; the decision accuracy of the first decision data sequence is higher than the decision accuracy of the second decision data sequence, and the decision accuracy of the second decision data sequence is higher than the decision accuracy of the third decision data sequence.

Thus, in this step, for example, behavioral intention feature decisions may be performed on the member cluster variables of the candidate training data member clusters to generate behavioral intention feature variables of the candidate training data member clusters, and frequent item variables may be extracted from the behavioral intention feature variables of the candidate training data member clusters to generate frequent item variables of the candidate training data member clusters.

On the basis, the network updating round number of the network is mined based on the first behavior intention, and a first influence factor corresponding to the first decision data sequence, a second influence factor corresponding to the second decision data sequence and a third influence factor corresponding to the third decision data sequence are output. Outputting a first training convergence index of a training data member cluster corresponding to the first decision data sequence based on first decision output information of the training data member cluster corresponding to the first decision data sequence, outputting a second training convergence index of the training data member cluster corresponding to the second decision data sequence based on second decision output information of the training data member cluster corresponding to the second decision data sequence, and outputting a third training convergence index of the training data member cluster corresponding to the third decision data sequence based on third decision output information of the training data member cluster corresponding to the third decision data sequence;

on this basis, outputting a cluster training convergence index of the user behavior training data at the time of the network update round number based on a weighted value based on the first training convergence index and the first influence factor, a weighted value based on the second training convergence index and the second influence factor, and a weighted value based on the third training convergence index and the third influence factor, determining an intention evaluation value of a candidate training data member in the candidate training data member cluster in a target intention evaluation dimension in accordance with behavior intention characteristic variables of the candidate training data member cluster, determining an intention evaluation variable of the candidate training data member in the target intention evaluation dimension in accordance with frequent item variables of the candidate training data member cluster, determining an intention evaluation variable of the candidate training data member in the target intention evaluation dimension in accordance with the intention evaluation value of the candidate training data member in the target intention evaluation dimension and the intention evaluation variable of the candidate training data member in the target intention evaluation dimension in accordance with the intention evaluation variable of the candidate training data member in the target intention evaluation dimension, and outputting the intention evaluation cost value of the candidate training data members in the network updating round number, and generating the combined training cost value of the user behavior training data based on the cluster training convergence index of the user behavior training data in the network updating round number and the intention evaluation cost value of the candidate training data members in the network updating round number.

Step S207, updating the network weight information of the first behavior intention mining network based on the combined training cost value, generating a second behavior intention mining network for performing behavior intention characteristic analysis on target user behavior data according to the first behavior intention mining network after updating the network weight information, and performing behavior intention characteristic analysis on user behavior big data by using the second behavior intention mining network as the target behavior intention mining network based on the second behavior intention mining network.

Further, referring to fig. 3, the cloud AI service system 100 may include: a processor 101 and a machine-readable storage medium 102. The machine-readable storage medium 102 is used for storing a program that enables the cloud AI service system 100 to execute the big data mining method applied to behavior intention analysis provided in any one of the foregoing embodiments, and the processor 101 is configured to execute the program stored in the machine-readable storage medium 102.

The program comprises one or more computer instructions, wherein the one or more computer instructions, when executed by the processor 101, enable all or part of the steps of any of the foregoing embodiments.

The architecture of the cloud AI service system 100 may further include a communication unit 103, which is used for the cloud AI service system 100 to communicate with other devices or systems (e.g., the cloud AI service system 200).

In addition, the present embodiment provides a computer storage medium for storing computer software instructions for the cloud AI service system 100, which includes a program for executing the big data mining method applied to behavior intention analysis in any one of the above method embodiments.

Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit the same; although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions in the embodiments of the present application.

Claims

1. A big data mining method applied to behavior intention analysis is applied to the cloud AI service system, and the method comprises the following steps:

2. The big data mining method applied to behavior intention analysis according to claim 1, wherein the step of obtaining the big data of the user behavior to be mined, analyzing the big data of the user behavior according to a target behavior intention mining network trained in advance, and outputting the corresponding behavior intention mining characteristics comprises:

updating network weight information of the first behavior intention mining network based on the combined training cost value, generating a second behavior intention mining network for performing behavior intention characteristic analysis on target user behavior data according to the first behavior intention mining network after updating the network weight information, performing behavior intention characteristic analysis on user behavior big data by taking the second behavior intention mining network as a target behavior intention mining network, and performing behavior intention characteristic analysis on the user behavior big data by taking the second behavior intention mining network as a target behavior intention mining network.

3. The big data mining method applied to behavioral intention analysis according to claim 2, wherein the mining outputs member cluster variables of the candidate training data member clusters according to the first behavioral intention mining network, loads the member cluster variables of the candidate training data member clusters to a training effectiveness decision network having a mapping relation with the first behavioral intention mining network, calls the training effectiveness decision network to perform training effectiveness decision on the candidate training data member clusters based on the network update round number of the first behavioral intention mining network and the set effective value evaluation configuration information of the training effectiveness decision network, and generates the training effectiveness decision information of the candidate training data member clusters, including:

loading the candidate training data member cluster to the first behavior intention mining network, carrying out variable coding on the candidate training data member cluster by the first behavior intention mining network, and taking a member cluster variable of the candidate training data member cluster mined by the first behavior intention mining network as a first updated member cluster variable;

when the number of network updating rounds of the first behavior intention mining network reaches the initial number of updating rounds, loading the first updating member cluster variable to a training effectiveness decision network in mapping connection with the first behavior intention mining network, and calling the training effectiveness decision network to output first effective value evaluation configuration information set by the training effectiveness decision network based on a first decision data sequence corresponding to the user behavior training data;

determining a first combined training cost value of the user behavior training data according to the first decision data sequence, updating network weight information of the first behavior intention mining network according to the first combined training cost value, taking the first behavior intention mining network after weight updating as a fuzzy behavior intention mining network, and updating the number of network updating rounds of the first behavior intention mining network;

and taking the member cluster variable of the candidate training data member cluster mined by the fuzzy behavior intention mining network as a second updating member cluster variable, loading the second updating member cluster variable to the training effectiveness decision network when the updated network updating round number reaches a preset updating round number, calling the training effectiveness decision network to evaluate configuration information based on a first effective value set by the training effectiveness decision network, and performing training effectiveness decision on the candidate training data member cluster to generate training effectiveness decision information of the candidate training data member cluster.

4. The big data mining method applied to behavioral intention analysis according to claim 3, wherein the loading of the candidate training data member clusters to the first behavioral intention mining network, the variable encoding of the candidate training data member clusters by the first behavioral intention mining network, the member cluster variables of the candidate training data member clusters mined by the first behavioral intention mining network as first updated member cluster variables, comprises:

loading the candidate training data member cluster to a behavior variable extraction unit of the first behavior intention mining network, and extracting an initial behavior variable of the candidate training data member cluster by the behavior variable extraction unit;

loading the extracted initial behavior variables to a coding unit of the first behavior intention mining network, and coding the initial behavior variables by the coding unit;

and taking the coded vector set as a first updating member cluster variable mined by the first behavior intention mining network.

5. The big data mining method applied to behavioral intention analysis according to claim 3, wherein when the number of network update rounds of the first behavioral intention mining network reaches an initial number of update rounds, loading the first update member cluster variable to a training effectiveness decision network in mapping relation with the first behavioral intention mining network, and invoking the training effectiveness decision network to output first effective value evaluation configuration information set by the training effectiveness decision network based on a first decision data sequence corresponding to the user behavioral training data, comprises:

when the number of network updating rounds of the first behavior intention mining network reaches the initial number of updating rounds, loading the first updating member cluster variable to a training effectiveness decision network in mapping connection with the first behavior intention mining network, and calling the training effectiveness decision network to take a decision data sequence corresponding to the candidate training data member cluster as a first decision data sequence corresponding to the user behavior training data;

updating and recording the initial effective value evaluation index of the negative unit behavior data in the candidate training data member cluster in the first decision data sequence;

obtaining a confidence threshold of the training effectiveness decision network in the initial updating round number, outputting the initial confidence threshold of the training effectiveness decision network in the initial updating round number according to an initial effective value evaluation index recorded in an updating way and the confidence threshold of the training effectiveness decision network in the initial updating round number, and taking effective value evaluation configuration information set by the training effectiveness decision network as first effective value evaluation configuration information according to the initial confidence threshold.

6. The big data mining method applied to behavioral intention analysis according to claim 5, wherein the updating of the initial valid value evaluation index recording the negative unit behavior data in the candidate training data member cluster in the first decision data sequence comprises:

in the first decision data sequence, outputting the total cluster number of the candidate training data member clusters participating in updating, acquiring anchor unit behavior data of a first target training data member cluster and negative unit behavior data of the first target training data member cluster from the candidate training data member clusters, and taking a unit behavior data cluster in which the anchor unit behavior data of the first target training data member cluster is located as a target unit behavior data cluster;

acquiring a sixth amount of unit behavior data in the target unit behavior data cluster, determining a behavior coding variable of each unit behavior data in the sixth amount of unit behavior data according to the first updating member cluster variable, and taking an average behavior coding variable determined by the behavior coding variable of each unit behavior data as a cluster center of the target unit behavior data cluster;

in the sixth amount of unit behavior data in the target unit behavior data cluster, using the searched unit behavior data having the minimum feature cost value with the negative unit behavior data in the first target training data member cluster as first target unit behavior data, using the feature cost value between the negative unit behavior data in the first target training data member cluster and the first target unit behavior data as a first target cost value, and using the feature cost value between the negative unit behavior data in the first target training data member cluster and the cluster center of the target unit behavior data cluster as a second target cost value;

taking the ratio of the first target cost value to the second target cost value as an evaluation index of an effective value to be updated of the negative unit behavior data of the first target training data member cluster;

and outputting an initial effective value evaluation index of the negative unit behavior data in the candidate training data member cluster according to the total cluster number and the effective value evaluation index to be updated of the negative unit behavior data of the first target training data member cluster.

7. The big data mining method applied to behavioral intention analysis according to claim 3, wherein the step of using the member cluster variables of the candidate training data member clusters mined by the fuzzy behavioral intention mining network as second updated member cluster variables, and when the updated number of network update rounds reaches a preset number of update rounds, loading the second updated member cluster variables to the training effectiveness decision network, calling the training effectiveness decision network to evaluate configuration information based on the first effective value set by the training effectiveness decision network, performing training effectiveness decision on the candidate training data member clusters, and generating the training effectiveness decision information of the candidate training data member clusters comprises:

taking the member cluster variable of the candidate training data member cluster mined by the fuzzy behavior intention mining network as a second updating member cluster variable, loading the second updating member cluster variable to the training effectiveness decision network when the updated network updating round number reaches a preset updating round number, calling the training effectiveness decision network to obtain the anchor unit behavior data of a second target training data member cluster and the negative unit behavior data of the second target training data member cluster from the candidate training data member cluster, and taking the unit behavior data cluster in which the anchor unit behavior data of the second target training data member cluster is located as a unit behavior data cluster to be learned;

acquiring a seventh amount of unit behavior data in the unit behavior data cluster to be learned, determining a behavior coding variable of each unit behavior data in the seventh amount of unit behavior data according to the second updating member cluster variable, and taking an average behavior coding variable determined by the behavior coding variable of each unit behavior data as a cluster center of the unit behavior data cluster to be learned;

outputting a cost average value corresponding to the unit behavior data cluster to be learned based on the characteristic cost values between the seventh number of unit behavior data in the unit behavior data cluster to be learned and the cluster center of the unit behavior data cluster to be learned respectively, and taking the cost average value as the intra-cluster cost value of the unit behavior data cluster to be learned;

in the seventh amount of unit behavior data in the unit behavior data cluster to be learned, using the searched unit behavior data having the minimum feature cost value with the negative unit behavior data in the second target training data member cluster as second target unit behavior data, using the feature cost value between the negative unit behavior data in the second target training data member cluster and the second target unit behavior data as a third target cost value, and using the feature cost value between the negative unit behavior data in the second target training data member cluster and the cluster center of the unit behavior data cluster to be learned as a fourth target cost value;

taking the ratio of the third target cost value to the fourth target cost value as a cost coefficient between the negative unit behavior data of the second target training data member cluster and the unit behavior data cluster to be learned;

and evaluating configuration information according to the cost coefficient, the intra-cluster cost value and a first effective value set by the training effectiveness decision network, performing training effectiveness decision on the second target training data member cluster, generating training effectiveness decision information of the second target training data member cluster, and generating the training effectiveness decision information of each training data member cluster in the candidate training data member cluster until each training data member cluster in the candidate training data member cluster is used as the second target training data member cluster.

8. The big data mining method applied to behavioral intention analysis according to claim 7, characterized in that the first decision data sequence corresponding to the user behavior training data includes a first decision data sequence, a second decision data sequence and a third decision data sequence; the decision accuracy of the first decision data sequence is higher than that of the second decision data sequence, and the decision accuracy of the second decision data sequence is higher than that of the third decision data sequence;

the method further comprises the following steps:

based on training effectiveness decision information of each training data member cluster in the candidate training data member clusters, outputting a training data member cluster corresponding to the first decision data sequence, a training data member cluster corresponding to the second decision data sequence and a training data member cluster corresponding to the third decision data sequence in each training data member cluster in the candidate training data member clusters;

and eliminating a training data member cluster corresponding to the third decision data sequence from a first decision data sequence corresponding to the user behavior training data, taking the first decision data sequence after eliminating the training data member cluster corresponding to the third decision data sequence as a target decision data sequence, and updating first effective value evaluation configuration information set by the training effectiveness decision network according to the training data member cluster in the target decision data sequence.

9. The big data mining method applied to behavioral intention analysis according to claim 2, wherein the training validity decision information of the candidate training data member cluster includes first decision output information of a training data member cluster corresponding to a first decision data sequence, second decision output information of a training data member cluster corresponding to a second decision data sequence, and third decision output information of a training data member cluster corresponding to a third decision data sequence;

the decision accuracy of the first decision data sequence is higher than that of the second decision data sequence, and the decision accuracy of the second decision data sequence is higher than that of the third decision data sequence;

performing behavior intention characteristic decision on the member cluster variables of the candidate training data member cluster to generate behavior intention characteristic variables of the candidate training data member cluster, performing combined training cost value evaluation on the user behavior training data based on the behavior intention characteristic variables of the candidate training data member cluster and the training effectiveness decision information, and outputting a combined training cost value of the user behavior training data, which includes:

performing behavior intention characteristic decision on the member cluster variables of the candidate training data member cluster to generate behavior intention characteristic variables of the candidate training data member cluster, and performing frequent item variable extraction on the behavior intention characteristic variables of the candidate training data member cluster to generate frequent item variables of the candidate training data member cluster;

outputting a first influence factor corresponding to the first decision data sequence, a second influence factor corresponding to the second decision data sequence and a third influence factor corresponding to the third decision data sequence based on the network updating round number of the first behavior intention mining network;

outputting a first training convergence index of a training data member cluster corresponding to the first decision data sequence based on first decision output information of the training data member cluster corresponding to the first decision data sequence, outputting a second training convergence index of the training data member cluster corresponding to the second decision data sequence based on second decision output information of the training data member cluster corresponding to the second decision data sequence, and outputting a third training convergence index of the training data member cluster corresponding to the third decision data sequence based on third decision output information of the training data member cluster corresponding to the third decision data sequence;

outputting a cluster training convergence index of the user behavior training data at the network update round number based on the weighted value of the first training convergence index and the first influence factor, the weighted value of the second training convergence index and the second influence factor, and the weighted value of the third training convergence index and the third influence factor;

determining intention evaluation values of candidate training data members in the candidate training data member clusters on a target intention evaluation dimension according to behavior intention characteristic variables of the candidate training data member clusters, determining intention evaluation variables of the candidate training data members on the target intention evaluation dimension according to frequent item variables of the candidate training data member clusters, and outputting intention evaluation cost values of the candidate training data members in the network updating rounds according to the intention evaluation values of the candidate training data members on the target intention evaluation dimension and the intention evaluation variables of the candidate training data members on the target intention evaluation dimension;

and generating a combined training cost value of the user behavior training data based on the cluster training convergence index of the user behavior training data in the network updating round number and the intention evaluation cost value of the candidate training data members in the network updating round number.

10. A cloud AI service system, comprising at least one processor; and a memory communicatively coupled to the at least one processor; wherein the memory stores instructions executable by the at least one processor to cause the at least one processor to perform to implement the big data mining method applied to behavioral intent analysis of any one of claims 1-9.