CN114757721A

CN114757721A - Service prediction analysis method and AI (Artificial Intelligence) mining system for joint big data mining

Info

Publication number: CN114757721A
Application number: CN202210572054.XA
Authority: CN
Inventors: 周正涛
Original assignee: Zibo Zhicheng E Commerce Co ltd
Current assignee: Guangzhou Xiaofei Information Technology Co ltd
Priority date: 2022-05-25
Filing date: 2022-05-25
Publication date: 2022-07-15
Anticipated expiration: 2042-05-25
Also published as: CN114757721B

Abstract

The embodiment of the invention provides a service prediction analysis method and an AI mining system for joint big data mining, which are used for analyzing a plurality of updating behavior activities in user updating big data of a new online service, determining joint basic data to be mined, which are used for loading the joint basic data to be mined to the joint mining online service, performing service situation analysis on the joint basic data to be mined of the joint mining online service and the basic data to be mined of the joint mining online service, determining service situation information of a service group formed by the new online service and the corresponding joint mining online service, determining service image information of the service group according to the service situation information of the service group, and pushing the service image information of the service group to a corresponding service development terminal, thereby performing joint big data mining on the new online service, outputting the service image information taking the service group as an analysis unit for a corresponding service developer to analyze audience service situations, thereby facilitating subsequent service optimization.

Description

Service prediction analysis method and AI (Artificial Intelligence) mining system for joint big data mining

Technical Field

The invention relates to the technical field of big data mining, in particular to a business prediction analysis method and an AI mining system for joint big data mining.

Background

In the current big data mining process, user behavior big data of a new online service is usually analyzed independently to determine the service audience condition of the new online service, so that related service developers can evaluate the new online service conveniently. However, in most practical scenarios, new online services are generated by relying on the existing online services, and currently, joint big data mining for the new online services is lacking.

Disclosure of Invention

In order to overcome at least the above disadvantages in the prior art, the present invention provides a business prediction analysis method and an AI mining system for joint big data mining.

In a first aspect, an embodiment of the present invention provides a business prediction analysis method for joint big data mining, which is applied to an AI mining system, and the method includes:

analyzing a plurality of updating behavior activities in the user updating big data of the new online service, and determining combined basic data to be mined, which are loaded to the combined mining online service;

Performing service situation analysis on the basic data to be jointly mined of the jointly mined online service and the basic data to be jointly mined of the jointly mined online service, and determining service situation information of a service group formed by the new online service and the corresponding jointly mined online service;

and determining the service portrait information of the service group according to the service situation information of the service group, and pushing the service portrait information of the service group to a corresponding service development terminal.

In a possible implementation manner of the first aspect, the step of analyzing a plurality of update behavior activities in the user update big data of the new online service and determining joint to-be-mined base data for loading to the joint mining online service includes:

acquiring a plurality of updating behavior activities in user updating big data aiming at new online business, analyzing activity joint mining values corresponding to the updating behavior activities, and determining a first updating behavior activity from the updating behavior activities based on the activity joint mining values;

acquiring a page attention point data set which has an association relation with the user updating big data, analyzing a page joint mining value corresponding to the first updating behavior activity based on the page attention point data set and the first updating behavior activity, and determining a second updating behavior activity from the first updating behavior activity based on the page joint mining value corresponding to the first updating behavior activity;

Analyzing the joint mining assistance value corresponding to the second updating behavior activity based on the page attention point data set and the second updating behavior activity, and analyzing the joint mining reference information corresponding to the second updating behavior activity based on the joint mining assistance value corresponding to the second updating behavior activity;

determining joint mining updating behavior activities from the second updating behavior activities based on the activity joint mining value corresponding to the second updating behavior activities, the page joint mining value corresponding to the second updating behavior activities, and the assisting joint mining value corresponding to the second updating behavior activities, and taking the joint mining updating behavior activities and the joint mining reference information corresponding to the joint mining updating behavior activities as joint to-be-mined basic data for loading to joint mining online services.

For example, the method further comprises:

acquiring a sample model training data set; the example model training data set comprises example user update data, an example page attention point data set of a prior page target having an association relation with the example user update data, data joint mining value labeling information corresponding to the example user update data, prior page joint mining value labeling information corresponding to the example user update data, and prior assistance joint mining value information corresponding to the example user update data;

Loading the example model training data set to an initial joint mining decision model, and analyzing a user updating big data joint mining prediction value corresponding to the example user updating data according to the initial joint mining decision model;

analyzing a joint mining prediction value of a page corresponding to the example user update data and a joint mining assistance prediction value corresponding to the example user update data based on the example page point of interest data set and the example user update data;

performing model weight information optimization on the initial joint mining decision model based on the data joint mining value labeling information, the prior page joint mining value labeling information, the prior assistance joint mining value information, the user updated big data joint mining prediction value, the page joint mining prediction value and the assistance joint mining prediction value, and determining a joint mining decision model;

the joint mining decision model is used for determining joint to-be-mined basic data of big data updated by a user; the joint basic data to be mined comprises joint mining updating behavior activities in the user updating big data and joint mining reference information corresponding to the joint mining updating behavior activities.

For example, the number of the example user update data is plural, the plurality of example user update data includes example user update data Po, o is a positive integer and o is not greater than the number of the plurality of example user update data;

the method further comprises the following steps:

performing a product operation on the service heat corresponding to the example user update data Po, the continuous state value corresponding to the example user update data Po and the average service interaction rate corresponding to the example user update data Po to determine a first prior evaluation value corresponding to the example user update data Po;

summing the number of passive behavior interaction resources corresponding to the example user update data Po and the number of passive behavior interaction resources corresponding to the example user update data Po to determine a second prior evaluation value corresponding to the example user update data Po;

obtaining a maximum first prior estimate among the first prior estimates corresponding to the sample user update data, and obtaining a maximum second prior estimate among the second prior estimates corresponding to the sample user update data;

determining a first ratio between a first prior evaluation value corresponding to the sample user update data Po and the maximum first prior evaluation value, and analyzing a second ratio between a second prior evaluation value corresponding to the sample user update data Po and the maximum second prior evaluation value;

Performing weight fusion on the first ratio and the second ratio, and determining candidate data joint mining value labeling information corresponding to the example user update data Po;

comparing candidate data joint mining value labeling information corresponding to the sample user updating data Po with a third preset mining value;

if the candidate data joint mining value labeling information corresponding to the example user update data Po is smaller than the third preset mining value, determining the candidate data joint mining value labeling information corresponding to the example user update data Po as data joint mining value labeling information corresponding to the example user update data Po;

and if the candidate data joint mining value labeling information corresponding to the example user update data Po is not less than the third preset mining value, determining the third preset mining value as the data joint mining value labeling information corresponding to the example user update data Po.

For example, the method further comprises:

acquiring a first service interaction rate of the prior page target aiming at the sample user update data, and comparing the first service interaction rate with a first set service interaction rate;

If the first service interaction rate is greater than the first set service interaction rate, determining that a first forward contact relationship exists between the sample page focus data and the sample user update data, and determining the first forward contact relationship as the priori page joint mining value labeling information;

and if the first service interaction rate is not greater than the first set service interaction rate, determining that a first negative relation exists between the sample page point of interest data and the sample user updating data, and determining the first negative relation as the prior page joint mining value labeling information.

For example, the example model training dataset further includes prior assistance topics corresponding to the example user update data; the prior assistance joint mining value information comprises assistance subject joint mining value information; the method further comprises the following steps:

acquiring a second service interaction rate of the prior page target aiming at the example user updating data, and comparing the second service interaction rate with a second set service interaction rate;

if the second service interaction rate is greater than the second set service interaction rate, determining that a second forward contact relationship exists among the prior assistance topic, the sample page point of interest data and the sample user update data, and determining the second forward contact relationship as the assistance topic joint mining value information;

And if the second service interaction rate is not greater than the second set service interaction rate, determining that a second negative direction relation exists among the prior assistance topic, the sample page point of interest data and the sample user update data, and determining the second negative direction relation as the assistance topic joint mining value information.

For example, the initial joint mining decision model includes a first initial joint mining decision unit for determining the user updated big data joint mining prediction value, a second initial joint mining decision unit for determining the page joint mining prediction value, and an assisted initial joint mining decision unit for determining the assisted joint mining prediction value; the model weight information in the initial joint mining decision model comprises model weight information in the first initial joint mining decision unit, model weight information in the second initial joint mining decision unit and model weight information in the assisted initial joint mining decision unit; the method for optimizing model weight information of the initial joint mining decision model based on the data joint mining value labeling information, the prior page joint mining value labeling information, the prior assisting joint mining value information, the user updated big data joint mining prediction value, the page joint mining prediction value and the assisting joint mining prediction value to determine a joint mining decision model comprises the following steps:

Determining a first loss function value between the data joint mining value labeling information and the user updated big data joint mining prediction value, optimizing model weight information of the first initial joint mining decision-making unit based on the first loss function value, and determining a first joint mining decision-making unit;

determining a second loss function value between the prior page joint mining value labeling information and the page joint mining prediction value, optimizing model weight information of the second initial joint mining decision unit based on the second loss function value, and determining a second joint mining decision unit;

determining a third loss function value between the prior assisted joint mining value information and the assisted joint mining prediction value, optimizing model weight information of the assisted initial joint mining decision unit based on the third loss function value, and determining an assisted joint mining decision unit;

and when the first joint mining decision-making unit, the second joint mining decision-making unit and the assistant joint mining decision-making unit meet a training termination condition, outputting a joint mining decision model comprising the first joint mining decision-making unit, the second joint mining decision-making unit and the assistant joint mining decision-making unit.

In a second aspect, an embodiment of the present invention further provides a business prediction analysis system for joint big data mining, where the business prediction analysis system for joint big data mining includes an AI mining system and multiple internet AI mining systems in communication connection with the AI mining system;

the AI mining system is configured to:

the method comprises the steps of obtaining a plurality of updating behavior activities in user updating big data aiming at new online business, analyzing activity joint mining values corresponding to the updating behavior activities, and determining a first updating behavior activity from the updating behavior activities based on the activity joint mining values;

acquiring a page concern point data set which has an association relation with the user update big data, analyzing a page joint mining value corresponding to the first update behavior activity based on the page concern point data set and the first update behavior activity, and determining a second update behavior activity from the first update behavior activity based on the page joint mining value corresponding to the first update behavior activity;

analyzing an assisted joint mining value corresponding to the second updating behavior activity based on the page attention point data set and the second updating behavior activity, and analyzing joint mining reference information corresponding to the second updating behavior activity based on the assisted joint mining value corresponding to the second updating behavior activity;

By adopting the embodiment scheme of any one aspect, a plurality of updating behaviors in the user updating big data of the new online service are analyzed, the joint to-be-mined basic data for loading to the joint mining online service is determined, the joint to-be-mined basic data of the joint mining online service and the basic to-be-mined data of the joint mining online service are subjected to service situation analysis, the service situation information of a service group formed by the new online service and the corresponding joint mining online service is determined, the service image information of the service group is determined according to the service situation information of the service group, and the service image information of the service group is pushed to the corresponding service development terminal, so that the joint big data mining is performed on the new online service, the service information taking the service group as an analysis unit is output for corresponding service developers to analyze the condition of service audiences, thereby facilitating subsequent service optimization.

Drawings

Fig. 1 is a schematic flowchart of a business prediction analysis method for joint big data mining according to an embodiment of the present invention.

Detailed Description

The architecture of the traffic prediction analysis system 10 for joint big data mining according to an embodiment of the present invention is described below, and the traffic prediction analysis system 10 for joint big data mining may include an AI mining system 100 and an internet AI mining system 200 communicatively connected to the AI mining system 100. The AI mining system 100 and the internet AI mining system 200 in the business prediction analysis system for joint big data mining 10 may cooperatively perform the business prediction analysis method for joint big data mining described in the following method embodiments, and the detailed description of the method embodiments may be referred to in the following steps of the specific AI mining system 100 and the internet AI mining system 200.

The business prediction analysis method for joint big data mining provided by this embodiment may be executed by the AI mining system 100, and is described in detail below with reference to fig. 1.

The Process100 analyzes multiple update behavior activities in the user update big data of the new online service, and determines the combined to-be-mined base data for loading to the combined mining online service.

In this embodiment, the new online service may be an online service whose online time is less than a preset time, and a specific type of the new online service may be determined based on a current popular user requirement, which is not specifically limited herein, and for example, a current live e-commerce service for a popular product may be used as a new online service. The joint mining online service may refer to an online service that has a connection with the new online service, such as an e-commerce shopping application service currently directed to a hot product. Therefore, combined big data mining can be conveniently carried out by combining the on-line service and the combined mining on-line service.

The user updating big data may refer to user behavior data of a user in a new online service, the user behavior data is new updated behavior data compared with an online service, and the user updating big data may specifically include a plurality of updating behavior activities.

And the Process200 is used for performing service situation analysis on the basic data to be jointly mined of the jointly mined online service and the basic data to be jointly mined of the jointly mined online service, and determining service situation information of a service group formed by the new online service and the corresponding jointly mined online service.

In this embodiment, the service situation may be used to represent a situation and a state of a current user using process of a service group formed by the new online service and the corresponding combined mining online service, such as a situation and a state of a change in user interest, a situation and a state of a change shared by the user to another user, and the like.

The basic data to be mined and the basic data to be mined can be input into a business situation analysis model, and business situation information of a business group formed by the new online business and the corresponding online business to be mined is determined. The service situation analysis model can be obtained by training based on previously collected sample data to be mined and corresponding prior service situation information, for example, the previously collected sample data to be mined is input into an initialized service situation analysis model to predict corresponding service situation prediction information, and then model parameters of the initialized service situation analysis model are adjusted based on a loss function value between the service situation prediction information and the corresponding prior service situation information and are iteratively trained, so that the service situation analysis model meeting model deployment requirements is output.

And the Process300 determines the service portrait information of the service group according to the service situation information of the service group, and pushes the service portrait information of the service group to a corresponding service development terminal.

In this embodiment, the service portrait information may be used to represent audience categories of the service situation information of the service group, such as a high audience category, a medium audience category, and a low audience category, or may be selected according to audience confidence, such as 0 to 100, where a larger value indicates a wider audience area.

The business situation information of the business group can be input into a business portrait analysis model, and the business portrait information of the business group is determined. The service portrait analysis model can be obtained by training based on previously collected sample service situation information and corresponding prior service portrait information, for example, the previously collected sample service situation information is input into an initialized service portrait analysis model to predict corresponding service portrait prediction information, then model parameters of the initialized service portrait analysis model are adjusted based on a loss function value between the service portrait prediction information and the corresponding prior service portrait information, and iterative training is performed, so that the service portrait analysis model meeting model deployment requirements is output.

Based on the steps, the embodiment of the application analyzes a plurality of updating behaviors in the user updating big data of the new online service, determines the joint to-be-mined basic data for loading to the joint mining online service, performs service situation analysis on the joint to-be-mined basic data of the joint mining online service and the basic to-be-mined data of the joint mining online service, determines the service situation information of a service group formed by the new online service and the corresponding joint mining online service, determines the service image information of the service group according to the service situation information of the service group, and pushes the service image information of the service group to the corresponding service development terminal, thereby performing joint big data mining on the new online service, outputting the service image information taking the service group as an analysis unit for corresponding service developers to analyze the service audience condition, thereby facilitating subsequent service optimization.

The Process101 acquires a plurality of update behavior activities in the user update big data, analyzes activity joint mining values corresponding to the plurality of update behavior activities, and determines a first update behavior activity from the plurality of update behavior activities based on the plurality of activity joint mining values.

For example, obtaining user update big data, dividing and treating the user update big data based on the behavior activity time trigger sequence, and determining a plurality of update behavior activities corresponding to the user update big data; the plurality of updating behavior activities comprise updating behavior activities ACTb, b is a positive integer and is not more than the global quantity corresponding to the plurality of updating behavior activities; acquiring S behavior activity themes from the updated behavior activity ACTb and frequency interaction data corresponding to the S behavior activity themes; s is a positive integer; performing frequent interaction intention analysis on the updating behavior activity ACTb, determining a frequent interaction intention, and extracting active behavior interaction resources and passive behavior interaction resources in the updating behavior activity ACTb; taking the frequent interaction intentions, the active behavior interaction resources and the passive behavior interaction resources as behavior activity knowledge points SNb corresponding to the updated behavior activity ACTb; generating joint mining analysis characteristics corresponding to the updated behavior activity ACTb based on the S behavior activity themes, the S frequent interaction data and the behavior activity knowledge points SNb; and analyzing the activity joint mining value corresponding to each of the plurality of updating behavior activities based on the joint mining analysis characteristics corresponding to each of the plurality of updating behavior activities.

The specific process of generating the joint mining analysis feature corresponding to the update behavior activity ACTb may include: acquiring a combined mining decision model; the joint mining decision model comprises a first joint mining decision unit; the first joint mining decision unit comprises a first feature mining branch, a second feature mining branch, a third feature mining branch and a fourth feature mining branch; respectively loading S behavior activity themes to a first feature mining branch, respectively performing feature mining on the S behavior activity themes through the first feature mining branch, determining operation behavior member features corresponding to the S behavior activity themes, aggregating the S operation behavior member features, and determining behavior activity features corresponding to the updated behavior activity ACTb; respectively loading S pieces of frequent interaction data to a second feature mining branch, respectively performing feature mining on the S pieces of frequent interaction data through the second feature mining branch, determining frequent interaction member features corresponding to the S pieces of frequent interaction data, aggregating the S pieces of frequent interaction member features, and determining frequent interaction features corresponding to the updating action ACTb; loading the behavior activity knowledge points SNb to a third feature mining branch, extracting key description data in the behavior activity knowledge points SNb through the third feature mining branch, performing feature mining on the key description data, and determining key description features corresponding to the key description data; and respectively loading the behavior activity characteristics, the frequent interaction characteristics and the key description characteristics to a fourth characteristic mining branch, aggregating the behavior activity characteristics, the frequent interaction characteristics and the key description characteristics through the fourth characteristic mining branch, and determining the joint mining analysis characteristics corresponding to the updated behavior activity ACTb.

The specific process of analyzing the activity joint mining value corresponding to each of the plurality of updating behavior activities may include: acquiring a combined mining decision model; the joint mining decision model comprises a first joint mining decision unit; the first joint mining decision unit comprises a first value prediction branch; loading the joint mining analysis features corresponding to the updating behavior activity ACTb into a first value prediction branch, performing joint mining value output on the joint mining analysis features corresponding to the updating behavior activity ACTb through the first value prediction branch, and determining a activity joint mining value corresponding to the updating behavior activity ACTb; the specific process of determining a first update behavioral activity from the plurality of update behavioral activities may include: respectively comparing activity joint mining values corresponding to the plurality of updating behavior activities with a first preset mining value; and in the plurality of updating behavior activities, combining the activities not less than the first preset mining value with the updating behavior activities corresponding to the mining value to serve as first updating behavior activities.

The AI mining system acquires the user update big data, segments the user update big data through the action activity time trigger sequence, and determines a plurality of update action activities of the user update big data. It is understood that, since the process of determining the activity joint mining value corresponding to each update behavior activity by the AI mining system is consistent, the embodiment of the present invention is described by taking the determination of the activity joint mining value corresponding to the update behavior activity ACT1 as an example, and the process of determining the activity joint mining value corresponding to the remaining update behavior activities in the plurality of update behavior activities is described below, where the update behavior activity ACT1 belongs to the update behavior activity ACTb described above. The AI mining system acquires S behavior activity themes from the updated behavior activity ACT1 and the respective corresponding frequent interaction data of the S behavior activity themes; the AI mining system performs frequent interaction intention analysis on the update behavior activity ACT1 to determine frequent interaction intention; extracting active behavior interaction resources in the update behavior activity ACT1 and extracting passive behavior interaction resources; in some embodiments, the AI mining system uses the frequent interaction intents, the active behavior interaction resources, and the passive behavior interaction resources as the behavior activity knowledge points SN1 corresponding to the updated behavior activity ACT 1.

The AI mining system obtains a first joint mining decision unit in the joint mining decision model, the first joint mining decision unit including a first feature mining branch 40a, a second feature mining branch 40b, a third feature mining branch 40c, a fourth feature mining branch 40e, and a first value predicting branch 40 f. The AI mining system loads S behavior activity topics to the first feature mining branch 40a, and if the S behavior activity topics include a first behavior activity topic and a second behavior activity topic, feature mining is performed on the first behavior activity topic through the first feature mining branch 40a to obtain first operation behavior member features corresponding to the first behavior activity topic, and feature mining is performed on the second behavior activity topic to obtain second operation behavior member features corresponding to the second behavior activity topic, so that the AI mining system can obtain operation behavior member features corresponding to the S behavior activity topics; aggregating the S operational behavior member characteristics 401a, the AI mining system may obtain the behavior activity characteristics 401d corresponding to the updated behavior activity ACT 1.

In addition, the AI mining system loads S frequent interaction data into the second feature mining branch 40b, respectively, assuming that the S frequent interaction data includes first frequent interaction data corresponding to the first behavioral activity topic, and second frequently interacted data corresponding to the second behavioral activity theme, then, via the second feature mining branch 40b, feature mining is carried out on the first frequent interaction data, so that first frequent interaction member features corresponding to the first frequent interaction data can be obtained, feature mining is carried out on the second frequent interaction data, so that second frequent interaction member features corresponding to the second frequent interaction data can be obtained, therefore, the AI mining system can obtain the frequent interaction member characteristics corresponding to the S frequent interaction data respectively, the S frequently-interacting member features 401b are aggregated to determine the frequently-interacting feature 402d corresponding to the update behavior activity ACT 1. The second feature mining branch 40b may include Conv-TasNet, BilSTM-TasNet, etc.

The AI mining system loads the behavior activity knowledge point SN1 to the third feature mining branch 40c, extracts key description data in the behavior activity knowledge point SN1 through the third feature mining branch 40c, performs feature mining on the key description data, and determines key description features corresponding to the key description data. The third feature mining branch 40c may be a bi-directional coding model or the like.

In some embodiments, the AI mining system loads the behavior activity characteristic 401d, the frequent interaction characteristic 402d, and the key description characteristic 403d to the fourth characteristic mining branch 40e, and aggregates the behavior activity characteristic 401d, the frequent interaction characteristic 402d, and the key description characteristic 403d through the fourth characteristic mining branch 40e, so as to obtain a joint mining analysis characteristic 401e corresponding to the updated behavior activity ACT 1. The AI mining system loads the joint mining analysis features 401e to the first value prediction branch 40f, and performs joint mining value output on the joint mining analysis features 401e through the first value prediction branch 40f to determine activity joint mining values corresponding to the updated behavior activities ACT 1. Wherein, based on the plurality of activities jointly mining the value, the AI mining system determines a specific process of the first updated behavior activity from the plurality of updated behavior activities, please refer to the foregoing description.

The Process102 acquires a page attention point data set having an association relation with user update big data, analyzes a page joint mining value corresponding to a first update behavior activity based on the page attention point data set and the first update behavior activity, and determines a second update behavior activity from the first update behavior activity based on the page joint mining value corresponding to the first update behavior activity.

For example, obtaining page concern data of a page target associated with user update big data, and obtaining page concern data of a combined mining online service associated with the page target; generating a page concern point data set based on page concern point data of a page target and page concern point data of combined mining online services; acquiring a joint mining decision model, and respectively loading a page focus point data set and first updating behavior activities into the joint mining decision model; the joint mining decision model comprises a second joint mining decision unit; the second joint mining decision unit comprises a first page attention feature mining branch; performing page attention feature mining on each page attention point data in the page attention point data set through a first page attention feature mining branch, and determining a first page attention feature corresponding to the page attention point data set; and acquiring a joint mining analysis characteristic corresponding to the first updating behavior activity, and analyzing a page joint mining value corresponding to the first updating behavior activity based on the first page attention characteristic and the joint mining analysis characteristic corresponding to the first updating behavior activity.

The second joint mining decision unit also comprises a first characteristic aggregation branch and a second value prediction branch; the specific process of determining the joint page mining value corresponding to the first updating behavior activity may include: respectively loading the first page attention feature and the joint mining analysis feature corresponding to the first updating behavior activity to a first feature aggregation branch; performing feature aggregation on the first page attention feature and the joint mining analysis feature corresponding to the first updating behavior activity through the first feature aggregation branch, and determining a first joint mining aggregation feature corresponding to the first updating behavior activity; loading the first joint mining aggregation feature to a second value prediction branch, performing joint mining value output on the first joint mining aggregation feature through the second value prediction branch, and determining a page joint mining value corresponding to a first updating behavior activity;

wherein the number of the first updating behavior activities is multiple; the specific process of determining the second update behavior activity from the first update behavior activity may include: comparing the joint mining value of the page corresponding to each of the plurality of first updating behavior activities with a second preset mining value; and in the plurality of first updating behavior activities, the first updating behavior activity corresponding to the joint mining value of the page larger than the second preset mining value is used as a second updating behavior activity.

The Process101 determines the first updating behavior activity with high joint mining value, and the first updating behavior activity is limited to the correlation of the attention points of the page in the step, so that the constructed first updating behavior activity is more fit with the attention points of the object, and the reliability of joint mining of the big data updated by the user can be further improved. The AI mining system obtains page focus data (abbreviated as new online page focus data) of the page target, and the new online page focus data can represent the focus of the page target.

The embodiment of the invention can provide two modes for acquiring the joint mining analysis characteristics corresponding to the first updating behavior activity, wherein the first mode comprises the following steps: the Process101 provides joint mining analysis features (including the joint mining analysis feature 401 e) corresponding to each of the plurality of update behavior activities, and the first update behavior activity belongs to the plurality of update behavior activities, so that the AI mining system can obtain the joint mining analysis feature corresponding to the first update behavior activity from the joint mining analysis features corresponding to each of the plurality of update behavior activities output by the first joint mining decision unit. The AI mining system may obtain, through the first joint mining decision unit 20e, the joint mining analysis feature corresponding to the updated behavior activity 201d, the joint mining analysis feature corresponding to the updated behavior activity 202d, the joint mining analysis feature corresponding to the updated behavior activity 203d, the joint mining analysis feature corresponding to the updated behavior activity 204d, the joint mining analysis feature corresponding to the updated behavior activity 205d, and the joint mining analysis feature corresponding to the updated behavior activity 206d, respectively, and determine that the updated behavior activity 201d, the updated behavior activity 202d, the updated behavior activity 203d, and the updated behavior activity 206d are the first updated behavior activity, and may directly perform, on the basis of the joint mining analysis feature for the updated behavior activity 201d, the joint mining analysis feature for the updated behavior activity 202d, which is output by the first joint mining decision unit 20e, The joint mining analysis feature for the update behavior activity 203d and the joint mining analysis feature for the update behavior activity 206d are used as the joint mining analysis feature corresponding to the first update behavior activity. The first method for obtaining the joint mining analysis characteristics corresponding to the first updating behavior activity can reduce the operation time and the operation cost of the joint mining decision model.

In order to improve the accuracy of the joint mining analysis feature corresponding to the first updated behavior activity, the AI mining system may adopt a second manner, and when training the second joint mining decision unit, the AI mining system takes the model weight information corresponding to the first feature mining branch 40a, the second feature mining branch 40b, the third feature mining branch 40c, and the fourth feature mining branch 40e in the trained first joint mining decision unit as the initial model weight information, and updates the initial model weight information according to the second example model training data set (including the example user update data, the example page attention data set, and the prior page joint mining value label information corresponding to the example user update data). The Process of obtaining the joint mining analysis feature 402e corresponding to the first updated behavior activity by the AI mining system through the initial model weight information is consistent with the Process of obtaining the joint mining analysis feature corresponding to each of the plurality of updated behavior activities through the first joint mining decision unit, so please refer to the technical scheme of the Process101 above. Since this initial model weight information is superior to the aforementioned model weight information, the joint mining analysis feature 402e is superior to a plurality of joint mining analysis features in the Process 101.

The second joint mining decision unit may include a first page attention feature mining branch 40g, a first feature aggregation branch 40h, and a second value prediction branch 40 i. Through the first page attention feature mining branch 40h, the AI mining system performs page attention feature mining on each page attention point data in the page attention point data set to determine a first page attention feature 401g corresponding to the page attention point data set; the AI mining system loads the first page attention feature 401g and the joint mining analysis feature (for example, the joint mining analysis feature 402 e) corresponding to the first update behavior activity to the first feature aggregation branch 40h, and performs feature aggregation on the first page attention feature 401g and the joint mining analysis feature 402e through the first feature aggregation branch 40h to obtain a first joint mining aggregation feature 401h corresponding to the first update behavior activity; in some embodiments, the AI mining system loads the first joint mining aggregate feature 401h into the second value prediction branch 40i, and the page joint mining value corresponding to the first updated behavior activity may be obtained through the second value prediction branch 40 i.

The AI mining system determines a process of a second update behavior activity from the first update behavior activity based on the page joint mining value corresponding to the first update behavior activity, please refer to the related description above.

The Process103 analyzes the joint mining assistance value corresponding to the second updating behavior activity based on the page focus data set and the second updating behavior activity, and analyzes the joint mining reference information corresponding to the second updating behavior activity based on the joint mining assistance value corresponding to the second updating behavior activity.

The AI mining system analyzes the assisted joint mining value corresponding to the second update behavior activity through the assisted joint mining decision unit in the joint mining decision model, and further determines the process of joint mining reference information, please refer to the related description above.

And the Process104 determines the joint mining updating behavior activity from the second updating behavior activity based on the activity joint mining value corresponding to the second updating behavior activity, the page joint mining value corresponding to the second updating behavior activity, and the assisting joint mining value corresponding to the second updating behavior activity, and uses the joint mining updating behavior activity and the joint mining reference information corresponding to the joint mining updating behavior activity as the joint to-be-mined base data for loading to the joint mining online service.

For example, the number of the second updating behavior activities is plural, the plurality of second updating behavior activities includes a second updating behavior activity Jm, m is a positive integer, and m is not greater than the number of the plurality of second updating behavior activities; performing weight fusion on the activity joint mining value corresponding to the second updating behavior activity Jm, the page joint mining value corresponding to the second updating behavior activity Jm and the assisting joint mining value corresponding to the second updating behavior activity Jm, and determining a global joint mining value corresponding to the second updating behavior activity Jm; acquiring a maximum global joint mining value from global joint mining values corresponding to the plurality of second updating behavior activities; determining the second updating behavior activity corresponding to the maximum global joint mining value as a joint mining updating behavior activity in the plurality of second updating behavior activities; and acquiring the joint mining reference information corresponding to the joint mining updating behavior activities from the joint mining reference information corresponding to the second updating behavior activities.

Therefore, the AI mining system determines the activity joint mining value corresponding to each of the plurality of updating behavior activities in the user updating big data, so that a first updating behavior activity can be determined from the plurality of updating behavior activities based on the plurality of activity joint mining values, and it can be understood that the first updating behavior activity belongs to the user updating big data and the joint mining value (quality) of the first updating behavior activity is superior to the joint mining value of the user updating big data; in some embodiments, the AI mining system obtains a page attention point data set having an association relationship with the user update big data, and analyzes a page joint mining value corresponding to a first update behavior activity based on the page attention point data set and the first update behavior activity, so that a second update behavior activity can be determined from the first update behavior activity based on the page joint mining value corresponding to the first update behavior activity, and it can be understood that the second update behavior activity is determined not only according to the user update big data content of the first update behavior activity, but also according to the page attention point data set, so that the joint mining value thereof is superior to the joint mining value of the first update behavior activity; in some embodiments, the AI mining system analyzes the joint mining assistance value corresponding to the second update behavior activity based on the page attention point data set and the second update behavior activity, and analyzes the joint mining reference information corresponding to the second update behavior activity based on the joint mining assistance value corresponding to the second update behavior activity, it being understood that the joint mining reference information is not only associated with the second update behavior activity, but also associated with the page attention point data set; in some embodiments, the AI mining system determines a joint mining update behavior activity from the second update behavior activity based on an activity joint mining value corresponding to the second update behavior activity, a page joint mining value corresponding to the second update behavior activity, and an assisted joint mining value corresponding to the second update behavior activity, and uses the joint mining update behavior activity and joint mining reference information corresponding to the joint mining update behavior activity as joint to-be-mined base data for loading to the joint mining online service. Therefore, the combined mining efficiency and the combined mining effect of the big data updated by the user can be improved by combining the basic data to be mined.

Another embodiment of the invention is further described below.

The Process201 obtains a plurality of updating behavior activities in the user updating big data, analyzes activity joint mining values corresponding to the plurality of updating behavior activities, and determines a first updating behavior activity from the plurality of updating behavior activities based on the plurality of activity joint mining values.

The Process202 acquires a page attention point data set having an association relation with user update big data, analyzes a page joint mining value corresponding to a first update behavior activity based on the page attention point data set and the first update behavior activity, and determines a second update behavior activity from the first update behavior activity based on the page joint mining value corresponding to the first update behavior activity.

For specific implementation processes of the processes 201 to 202, please refer to the processes 101 to 102 in the foregoing embodiments, which are not described herein again.

The Process203 obtains joint mining theme characteristics corresponding to the multiple behavior activity themes in the second updating behavior activity.

For example, generating a time-space domain based on the theme, performing theme output on the second updating behavior activity, and determining a plurality of behavior activity themes in the second updating behavior activity; acquiring a combined mining decision model; the joint mining decision model comprises a third joint mining decision unit; the third joint mining decision unit comprises a theme analysis branch; and respectively loading the plurality of behavior activity themes to a theme analysis branch, respectively carrying out feature mining on the plurality of behavior activity themes through the theme analysis branch, and determining the joint mining theme features corresponding to the plurality of behavior activity themes.

The AI mining system may obtain a plurality of behavior activity topics from the second updated behavior activity through a topic generation time-space domain, the plurality of behavior activity topics are all used as candidate assistance topics, the AI mining system needs to determine topic joint mining values corresponding to the plurality of behavior activity topics, and further determine topic joint mining values corresponding to the second updated behavior activity, a third joint mining decision unit is introduced below, and a process of obtaining the topic joint mining values corresponding to each behavior activity topic by the AI mining system through the third joint mining decision unit is consistent, so the embodiment of the present invention is described by taking obtaining the topic joint mining value corresponding to the behavior activity topic TH1 as an example, and an implementation manner of remaining behavior activity topics among the plurality of behavior activity topics is please refer to the description below.

The AI mining system loads a behavior activity topic TH1 (belonging to a behavior activity topic THg in the Process 204) to a topic parsing branch 70a in the third joint mining decision unit, and performs feature mining on the behavior activity topic TH1 through the topic parsing branch 70a to determine a joint mining topic feature 701a corresponding to the behavior activity topic TH 1.

The Process204 analyzes a theme joint mining value corresponding to the second updating behavior activity based on the multiple joint mining theme features, the second updating behavior activity and the page focus point data set, and analyzes an assistance theme corresponding to the second updating behavior activity based on the theme joint mining value corresponding to the second updating behavior activity; and the assistance theme corresponding to the second updating action belongs to a plurality of action themes.

For example, the plurality of behavior activity themes include a behavior activity theme THg, and the plurality of joint mining theme features include joint mining theme features corresponding to the behavior activity theme THg; g is a positive integer and is not greater than the global quantity corresponding to the plurality of behavior activity themes; acquiring a joint mining analysis characteristic corresponding to the second updating behavior activity, and acquiring a second page attention characteristic corresponding to the page attention point data set; acquiring a combined mining decision model; the joint mining decision model comprises a third joint mining decision unit; the third joint mining decision unit comprises a second feature aggregation branch; respectively loading the joint mining theme characteristics corresponding to the behavior activity theme THg, the joint mining analysis characteristics corresponding to the second updated behavior activity and the second page attention characteristics to a second characteristic aggregation branch; performing feature aggregation on the joint mining subject feature corresponding to the behavior activity subject THg, the joint mining analysis feature corresponding to the second updated behavior activity and the second page attention feature through a second feature aggregation branch, and determining a second joint mining aggregation feature corresponding to the behavior activity subject THg; and analyzing the topic joint mining value corresponding to the second updating behavior activity based on the second joint mining aggregation characteristics corresponding to the plurality of behavior activity topics, and analyzing the assistance topic corresponding to the second updating behavior activity based on the topic joint mining value corresponding to the second updating behavior activity.

The third joint mining decision unit also comprises a third valence prediction branch; the specific process of determining the assistance topic corresponding to the second updating behavior activity may include: loading the second combined mining aggregation characteristics corresponding to the behavior activity theme THg to a third value prediction branch, and performing combined mining value output on the second combined mining aggregation characteristics corresponding to the behavior activity theme THg through the third value prediction branch to determine a theme combined mining value corresponding to the behavior activity theme THg; acquiring the maximum theme joint mining value from theme joint mining values corresponding to the themes of the plurality of behavior activities respectively, and determining the maximum theme joint mining value as the theme joint mining value corresponding to the second updated behavior activity; and determining the behavior activity theme corresponding to the maximum theme joint mining value as an assistance theme corresponding to the second updated behavior activity in the plurality of behavior activity themes.

The embodiment of the present invention may provide 3 different manners for obtaining the joint mining analysis feature corresponding to the second update behavior activity, and the first obtaining manner may refer to the description of the Process102 in the foregoing embodiment about obtaining the joint mining analysis feature corresponding to the first update behavior activity, and the principles of the two are consistent; the second obtaining manner is similar to the first obtaining manner, the Process102 already provides the joint mining analysis feature (including the joint mining analysis feature 402 e) corresponding to the first updating behavior activity, and the second updating behavior activity belongs to the first updating behavior activity, so that the AI mining system can obtain the joint mining analysis feature corresponding to the second updating behavior activity from the joint mining analysis feature corresponding to the first updating behavior activity output by the second joint mining decision unit.

In order to improve the accuracy of the joint mining analysis feature corresponding to the second update behavior activity, the AI mining system may adopt a third manner, and a third joint mining decision unit is introduced below, which is the same as the parameter layer in the second joint mining decision unit but is inconsistent with the model weight information between the second joint mining decision unit and the third joint mining decision unit, because when the third joint mining decision unit is trained, the AI mining system uses the model weight information in the trained second joint mining decision unit as the initial model weight information, and performs fine tuning on the initial model weight information according to a third example model training data set (including example user update data, an example page attention point data set, a prior assistance topic corresponding to the example user update data, and assistance topic joint mining value information corresponding to the example user update data). It can be understood that the Process of obtaining the joint mining analysis feature corresponding to the second updated behavior activity by the AI mining system through the initial model weight information is the same as the Process of obtaining the joint mining analysis feature 402e through the second joint mining decision unit, so please refer to the above technical scheme of the Process 101. Since the model weight information of this time is superior to the model weight information, the joint mining analysis feature corresponding to the second update behavior activity output this time is superior to the joint mining analysis feature 402 e.

The AI mining system loads the joint mining topic feature 701a corresponding to the behavior activity topic TH1, the joint mining analysis feature corresponding to the second updated behavior activity, and the second page attention feature to the second feature aggregation branch 70b, respectively; through the second feature aggregation branch 70b, feature aggregation can be performed on the joint mining topic feature 701a, the joint mining analysis feature corresponding to the second updated behavior activity, and the second page attention feature, so that a second joint mining aggregation feature 701b corresponding to the behavior activity topic TH1 can be obtained; in some embodiments, the AI mining system loads the second combined mining aggregation feature 701b into the third valence prediction branch 70c, and the third value prediction branch 70c may output the combined mining value to the second combined mining aggregation feature 701b, so that the topic combined mining value corresponding to the behavior activity topic TH1 may be obtained. According to the above description, the AI mining system may obtain topic joint mining values corresponding to each of the plurality of behavioral activity topics.

The Process205 analyzes the joint mining value of the frequent item corresponding to the second updating behavior activity and the assisting frequent item corresponding to the second updating behavior activity based on the page focus data set and the behavior activity knowledge point corresponding to the second updating behavior activity.

For example, the assistance frequent item is composed of N joint mining behavior instances; acquiring a combined mining decision model; the joint mining decision model comprises a fourth joint mining decision unit; the fourth joint mining decision unit comprises a second page attention feature mining branch, a third page attention feature mining branch, a frequent evaluation branch and a frequent item decision branch; loading the behavior activity knowledge points corresponding to the second updating behavior activity to a second page attention feature mining branch, performing page attention feature mining on the behavior activity knowledge points corresponding to the second updating behavior activity through the second page attention feature mining branch, and determining behavior activity knowledge point features; loading the page concern point data set to a third page concern feature mining branch, and performing page concern feature mining on the page concern point data set through the third page concern feature mining branch to determine a third page concern feature; respectively loading the behavior activity knowledge point feature, the frequent item feature Fi to be evaluated corresponding to the second updated behavior activity and the third page attention feature to a frequent evaluation branch, and aggregating the behavior activity knowledge point feature, the frequent item feature Fi to be evaluated and the third page attention feature through the frequent evaluation branch to determine a frequent evaluation coefficient corresponding to the behavior activity knowledge point feature; i is a non-negative integer less than N; analyzing the characteristic Fi +1 of the frequent item to be evaluated, which corresponds to the second updated behavior activity, based on the frequent evaluation coefficient corresponding to the behavior activity knowledge point characteristic; the joint mining behavior example indicated by the frequent item feature Fi to be evaluated is the last joint mining behavior example of the joint mining behavior example indicated by the frequent item feature Fi +1 to be evaluated; when i +1 is equal to N, respectively loading the N frequent item features to be evaluated to the frequent item decision branches, generating joint mining behavior examples respectively indicated by the N frequent item features to be evaluated through the frequent item decision branches, and outputting the N joint mining behavior examples as the assisted frequent items corresponding to the second updating behavior activity; and analyzing the joint mining value of the frequent items corresponding to the second updating behavior activity based on the N frequent item characteristics to be evaluated.

The definition of the behavior activity knowledge point corresponding to the second updated behavior activity refers to the definition of the behavior activity knowledge point SN1, the definition of the second page attention feature mining branch and the third page attention feature mining branch, and the definition of the first page attention feature mining branch.

The Process206 analyzes the assisted joint mining value corresponding to the second updating behavior activity based on the topic joint mining value corresponding to the second updating behavior activity and the frequent item joint mining value corresponding to the second updating behavior activity; and analyzing the joint mining reference information corresponding to the second updating behavior activity based on the assistance theme corresponding to the second updating behavior activity and the assistance frequent item corresponding to the second updating behavior activity.

And the Process207 determines the joint mining updating behavior activity from the second updating behavior activity based on the activity joint mining value corresponding to the second updating behavior activity, the page joint mining value corresponding to the second updating behavior activity and the assisting joint mining value corresponding to the second updating behavior activity, and uses the joint mining updating behavior activity and the joint mining reference information corresponding to the joint mining updating behavior activity as the joint to-be-mined basic data for loading to the joint mining online service.

For example, the activity joint mining value corresponding to the second updating behavior activity, the page joint mining value corresponding to the second updating behavior activity, the topic joint mining value corresponding to the second updating behavior activity, and the frequent item joint mining value corresponding to the second updating behavior activity are subjected to weight fusion to determine the global joint mining value corresponding to the second updating behavior activity, and the subsequent Process may refer to the technical scheme of the Process104 in the foregoing embodiment.

Another embodiment of the invention is described below.

A Process301, obtaining a training data set of the example model; the example model training data set comprises example user update data, an example page attention data set of a prior page target in an association relationship with the example user update data, data joint mining value labeling information corresponding to the example user update data, prior page joint mining value labeling information corresponding to the example user update data, and prior assistance joint mining value information corresponding to the example user update data.

For example, the number of the example user update data is plural, the plurality of example user update data includes example user update data Po, o is a positive integer and o is not greater than the number of the plurality of example user update data; performing product operation on the service heat corresponding to the example user update data Po, the continuous state value corresponding to the example user update data Po and the average service interaction rate corresponding to the example user update data Po to determine a first prior evaluation value corresponding to the example user update data Po; summing the number of the passive behavior interaction resources corresponding to the example user update data Po and the number of the passive behavior interaction resources corresponding to the example user update data Po to determine a second prior evaluation value corresponding to the example user update data Po; obtaining a maximum first prior evaluated value from among first prior evaluated values corresponding to the sample user update data, and obtaining a maximum second prior evaluated value from among second prior evaluated values corresponding to the sample user update data; determining a first ratio between a first prior evaluation value corresponding to the example user update data Po and a maximum first prior evaluation value, and analyzing a second ratio between a second prior evaluation value corresponding to the example user update data Po and a maximum second prior evaluation value; performing weight fusion on the first ratio and the second ratio, and determining candidate data joint mining value labeling information corresponding to the sample user update data Po; comparing the candidate data joint mining value labeling information corresponding to the example user updating data Po with a third preset mining value; if the candidate data joint mining value labeling information corresponding to the example user update data Po is smaller than a third preset mining value, determining the candidate data joint mining value labeling information corresponding to the example user update data Po as data joint mining value labeling information corresponding to the example user update data Po; and if the candidate data joint mining value labeling information corresponding to the example user updating data Po is not less than the third preset mining value, determining the third preset mining value as the data joint mining value labeling information corresponding to the example user updating data Po.

For example, a first service interaction rate of the prior page target for the update data of the sample user is obtained, and the first service interaction rate is compared with a first set service interaction rate; if the first service interaction rate is greater than the first set service interaction rate, determining that a first forward contact relation exists between the sample page attention point data and the sample user updating data, and determining the first forward contact relation as the priori page joint mining value labeling information; and if the first service interaction rate is not greater than the first set service interaction rate, determining that a first negative direction relation exists between the sample page concern data and the sample user updating data, and determining the first negative direction relation as the prior page combined mining value labeling information.

For example, the example model training data set further includes a priori assistance topics corresponding to the example user update data; the priori assistance combined mining value information comprises assistance topic combined mining value information; acquiring a second service interaction rate of the prior page target aiming at the update data of the example user, and comparing the second service interaction rate with a second set service interaction rate; if the second service interaction rate is greater than the second set service interaction rate, determining that a second forward contact relation exists among the prior assistance theme, the example page focus data and the example user update data, and determining the second forward contact relation as assistance theme joint mining value information; and if the second service interaction rate is not greater than the second set service interaction rate, determining that a second negative direction relation exists among the prior assistance topic, the example page attention point data and the example user updating data, and determining the second negative direction relation as the assistance topic joint mining value information.

The example model training data set may include a first example model training data set for training the first initial joint mining decision unit, a second example model training data set for training the second initial joint mining decision unit, and an assistance example model training data set for training the assistance initial joint mining decision unit, and when the joint mining reference information includes only assistance topics, the assistance initial joint mining decision unit includes a third initial joint mining decision unit, and the assistance example model training data set is the third example model training data set; when the joint mining reference information only comprises a frequent assisting item, the initial assisting joint mining decision unit comprises a fourth initial joint mining decision unit, and the assisting example model training data set is a fourth example model training data set; when the joint mining reference information comprises the assistance topic and the assistance frequent item, the assistance initial joint mining decision unit comprises a third initial joint mining decision unit and a fourth initial joint mining decision unit, and the assistance case model training data set comprises a third case model training data set and a fourth case model training data set. The first example model training data set comprises example user update data and data joint mining value labeling information corresponding to the example user update data; the fourth example model training data set includes example user update data, an example page attention data set of a prior page target associated with the example user update data, and assistance frequent item quality tags corresponding to the example user update data.

It can be understood that the third preset mining value, the first set service interaction rate, the second set service interaction rate, and the third set service interaction rate can all be set based on actual requirements.

The Process302 loads the example model training data set to the initial joint mining decision model, and analyzes the joint mining prediction value of the big user update data corresponding to the example user update data through the initial joint mining decision model.

For example, the AI mining system may load a first example model training dataset in the Process301 to a first initial joint mining decision unit in the initial joint mining decision model, where the AI mining system determines, through the first initial joint mining decision unit, an implementation manner of a user update big data joint mining prediction value corresponding to the example user update data, and the implementation manner of an activity joint mining prediction value corresponding to the update behavior activity is determined through the first joint mining decision unit, so please refer to the technical solution of the Process101 in the foregoing embodiment.

The Process303 analyzes the joint page mining prediction value corresponding to the example user update data and the assisted joint mining prediction value corresponding to the example user update data based on the example page point of interest data set and the example user update data.

For example, the AI mining system may load a second example model training dataset in the Process301 to a second initial joint mining decision unit in the initial joint mining decision model, where the AI mining system determines, through the second initial joint mining decision unit, an implementation manner of a page joint mining prediction value corresponding to the example user update data, and the implementation manner of a page joint mining prediction value corresponding to the update behavior activity is determined through the second joint mining decision unit, which is consistent with the implementation manner of the page joint mining value corresponding to the update behavior activity, so please refer to the technical solution of the Process102 in the foregoing embodiment.

The AI mining system may load the assisting example model training data set in the Process301 to an assisting initial joint mining decision unit in the initial joint mining decision model, wherein the AI mining system determines, through the assisting initial joint mining decision unit, an implementation manner of an assisting joint mining prediction value corresponding to the example user update data, and determines, through the assisting joint mining decision unit, an implementation manner of an assisting joint mining value corresponding to the update behavior activity, which is the same, so please refer to the technical scheme of the Process103 in the foregoing embodiment.

The Process304 is used for optimizing model weight information of the initial joint mining decision model and determining a joint mining decision model based on the data joint mining value labeling information, the prior page joint mining value labeling information, the prior assisted joint mining value information, the user updated big data joint mining prediction value, the page joint mining prediction value and the assisted joint mining prediction value; the joint mining decision model is used for determining joint to-be-mined basic data of the big data updated by the user; the joint basic data to be mined comprises joint mining updating behavior activities in the user updating big data and joint mining reference information corresponding to the joint mining updating behavior activities.

For example, the initial joint mining decision model comprises a first initial joint mining decision unit for determining the user update big data joint mining prediction value, a second initial joint mining decision unit for determining the page joint mining prediction value, and an assisted initial joint mining decision unit for determining the assisted joint mining prediction value; the model weight information in the initial joint mining decision model comprises model weight information in a first initial joint mining decision unit, model weight information in a second initial joint mining decision unit and model weight information in an auxiliary initial joint mining decision unit; determining a first loss function value between the data joint mining value labeling information and the user updated big data joint mining prediction value, optimizing model weight information of the first initial joint mining decision unit based on the first loss function value, and determining a first joint mining decision unit; determining a second loss function value between the prior page joint mining value labeling information and the page joint mining prediction value, optimizing model weight information of a second initial joint mining decision unit based on the second loss function value, and determining a second joint mining decision unit; determining a third loss function value between the prior assisted joint mining value information and the assisted joint mining prediction value, optimizing model weight information of the assisted initial joint mining decision unit based on the third loss function value, and determining an assisted joint mining decision unit; and when the first joint mining decision unit, the second joint mining decision unit and the assistant joint mining decision unit meet the training termination condition, outputting a joint mining decision model comprising the first joint mining decision unit, the second joint mining decision unit and the assistant joint mining decision unit.

The embodiment of the invention deeply models the first initial joint mining decision unit through the first example model training data set so that the first joint mining decision unit can determine the first updating behavior activity with high joint mining value in a plurality of updating behavior activities, deeply models the second initial joint mining decision unit through the second example model training data set so that the second joint mining decision unit can determine the second updating behavior activity with high joint mining value in the first updating behavior activity, deeply models the assisting initial joint mining decision unit through the assisting example model training data set so that the assisting joint mining decision unit can determine the assisting joint mining value and the joint mining reference information corresponding to the second updating behavior activity, and further can obtain the joint mining values with different dimensions, the joint mining updating behavior activities and the corresponding joint mining reference information are analyzed, and then the joint to-be-mined basic data can be generated.

In some embodiments, the AI-mining system 100 may include a processor 110, a machine-readable storage medium 120, a bus 130, and a communication unit 140.

The processor 110 may perform various suitable actions and processes based on a program stored in the machine-readable storage medium 120, such as program instructions associated with the business prediction analysis methods for federated big data mining described in the foregoing embodiments. The processor 110, the machine-readable storage medium 120, and the communication unit 140 perform signal transmission through the bus 130.

In particular, the processes described in the above exemplary flow diagrams may be implemented as computer software programs, according to embodiments of the present invention. For example, embodiments of the invention include a computer program product comprising a computer program embodied on a computer-readable medium, the computer program comprising program code for performing the method illustrated in the flow chart. In such an embodiment, the computer program may be downloaded and installed from a network via the communication unit 140, and when executed by the processor 110, performs the above-described functions defined in the methods of the embodiments of the present invention.

Yet another embodiment of the present invention further provides a computer-readable storage medium, in which computer-executable instructions are stored, and when the computer-executable instructions are executed by a processor, the method for business prediction analysis for joint big data mining is implemented as described in any of the above embodiments.

Yet another embodiment of the present invention further provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, the method for business prediction analysis for joint big data mining according to any of the above embodiments is implemented.

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

Claims

1. A business prediction analysis method for joint big data mining, wherein the method is executed by an AI mining system, and the method comprises the following steps:

analyzing a plurality of updating behavior activities in the user updating big data of the new online service, and determining combined basic data to be mined, which are used for being loaded to the combined mining online service;

Analyzing the service situation of the basic data to be jointly mined of the jointly mined online service and the basic data to be jointly mined of the jointly mined online service, and determining the service situation information of a service group formed by the new online service and the corresponding jointly mined online service;

2. The method for predictive analysis of services by federated big data mining of claim 1, wherein the step of analyzing a plurality of update behavior activities in the user update big data for new online services to determine the federated to-be-mined base data for loading to the federated mining online services comprises:

3. The business prediction analysis method for joint big data mining according to claim 2, wherein the step of obtaining a plurality of updating behavior activities in the user updating big data for the new online business and analyzing activity joint mining values corresponding to the plurality of updating behavior activities comprises:

Acquiring user update big data, dividing and treating the user update big data based on a behavior activity time trigger sequence, and determining a plurality of update behavior activities corresponding to the user update big data; the plurality of update behavior activities comprise update behavior activities ACTb, b is a positive integer and is not greater than the global quantity corresponding to the plurality of update behavior activities;

s behavior activity themes and frequent interaction data corresponding to the S behavior activity themes are obtained from the updated behavior activity ACTb; s is a positive integer;

performing frequent interaction intention analysis on the updating behavior activity ACTb, determining a frequent interaction intention, and extracting active behavior interaction resources and passive behavior interaction resources in the updating behavior activity ACTb;

taking the frequent interaction intention, the active behavior interaction resource and the passive behavior interaction resource as behavior activity knowledge points SNb corresponding to the updated behavior activity ACTb;

generating joint mining analysis characteristics corresponding to the updated behavior activity ACTb based on the S behavior activity themes, the S frequent interaction data and the behavior activity knowledge points SNb;

analyzing activity joint mining values corresponding to the plurality of updating behavior activities based on joint mining analysis characteristics corresponding to the plurality of updating behavior activities;

Generating joint mining analysis characteristics corresponding to the updated behavior activity ACTb based on the S behavior activity topics, the S frequent interaction data and the behavior activity knowledge points SNb, wherein the joint mining analysis characteristics comprise:

acquiring a combined mining decision model; the joint mining decision model comprises a first joint mining decision unit; the first joint mining decision unit comprises a first feature mining branch, a second feature mining branch, a third feature mining branch and a fourth feature mining branch;

loading the S behavior activity themes to the first feature mining branch respectively, performing feature mining on the S behavior activity themes according to the first feature mining branch, determining operation behavior member features corresponding to the S behavior activity themes respectively, aggregating the S operation behavior member features, and determining behavior activity features corresponding to the updated behavior activity ACTb;

respectively loading the S pieces of frequent interaction data to the second feature mining branch, respectively performing feature mining on the S pieces of frequent interaction data according to the second feature mining branch, determining frequent interaction member features corresponding to the S pieces of frequent interaction data, aggregating the S pieces of frequent interaction member features, and determining frequent interaction features corresponding to the updating action ACTb;

Loading the behavior activity knowledge points SNb to the third feature mining branch, extracting key description data in the behavior activity knowledge points SNb according to the third feature mining branch, performing feature mining on the key description data, and determining key description features corresponding to the key description data;

and respectively loading the behavior activity characteristics, the frequent interaction characteristics and the key description characteristics to the fourth characteristic mining branch, aggregating the behavior activity characteristics, the frequent interaction characteristics and the key description characteristics according to the fourth characteristic mining branch, and determining the joint mining analysis characteristics corresponding to the updated behavior activity ACTb.

4. The business prediction analysis method for joint big data mining according to claim 3, wherein the analyzing joint mining values of activities corresponding to the plurality of updating activities based on the joint mining analysis features corresponding to the plurality of updating activities comprises:

acquiring a combined mining decision model; the joint mining decision model comprises a first joint mining decision unit; the first joint mining decision unit comprises a first value prediction branch;

Loading the joint mining analysis features corresponding to the updating behavior activity ACTb to the first price value prediction branch, and performing joint mining value output on the joint mining analysis features corresponding to the updating behavior activity ACTb according to the first price value prediction branch to determine a joint mining value of the activity corresponding to the updating behavior activity ACTb;

then the determining a first updated behavioral activity from the plurality of updated behavioral activities based on the plurality of activity joint mining values comprises:

comparing the activity joint mining values corresponding to the plurality of updating behavior activities with a first preset mining value respectively;

and in the plurality of updating behavior activities, combining the activities not less than the first preset mining value with the updating behavior activities corresponding to the mining value to serve as first updating behavior activities.

5. The business prediction analysis method for combined big data mining according to claim 2, wherein the step of obtaining a page interest point data set having an association relationship with the user updated big data and analyzing a page combined mining value corresponding to the first update activity based on the page interest point data set and the first update activity comprises:

Acquiring page concern point data of a page target having an association relation with the user update big data, and acquiring page concern point data of the combined mining online service having an association relation with the page target;

generating a page concern point data set based on the page concern point data of the page target and the page concern point data of the combined mining online service;

acquiring a joint mining decision model, and respectively loading the page attention point data set and the first updating behavior activity to the joint mining decision model; the joint mining decision model comprises a second joint mining decision unit; the second joint mining decision unit comprises a first page attention feature mining branch;

according to the first page concern feature mining branch, performing page concern feature mining on each page concern point data in the page concern point data set, and determining a first page concern feature corresponding to the page concern point data set;

and acquiring a joint mining analysis characteristic corresponding to the first updating behavior activity, and analyzing a page joint mining value corresponding to the first updating behavior activity based on the first page attention characteristic and the joint mining analysis characteristic corresponding to the first updating behavior activity.

6. The traffic prediction analysis method for joint big data mining according to claim 5, characterized in that the second joint mining decision unit further comprises a first feature aggregation branch and a second value prediction branch;

analyzing a joint mining value of a page corresponding to the first updated behavior activity based on the first page attention feature and the joint mining analysis feature corresponding to the first updated behavior activity, including:

loading the first page attention feature and the joint mining analysis feature corresponding to the first updating behavior activity to the first feature aggregation branch respectively;

according to the first feature aggregation branch, feature aggregation is carried out on the first page attention feature and the joint mining analysis feature corresponding to the first updating behavior activity, and a first joint mining aggregation feature corresponding to the first updating behavior activity is determined;

loading the first joint mining aggregation characteristic to the second value prediction branch, performing joint mining value output on the first joint mining aggregation characteristic according to the second value prediction branch, and determining a page joint mining value corresponding to the first updating behavior activity; wherein the number of the first updating behavior activities is multiple;

Determining a second updated behavior activity from the first updated behavior activity based on the joint mining value of the page corresponding to the first updated behavior activity, including:

comparing the joint mining value of the page corresponding to each of the plurality of first updating behavior activities with a second preset mining value;

and in the plurality of first updating behavior activities, the first updating behavior activity corresponding to the page joint mining value larger than the second preset mining value is used as a second updating behavior activity.

7. The business prediction analysis method for joint big data mining according to claim 2, wherein the assistance joint mining value corresponding to the second updating behavior activity comprises a topic joint mining value corresponding to the second updating behavior activity and a frequent item joint mining value corresponding to the second updating behavior activity; the joint mining reference information corresponding to the second updating behavior activity comprises an assistance theme corresponding to the second updating behavior activity and an assistance frequent item corresponding to the second updating behavior activity;

analyzing a joint mining assistance value corresponding to the second updating behavior activity based on the page attention point data set and the second updating behavior activity, and analyzing joint mining reference information corresponding to the second updating behavior activity based on the joint mining assistance value corresponding to the second updating behavior activity, wherein the analyzing comprises:

Acquiring joint mining theme characteristics corresponding to a plurality of behavior activity themes in the second updating behavior activity, analyzing theme joint mining values corresponding to the second updating behavior activity based on the joint mining theme characteristics, the second updating behavior activity and the page attention point data set, and analyzing assistance themes corresponding to the second updating behavior activity based on the theme joint mining values corresponding to the second updating behavior activity; an assistance topic corresponding to the second updating behavior activity belongs to the plurality of behavior activity topics;

analyzing a frequent item joint mining value corresponding to the second updating behavior activity and an assistance frequent item corresponding to the second updating behavior activity based on the page attention point data set and a behavior activity knowledge point corresponding to the second updating behavior activity, wherein the assistance frequent item is used for representing joint mining reference information corresponding to the second updating behavior activity;

the obtaining of joint mining theme characteristics corresponding to each of a plurality of behavior activity themes in the second updating behavior activity, analyzing a theme joint mining value corresponding to the second updating behavior activity based on the joint mining theme characteristics, the second updating behavior activity and the page attention point data set, and analyzing an assistance theme corresponding to the second updating behavior activity based on the theme joint mining value corresponding to the second updating behavior activity includes:

Generating a time-space domain based on a theme, performing theme output on the second updating behavior activity, and determining a plurality of behavior activity themes in the second updating behavior activity;

acquiring a combined mining decision model; the joint mining decision model comprises a third joint mining decision unit; the third joint mining decision unit comprises a theme analyzing branch and a second feature aggregation branch;

respectively loading the plurality of behavior activity themes to the theme analysis branch, respectively performing feature mining on the plurality of behavior activity themes according to the theme analysis branch, and determining joint mining theme features corresponding to the plurality of behavior activity themes; the plurality of behavior activity themes comprise behavior activity themes THg, and the plurality of joint mining theme features comprise joint mining theme features corresponding to the behavior activity themes THg; g is a positive integer and is not greater than the global quantity corresponding to the plurality of behavior activity topics;

acquiring a joint mining analysis characteristic corresponding to the second updating behavior activity, and acquiring a second page attention characteristic corresponding to the page attention point data set;

respectively loading the joint mining theme features corresponding to the behavior activity theme THg, the joint mining analysis features corresponding to the second updating behavior activity and the second page attention features to the second feature aggregation branch;

According to the second feature aggregation branch, feature aggregation is carried out on the joint mining subject feature corresponding to the behavior activity subject THg, the joint mining analysis feature corresponding to the second updated behavior activity and the second page attention feature, and a second joint mining aggregation feature corresponding to the behavior activity subject THg is determined;

and analyzing the topic joint mining value corresponding to the second updating behavior activity based on the second joint mining aggregation characteristics corresponding to the plurality of behavior activity topics, and analyzing the assistance topic corresponding to the second updating behavior activity based on the topic joint mining value corresponding to the second updating behavior activity.

8. The business predictive analysis method for federated big data mining as in claim 7, wherein the assistance frequent term is composed of N federated mining behavior instances;

analyzing a joint mining value of frequent items corresponding to the second updating behavior activity and an assistance frequent item corresponding to the second updating behavior activity based on the page attention point data set and the behavior activity knowledge point corresponding to the second updating behavior activity, wherein the joint mining value of frequent items corresponding to the second updating behavior activity and the assistance frequent item corresponding to the second updating behavior activity comprise:

acquiring a combined mining decision model; the joint mining decision model comprises a fourth joint mining decision unit; the fourth joint mining decision unit comprises a second page attention feature mining branch, a third page attention feature mining branch, a frequent evaluation branch and a frequent item decision branch;

Loading the behavior activity knowledge points corresponding to the second updating behavior activity to the second page attention feature mining branch, performing page attention feature mining on the behavior activity knowledge points corresponding to the second updating behavior activity according to the second page attention feature mining branch, and determining behavior activity knowledge point features;

loading the page concern point data set to the third page concern feature mining branch, performing page concern feature mining on the page concern point data set according to the third page concern feature mining branch, and determining a third page concern feature;

loading the behavior activity knowledge point feature, the to-be-evaluated frequent item feature Fi corresponding to the second updated behavior activity, and the third page attention feature to the frequent evaluation branch, respectively, aggregating the behavior activity knowledge point feature, the to-be-evaluated frequent item feature Fi, and the third page attention feature according to the frequent evaluation branch, and determining a frequent evaluation coefficient corresponding to the behavior activity knowledge point feature; i is a non-negative integer less than N;

analyzing the characteristic Fi +1 of the frequent item to be evaluated corresponding to the second updating behavior activity based on the frequent evaluation coefficient corresponding to the behavior activity knowledge point characteristic; the joint mining behavior example indicated by the frequent item feature Fi to be evaluated is the last joint mining behavior example of the joint mining behavior example indicated by the frequent item feature Fi +1 to be evaluated;

When i +1 is equal to N, respectively loading N to-be-evaluated frequent item features to the frequent item decision branches, generating joint mining behavior examples respectively indicated by the N to-be-evaluated frequent item features according to the frequent item decision branches, and outputting the N joint mining behavior examples as assisted frequent items corresponding to the second updating behavior activity;

and analyzing the joint mining value of the frequent items corresponding to the second updating behavior activity based on the N frequent item characteristics to be evaluated.

9. The traffic prediction analysis method for federated big data mining according to claim 2, wherein the number of the second updating behavior activities is plural, the plurality of second updating behavior activities includes a second updating behavior activity Jm, m is a positive integer, and m is not greater than the number of the plurality of second updating behavior activities;

determining a joint mining update behavior activity from the second update behavior activity based on the activity joint mining value corresponding to the second update behavior activity, the page joint mining value corresponding to the second update behavior activity, and the assisted joint mining value corresponding to the second update behavior activity, including:

Performing weight fusion on the activity joint mining value corresponding to the second updating behavior activity Jm, the page joint mining value corresponding to the second updating behavior activity Jm and the assisted joint mining value corresponding to the second updating behavior activity Jm, and determining a global joint mining value corresponding to the second updating behavior activity Jm;

obtaining a maximum global joint mining value from global joint mining values corresponding to the plurality of second updating behavior activities;

determining the second updating behavior activity corresponding to the maximum global joint mining value as the joint mining updating behavior activity in the plurality of second updating behavior activities;

and acquiring the joint mining reference information corresponding to the joint mining updating behavior activities from the joint mining reference information corresponding to the second updating behavior activities.

10. An AI mining system comprising a processor and a memory for storing a computer program operable on the processor, the processor being configured to perform the traffic prediction analysis method for joint big data mining of any of claims 1 to 9 when executing the computer program.