CN114564648A

CN114564648A - Personalized service content optimization method based on big data and artificial intelligence cloud system

Info

Publication number: CN114564648A
Application number: CN202210206117.XA
Authority: CN
Inventors: 赵尚益
Original assignee: Individual
Current assignee: Individual
Priority date: 2022-03-03
Filing date: 2022-03-03
Publication date: 2022-05-31

Abstract

After a target offline user demand data set corresponding to offline behavior big data is obtained, corresponding online personalized service content data is pushed to a 5G internet terminal based on each target offline user demand, tracking offline behavior data specific to the online personalized service content data is obtained, derived demand characteristics corresponding to each target offline user demand are obtained through analysis, and content optimization is conducted on the next pushed online personalized service content data based on the derived demand characteristics corresponding to each target offline user demand. Therefore, the tracking offline behavior data aiming at the online personalized service content data is further tracked and considered, and the content of the online personalized service content data pushed next time is optimized after the tracking offline behavior data is compared and analyzed with the original basic offline behavior data, so that the accuracy of the subsequent pushed content can be improved.

Description

Personalized service content optimization method based on big data and artificial intelligence cloud system

Technical Field

The application relates to the technical field of personalized internet service information, in particular to a personalized service content optimization method based on big data and an artificial intelligent cloud system.

Background

In the current big data era, internet service providers have more opportunities to know users and even know the needs of the users than the users themselves. However, in fact, only a few customers really obtain precise and attentive personalized services, and how internet service providers quickly grasp the personalized needs and psychological expectations of the customers is a problem to be researched urgently.

In the related art, the content data of the online personalized service is usually pushed after the offline user requirements are mined, so that the un-mined user requirements of the offline scene except the online scene in the related art can be timely made up, and the personalized service experience of the user is improved. However, due to the mining precision problem, not all the mining offline user requirements completely meet the actual interests of users, and the related technologies lack a tracking verification content optimization process, and still have a space for improving the content pushing precision.

Disclosure of Invention

In order to overcome at least the above disadvantages in the prior art, the present application aims to provide a personalized service content optimization method based on big data and an artificial intelligence cloud system.

In a first aspect, the present application provides a big data-based personalized service content optimization method, which is applied to an artificial intelligence cloud system, where the artificial intelligence cloud system is in communication connection with a plurality of 5G internet terminals, and the method includes:

acquiring a target offline user demand data set corresponding to the offline behavior big data of the 5G internet terminal, wherein the target offline user demand data set comprises each basic offline behavior data and a target offline user demand corresponding to each basic offline behavior data;

pushing corresponding online personalized service content data to the 5G internet terminal based on the requirement of each target offline user, and acquiring tracking offline behavior data of the 5G internet terminal aiming at the online personalized service content data;

analyzing the basic offline behavior data and the tracking offline behavior data corresponding to each target offline user requirement to obtain derived requirement characteristics corresponding to each target offline user requirement;

and optimizing the content of the next pushed online personalized service content data based on the derived demand characteristics corresponding to the demand of each target offline user.

In a second aspect, an embodiment of the present application further provides a big data-based personalized service content optimization system, where the big data-based personalized service content optimization system includes an artificial intelligence cloud system and a plurality of 5G internet terminals communicatively connected to the artificial intelligence cloud system;

the artificial intelligence cloud system is used for:

acquiring a target offline user requirement data set corresponding to the offline behavior big data of the 5G internet terminal, wherein the target offline user requirement data set comprises each basic offline behavior data and a target offline user requirement corresponding to each basic offline behavior data;

According to any one of the aspects, in the embodiment provided by the application, after a target offline user requirement data set corresponding to offline behavior big data is obtained, corresponding online personalized service content data is pushed to a 5G internet terminal based on each target offline user requirement, tracking offline behavior data for the online personalized service content data is obtained, then derived requirement characteristics corresponding to each target offline user requirement are obtained through analysis, and content optimization is performed on the next pushed online personalized service content data based on the derived requirement characteristics corresponding to each target offline user requirement. Therefore, the tracking offline behavior data aiming at the online personalized service content data are further tracked and considered, and the content of the online personalized service content data pushed next time is optimized after the tracking offline behavior data are compared and analyzed with the original basic offline behavior data, so that the accuracy of the subsequent pushed content can be improved.

Drawings

FIG. 1 is a schematic application environment diagram of a big data-based personalized service content optimization system according to an embodiment of the present application;

fig. 2 is a schematic flowchart of a method for optimizing personalized service content based on big data according to an embodiment of the present application;

fig. 3 is a schematic block diagram of a structure of an artificial intelligence cloud system for implementing the above-described personalized service content optimization method based on big data according to an embodiment of the present application.

Detailed Description

Fig. 1 is a schematic application environment diagram of a big data-based personalized service content optimization system 10 according to an embodiment of the present application. The big data based personalized service content optimization system 10 may include an artificial intelligence cloud system 100 and a 5G internet terminal 200 communicatively connected to the artificial intelligence cloud system 100. The big data based personalized service content optimization system 10 shown in fig. 1 is only one possible example, and in other possible embodiments, the big data based personalized service content optimization system 10 may also include only at least some of the components shown in fig. 1 or may also include other components.

In an embodiment that can be implemented independently, the artificial intelligence cloud system 100 and the 5G internet terminal 200 in the big data based personalized service content optimization system 10 can cooperatively perform the big data based personalized service content optimization method described in the following method embodiment, and the detailed description of the method embodiment can be referred to in the following steps of the artificial intelligence cloud system 100 and the 5G internet terminal 200.

The big data-based personalized service content optimization method provided by the present embodiment may be executed by the artificial intelligence cloud system 100 shown in fig. 1, and the details of the big data-based personalized service content optimization method are described below.

And step S110, acquiring a target offline user demand data set corresponding to the offline behavior big data.

In one design approach, the target offline user requirement data set includes each piece of basic offline behavior data and a target offline user requirement corresponding to each piece of basic offline behavior data.

Step S120, pushing corresponding online personalized service content data to the 5G Internet terminal based on the requirement of each target offline user, and acquiring tracking offline behavior data of the 5G Internet terminal aiming at the online personalized service content data.

Step S130, analyzing the basic offline behavior data and the tracking offline behavior data corresponding to each target offline user requirement to obtain the derived requirement characteristics corresponding to each target offline user requirement.

And step S140, optimizing the content of the next pushed online personalized service content data based on the derived demand characteristics corresponding to the demand of each target offline user.

Based on the above steps, after a target offline user demand data set corresponding to offline behavior big data is obtained, corresponding online personalized service content data is pushed to the 5G internet terminal based on each target offline user demand, tracking offline behavior data of the 5G internet terminal for the online personalized service content data is obtained, then, basic offline behavior data and tracking offline behavior data corresponding to each target offline user demand are analyzed, derived demand characteristics corresponding to each target offline user demand are obtained, and content optimization is performed on next pushed online personalized service content data based on the derived demand characteristics corresponding to each target offline user demand. Therefore, the tracking offline behavior data aiming at the online personalized service content data is further tracked and considered, and the content of the online personalized service content data pushed next time is optimized after the tracking offline behavior data is compared and analyzed with the original basic offline behavior data, so that the accuracy of the subsequent pushed content can be improved.

In an independently implementable embodiment, as for step S120, in a process of pushing corresponding online personalized service content data to the 5G internet terminal based on each target offline user requirement, an embodiment of the present application further provides a personalized service content pushing method based on big data, including the following steps.

Step S210, acquiring a basic content distribution knowledge network generated by a previous content distribution task in a content distribution process of online user required service content corresponding to the internet user image of the 5G internet terminal, wherein the basic content distribution knowledge network comprises a content distribution knowledge entity which is distributed to the previous content distribution task in the online user required service content.

Step S220, according to the basic content distribution knowledge network, determining an initial reference knowledge unit which is transmitted to a content pushing task and used for referring to a content pushing node of the current content distribution knowledge network when a content distribution task at the current pushing stage in a content distribution process of the on-line user demand service content distributes and calls the current content distribution knowledge network in the on-line user demand service content.

Step S230, determining a target reference knowledge unit transferred to the content push task by the content distribution task at the current push stage in the process of distributing and calling the content of the on-line user demand service according to the fusion reference knowledge unit generated by the initial reference knowledge unit and the content distribution task at the current push stage and aiming at the next group of acquired content distribution knowledge networks.

Step S240, according to the target reference knowledge unit transmitted to the content push task in the process of distributing and calling the content required by the online user by the content distribution task at the current push stage, executing a push behavior of corresponding online personalized service content data to the content push task from the content required by the online user.

Based on the above steps, the embodiment obtains the basic content distribution knowledge network generated by the previous content distribution task in the content distribution process of the internet user representation of the 5G internet terminal corresponding to the online user demand service content, then determines the initial reference knowledge unit of the content push node for referring to the current content distribution knowledge network, which is transmitted to the content push task in the content distribution process of the online user demand service content when the content distribution task in the current push stage in the content distribution process of the online user demand service content distributes and calls the current content distribution knowledge network in the online user demand service content, thereby determining the target reference knowledge unit transmitted to the content push task by the content distribution task in the current push stage in the process of distributing and calling the online user demand service content, executing the push behavior of the corresponding online personalized service content data to the content push task from the online user demand service content, thereby improving the matching degree of content push and user requirements.

In an embodiment that can be implemented independently, with respect to step S140, the present application provides a service content optimization method based on artificial intelligence, which includes the following steps.

Step S310, acquiring knowledge map data of a derived demand knowledge map generated by derived demand knowledge points required by a target offline user, wherein the knowledge map data comprises frequent knowledge point categories of the derived demand knowledge map, derived demand knowledge point information of the derived demand knowledge map and a reference demand knowledge map of the derived demand knowledge map.

In one design approach, the derived requirement knowledge map may be any form of requirement knowledge map information related to the derived requirement knowledge points.

In one design approach, a required knowledge map information may have one or more frequent knowledge point categories. The frequent knowledge point category may represent a required knowledge map information derived required category of the required knowledge map information or a frequent feature in the required knowledge map information.

For another example, the derived demand knowledge point information may be derived demand knowledge points having a content association relationship with the derived demand knowledge map.

For another example, the derived demand knowledge point information may further include one or more of a target derived demand knowledge point of the derived demand knowledge map and an associated derived demand knowledge point of the derived demand knowledge map; the reference demand knowledge map may include a derived demand knowledge map corresponding to a target derived demand knowledge point of the derived demand knowledge map, and the derived demand knowledge map may include one or more demand knowledge map information.

In one design approach, the knowledge map data may be related data or information having knowledge map properties with specific content information of the derived demand knowledge map, for example, it may include frequent knowledge point categories of the derived demand knowledge map, derived demand knowledge point information of the derived demand knowledge map, and a reference demand knowledge map of the derived demand knowledge map.

Step S320, obtaining the derived demand knowledge map and the interest prediction information of each knowledge map unit in the knowledge map data.

The interest prediction information of the derived demand knowledge map can represent content characteristics of the derived demand knowledge map, and the interest prediction information of each knowledge map unit in the knowledge map data can identify the content characteristics of each knowledge map unit.

Step S330, determining the knowledge map attribute corresponding to the derived demand knowledge map. And the knowledge map attribute represents knowledge point connection relation between the derived demand knowledge map and each knowledge map unit in the knowledge map data.

Step S340, according to the interest prediction information and the knowledge map attribute, determining interest point information of the derived demand knowledge map, performing interest point feature analysis on the derived demand knowledge map according to the interest point information, and performing content optimization on the next pushed online personalized service content data based on the interest point feature of the derived demand knowledge map.

In a design idea, the interest point information of the derived requirement knowledge map may include various types of interest point information related to the derived requirement knowledge map, for example, the interest point information may be mapped into a corresponding interest instantiated feature space through a predetermined interest instantiated feature mapping function, so as to implement interest point feature expression for the derived requirement knowledge map, and further obtain the interest point information.

In this way, when the interest resource information of the required knowledge map information is acquired and the derived required knowledge map is analyzed, the knowledge map data including the frequent knowledge point category of the derived required knowledge map, the derived required knowledge point information of the derived required knowledge map and the reference required knowledge map data of the derived required knowledge map are analyzed. The type of the frequent knowledge points can represent the relevant characteristics of the required knowledge map information, the quoted required knowledge map can represent the characteristics of the other dimension of the target derived required knowledge point, and the interest point information obtained by analysis has the characteristics of the other dimension related to the required knowledge map information besides the characteristics of the required knowledge map information, so that the obtained interest point information can accurately express the interest instantiation characteristics of the required knowledge map information, the content optimization is carried out on the content data of the next pushed online personalized service based on the interest point characteristics of the derived required knowledge map, and the accuracy of the content optimization can be improved.

Specific implementations of the above steps are described in detail below with reference to specific embodiments.

First, in step S330, the determining the knowledge map attribute corresponding to the derived requirement knowledge map may include the following steps.

Step S3301, generating a derived demand guide map corresponding to the derived demand knowledge map according to the knowledge map data and the derived demand knowledge map, where the derived demand guide map is a knowledge map obtained by performing attribute feature integration on the attributes of the knowledge map.

In one design approach, the derived demand graph may include a plurality of graph elements directly or indirectly connected to each other. The map guide unit may include a map guide unit corresponding to the derived demand knowledge map, and a map guide unit corresponding to each knowledge map unit in the knowledge map data, and the map guide connection attribute in the derived demand knowledge map includes connection relationship information between the derived demand knowledge map and the map guide unit corresponding to each knowledge map unit in the knowledge map data. The map unit can be understood as different map nodes and the like constituting the derived demand map. Characteristic connection relations can exist among the nodes of the guide graph. The join relationship information may be a multi-dimensional array for expressing the correlation information between the map units having the characteristic join relationship.

Based on the above, in step S340, the determining, according to each of the interest prediction information and the knowledge map attribute, interest point information of interest of the derived demand knowledge map includes:

step S3401, according to each piece of interest prediction information and the derivative demand guide map, determining the interest point information of the derivative demand knowledge map. The knowledge map data comprises one or more frequent knowledge point categories, and the map joint attributes in the derived demand map further comprise joint relation information between map units associated with the frequent knowledge point categories.

In a design idea, a knowledge map attribute between knowledge map data and a derived demand knowledge map is expressed by a derived demand map. The derived demand guide map not only comprises the information of each guide map unit, but also comprises guide map connection attributes in the derived demand guide map as knowledge map attributes, so that the expression of the knowledge map attributes is realized according to the derived demand guide map, and the corresponding characteristics of the derived demand knowledge map can be accurately expressed.

In detail, in step S3401, determining interest point information of interest of the derived demand knowledge map according to each of the interest prediction information and the derived demand guide map may include:

acquiring first interest resource information corresponding to required knowledge map information under various derived demand categories according to interest prediction information corresponding to each cascaded map guide unit under various derived demand categories of a target map guide unit in the derived demand map, wherein the target map guide unit is a map guide unit corresponding to the derived demand knowledge map, and map guide units corresponding to various types of data in knowledge map data respectively correspond to map guide units under one derived demand category;

and obtaining interest point information of interest of the derived demand knowledge map according to the first interest resource information corresponding to the target map guiding unit and the interest prediction information of the target map guiding unit.

In a design idea, a cascaded map guide unit of a target map guide unit may be a map guide unit having information about a connection relationship with the target map guide unit, and the cascaded map guide unit may express information about an associated map guide unit of the target map guide unit.

It should be understood that, for each map guide unit in the derived demand map, first interest resource information corresponding to demand knowledge map information in various derived demand categories corresponding to the map guide unit needs to be acquired according to interest prediction information corresponding to the demand knowledge map information in various derived demand categories of the map guide unit.

In addition, the cascade map guide units under each derivative requirement category correspondingly express different relevant information of the map guide units, so that when acquiring the map guide unit interest resource information (first interest resource information) of each cascade map guide unit, the cascade map guide units can be acquired according to the derivative requirement categories of the cascade map guide units. The demand knowledge map information under the same derived demand category corresponds to a first interest resource information.

In detail, in order to obtain the materialization description information of each map unit, the following steps may be performed in a walking loop for each map unit in the derived demand map:

(1) and acquiring second interest resource information corresponding to the demand knowledge map information under the derivative demand type according to the current interest resource information of each cascaded map guide unit under various derivative demand types of the map guide unit.

(2) And obtaining target interest resource information of the map guide unit according to the current interest resource information of the map guide unit and the second interest resource information corresponding to the map guide unit. In a design idea, the current interest resource information is preset resource information when the picture is first walked in a loop, and the target interest resource information is interest resource information of the picture guide unit. Correspondingly, when the first walking circulation is not performed, the current interest resource information corresponding to the first walking circulation is the interest prediction information, the current interest resource information corresponding to the first walking circulation is the target interest resource information obtained by the previous walking circulation, and the interest resource information of the map guide unit is the target interest resource information obtained by the last walking circulation. For example, the preset resource information may be resource information obtained based on the interest prediction information.

On the basis of the above, in step S3401, according to interest prediction information corresponding to each cascaded map guide unit in various derived demand categories of the target map guide unit, first interest resource information corresponding to demand knowledge map information in each derived demand category is acquired, and a specific implementation manner may include:

aiming at each derived demand category, carrying out feature integration on the map guide unit interest resource information of each cascaded map guide unit in the derived demand category of the target map guide unit to obtain first interest resource information corresponding to the demand knowledge map information in the derived demand category;

correspondingly, on the basis of the above content, in step S3401, obtaining interest point information of interest of the derived demand knowledge map according to each piece of first interest resource information corresponding to the target map guiding unit and interest prediction information of the target map guiding unit, which may specifically include:

(11) acquiring a first category influence coefficient corresponding to demand knowledge map information under each derived demand category and a second category influence coefficient corresponding to the derived demand knowledge map;

(12) according to a first category influence coefficient corresponding to the demand knowledge map information under each derived demand category, fusing first interest resource information corresponding to the demand knowledge map information under each derived demand category to obtain second interest resource information corresponding to the demand knowledge map information under each derived demand category;

(13) according to the second category influence coefficient, fusing the interest resource information of the map guide unit of the target map guide unit to obtain third interest resource information;

(14) splicing the second interest resource information and the third interest resource information corresponding to the demand knowledge map information under each derivative demand category to obtain spliced interest resource information;

(15) and obtaining the interest point information of interest of the derived demand knowledge map according to the spliced interest resource information.

The method comprises the steps of obtaining the interest resource information of each map guide unit of each cascade map guide unit, and obtaining the interest resource information of each cascade map guide unit corresponding to each cascade map guide unit under each derivative requirement category. Each cascaded map guide unit under various derived demand categories of a target map guide unit in the derived demand map guide can be correspondingly processed to obtain first interest resource information corresponding to demand knowledge map information under various derived demand categories of the target map guide unit.

The method includes the steps that the requirement knowledge map information under various derivative requirement categories has different category influence degrees aiming at interest point information of a derivative requirement knowledge map, and therefore the first interest resource information corresponding to the requirement knowledge map information under various derivative requirement categories and the interest resource information of a map guide unit of a target map guide unit can be spliced according to a first category influence coefficient corresponding to the requirement knowledge map information under various derivative requirement categories and a second category influence coefficient corresponding to the derivative requirement knowledge map, so that the influence of the requirement knowledge map information under the derivative requirement categories on the interest point information is particularly considered in the obtained spliced interest resource information, and the obtained interest point information is further improved. In one design idea, the first category influence coefficients corresponding to the demand knowledge map information under each derived demand category of different map units in the derived demand map may be the same or different.

After the first interest resource information corresponding to the demand knowledge map information under various derived demand categories is obtained, the first interest resource information corresponding to the demand knowledge map information under various derived demand categories and the interest resource information of the map guide unit of the target map guide unit can be spliced to obtain spliced interest resource information, the spliced interest resource information contains the interest resource information of each cascaded map guide unit and the interest resource information of the target map guide unit, and the interest point information of the derived demand knowledge map obtained by further analyzing the spliced interest resource information is more practical.

As an example, the knowledge map data and the derived demand knowledge map respectively correspond to demand knowledge map information under different derived demand categories, and this embodiment may express the above features through a derived demand guide map, based on which the derived demand guide map may be a derived demand guide map having a guide map unit under the derived demand category, for example, the derived demand guide map may be N { U, S, C }, where U denotes a guide map unit, S denotes a guide map join attribute, and C denotes a guide map unit derived demand category. The map guidance units in the knowledge map data and the related information thereof correspond to map guidance units in a derived demand category. For example, the knowledge map data includes frequent knowledge point categories and derivative requirement knowledge point information, each frequent knowledge point category corresponds to one map guide unit, each derivative requirement knowledge point information corresponds to one map guide unit, the map guide unit corresponding to each frequent knowledge point category is a map guide unit under a corresponding derivative requirement category, and the map guide unit corresponding to each derivative requirement knowledge point information is a map guide unit under another different derivative requirement category.

Illustratively, the knowledge map data of the derived demand knowledge map includes a frequent knowledge point category and derived demand knowledge point information. The derived demand knowledge point information comprises a first target derived demand knowledge point, a second target derived demand knowledge point, an associated derived demand knowledge point, a first quote demand knowledge map and a second quote demand knowledge map. Then, the map guide units in the derived demand map corresponding to the derived demand knowledge map comprise corresponding first map guide units, second map guide units corresponding to the categories of the frequent knowledge points, third map guide units corresponding to the first target derived demand knowledge points, fourth map guide units corresponding to the second target derived demand knowledge points, fifth map guide units corresponding to the associated derived demand knowledge points, sixth map guide units corresponding to the first quoted demand knowledge map, and seventh map guide units corresponding to the second quoted demand knowledge map; the map connection attribute in the derived demand map comprises connection relation information between the first map unit and the second map unit, the third map unit, the fourth map unit, the fifth map unit, the sixth map unit and the seventh map unit respectively.

For step S320, the obtaining of the interest prediction information of the derived demand knowledge map and each knowledge map unit in the knowledge map data may include the following steps.

Step S3201, key knowledge map information of the derived demand knowledge map is obtained, key interest resource information corresponding to the key knowledge map information is collected, and the key interest resource information is used as interest prediction information of the derived demand knowledge map.

In an embodiment that can be implemented independently, the derived demand knowledge point information includes one or more of a target derived demand knowledge point of a derived demand knowledge map or an associated derived demand knowledge point of the derived demand knowledge map, if the derived demand knowledge point information includes the target derived demand knowledge point, the past demand knowledge map information is demand knowledge map information interactively generated by the target derived demand knowledge point in a first preset business progress stage before a current business progress stage, and if the derived demand knowledge point information includes the associated derived demand knowledge point, the past demand knowledge map information is demand knowledge map information generated by the associated derived demand knowledge point in a second preset business progress stage before the current business progress stage. The first preset service progress stage and the second preset service progress stage may be the same or different.

In another embodiment, which can be implemented independently, if the knowledge map data includes frequent knowledge point categories of the derivative demand knowledge map, where the frequent knowledge point categories are identification IDs, for a map guide unit corresponding to each frequent knowledge point category, identifying interest resource information corresponding to the frequent knowledge point categories can be obtained in a manner similar to that of the key knowledge map information, and the identifying interest resource information is used as interest prediction information of the map guide unit corresponding to the frequent knowledge point category. If the frequent knowledge point category is not the ID identification, the frequent knowledge point category can be mapped to the ID identification, and then the interest prediction information of the guide picture unit corresponding to the frequent knowledge point category is determined.

Step S3202, when the knowledge map data includes the derived demand knowledge point information, for each derived demand knowledge point information, obtaining past demand knowledge map information corresponding to the derived demand knowledge point information, and determining interest prediction information of the derived demand knowledge point information according to the past demand knowledge map information.

In a design idea, the key knowledge map information may represent key features of the required knowledge map information, for example, feature information related to a preset dimension, so that key interest resource information of the key knowledge map information may be used as interest prediction information of a map guide unit corresponding to the derived required knowledge map.

Wherein the derived demand knowledge point information comprises one or more of target derived demand knowledge points of the derived demand knowledge map and associated derived demand knowledge points of the derived demand knowledge map; the quote demand knowledge map comprises a derived demand knowledge map corresponding to a target derived demand knowledge point of the derived demand knowledge map, and the derived demand knowledge map is one or more demand knowledge map information corresponding to the target derived demand knowledge point in a business progress stage for generating the derived demand knowledge map; when the derived demand knowledge point information includes a target derived demand knowledge point, the past demand knowledge map information is demand knowledge map information generated by the target derived demand knowledge point in a first preset business progress stage before a current business progress stage, and when the derived demand knowledge point information includes an associated derived demand knowledge point, the past demand knowledge map information is demand knowledge map information generated by the associated derived demand knowledge point in a second preset business progress stage before the current business progress stage.

The related derivative demand knowledge points can be derivative demand knowledge points related to a derivative demand knowledge map, and the reference demand knowledge map is one or more pieces of demand knowledge map information generated by interaction before and after the business progress stage of generating the derivative demand knowledge map.

On the basis of the above, in step S340, the performing of the interest point feature analysis of the derived demand knowledge map according to the interest point information may include any one of the following two ways.

Firstly, determining a quotation requirement knowledge map from a first requirement knowledge map information base according to the attention interest point information of the derivative requirement knowledge map and a first interest index parameter of the attention interest point information of each candidate requirement knowledge map information in the first requirement knowledge map information base, and issuing the quotation requirement knowledge map to a target derivative requirement knowledge point, wherein the derivative requirement knowledge map is associated requirement knowledge map information of past requirement knowledge map information of the target derivative requirement knowledge point.

Secondly, clustering the information of the various required knowledge maps in a second required knowledge map information base according to a second interest index parameter between the concerned interest point information of the various required knowledge maps in the second required knowledge map information base, wherein the derived required knowledge map is one or more pieces of information of the required knowledge maps in the second required knowledge map information base.

For example, in another embodiment that can be implemented independently, in step S340, a plurality of pieces of demand knowledge map information (derived demand knowledge maps) may be clustered according to the interest point information corresponding to each derived demand knowledge map to obtain demand knowledge map information sets corresponding to different clusters. In detail, the determining, according to each of the interest prediction information and the knowledge map attribute, interest point information of the derived demand knowledge map, so as to perform interest point feature analysis of the derived demand knowledge map according to the interest point information, may include the following:

firstly, acquiring at least one interest resource hotspot characteristic corresponding to a plurality of pieces of undetermined required knowledge map information; each interest resource hotspot characteristic is obtained by analyzing interest resource information corresponding to different pieces of demand knowledge map information, and each piece of pending demand knowledge map information is one derived demand knowledge map;

secondly, for each piece of required knowledge map information, acquiring grouping label information of each interest resource hotspot feature corresponding to the required knowledge map information;

then, for each piece of required knowledge map information, determining target grouping label information of the required knowledge map information according to each grouping label information of the required knowledge map information corresponding to each interest resource hotspot characteristic to obtain required knowledge map information grouping label information of each piece of undetermined required knowledge map information;

then, clustering the undetermined required knowledge map information according to required knowledge map information clustering label information of the undetermined required knowledge map information;

and finally, clustering the multiple pieces of undetermined required knowledge map information according to clustering results of the multiple pieces of undetermined required knowledge map information to obtain required knowledge map information sets respectively corresponding to the multiple pieces of required knowledge map information clusters.

The acquiring of the at least one interest resource hotspot characteristic of the pending demand knowledge map information comprises the following steps:

acquiring at least one type of required knowledge item information from the undetermined required knowledge map information, acquiring interest resource information corresponding to each required knowledge item information, and acquiring interest instantiation characteristics of each required knowledge item information according to the interest resource information;

and performing feature fusion on the interest instantiation features of the required knowledge item information to obtain interest resource hot spot features of the undetermined required knowledge map information, wherein the required knowledge item information is one item in knowledge map data of the required knowledge map information.

Correspondingly, the obtaining of at least one interest resource hotspot feature corresponding to each of the plurality of pieces of undetermined demand knowledge map information includes:

acquiring at least one required knowledge item information from each pending required knowledge map information;

performing interest instantiation feature extraction on interest resource information of the pending required knowledge map information to obtain at least one global interest resource hot spot feature of the pending required knowledge map information;

performing interest instantiation feature extraction on each required knowledge item information to obtain at least one local interest resource hotspot feature of each required knowledge item information; and

and for each local interest resource hot spot feature of each required knowledge item information, respectively performing feature fusion on each local interest resource hot spot feature and a global interest resource hot spot feature in the undetermined required knowledge map information to obtain an interest resource hot spot feature of the undetermined required knowledge map information.

Further, also in step S340, the step of determining the interesting point information of the derived demand knowledge map according to each piece of interest prediction information and the knowledge map attribute may be implemented in a deep learning network manner, for example, the interest prediction information and the knowledge map attribute may be input to an interesting point analysis network to obtain the interesting point information of the derived demand knowledge map.

For example, in an embodiment that can be implemented independently, the present application provides a method for interest analysis based on artificial intelligence, which may specifically include the following steps.

Step S401, a network convergence basic database is obtained, wherein the network convergence basic database comprises a plurality of basic attention interest training data. The basic attention interest training data may be used as training sample data, and may include an example derived demand guide map corresponding to example demand knowledge map information and interest prediction information of each guide map unit in the example derived demand guide map. The method comprises the steps that each map guide unit in each example derived demand map comprises a first map guide unit corresponding to example demand knowledge map information and a second map guide unit corresponding to each first associated demand knowledge map information, the first associated demand knowledge map information is different types of data included in knowledge map data of the example demand knowledge map information, and the map guide connection attribute in the example derived demand map comprises connection relation information between the first map guide unit and each second map guide unit.

Step S402, loading each basic attention interest training data to an initial attention interest point analysis network to obtain the predicted interest resource information of each map guiding unit corresponding to each basic attention interest training data. In one design approach, the initial interest point analysis network may be, but is not limited to, a convolutional neural network model, a cyclic convolutional network model, and the like.

Step S403, for each basic attention interest training data, determining a first loss function value corresponding to the basic attention interest training data according to a second interest index parameter between the predicted interest resource information of the first map unit and the predicted interest resource information of each second map unit in the example derived demand map of the basic attention interest training data.

Step S404, determining a global loss function value corresponding to the interest point analysis network according to the first loss function value corresponding to each of the basic interest training data.

Step S405, carrying out circular training on the concerned interest point analysis network according to the global loss function value until the global loss function value is converged.

The example derived demand guide map may be a derived demand guide map corresponding to the example demand knowledge map information, and the interest prediction information of each guide map unit in the example derived demand guide map may be obtained according to the interest prediction information of each guide map unit in the derived demand guide map corresponding to the derived demand knowledge map, which is not described in detail herein.

The method for obtaining the predicted interest resource information of the map guide unit (first map guide unit) corresponding to the example requirement knowledge map information corresponding to the basic interest training data through the initial interest point analysis network for each piece of basic interest training data is the same as or similar to the method for determining the interest point information of the derived requirement knowledge map according to the first interest resource information and the knowledge map attributes. For other map units in the example derived demand map corresponding to the basic attention interest training data, the predicted interest resource information corresponding to each map unit in the other map units may also be obtained in a similar manner, for example, for each map unit in the other map units except the map unit corresponding to the example demand knowledge map information in the example derived demand map, the predicted interest resource information of the map unit may be determined according to the map unit interest resource information of the map unit and the first interest resource information corresponding to the cascaded map unit of the map unit.

For each base interest training data, the first loss function value may represent a loss parameter between the predicted interest resource information of the first map unit and the predicted interest resource information of the second map units in the example derived demand map of the base interest training data. The smaller the global loss function value is, the better the training performance is, and the method is correspondingly more suitable for acquiring the information of the interest points.

For example, based on the above, in a design idea, for each of the example derived demand maps, the example derived demand map further includes a third map unit corresponding to each of second associated demand knowledge map information, and the second associated demand knowledge map information includes data having no map link attribute with the example demand knowledge map information and having a map link attribute with one or more of the first associated demand knowledge map information. For each piece of the first associated requirement knowledge map information, the map connection attribute in the example derived requirement map may further include connection relationship information between a second map unit corresponding to the first associated requirement knowledge map information and a third map unit having a map connection attribute with the first associated requirement knowledge map information in the second associated requirement knowledge map information. On this basis, the step of network convergence may further include:

step S406, for each basic attention interest training data, determining a second loss function value corresponding to the basic attention interest training data according to a second interest index parameter between the predicted interest resource information of each third map unit and the predicted interest resource information of the first map unit.

On the basis of step S406, in step S404, the determining a global loss function value corresponding to the interest point analysis network according to the first loss function value corresponding to each of the basic interest training data may include: and performing weighted calculation according to the first loss function value and the second loss function value corresponding to each piece of basic attention interest training data to obtain the global loss function value. For example, the global loss function value may be obtained by performing influence weight calculation (for example, weighted multiplication) on the first loss function value and the second loss function value based on an influence weight of a first loss function value and a second loss function value set in advance.

Wherein the second associated requirement knowledge map information comprises data having no map junction attribute with the example requirement knowledge map information and having a map junction attribute with one or more of the first associated requirement knowledge map information. In this way, the global loss function value is taken as another constraint condition to be considered for network convergence by the second loss function value, and the second loss function value can express the matching or difference between the predicted interest resource information of the third map unit and the predicted interest resource information of the first map unit corresponding to each piece of second associated demand knowledge map information.

In an embodiment that can be implemented independently, the step of obtaining the target offline user requirement data set corresponding to the offline behavior big data of the 5G internet terminal in step S110 may be implemented by the following steps.

And R101, acquiring a reference offline behavior data sequence.

Wherein the reference offline behavior data sequence comprises reference offline behavior data of a plurality of offline environment tags. The offline environment tag may be understood as reference offline behavior data obtained in different offline environments, for example, reference offline behavior data corresponding to different interactive offline environments and reference offline behavior data corresponding to different non-interactive offline environments, and it may be found that the interactive offline environment and the non-interactive offline environment may be the offline environment tags referring to the offline behavior data.

The reference offline behavior data sequence may be obtained in a variety of ways, for example, the reference offline behavior data sequence may be directly obtained, or when the reference offline behavior data number is large or the occupied space is large, the reference offline behavior data sequence may also be indirectly obtained, which may specifically be as follows:

(1) and directly acquiring a reference offline behavior data sequence.

For example, the reference offline behavior data uploaded by the 5G internet terminal or the 5G internet service platform may be directly received, so as to obtain the reference offline behavior data sequence, or at least one reference offline behavior data may be screened from the reference offline behavior database, so as to obtain the reference offline behavior data sequence.

(2) Indirectly obtaining a reference offline behavior data sequence

For example, an offline demand mining service instruction may be responded, the offline demand mining service instruction carries index node information of reference offline behavior data or a reference offline behavior data sequence, and the reference offline behavior data or the reference offline behavior data sequence is obtained according to the index node information, so as to obtain the reference offline behavior data sequence.

In a possible design idea, after the reference offline behavior data is obtained, the offline environment tag of the reference offline behavior data may be further identified, for example, the attached information of the reference offline behavior data may be obtained, the data source tag information of the reference offline behavior data is identified in the attached information, and the offline environment tag of the reference offline behavior data is determined according to the data source tag information, or the data source tag information of the reference offline behavior data may be directly obtained, and the corresponding offline environment tag of the reference offline behavior data is identified in the data source tag information.

And step R102, clustering the reference offline behavior data according to the offline environment tags of the reference offline behavior data to obtain a plurality of reference offline behavior data clusters, and respectively performing network weight optimization on the first AI network by using the reference offline behavior data clusters to obtain a second AI network corresponding to each reference offline behavior data cluster.

For example, in a separately implementable embodiment, step R102 may be implemented by the following exemplary steps:

and A1, clustering the reference offline behavior data according to the offline environment tags of the reference offline behavior data to obtain a plurality of reference offline behavior data clusters.

For example, the reference offline behavior data belonging to the same offline environment tag may be classified into one type, so as to obtain a reference offline behavior data group corresponding to the offline environment tag, or the reference offline behavior data in the reference offline behavior data sequence may be clustered, so as to obtain a plurality of reference offline behavior data groups, and according to the offline environment tag of the reference offline behavior data, the group tag corresponding to each reference offline behavior data group is determined, and the group tag is added to the corresponding reference offline behavior data group, so as to obtain the reference offline behavior data group.

And A2, respectively carrying out network weight optimization on the first AI network by adopting the reference offline behavior data cliques to obtain a second AI network corresponding to each reference offline behavior data clique.

For example, a first AI network may be used to perform offline user requirement mining on reference offline behavior data in a reference offline behavior data group to obtain a predicted offline user requirement mining result of the reference offline behavior data, perform loss calculation on the predicted offline user requirement mining result and a calibrated user requirement type in the reference offline behavior data, determine risk assessment coefficient information of the reference offline behavior data, and update parameters of the first AI network according to the risk assessment coefficient information of the reference offline behavior data to converge the first AI network, so as to obtain a second AI network corresponding to each reference offline behavior data group.

In a design idea, different reference offline behavior data clusters are used to perform network weight optimization on a first AI network, so as to obtain different second AI networks.

And step R103, extracting the offline behavior description of the reference offline behavior data in the reference offline behavior data sequence by adopting the second AI network, obtaining the reference offline behavior description corresponding to each second AI network, and determining the relation description component of the reference offline behavior description.

And the relationship description component is used for representing behavior description relationship between the reference offline behavior descriptions corresponding to different second AI networks.

The reference offline behavior description may be a behavior description feature composed of transfer relationship features extracted by the second AI network from the reference offline behavior data sequence, where the transfer relationship features may include offline behavior features and reference relationship features of each reference offline behavior data.

In an embodiment that can be implemented independently, step R103 can be implemented by the following exemplary steps.

And B1, extracting the offline behavior description of the reference offline behavior data in the reference offline behavior data sequence by adopting the second AI network to obtain the reference offline behavior description corresponding to each second AI network.

For example, the second AI network may be used to perform feature extraction on each reference offline behavior data in the reference offline behavior data sequence to obtain offline behavior features of the reference offline behavior data, determine an offline user demand tendency value of an offline behavior in the corresponding reference offline behavior data according to the offline behavior features, and use the offline behavior features of the reference offline behavior data and the offline user demand tendency value as a reference offline behavior description corresponding to the second AI network.

The offline behavior features may be features of an intermediate layer in a process of extracting features from reference offline behavior data, and may include behavior type features, behavior path features, behavior trajectory features, behavior association relationship features, and the like of the reference offline behavior data.

The offline user demand tendency value may be a predicted value of a user demand tendency confidence of the offline behavior in the reference offline behavior data.

And B2, determining a relation description component described by referring to the offline behavior.

For example, the relationship description component described with reference to the offline behavior may be determined by the feature attention model, which may be as follows:

and converting the reference offline behavior description into behavior relation characteristics by adopting a characteristic attention model (attention network), acquiring preset behavior relation parameters corresponding to the behavior relation characteristics, and fusing the preset behavior relation parameters and the behavior relation characteristics to obtain the reference behavior relation description.

For example, the reference offline behavior description may be converted into a behavior description feature vector, then a behavior entity vector (k) and a behavior relation value vector (v) are constructed based on the behavior description feature vector, then the behavior entity vector (k) and the behavior relation value vector (v) are made to be the same as the behavior description feature vector, and the assigned behavior entity vector (k) and the assigned behavior relation value vector (v) are used as the behavior relation feature.

After the behavior relation characteristic is converted into the behavior relation characteristic, the preset behavior relation parameter corresponding to the behavior relation characteristic may be obtained, for example, the association relation parameter (q) corresponding to the behavior relation characteristic may be screened from the preset behavior relation parameter (q) sequence as the preset behavior relation parameter, or the association relation parameter (q) corresponding to the reference offline behavior description may be screened from the preset behavior relation parameter (q) sequence as the preset behavior relation parameter.

After the preset behavior relation parameter is obtained, the implementation manner of fusing the preset behavior relation parameter and the behavior relation feature may be: and performing dot product on the association relation parameters and the behavior entity vector (k) in each behavior relation characteristic to obtain attention relation description of the reference offline behavior description to other reference offline behavior descriptions, wherein the attention relation description is used as the reference behavior relation description.

After the reference behavior relationship description is obtained, the reference behavior relationship description may be further subjected to regularization processing, so as to obtain a relationship description component of the reference offline behavior description, for example, the attention relationship description of the reference offline behavior description may be subjected to regularization processing by a softmax function, or other normalization functions may be further adopted to perform regularization processing, so as to obtain a relationship description component corresponding to each reference offline behavior description.

The method comprises the steps of determining a relation description component described by referring to offline behaviors by adopting a feature attention model, and mainly aiming at forming a transfer relation feature with generalization.

In a possible design idea, the relationship description component of the reference offline behavior description is determined, and may also be determined according to the reference offline behavior data group, for example, loss calculation may be performed according to the number of reference offline behavior data in the reference offline behavior data group corresponding to the reference offline behavior description or the total number of reference offline behavior data of the offline environment tag and the reference offline behavior data of the reference offline behavior data sequence or the total offline environment tag, and the relationship description component of the reference offline behavior description is determined according to a loss calculation result, or other algorithms may also be used to determine the relationship description component of the reference offline behavior description.

And step R104, aggregating the reference offline behavior descriptions according to the relation description component to obtain the aggregated reference offline behavior descriptions.

For example, a first relationship description component corresponding to the offline behavior feature and a second relationship description component corresponding to the offline user demand tendency value may be extracted from the relationship description components, the offline behavior features of each piece of reference offline behavior data are feature-enriched according to the first relationship description component, the feature-enriched offline behavior features are aggregated to obtain an aggregated offline behavior feature of each piece of reference offline behavior data, the feature enrichment is performed on each offline user demand tendency value based on the second relationship description component, and the relationship between the feature-enriched offline user demand tendency values is determined to obtain the reference relationship feature of each piece of reference offline behavior data.

And step R105, performing network weight optimization on the first AI network by adopting the aggregated reference offline behavior description and the reference offline behavior data sequence to obtain a third AI network, and performing offline user demand mining on the offline behavior big data to be mined based on the third AI network.

For example, in an embodiment that can be implemented independently, step R105 can be implemented by the following exemplary steps.

And C1, performing network weight optimization on the first AI network by adopting the aggregated reference offline behavior description and the reference offline behavior data sequence to obtain a third AI network.

For example, the first AI network may be used to extract an offline behavior description of each reference offline behavior data in the reference offline behavior data sequence to obtain a target reference offline behavior description corresponding to the first AI network, determine risk assessment coefficient information corresponding to the reference offline behavior data sequence according to the aggregate reference offline behavior description and the target reference offline behavior description, and perform network weight optimization on the first AI network based on the risk assessment coefficient information to obtain a third AI network, which may specifically be as follows:

and D1, extracting the offline behavior description of each reference offline behavior data in the reference offline behavior data sequence by adopting the first AI network to obtain the target reference offline behavior description corresponding to the first AI network.

For example, a first AI network is adopted to extract the characteristics of each reference offline behavior data in the reference offline behavior data sequence, extract the reference offline behavior data for multiple times, aggregate the extracted offline behavior characteristics to obtain the target offline behavior characteristics of each reference offline behavior data, determine the target offline user demand tendency value of the offline behavior in the corresponding reference offline behavior data according to the target offline behavior characteristics, calculate the relationship between the target offline user demand tendency values respectively to obtain target reference relationship characteristics, and use the target offline behavior characteristics and the target reference relationship characteristics of the reference offline behavior data as the target reference offline behavior description corresponding to the first AI network.

D2, determining risk assessment coefficient information corresponding to the reference offline behavior data sequence according to the aggregate reference offline behavior description and the target reference offline behavior description.

For example, the aggregate reference offline behavior description and the target reference offline behavior description may be subjected to loss calculation to obtain first risk assessment coefficient information corresponding to the reference offline behavior data sequence, user demand prediction information of each reference offline behavior data in the reference offline behavior data sequence is extracted from the target reference offline behavior description, the user demand prediction information and user demand calibration information in the reference offline behavior data are subjected to loss calculation to obtain second risk assessment coefficient information corresponding to the reference offline behavior data sequence, and the first risk assessment coefficient information and the second risk assessment coefficient information are subjected to weighted calculation to obtain risk assessment coefficient information corresponding to the reference offline behavior data sequence, which may specifically be as follows:

(1) and performing loss calculation on the aggregated reference offline behavior description and the target reference offline behavior description to obtain first risk assessment coefficient information corresponding to the reference offline behavior data sequence.

For example, a target offline behavior feature and a target reference relationship feature corresponding to each reference offline behavior data may be extracted from the target reference offline behavior description, the target offline behavior feature and an aggregation offline behavior feature corresponding to the aggregation reference offline behavior description are subjected to loss calculation to obtain risk assessment coefficient information of the offline behavior feature corresponding to the reference offline behavior data sequence, the target reference relationship feature and an aggregation reference relationship feature corresponding to the aggregation reference offline behavior description are subjected to loss calculation to obtain risk assessment coefficient information of the relationship feature corresponding to the reference offline behavior data sequence, and the risk assessment coefficient information of the offline behavior feature and the risk assessment coefficient information of the relationship feature are used as first risk assessment coefficient information corresponding to the reference offline behavior data sequence.

For example, a feature vector loss value between the mean feature vector of the target offline behavior feature and the mean feature vector of the aggregated offline behavior feature in the aggregated reference offline behavior description may be calculated, so as to obtain a first feature vector loss value. And determining risk evaluation coefficient information of the off-line behavior characteristics corresponding to the reference off-line behavior data sequence based on the first feature vector loss value.

The loss calculation method may include a plurality of loss calculation methods for the target reference relationship feature and the aggregation reference relationship feature corresponding to the aggregation reference offline behavior description, for example, a loss value of a feature vector of the target reference relationship feature and the aggregation sample feature may be calculated to obtain a second feature vector loss value, for example, the second feature vector loss value may be a difference value of cosine angles, a weighting calculation is performed on a second feature vector loss value corresponding to each reference offline behavior data in the reference offline behavior data sequence, and the weighted calculated feature vector loss value is used as risk assessment coefficient information of the relationship feature corresponding to the reference offline behavior data sequence.

It should be noted that calculating the risk assessment coefficient information of the offline behavior feature and the risk assessment coefficient information of the relationship feature may be regarded as transmitting the aggregated reference offline behavior description transfer relationship feature to the student network by a dark knowledge extraction method, so as to obtain the second AI network, and different dark knowledge extraction strategies are adopted for different types of transfer relationship features, when the transfer relationship feature is the offline behavior feature of the reference offline behavior data, the MMD (maximum mean difference) dark knowledge extraction method may be adopted for transfer, and when the transfer relationship feature is the reference relationship feature, the structural dark knowledge extraction method may be adopted for transfer. Of course, different dark knowledge extraction modes, such as variation-based feature dark knowledge extraction, can be used for extracting the dark knowledge of the transfer relationship features and transferring the extracted dark knowledge to the student network, and the main purpose is to ensure the effectiveness of the extraction of the dark knowledge.

(2) And extracting user demand prediction information of each reference offline behavior data in the reference offline behavior data sequence from the target reference offline behavior description, and performing loss calculation on the user demand prediction information and user demand calibration information in the reference offline behavior data to obtain second risk assessment coefficient information corresponding to the reference offline behavior data sequence.

For example, the following may be specifically mentioned:

and E1, extracting user demand prediction information of each reference offline behavior data in the reference offline behavior data sequence from the target reference offline behavior description.

For example, a target user demand characteristic may be extracted from the target offline behavior characteristic corresponding to each reference offline behavior data, a target offline user demand tendency value may be extracted from the target reference relationship characteristic corresponding to each reference offline behavior data, a predicted user demand type of the corresponding reference offline behavior data may be determined according to the target offline user demand tendency value, and the predicted user demand type and the target user demand characteristic may be used as user demand prediction information of the reference offline behavior data.

For example, at least one offline user demand tendency value of the reference offline behavior data is calculated according to the relationship between the offline user demand tendency values in the target reference relationship feature, and the offline user demand tendency values are fused to obtain the target offline user demand tendency value.

After the target offline user demand tendency value is obtained, the predicted user demand type of the reference offline behavior data may be determined, for example, the target offline user demand tendency value may be compared with a preset demand tendency value threshold, when the target offline user demand tendency value is greater than the preset demand tendency value threshold, the predicted user demand type of the reference offline behavior data may be determined to be an actual user demand type, and when the target offline user demand tendency value is not greater than the preset demand tendency value threshold, the predicted user demand type of the reference offline behavior data may be determined to be a non-actual user demand type.

And E2, performing loss calculation on the user demand prediction information and the user demand calibration information in the reference offline behavior data to obtain second risk assessment coefficient information corresponding to the reference offline behavior data sequence.

For example, user requirement calibration information of each reference offline behavior data in the reference offline behavior data sequence may be obtained, where the user requirement calibration information may include a calibrated user requirement characteristic and a calibrated user requirement type, loss calculation is performed on a target user requirement characteristic and the user requirement characteristic calibrated by the corresponding reference offline behavior data to obtain risk assessment coefficient information of the user requirement characteristic, loss calculation is performed on a predicted user requirement type and the user requirement type calibrated by the corresponding reference offline behavior data to obtain risk assessment coefficient information of the user requirement type, and the risk assessment coefficient information of the user requirement characteristic and the risk assessment coefficient information of the user requirement type are used as second risk assessment coefficient information corresponding to the reference offline behavior data sequence.

The risk assessment coefficient information of the user demand characteristics may be determined in various ways, for example, the risk assessment coefficient information of the user demand characteristics referring to the offline behavior data may be calculated by using a mean square error loss function according to the predicted user demand characteristics and the calibrated user demand characteristics, or the risk assessment coefficient information of the user demand characteristics referring to the offline behavior data may be calculated by using other types of loss functions.

For example, the binary cross entropy loss function may be used to calculate the risk assessment coefficient information of the user demand type referring to the offline behavior data according to the predicted user demand type and the calibrated user demand type, or another classification loss function may be used to calculate the risk assessment coefficient information of the user demand type referring to the offline behavior data.

(3) And performing weighted calculation on the first risk evaluation coefficient information and the second risk evaluation coefficient information to obtain risk evaluation coefficient information corresponding to the reference offline behavior data sequence.

For example, loss weights corresponding to the first risk assessment coefficient information and the second risk assessment coefficient information may be obtained, feature enrichment may be performed on the first risk assessment coefficient information and the second risk assessment coefficient information according to the loss weights, weighted calculation may be performed on the first risk assessment coefficient information after the feature enrichment and the second risk assessment coefficient information after the feature enrichment to obtain risk assessment coefficient information corresponding to the reference offline behavior data sequence, or the second risk assessment coefficient information may be performed on the first risk assessment coefficient information to perform splicing to obtain risk assessment coefficient information corresponding to the reference offline behavior data sequence, or the first risk assessment coefficient information and the second risk assessment coefficient information may be combined to obtain risk assessment coefficient information corresponding to the reference offline behavior data sequence.

And D3, based on the risk assessment coefficient information, performing network weight optimization on the first AI network to obtain a third AI network.

For example, the network parameters of the first AI network may be updated based on the risk assessment coefficient information, for example, the network parameters of the first AI network may be updated using a gradient descent algorithm or other algorithms to converge the first AI network, and when convergence is completed, the third AI network may be obtained.

And C2, performing offline user requirement mining on the offline behavior big data to be mined based on the third AI network.

For example, offline behavior big data to be mined can be obtained, the offline behavior big data to be mined comprises at least one offline behavior data used for offline user demand mining, feature extraction is performed on the offline behavior big data to be mined by adopting a third AI network to obtain offline user demand related features of the offline behavior data, a user demand tendency confidence coefficient of the offline behavior data is calculated according to the offline user demand related features, the user demand tendency confidence coefficient is used for representing the possibility that the offline behavior data is related to candidate offline user demands, and when the user demand tendency confidence coefficient is larger than a preset confidence coefficient threshold, the corresponding candidate offline user demands are determined as mined target offline user demands.

After the offline behavior big data to be mined is obtained, feature extraction may be performed on the offline behavior big data to be mined by using the third AI network, and the feature extraction manner may be various, for example, the offline behavior feature of the offline behavior big data to be mined may be directly extracted, the offline behavior relation feature corresponding to the offline behavior big data to be mined is determined according to the offline behavior feature, and the offline behavior feature and the offline behavior relation feature are fused to obtain the offline user demand related feature of the offline behavior data.

In one possible design approach, the artificial intelligence cloud system may further include storing the aggregated reference offline behavior description onto the blockchain.

As can be seen from the above, in the embodiments of the present invention, after the reference offline behavior data sequence is obtained, the reference offline behavior data is clustered according to the offline environment tag of the reference offline behavior data in the reference offline behavior data sequence to obtain a plurality of reference offline behavior data clusters, the first AI network is respectively subjected to network weight optimization by using the reference offline behavior data clusters to obtain the second AI network corresponding to each reference offline behavior data cluster, then, the offline behavior descriptions of the reference offline behavior data in the reference offline behavior data sequence are extracted by using the second AI network to obtain the reference offline behavior description corresponding to each second AI network, the relationship description component of the reference offline behavior description is determined, then, the reference offline behavior descriptions are aggregated according to the relationship description component to obtain an aggregated reference offline behavior description, and finally, the aggregated reference offline behavior description and the reference offline behavior data sequence are used to perform network on the first AI network Optimizing the weight to obtain a third AI network, and performing offline user demand mining on the offline behavior big data to be mined based on the third AI network; therefore, the teacher network (second AI network) is trained through the reference offline behavior data groups corresponding to the plurality of offline environment labels, part of transfer relationship features of the reference offline behavior data sequence are extracted through different second AI networks, and aggregation is performed according to the relationship description components of the transfer relationship features, so that aggregate reference offline behavior descriptions are obtained, the aggregate reference offline behavior descriptions are transferred to the student network (third AI network) by means of dark knowledge extraction, so that the feature connection characteristics of the transfer relationship features are effectively learned, the accuracy of third AI network prediction is improved, and finally the accuracy of offline user demand mining can be improved.

The method described in the above examples is further illustrated in detail below by way of example. In one design approach, the second AI network is a teacher network, and the third AI network is a student network.

For example, an embodiment of the present application provides a method for optimizing personalized service content based on big data, where the specific flow is as follows:

and step R201, acquiring a reference offline behavior data sequence.

And R202, clustering the reference offline behavior data according to the offline environment labels of the reference offline behavior data to obtain a plurality of reference offline behavior data clusters.

And step R203, performing network weight optimization on the first AI network by adopting the reference offline behavior data clusters respectively to obtain a teacher network corresponding to each reference offline behavior data cluster.

For example, the first AI network may be used to perform offline user requirement mining on reference offline behavior data in the reference offline behavior data group to obtain a predicted offline user requirement mining result of the reference offline behavior data, perform loss calculation on the predicted offline user requirement mining result and a calibrated user requirement type in the reference offline behavior data, determine risk assessment coefficient information of the reference offline behavior data, and update parameters of the first AI network according to the risk assessment coefficient information of the reference offline behavior data to converge the first AI network, so as to obtain a teacher network corresponding to each reference offline behavior data group.

And step R204, extracting the offline behavior description of the reference offline behavior data in the reference offline behavior data sequence by adopting the teacher network to obtain the reference offline behavior description corresponding to each teacher network.

For example, a teacher network may be used to perform feature extraction on each reference offline behavior data in the reference offline behavior data sequence to obtain offline behavior features of the reference offline behavior data, determine an offline user demand tendency value of an offline behavior in the corresponding reference offline behavior data according to the offline behavior features, and use the offline behavior features of the reference offline behavior data and the offline user demand tendency value as reference offline behavior descriptions corresponding to the teacher network.

And step R205, determining a relation description component described by referring to the offline behavior.

And step R206, aggregating the reference offline behavior descriptions according to the relation description component to obtain an aggregated reference offline behavior description.

For example, a first relationship description component corresponding to the offline behavior feature and a second relationship description component corresponding to the offline user demand tendency value may be extracted from the relationship description components, the offline behavior features of each piece of reference offline behavior data are feature-enriched according to the first relationship description component, the feature-enriched offline behavior features are aggregated to obtain the aggregated offline behavior feature of each piece of reference offline behavior data, the feature-enriched offline user demand tendency value is converted into a corresponding one-dimensional vector based on the second relationship description component, and cosine angles between the one-dimensional vectors are respectively calculated to obtain the reference relationship features corresponding to the reference offline behavior data.

And step R207, performing network weight optimization on the first AI network by adopting the aggregate reference offline behavior description and the reference offline behavior data sequence to obtain the student network.

The target offline behavior feature and the target reference relationship feature corresponding to each piece of reference offline behavior data can be extracted from the target reference offline behavior description, the target offline behavior feature and the average feature vector of the aggregate offline behavior feature in the aggregate reference offline behavior description are calculated, and the feature vector loss value between the average feature vector of the target offline behavior feature and the average feature vector of the aggregate offline behavior feature is calculated to obtain the first feature vector loss value. And determining risk evaluation coefficient information of the offline behavior characteristics corresponding to the reference offline behavior data sequence based on the first feature vector loss value.

And calculating a feature vector loss value of the target reference relation feature and the aggregation sample feature to obtain a second feature vector loss value, for example, the second feature vector loss value may be a difference value of cosine angles, performing weighted calculation on the second feature vector loss value corresponding to each reference offline behavior data in the reference offline behavior data sequence, and taking the feature vector loss value obtained by weighted calculation as risk evaluation coefficient information of the relation feature corresponding to the reference offline behavior data sequence. And taking the risk evaluation coefficient information of the offline behavior characteristics and the risk evaluation coefficient information of the relationship characteristics as first risk evaluation coefficient information corresponding to the reference offline behavior data sequence.

Target user demand characteristics can be extracted from the target offline behavior characteristics corresponding to each reference offline behavior data, at least one offline user demand tendency value of the reference offline behavior data is calculated according to the relation between the offline user demand tendency values in the target reference relation characteristics, and the offline user demand tendency values are fused to obtain the target offline user demand tendency value. And comparing the target offline user demand tendency value with a preset demand tendency value threshold, determining that the predicted user demand type of the reference offline behavior data is the actual user demand type when the target offline user demand tendency value is greater than the preset demand tendency value threshold, and determining that the predicted user demand type of the reference offline behavior data is the non-actual user demand type when the target offline user demand tendency value is not greater than the preset demand tendency value threshold. And taking the predicted user demand type and the target user demand characteristics as user demand prediction information of the reference offline behavior data.

The method comprises the steps of obtaining user demand calibration information of each piece of reference offline behavior data in a reference offline behavior data sequence, wherein the user demand calibration information can comprise calibrated user demand characteristics and calibrated user demand types, calculating risk assessment coefficient information of the user demand characteristics of the reference offline behavior data according to predicted user demand characteristics and calibrated user demand characteristics by adopting a mean square error loss function, calculating risk assessment coefficient information of the user demand types of the reference offline behavior data according to predicted user demand types and calibrated user demand types by adopting a binary cross entropy loss function, and taking the risk assessment coefficient information of the user demand characteristics and the risk assessment coefficient information of the user demand types as second risk assessment coefficient information corresponding to the reference offline behavior data sequence.

The method includes the steps of obtaining loss weights corresponding to first risk evaluation coefficient information and second risk evaluation coefficient information, performing feature enrichment on the first risk evaluation coefficient information and the second risk evaluation coefficient information according to the loss weights, performing weighted calculation on the first risk evaluation coefficient information after the feature enrichment and the second risk evaluation coefficient information after the feature enrichment to obtain risk evaluation coefficient information corresponding to a reference offline behavior data sequence, or splicing the second risk evaluation coefficient information of the first risk evaluation coefficient information to obtain risk evaluation coefficient information corresponding to the reference offline behavior data sequence, or combining the first risk evaluation coefficient information and the second risk evaluation coefficient information to obtain risk evaluation coefficient information corresponding to the reference offline behavior data sequence.

A gradient descent algorithm or other algorithm may be used to update the network parameters of the first AI network to converge the first AI network, and when convergence is complete, the student network is obtained.

And R208, performing offline user requirement mining on the offline behavior big data to be mined based on the student network.

For example, offline behavior big data to be mined may be directly acquired. The method comprises the steps that an offline behavior feature of offline behavior big data to be mined is directly extracted by a feature extraction network of a student network, an offline behavior relation feature corresponding to the offline behavior big data to be mined is determined according to the offline behavior feature, the offline behavior feature and the offline behavior relation feature are fused to obtain an offline user demand related feature of the offline behavior data, a user demand tendency confidence coefficient of the offline behavior data is calculated according to the offline user demand related feature, and the user demand tendency confidence coefficient is used for representing the possibility that the offline behavior data is related to candidate offline user demands; and when the user requirement tendency confidence coefficient is larger than a preset confidence coefficient threshold value, determining the corresponding candidate offline user requirement as the mined target offline user requirement.

As can be seen from the above, after the reference offline behavior data sequence is obtained, the reference offline behavior data is clustered according to the offline environment tag of the reference offline behavior data in the reference offline behavior data sequence to obtain a plurality of reference offline behavior data clusters, the first AI network is respectively subjected to network weight optimization by using the reference offline behavior data clusters to obtain the second AI network corresponding to each reference offline behavior data cluster, then, the offline behavior descriptions of the reference offline behavior data in the reference offline behavior data sequence are extracted by using the second AI network to obtain the reference offline behavior description corresponding to each second AI network, the relationship description component of the reference offline behavior description is determined, then, the reference offline behavior descriptions are aggregated according to the relationship description component to obtain an aggregated reference offline behavior description, and finally, the first AI network is subjected to network weight optimization by using the aggregated reference offline behavior description and the reference behavior data sequence, obtaining a third AI network, and performing offline user demand mining on the offline behavior big data to be mined based on the third AI network; therefore, the teacher network (second AI network) is trained through the reference offline behavior data groups corresponding to the plurality of offline environment labels, part of transfer relationship features of the reference offline behavior data sequence are extracted through different second AI networks, and aggregation is performed according to the relationship description components of the transfer relationship features, so that aggregate reference offline behavior descriptions are obtained, the aggregate reference offline behavior descriptions are transferred to the student network (third AI network) by means of dark knowledge extraction, so that the feature connection characteristics of the transfer relationship features are effectively learned, the accuracy of third AI network prediction is improved, and finally the accuracy of offline user demand mining can be improved.

Fig. 3 illustrates a hardware structural diagram of an artificial intelligence cloud system 100 for implementing the big data based personalized service content optimization method, according to an embodiment of the present application, and as shown in fig. 3, the artificial intelligence cloud system 100 may include a processor 110, a machine-readable storage medium 120, a bus 130, and a communication unit 140.

In a specific implementation process, at least one processor 110 executes computer-executable instructions stored in the machine-readable storage medium 120, so that the processor 110 may execute the method for optimizing personalized service content based on big data according to the above method embodiment, the processor 110, the machine-readable storage medium 120, and the communication unit 140 are connected through the bus 130, and the processor 110 may be configured to control transceiving actions of the communication unit 140, so as to perform data transceiving with the aforementioned 5G internet terminal 200.

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

In addition, an embodiment of the present application further provides a readable storage medium, where the readable storage medium has preset computer-executable instructions, and when a processor executes the computer-executable instructions, the method for optimizing personalized service content based on big data is implemented.

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

Claims

1. A personalized service content optimization method based on big data is applied to an artificial intelligence cloud system, the artificial intelligence cloud system is in communication connection with a plurality of 5G internet terminals, and the method comprises the following steps:

analyzing the basic offline behavior data and the tracking offline behavior data corresponding to each target offline user requirement to obtain derivative requirement characteristics corresponding to each target offline user requirement;

2. The method for optimizing personalized service content based on big data according to claim 1, wherein the step of analyzing the basic offline behavior data and the tracking offline behavior data corresponding to each target offline user requirement to obtain the derived requirement characteristics corresponding to each target offline user requirement comprises:

extracting a first offline behavior feature of basic offline behavior data corresponding to each target offline user requirement and a second offline behavior feature of the tracking offline behavior data;

calculating a differential offline behavior signature between the first offline behavior signature and the second offline behavior signature;

and carrying out requirement characteristic mapping on the difference offline behavior characteristics to obtain derived requirement characteristics corresponding to the requirements of the target offline users.

3. The method for optimizing the content of the big data-based personalized service, according to claim 1, wherein the step of optimizing the content of the next pushed online personalized service content data based on the derived demand characteristics corresponding to each of the target offline user demands comprises:

acquiring derived demand knowledge points of derived demand characteristics corresponding to each target offline user demand, and acquiring knowledge map data of each derived demand knowledge map generated by the derived demand knowledge points, wherein the knowledge map data comprises a frequent knowledge point category of the derived demand knowledge map, derived demand knowledge point information of the derived demand knowledge map and a quotation demand knowledge map of the derived demand knowledge map;

acquiring the derived demand knowledge map and interest prediction information of each knowledge map unit in the knowledge map data;

determining knowledge map attributes corresponding to the derived demand knowledge map, wherein the knowledge map attributes represent knowledge point connection relations between the derived demand knowledge map and each knowledge map unit in the knowledge map data;

and according to the interest prediction information and the knowledge map attribute, determining interest point information of the derivative demand knowledge map, performing interest point feature analysis on the derivative demand knowledge map according to the interest point information, and performing content optimization on the next pushed online personalized service content data based on the interest point feature of the derivative demand knowledge map.

4. The method for optimizing personalized service content based on big data according to claim 3, wherein the determining the knowledge map attributes corresponding to the derived demand knowledge map comprises:

generating a derived demand guide map corresponding to the derived demand knowledge map according to the knowledge map data and the derived demand knowledge map, wherein the derived demand guide map is a knowledge map which is subjected to attribute feature integration on the attributes of the knowledge map; the map guide units in the derived demand map comprise map guide units corresponding to the derived demand knowledge map and map guide units corresponding to all knowledge map units in the knowledge map data respectively, and the map guide connection attribute in the derived demand map comprises connection relation information between the derived demand knowledge map and the map guide units corresponding to all knowledge map units in the knowledge map data;

the step of determining interest point information of interest of the derived demand knowledge map according to the interest prediction information and the knowledge map attributes comprises the following steps:

according to the interest prediction information and the derived demand guide map, determining interest point information of the derived demand knowledge map;

the knowledge map data comprises one or more frequent knowledge point categories, and the map joint attributes in the derived demand map further comprise joint relation information between map units associated with the frequent knowledge point categories.

5. The method as claimed in claim 4, wherein the determining interest point information of the derived demand knowledge map according to the interest prediction information and the derived demand guide map comprises:

aiming at a target map guiding unit in the derived demand map, acquiring first interest resource information corresponding to demand knowledge map information under various derived demand categories according to interest prediction information corresponding to each cascade map guiding unit under various derived demand categories of the target map guiding unit, wherein the target map guiding unit is a map guiding unit corresponding to the derived demand knowledge map, and map guiding units respectively corresponding to various types of data in the knowledge map data correspond to map guiding units under one derived demand category;

6. The big data based personalized service content optimization method according to claim 5, further comprising:

aiming at each map guide unit in the derived demand map guide, the following steps are executed in a walking cycle to collect the interest resource information of the map guide unit:

acquiring second interest resource information corresponding to the demand knowledge map information under various derived demand categories of the map guide unit according to the current interest resource information of each cascaded map guide unit under various derived demand categories;

obtaining target interest resource information of the map guide unit according to the current interest resource information of the map guide unit and each second interest resource information corresponding to the map guide unit; when the first wandering cycle is not performed, the current interest resource information corresponding to the non-first wandering cycle is the target interest resource information obtained by the previous wandering cycle, and the interest resource information of the map guide unit is the target interest resource information obtained by the last wandering cycle;

the acquiring first interest resource information corresponding to the demand knowledge map information under each derived demand category according to the interest prediction information corresponding to each cascaded map guide unit under each derived demand category of the target map guide unit includes:

the obtaining of interest point information of interest of the derived demand knowledge map according to the first interest resource information corresponding to the target map guide unit and the interest prediction information of the target map guide unit includes:

acquiring a first category influence coefficient corresponding to demand knowledge map information under each derived demand category and a second category influence coefficient corresponding to the derived demand knowledge map;

according to a first category influence coefficient corresponding to the demand knowledge map information under each derived demand category, fusing first interest resource information corresponding to the demand knowledge map information under each derived demand category to obtain second interest resource information corresponding to the demand knowledge map information under each derived demand category;

according to the second category influence coefficient, fusing the interest resource information of the map guide unit of the target map guide unit to obtain third interest resource information;

splicing second interest resource information and third interest resource information corresponding to the demand knowledge map information under each derived demand category to obtain spliced interest resource information;

and obtaining the interest point information of interest of the derived demand knowledge map according to the spliced interest resource information.

7. The method for optimizing big data based personalized service content according to any one of claims 3 to 6, wherein the obtaining of the derived demand knowledge map and the interest prediction information of each knowledge map unit in the knowledge map data comprises:

acquiring key knowledge map information of the derived demand knowledge map, acquiring key interest resource information corresponding to the key knowledge map information, and taking the key interest resource information as interest prediction information of the derived demand knowledge map;

when the knowledge map data comprises the derived demand knowledge point information, acquiring past demand knowledge map information corresponding to the derived demand knowledge point information for each derived demand knowledge point information, and determining interest prediction information of the derived demand knowledge point information according to the past demand knowledge map information;

wherein the derived demand knowledge point information comprises one or more of target derived demand knowledge points of the derived demand knowledge map and associated derived demand knowledge points of the derived demand knowledge map; the quote demand knowledge map comprises a derived demand knowledge map corresponding to a target derived demand knowledge point of the derived demand knowledge map, and the derived demand knowledge map is one or more demand knowledge map information corresponding to the target derived demand knowledge point in a service progress stage for generating the derived demand knowledge map; when the derived demand knowledge point information includes a target derived demand knowledge point, the past demand knowledge map information is demand knowledge map information generated by the target derived demand knowledge point in a first preset business progress stage before a current business progress stage, and when the derived demand knowledge point information includes an associated derived demand knowledge point, the past demand knowledge map information is demand knowledge map information generated by the associated derived demand knowledge point in a second preset business progress stage before the current business progress stage.

8. The method for optimizing personalized service content based on big data according to any one of claims 3 to 6, wherein the performing interest point feature analysis of the derived demand knowledge map according to the interest point information comprises:

determining a reference demand knowledge map from a first demand knowledge map information base according to the interest point information of the derived demand knowledge map and a first interest index parameter of the interest point information of each candidate demand knowledge map information in the first demand knowledge map information base, and issuing the reference demand knowledge map to a target derived demand knowledge point, wherein the derived demand knowledge map is associated demand knowledge map information of past demand knowledge map information of the target derived demand knowledge point; or

Clustering the information of the various required knowledge maps in the second required knowledge map information base according to a second interest index parameter between the concerned interest point information of the various required knowledge maps in the second required knowledge map information base, wherein the derived required knowledge map is one or more pieces of required knowledge map information in the second required knowledge map information base.

9. The method for optimizing personalized service content based on big data according to any one of claims 3 to 6, wherein the determining interest point information of interest of the derived demand knowledge map according to each interest prediction information and the knowledge map attribute comprises:

inputting the interest prediction information and the knowledge map attribute into an interest point analysis network to obtain interest point information of the derived demand knowledge map;

wherein the method further comprises a step of network convergence for the point of interest analysis network, the step comprising:

acquiring a network convergence basic database, wherein the network convergence basic database comprises a plurality of basic attention interest training data, the basic attention interest training data comprises an example derived demand guide map corresponding to the example demand knowledge map information and interest prediction information of each guide map unit in the example derived demand guide map, wherein each map guide unit in each example derived demand map comprises a first map guide unit corresponding to the example demand knowledge map information and a second map guide unit corresponding to each first associated demand knowledge map information, the first associated required knowledge map information is different types of data included in knowledge map data of example required knowledge map information, the map join attribute in the example derived demand map comprises join relationship information between the first map guide unit and each second map guide unit;

loading each basic attention interest training data to an initial attention interest point analysis network to obtain predicted interest resource information of each map guide unit corresponding to each basic attention interest training data;

for each basic attention interest training data, determining a first loss function value corresponding to the basic attention interest training data according to a second interest index parameter between predicted interest resource information of a first map unit and predicted interest resource information of each second map unit in an example derived demand map of the basic attention interest training data;

determining a global loss function value corresponding to the interest point analysis network according to a first loss function value corresponding to each piece of basic interest training data;

and carrying out cyclic training on the concerned interest point analysis network according to the global loss function value until the global loss function value is converged.

10. An artificial intelligence cloud system, comprising a processor and a machine-readable storage medium, wherein the machine-readable storage medium stores a computer program, and the computer program is loaded and executed by the processor to implement the big data based personalized service content optimization method according to any one of claims 1 to 9.