CN115757979A

CN115757979A - User data recommendation method and system based on artificial intelligence

Info

Publication number: CN115757979A
Application number: CN202211638446.8A
Authority: CN
Inventors: 杨海龙; 闫志超
Original assignee: Jinan Yilin Chengda Network Co ltd
Current assignee: Hainan Darun Feng Enterprise Management Partnership LP
Priority date: 2022-12-20
Filing date: 2022-12-20
Publication date: 2023-03-07
Anticipated expiration: 2042-12-20
Also published as: CN115757979B

Abstract

The invention provides a user data recommendation method and system based on artificial intelligence, which comprises the following steps: respectively acquiring rough transaction characteristics of the merchant to be recommended, the first-level related merchant and the second-level related merchant aiming at the commodity transaction data; performing a first-level feature correlation selection operation on the rough transaction features of the to-be-recommended merchants and the rough transaction features of the first-level related merchants to obtain first-level transaction features of the to-be-recommended merchants; performing second-level feature correlation selection operation on the rough transaction features of the to-be-recommended merchants, the rough transaction features of second-level related merchants and the rough transaction features of first-level related merchants to obtain second-level transaction features of the to-be-recommended merchants; generating target transaction characteristics of the commercial tenant to be recommended based on the primary transaction characteristics and the secondary transaction characteristics; the target user type of the merchant to be recommended is determined based on the target transaction characteristics, user data is recommended to the merchant to be recommended according to the target user type, and the user data can be effectively recommended to the merchant.

Description

User data recommendation method and system based on artificial intelligence

Technical Field

The invention relates to the technical field of electronic commerce management, in particular to a user data recommendation method and system based on artificial intelligence.

Background

With the development of electronic commerce, a large number of transactions, and transaction data generated in the internet, competition among merchants is increasingly understood. In the prior art, a merchant needs to pay a large cost for acquiring user data, and the acquired user data is not necessarily a target user. Based on this, how to recommend the user data to the merchant more quickly and effectively is a problem to be solved urgently.

Disclosure of Invention

The invention aims to provide a user data recommendation method and system based on artificial intelligence.

In a first aspect, an embodiment of the present invention provides a user data recommendation method based on artificial intelligence, which obtains rough transaction characteristics of merchants to be recommended, primary relevant merchants, and secondary relevant merchants for commodity transaction data, respectively;

a first-level related merchant label is arranged between the merchant to be recommended and the first-level related merchant, and a second-level related merchant label is arranged between the merchant to be recommended and the second-level related merchant;

performing a first-level feature correlation selection operation on the rough transaction features of the to-be-recommended merchants and the rough transaction features of the first-level related merchants to obtain first-level transaction features of the to-be-recommended merchants;

performing second-level feature correlation selection operation on the rough transaction features of the to-be-recommended merchants, the rough transaction features of the second-level related merchants and the rough transaction features of the first-level related merchants to obtain second-level transaction features of the to-be-recommended merchants;

generating target transaction characteristics of the merchant to be recommended based on the primary transaction characteristics and the secondary transaction characteristics;

and determining the target user type of the merchant to be recommended based on the target transaction characteristics, and recommending user data to the merchant to be recommended according to the target user type.

In a second aspect, an embodiment of the present invention provides a server system, which includes a server, and the server is configured to perform the embodiment of the first aspect.

Compared with the prior art, the beneficial effects provided by the invention comprise: the rough transaction characteristics of the merchant to be recommended, the primary related merchant and the secondary related merchant are respectively obtained, and the rough transaction characteristics are used for representing merchant attribute information and explicit transaction information of the merchant. Further, the rough transaction characteristics of the to-be-recommended merchant and the rough transaction characteristics of the first-level related merchant can be subjected to first-level feature correlation selection operation to obtain first-level transaction characteristics, and the rough transaction characteristics of the to-be-recommended merchant, the rough transaction characteristics of the first-level related merchant and the rough transaction characteristics of the second-level related merchant are subjected to second-level feature correlation selection operation to obtain second-level transaction characteristics. The implicit transaction information of the merchant to be recommended can be mined through the feature correlation selection operation of the first level and the feature correlation selection operation of the second level, that is, the first-level transaction characteristics and the second-level transaction characteristics can represent merchant attribute information and explicit transaction information of the merchant to be recommended and can also represent the implicit transaction information of the merchant to be recommended. The target transaction characteristics of the to-be-recommended merchant are generated based on the primary transaction characteristics and the secondary transaction characteristics, and the target transaction characteristics can represent rich transaction information of the to-be-recommended merchant, so that the recommendation accuracy of user data can be improved by recommending the user data for the to-be-recommended merchant based on the target transaction characteristics.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required to be used in the embodiments will be briefly described below. It is appreciated that the following drawings depict only certain embodiments of the invention and are therefore not to be considered limiting of its scope. It is obvious to a person skilled in the art that other relevant figures can also be obtained on the basis of these figures without inventive effort.

FIG. 1 is a schematic flowchart illustrating steps of a method for recommending user data based on artificial intelligence according to an embodiment of the present invention;

fig. 2 is a schematic block diagram of a server structure according to an embodiment of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

It should be noted that all the data acquisition actions in the present application are performed under the premise of complying with the corresponding data protection regulation policy of the country of the location and obtaining the authorization given by the owner of the corresponding device.

In order to solve the technical problem in the foregoing background art, fig. 1 is a schematic flowchart of a method for recommending user data based on artificial intelligence according to an embodiment of the present disclosure, where the method for recommending user data based on artificial intelligence according to this embodiment may be executed by a server, and a detailed description is performed on the method for recommending user data based on artificial intelligence.

The user data recommendation method based on artificial intelligence can comprise the following steps S101-S105:

s101, rough transaction characteristics of the commercial tenant to be recommended, the first-level related commercial tenant and the second-level related commercial tenant aiming at the commodity transaction data are respectively obtained.

And a first-stage related merchant label is arranged between the merchant to be recommended and the first-stage related merchant, and a second-stage related merchant label is arranged between the merchant to be recommended and the second-stage related merchant.

The first-level relevant merchant label and the second-level relevant merchant label may be determined based on a degree of correlation between merchants, where the degree of correlation may be determined based on information such as commodities, prices, and the like between merchants, and the degree of correlation corresponding to the first-level relevant merchant label is greater than the degree of correlation corresponding to the second-level relevant merchant label.

It will be appreciated that the specific content to which the merchandise transaction data refers may vary, for example, the merchandise transaction data may include merchandise deal data, merchandise return data, merchandise price reduction data, purchaser data, and the like. The coarse transaction characteristics may include a first characteristic characterizing merchant attribute information for the merchant and a transaction characteristic vector characterizing a first amount of transaction information for the merchant. The merchant attribute information may include inherent basic attribute elements of the merchant such as a registration address, a shipping address, a category of merchandise sold, and the like. In the application, the basic merchant information and the historical merchant transaction data of the merchant to be recommended can be obtained. And generating rough transaction characteristics of the merchant to be recommended for the commodity transaction data based on the basic merchant information of the merchant to be recommended and the historical merchant transaction data. The rough transaction characteristics of the to-be-recommended merchant comprise first characteristics used for representing merchant attribute information of the to-be-recommended merchant and transaction characteristic vectors of transaction information of the to-be-recommended merchant.

Similarly, the basic merchant information and the historical merchant transaction data of the first-level relevant merchant can be obtained, and the rough transaction characteristics of the first-level relevant merchant for the commodity transaction data are generated based on the basic merchant information and the historical merchant transaction data of the first-level relevant merchant. The second-level relevant merchants are in the same way.

Optionally, step S101 includes: acquiring a first related merchant map corresponding to a merchant to be recommended, wherein the first related merchant map comprises a first map element used for representing rough transaction characteristics of the merchant to be recommended, a second map element used for representing rough transaction characteristics of the merchant to be recommended, and a correlation inference line obtained by performing correlation inference on the first map element and the second map element; the merchant to be recommended and the pending related merchant have related merchant labels; according to a correlation inference path taking the first map element as an inference base point, sampling and inquiring second map elements in the first related merchant map one by one to obtain a first-stage related merchant having a first-stage related merchant label with the merchant to be recommended and a second-stage related merchant having a second-stage related merchant label with the merchant to be recommended; and rough transaction characteristics of the merchant to be recommended, the first-stage related merchant and the second-stage related merchant aiming at the commodity transaction data are respectively obtained from the first related merchant map.

The undetermined related merchants comprise undetermined related merchants with primary related merchant labels of merchants to be recommended and undetermined related merchants with secondary related merchant labels of merchants to be recommended, the server can take all or part of the first related merchant map and the undetermined merchants with the primary related merchant labels of the merchants to be recommended as primary related merchants, and take all or part of the first related merchant map and the undetermined merchants with the secondary related merchant labels of the merchants to be recommended as secondary related merchants; and rough transaction characteristics of the to-be-recommended merchant, the first-level relevant merchant and the second-level relevant merchant are respectively obtained from the first relevant merchant map.

For example, the first related merchant map includes a first map element including map element 1, a second map element including map element 2, map element 3, map element 4, map element 5, map element 6, map element 7 and map element 8, map element 9 is used for representing rough transaction characteristics of the merchant to be recommended, map element 2, map element 3 and map element 4 are respectively used for representing rough transaction characteristics of a second predetermined related merchant having a first-level related merchant tag with the merchant to be recommended, and map element 5, map element 6, map element 7 and map element 8 are respectively used for representing rough transaction characteristics of a second predetermined related merchant having a second-level related merchant tag with the merchant to be recommended. According to a correlation inference path taking the map element 1 as an inference base point, traversing second map elements in the first related merchant map one by one to obtain a first to-be-referred merchant having a first-level related merchant label with the to-be-referred merchant and a second to-be-referred related merchant having a second-level related merchant label with the to-be-referred merchant. Further, a part of the to-be-determined relevant merchants are sampled from the first to-be-recommended relevant merchants to serve as first-stage relevant merchants having first-stage relevant merchant labels with the to-be-recommended merchants, for example, the first-stage relevant merchants include first to-be-determined relevant merchants corresponding to map elements 2 and 3, respectively, where the first to-be-determined relevant merchants corresponding to map elements 4 do not satisfy the sampling query condition. And sampling and querying part of the undetermined related merchants from the second undetermined related merchants to serve as second-level related merchants with secondary related merchant labels for the merchants to be recommended, wherein the second-level related merchants comprise second undetermined related merchants respectively corresponding to the map elements 5, 6, 7 and 8, and the second undetermined related merchants respectively corresponding to the map elements 5, 6, 7 and 8 meet sampling and querying conditions. The rough transaction characteristics of the undetermined related merchants corresponding to the first-level related merchants in the first related merchant map are determined as the rough transaction characteristics of the first-level related merchants, the rough transaction characteristics of the undetermined related merchants corresponding to the second-level related merchants in the first related merchant map are determined as the rough transaction characteristics of the second-level related merchants, and the rough transaction characteristics of the merchants to be recommended are obtained from the first related merchant map. By performing sampling query on the to-be-recommended merchants with the related labels, the problem of resource waste caused by excessive number of the to-be-recommended merchants can be solved, and the problem of low accuracy of target transaction characteristics of the to-be-recommended merchants caused by unbalanced distribution of the to-be-recommended merchants can be solved.

Optionally, the rough transaction feature includes a transaction feature vector corresponding to a first quantity of transaction information, and the performing, according to a correlation inference path using the first map element as an inference base point, a sampling query on second map elements in the first related merchant map one by one to obtain a first-stage related merchant having a first-stage related merchant tag with the merchant to be recommended, and a second-stage related merchant having a second-stage related merchant tag with the merchant to be recommended, includes: according to a correlation inference path which takes the first map element as an inference base point, sampling and inquiring second map elements in the first related merchant map one by one to obtain a second number of first to-be-related merchants which have first-level related merchant labels with the to-be-recommended merchants and a third number of second to-be-related merchants which have second-level related merchant labels with the to-be-recommended merchants; acquiring the number of first merchants of the first to-be-correlated merchants, in which the transaction feature vectors corresponding to the first number of transaction information in the second number of first to-be-correlated merchants do not meet a preset index; obtaining the number of second commercial tenants of second undetermined related commercial tenants, in which the transaction feature vectors corresponding to the first number of transaction information do not meet preset indexes, in the third number of second undetermined related commercial tenants; if the number of the first merchants is smaller than a first calibration number, determining the second number of first to-be-calibrated related merchants as first-level related merchants with first-level related merchant labels for the to-be-recommended merchants; and if the number of the second merchants is smaller than a second calibrated number, determining the third number of second pending related merchants as the first-level related merchants with the first-level related merchant labels of the to-be-recommended merchants. The second number and the third number may be the same or different, for example, the second number may be greater than or equal to the third number, or the second number may be less than or equal to the third number.

The rough transaction characteristics comprise transaction characteristic vectors corresponding to a first quantity of transaction information, the transaction information is used for representing one type of commodity transaction data or transaction information of one commodity transaction data, the transaction characteristic vectors can meet preset indexes or can not meet the preset indexes, the transaction information meeting the preset indexes is used for representing that a merchant to be recommended has the transaction information, and the transaction information not meeting the preset indexes is used for representing that the merchant to be recommended does not have the transaction information. For example, the meeting preset index may be that the characteristic value corresponding to the transaction information is 1, the not meeting preset index is that the characteristic value corresponding to the transaction information is 0, the transaction information is the transaction information of a user purchasing a certain type of product, if the characteristic value of the transaction characteristic of the merchant to be recommended for the user purchasing a certain type of product is 1, it indicates that the merchant to be recommended is interested in the user purchasing a certain type of product, that is, the merchant to be recommended has the transaction information of the user purchasing a certain type of product; if the characteristic value of the transaction characteristic of the merchant to be recommended for the user to purchase a certain product is 0, the merchant to be recommended is not interested in the user to purchase the certain product, namely the merchant to be recommended does not have the transaction information of the user to purchase the certain product. The transaction characteristic vectors corresponding to the first quantity of transaction information in the rough transaction characteristics of the merchants do not meet the preset index, and the merchants can be called as merchants without transaction information.

In order to avoid imbalance of various transaction information in first-level relevant merchants and second-level relevant merchants of merchants to be recommended, the server may perform sampling query on second map elements in the first-level relevant merchant map one by one according to a relevance inference path taking the first map element as an inference base point to obtain a second number of first to-be-recommended relevant merchants having first-level relevant merchant labels with the merchants to be recommended and a third number of second to-be-determined relevant merchants having second-level relevant merchant labels with the merchants to be recommended, that is, a limited number of first-level relevant merchants and second-level relevant merchants are obtained from the first relevant merchant map, so that waste of resources due to an excessive number of merchants in the first-level relevant merchants or the second-level relevant merchants can be avoided. Further, the number of first merchants of the first to-be-determined relevant merchants, for which the transaction feature vectors corresponding to the first quantity of transaction information do not satisfy the preset index, in the second number of first to-be-determined relevant merchants may be obtained, and the number of second merchants of the second to-be-determined relevant merchants, for which the transaction feature vectors corresponding to the first quantity of transaction information do not satisfy the preset index, in the third number of second to-be-determined relevant merchants may be obtained; that is, the first number of merchants is used to represent the number of merchants without some transaction information in the second number of first to-be-correlated merchants, and the second number of merchants is used to represent the number of merchants without some transaction information in the third number of second to-be-correlated merchants.

S102, performing first-level feature correlation selection operation on the rough transaction feature of the merchant to be recommended and the rough transaction feature of the merchant related to the first level to obtain the first-level transaction feature of the merchant to be recommended.

In the application, a first-level feature correlation selection operation can be executed on the rough transaction feature of the merchant to be recommended and the rough transaction feature of the first-level related merchant to obtain a first-level transaction feature of the merchant to be recommended; the first-level feature correlation selection operation refers to mining implicit transaction features of the merchants to be recommended based on the rough transaction features of the merchants to be recommended and the rough transaction features of the first-level relevant merchants; namely, the primary transaction characteristics are used for representing the explicit transaction characteristics and the implicit transaction characteristics of the merchant to be recommended and the merchant attribute information of the merchant to be recommended.

Optionally, step S102 may include: the method comprises the steps of performing feature selection operation on rough transaction features of at least two first-level related merchants through a first-level correlation feature selection branch of a transaction feature extraction model to obtain first correlation features, determining first correlation parameters aiming at the at least two first-level related merchants based on the first correlation features, and determining the first-level transaction features of the merchants to be recommended based on the first correlation parameters and the rough transaction features of the merchants to be recommended.

The rough transaction characteristics of at least two first-level related merchants are subjected to weighted characteristic selection operation through the relevant characteristic selection branch of the first level of the transaction characteristic extraction model based on the correlation degree between the merchant to be recommended and each first-level related merchant respectively, so that first relevant characteristics are obtained. Further, a first correlation parameter for at least two primary correlated merchants is determined based on the first correlation characteristic.

Optionally, the primary transaction characteristics may be determined by any one or more of the following in combination:

1) The first correlation parameter comprises first correlation information, first invisible transaction information of the merchant to be recommended is mined based on the first correlation information and the rough transaction characteristics of the merchant to be recommended, and a transaction characteristic vector corresponding to the first invisible transaction information in the rough transaction characteristics of the merchant to be recommended is adjusted to meet a preset index, so that first-level transaction characteristics are obtained. For example, the first correlation information indicates that merchant attribute information indicating that the shipping address is a poval and that the operation product is footwear is correlated with transaction information corresponding to a footwear transaction, that is, that most of the merchants having the goods address of the poval and the operation product is footwear have transaction information corresponding to a cost-effective transaction. If the delivery address of the merchant to be recommended is determined to be a Putian and the operation commodity is the shoes based on the rough transaction characteristics of the merchant to be recommended, the transaction information corresponding to the high-cost-performance transaction is determined to be first invisible transaction information of the merchant to be recommended, the transaction characteristic vector corresponding to the first invisible transaction information in the rough transaction characteristics of the merchant to be recommended is adjusted to meet a preset index, and first-level transaction characteristics of the merchant to be recommended are obtained.

2) The first correlation parameter comprises first transaction related information, second implicit transaction information of the merchant to be recommended is mined based on the first transaction related information and the rough transaction characteristics of the merchant to be recommended, and a transaction characteristic vector corresponding to the second implicit transaction information in the rough transaction characteristics of the merchant to be recommended is adjusted to meet a preset index, so that first-level transaction characteristics are obtained. For example, the first transaction-related information represents that the transaction information corresponding to the high cost performance transaction is associated with the transaction information corresponding to the low unit price transaction, that is, most of the first-level related merchants having the transaction information corresponding to the high cost performance transaction also have the transaction information corresponding to the low unit price transaction. If the to-be-recommended merchant has the transaction information corresponding to the high cost performance transaction based on the rough transaction characteristics of the to-be-recommended merchant, determining the transaction information corresponding to the low-price transaction as second implicit transaction information of the to-be-recommended merchant, and adjusting a transaction characteristic vector corresponding to the second implicit transaction information in the rough transaction characteristics of the to-be-recommended merchant to meet a preset index to obtain first-level transaction characteristics.

S103, performing second-level feature correlation selection operation on the rough transaction feature of the merchant to be recommended, the rough transaction feature of the second-level related merchant and the rough transaction feature of the first-level related merchant to obtain the second-level transaction feature of the merchant to be recommended.

In the application, a second-level feature correlation selection operation can be executed on the rough transaction feature of the merchant to be recommended, the rough transaction feature of the second-level relevant merchant and the rough transaction feature of the first-level relevant merchant to obtain a second-level transaction feature of the merchant to be recommended, wherein the second-level feature correlation selection operation is to excavate implicit transaction features of the merchant to be recommended based on the rough transaction feature of the merchant to be recommended, the rough transaction feature of the first-level relevant merchant and the rough transaction feature of the second-level relevant merchant; namely, the secondary transaction characteristics are used for representing the dominant transaction characteristics and the recessive transaction characteristics of the merchant to be recommended and the attribute information of the merchant to be recommended. The implicit transaction information of the merchant can be mined through the feature correlation selection operation of the second level, and the accuracy of obtaining the target transaction features of the merchant to be recommended is improved.

Optionally, the server may obtain the secondary transaction characteristics of the merchant to be recommended by any one of the following two examples:

example one: performing feature selection operation on the rough transaction features of the to-be-recommended merchant and the rough transaction features of the second-level relevant merchants through a second-level relevant feature selection branch of the transaction feature extraction model to obtain second relevant features, and determining second relevant parameters between the to-be-recommended merchant and the second-level relevant merchants based on the second relevant features; and determining the secondary transaction characteristics of the merchant to be recommended based on the second correlation parameters and the rough transaction characteristics of the primary relevant merchant.

In example one: and performing feature selection operation on the rough transaction features of the merchant to be recommended and the rough transaction features of the second-level relevant merchants based on the correlation degree between the merchant to be recommended and the second-level relevant merchants through the correlation feature selection branch of the second level of the transaction feature extraction model to obtain second relevant features. Further, a second correlation parameter between the to-be-recommended merchant and the second-level correlation merchant is determined based on the second correlation characteristic, and the second correlation parameter comprises second correlation information and/or second transaction related information. The second related relation information is used for representing the related relation between the transaction information of the second-level related merchants and the merchants to be recommended and the merchant attribute information, and the second transaction related information is used for representing the related relation between the transaction information.

And S104, generating target transaction characteristics of the merchant to be recommended based on the primary transaction characteristics and the secondary transaction characteristics.

In the embodiment of the application, the first-level transaction characteristics are used for representing merchant attribute information of a merchant to be recommended, the explicit transaction characteristics and the first implicit transaction characteristics of the merchant to be recommended, the second-level transaction characteristics are used for representing merchant attribute information of the merchant to be recommended, the explicit transaction characteristics and the second implicit transaction characteristics of the merchant to be recommended, or the second-level transaction characteristics are used for representing merchant attribute information of the merchant to be recommended, the explicit transaction characteristics, the second implicit transaction characteristics and the third implicit transaction characteristics of the merchant to be recommended. Further, the first-level transaction characteristics and the second-level transaction characteristics can be subjected to characteristic fusion and other processing to obtain target transaction characteristics of the to-be-recommended merchant, and the target transaction characteristics can represent rich transaction characteristics of the to-be-recommended merchant, so that user data can be recommended for the merchant based on the target transaction characteristics, and accuracy of the recommended user data is improved.

Alternatively, step S104 may include: performing feature selection operation on the secondary transaction features respectively corresponding to at least two secondary related merchants through a target correlation feature selection branch of a transaction feature extraction model to obtain third correlation features; and performing characteristic fusion operation on the first-level transaction characteristic and the third relevant characteristic to obtain a target transaction characteristic of the merchant to be recommended.

The method comprises the steps of executing feature selection operation on at least two secondary transaction features respectively corresponding to two secondary related merchants through a target correlation feature selection branch of a transaction feature extraction model to obtain a third correlation feature, and executing feature fusion operation on the primary transaction feature and the third correlation feature to obtain a fused transaction feature. The feature fusion operation may refer to adding a feature value in the primary transaction feature to the third relevant feature to obtain a transaction feature vector after the feature fusion operation.

S105, determining the target user type of the merchant to be recommended based on the target transaction characteristics, and recommending user data to the merchant to be recommended according to the target user type.

Optionally, step S105 may specifically include the following steps:

1) And acquiring a plurality of feature characteristics of the target transaction features and a user type matrix of the target transaction features, wherein the plurality of feature characteristics of the target transaction features are obtained by performing traversal retrieval on the plurality of feature characteristics of the target transaction features and feature characteristics in a feature characteristic library, and the user type matrix of the target transaction features is used for indicating user type probability distribution of the target transaction features.

2) Determining an inefficiency characteristic representation corresponding to the user type probability distribution of the target transaction characteristic based on the user type matrix and the inefficiency characteristic set of the target transaction characteristic, wherein the inefficiency characteristic representation corresponding to the user type probability distribution of the target transaction characteristic comprises a characteristic representation with the user type probability distribution influence of the target transaction characteristic lower than a threshold value, and the inefficiency characteristic set comprises a degree of matching between the user type probability distribution of the target transaction characteristic and the inefficiency characteristic corresponding to the user type probability distribution of the target transaction characteristic.

3) And removing the inefficient characteristic representation corresponding to the user type probability distribution of the target transaction characteristic from the plurality of characteristic representations of the target transaction characteristic to obtain the target characteristic representation of the target transaction characteristic.

4) And determining a target user type based on the target characteristic characterization of the target transaction characteristic and a user type set, wherein the user type set comprises a plurality of user types.

Optionally, the feature characterization library may be obtained by clustering feature characterizations of the sample transaction features, which is not limited in this application.

In one possible implementation, a user type matrix for the target transaction characteristics may be determined based on the target transaction characteristics and a preconfigured classification algorithm.

Optionally, a plurality of feature characterizations of each sample transaction feature in the sample transaction feature set, a user type matrix of each sample transaction feature, and a first sample transaction feature set may be obtained, where the plurality of feature characterizations of each sample transaction feature are obtained by performing traversal search on the plurality of feature characterizations of each sample transaction feature and the feature characterizations in the feature characterization library, the user type matrix of each sample transaction feature is used to indicate a user type probability distribution of each sample transaction feature, and the first sample transaction feature set includes sample transaction features of the same cluster in the sample transaction feature set subjected to clustering processing; generating an inefficiency feature set based on the plurality of feature characterizations of each sample transaction feature, the user type matrix of each sample transaction feature, and the first sample transaction feature set, the inefficiency feature set comprising a degree of match between a user type probability distribution of each sample transaction feature and an inefficiency feature characterization corresponding to the user type probability distribution of each sample transaction feature, the inefficiency feature characterization corresponding to the user type probability distribution of each sample transaction feature comprising a feature characterization having a user type probability distribution impact of less than a threshold value with respect to the each sample transaction feature.

Specifically, the inefficiency feature set may be generated based on the plurality of feature characterizations of each sample transaction feature, the user type matrix of each sample transaction feature, and the first sample transaction feature set, and an association index between a plurality of user type probability distributions of the sample transaction feature set and a plurality of feature characterizations of the sample transaction feature set may be determined based on the plurality of feature characterizations corresponding to each sample transaction feature, the user type matrix of each sample transaction feature, and the first sample transaction feature set, the plurality of user type probability distributions of the sample transaction feature set including the user type probability distribution of each sample transaction feature, the plurality of feature characterizations of the sample transaction feature set including the plurality of feature characterizations of each sample transaction feature; and generating the inefficient feature set based on the association index between the probability distribution of the multiple user types of the sample transaction feature set and the feature characterization of the sample transaction feature set.

In an example, an association indicator between a first user type probability distribution and a first feature characterization may be determined based on a support degree that a first condition is satisfied, a support degree that a second condition is satisfied, and a support degree that the first condition is satisfied at the same time as the second condition, the plurality of user type probability distributions of the sample transaction feature set including the first user type probability distribution, the plurality of feature characterizations of the sample transaction feature set including the first feature characterization, the first condition representing that the plurality of feature characterizations of the first sample transaction feature of the sample transaction feature set and the plurality of feature characterizations of the second sample transaction feature of the sample transaction feature set each include the first feature characterization, the user type probability distribution of the first sample transaction feature being the first user type probability distribution, and the second condition representing that the first sample transaction feature and the second sample transaction feature belong to the same cluster.

In a possible implementation manner, at least one feature representation of the plurality of feature representations of the sample transaction feature set, which has the smallest numerical value of the associated index of each user type probability distribution, is used as the low-efficiency feature representation corresponding to the user type probability distribution.

Optionally, at least one of the plurality of feature characterizations of the sample transaction feature set, in which the associated index of the probability distribution of each user type is smaller than a first preset threshold, may be used as an inefficient feature characterization corresponding to the probability distribution of the user type.

Optionally, the user type set includes a plurality of user types and a target feature representation corresponding to each of the plurality of user types, where the target feature representation corresponding to each user type is obtained by removing an inefficient feature representation corresponding to the user type probability distribution of each user type from the plurality of feature representations corresponding to each user type.

In an example, the set of user types may be generated based on a plurality of feature characterizations for each of the plurality of sample transaction features, a user type matrix for the each sample transaction feature, and the set of inefficiency features.

Specifically, an inefficiency characteristic representation corresponding to the user type probability distribution of each sample transaction characteristic may be determined based on the user type matrix of each sample transaction characteristic and the inefficiency characteristic set, the inefficiency characteristic representation corresponding to the user type probability distribution of each sample transaction characteristic is removed from the plurality of characteristic representations of each sample transaction characteristic, a target characteristic representation of each sample transaction characteristic is obtained, and the target characteristic representation of each sample transaction characteristic is added to the user type set.

In one possible embodiment, a matching index between a plurality of feature characterizations of the target transaction feature and a plurality of feature characterizations of user types in the user type set may be calculated, and at least one user type having a highest matching index with the target transaction feature may be determined as the target user type.

In another possible implementation, a matching index between a plurality of feature characterizations of the target transaction feature and a plurality of feature characterizations of user types in the user type set may be calculated, and at least one user type with a matching index greater than a preset matching threshold may be determined as a target user type.

Based on the above scheme, the target feature characterization of the target transaction feature is obtained by removing an inefficient feature characterization corresponding to the user type probability distribution of the target transaction feature from the plurality of feature characterizations of the target transaction feature, that is, a feature characterization with low energy efficiency for determining the target user type is removed from the plurality of feature characterizations of the target transaction feature, that is, the target feature characterization of the target transaction feature has a significant effect on determining the target transaction feature. Therefore, the target characteristic representation of the target transaction characteristic and the user type set are jointly determined, and the determination efficiency and accuracy of the target user type corresponding to the merchant to be recommended are improved.

In the application, the rough transaction characteristics of the merchant to be recommended, the primary relevant merchant and the secondary relevant merchant are respectively obtained, and the rough transaction characteristics are used for representing merchant attribute information and dominant transaction information of the merchant. Further, the rough transaction characteristics of the to-be-recommended merchant and the rough transaction characteristics of the first-level related merchant can be subjected to first-level feature correlation selection operation to obtain first-level transaction characteristics, and the rough transaction characteristics of the to-be-recommended merchant, the rough transaction characteristics of the first-level related merchant and the rough transaction characteristics of the second-level related merchant are subjected to second-level feature correlation selection operation to obtain second-level transaction characteristics. The implicit transaction information of the merchant to be recommended can be mined through the feature correlation selection operation of the first level and the feature correlation selection operation of the second level, that is, the first-level transaction characteristics and the second-level transaction characteristics can represent merchant attribute information and explicit transaction information of the merchant to be recommended and can also represent the implicit transaction information of the merchant to be recommended. The target transaction characteristics of the to-be-recommended merchant are generated based on the primary transaction characteristics and the secondary transaction characteristics, and the target transaction characteristics can represent rich transaction information of the to-be-recommended merchant, so that the recommendation accuracy of user data can be improved by recommending the user data for the to-be-recommended merchant based on the target transaction characteristics.

Further, the method may comprise the steps of:

d1 The server respectively acquires pre-labeled transaction characteristics of the first merchant example, the first-level related merchant example and the second-level related merchant example aiming at the commodity transaction data; the first and second related business examples have a second related business label therebetween.

In the embodiment of the application, the server may obtain the pre-labeled transaction characteristics of the first merchant example, the first-level related merchant example, and the second-level related merchant example for the commodity transaction data, where the pre-labeled transaction characteristics of the first merchant example, the first-level related merchant example, and the second-level related merchant example for the commodity transaction data may be manually labeled and verified in advance, and the pre-labeled transaction characteristics are used to represent the real transaction information of the merchant.

Optionally, D1) above includes: acquiring a second relevant merchant map Guan Shanghu, wherein the second relevant merchant map comprises map elements used for representing pre-labeled transaction characteristics of pending relevant merchant examples and a correlation inference line obtained by performing correlation inference on the map elements corresponding to the pending relevant merchant examples with relevant merchant labels, and the pre-labeled transaction characteristics comprise transaction characteristic vectors corresponding to a first amount of transaction information; according to the correlation reasoning path in the second related merchant map, sampling query is performed on map elements in the second related merchant map one by one to obtain a fourth number of target undetermined related merchant examples corresponding to the first number of transaction information respectively; a target undetermined related merchant example corresponding to one transaction information is an undetermined related merchant example in which the transaction characteristic vector corresponding to the transaction information meets a preset index in the second related merchant map; determining a first commercial tenant example, a first-level related commercial tenant example and a second-level related commercial tenant example based on a fourth number of target undetermined related commercial tenant examples and the second-level Guan Shanghu atlases corresponding to the first number of transaction information respectively; and obtaining the pre-labeled transaction characteristics of the first merchant example, the first-level relevant merchant example and the second-level relevant merchant example respectively aiming at the commodity transaction data from the second relevant merchant map.

For example, a second-phase Guan Shanghu map may be obtained, and the first merchant example, the first-level related merchant example, and the second-level related merchant example are obtained from the second related merchant map in a sampling query manner; the constant sampling query mode is to perform sampling query on map elements in the second relevant merchant map one by one according to the relevance reasoning path in the second relevant merchant map to obtain a fourth number of target undetermined relevant merchant examples respectively corresponding to the first number of transaction information; that is to say, a fourth quantity of target undetermined related business examples with the transaction information are sampled and inquired for each transaction information, so that the quantity of the business examples corresponding to each transaction information is uniform. A part of target undetermined related merchant examples can be selected from a fourth number of target undetermined related merchant examples respectively corresponding to the first number of transaction information as a first merchant example, and a part of the target undetermined related merchant examples as a verified merchant example, and then, a first-level related merchant example having a first-level related merchant tag with the first merchant example and a second-level related merchant example having a second-level related merchant tag with the first merchant example are determined from the second related merchant map. Further, the first business example, the first-level relevant business example and the second-level relevant business example are respectively obtained from the second relevant business map according to the pre-labeled transaction characteristics of the commodity transaction data. By performing sampling query on the first relevant merchant map, the number of merchants corresponding to each transaction information is uniform, and the generalization capability of the transaction characteristic extraction model is improved, namely the inductive learning capability of the transaction characteristic extraction model is improved.

A merchant example set and a verified merchant example set may be divided according to a scale, a pending related merchant example in the merchant example set is used for performing training on the initial transaction feature extraction model, and the pending related merchant example in the merchant example set may refer to the first merchant example; the pending related merchant paradigm in the example set of validated merchants validates the training effect of the initial transaction feature extraction model, and the pending related merchant paradigm in the example set of validated merchants may be referred to as the validated merchant paradigm below.

Optionally, the determining a first business example, a first-level relevant business example, and a second-level relevant business example based on a fourth number of target pending relevant business examples and the second-level Guan Shanghu atlases respectively corresponding to the first number of transaction information includes: determining a first-stage undetermined related merchant example with a first-stage related merchant label from the second related merchant map; the target pending related commercial tenant models belong to a fourth number of target pending related commercial tenant models corresponding to the first number of transaction information respectively; determining a second undetermined related merchant example with a second related merchant label from the second related merchant map; taking the target undetermined related business example as a first business example, and sampling and inquiring a fifth number of primary undetermined related business examples from the primary undetermined related business example to be taken as the primary related business example; and sampling and inquiring a sixth number of secondary undetermined related business examples from the secondary undetermined related business examples to serve as the secondary related business examples.

A first undetermined related merchant example with a first related merchant label relative to a target undetermined related merchant example can be determined from the second related merchant map, and a second undetermined related merchant example with a second related merchant label relative to the target undetermined related merchant example is determined from the second related merchant map; a fifth number of primary undetermined related business examples can be sampled and inquired from the primary undetermined related business examples and used as the primary related business examples; sampling and inquiring a sixth number of secondary undetermined related merchant examples from the secondary undetermined related merchant examples to serve as secondary related merchant examples; the fifth number may be greater than or equal to the sixth number, or the fifth number may also be smaller than the sixth number, and the fifth number and the sixth number may be determined by the performance of the server, and the target pending relevant business example is taken as the first business example. Through sampling query, the number of the merchants respectively corresponding to the primary undetermined related merchant example and the secondary undetermined related merchant example of the target undetermined related merchant example is not excessive, and the problem of insufficient memory in the training process of the initial transaction characteristic extraction model due to the super merchant example is avoided.

The number of the merchants with the transaction information in the first-level related merchant example is larger than that of the merchants without the transaction information, and the number of the merchants with the transaction information in the second-level related merchant example is larger than that of the merchants without the transaction information; the merchant example has transaction information, namely transaction characteristic vectors of pre-labeled transaction characteristics of the merchant example partially meet preset indexes, or the merchant example has transaction information, namely transaction characteristic vectors of pre-labeled transaction characteristics of the merchant example all meet the preset indexes, and the merchant example does not have the transaction information, namely transaction characteristic vectors of pre-labeled transaction characteristics of the merchant example do not meet the preset indexes, so that the condition that the merchant example without the transaction information is completely dominated in the training process of an initial transaction characteristic extraction model is avoided, and the capability of mining implicit transaction information of the first merchant example by means of pre-labeled transaction characteristics of other related merchants is lost.

That is to say, in the training and verification process, in order to improve the training efficiency of the initial transaction feature extraction model, all the primary undetermined related merchant examples and the secondary undetermined related merchant examples of the first merchant example are not input into the initial transaction feature extraction model, but the sampling query is also performed in advance on the primary undetermined related merchant examples and the second undetermined related merchant examples. In the sampling query process, the maximum number of the first-level related business examples and the maximum number of the second-level related business examples of each first business example are set, so that the problem of insufficient memory in the training process caused by the super map elements (namely, too many related business examples of the first business examples) is solved. Meanwhile, in the first-stage related merchant example and the second-stage related merchant example of each merchant example, the number of merchants of the related merchant examples with transaction information cannot be less than that of merchants of related merchant examples without transaction labels, so that the merchant dominance of the merchant without transaction labels in the training process of the initial transaction feature extraction model is avoided, and the capability of mining the implicit transaction information of the implicit first merchant example by means of the pre-labeled transaction features of the related merchant examples is lost.

D2 The server performs feature interference operations on the pre-labeled transaction features of the first business instance, the first-level related business instance, and the second-level related business instance, respectively, through an initial transaction feature extraction model, so as to obtain pre-labeled transaction features after the feature interference operations corresponding to the first business instance, the first-level related business instance, and the second-level related business instance, respectively.

In the embodiment of the present application, a feature interference operation may be respectively performed on the pre-labeled transaction feature of the first merchant example, the pre-labeled transaction feature of the second-level related merchant example, and the pre-labeled transaction feature of the first-level related merchant example, so as to obtain the pre-labeled transaction feature after the feature interference operation corresponding to the first merchant example, the pre-labeled transaction feature after the feature interference operation corresponding to the second-level related merchant example, and the pre-labeled transaction feature after the feature interference operation corresponding to the first-level related merchant example. The feature interference operation may be to perform iteration on the transaction feature vector in the pre-labeled transaction features, and the feature interference operation may be to introduce noise into the pre-labeled transaction features of the first merchant example and the pre-labeled transaction features of the second-level related merchant example and the pre-labeled transaction features of the first-level related merchant example by a noise introducing module in an initial transaction feature extraction model. The feature interference operation is executed through the pre-labeled transaction features, so that the generalization capability of the first example features of the first merchant example is improved, namely the generalization capability of the transaction feature extraction model is improved.

Optionally, D2) may include: the server randomly changes the transaction characteristic vector in the pre-labeled transaction characteristics of the first merchant example through an initial transaction characteristic extraction model to obtain the pre-labeled transaction characteristics after the characteristic interference operation corresponding to the first merchant example; adjusting the transaction characteristic vector in the pre-labeled transaction characteristics of the first-level related merchant example to obtain pre-labeled transaction characteristics after characteristic interference operation corresponding to the first-level related merchant example; and adjusting the transaction characteristic vector in the pre-labeled transaction characteristics of the second-level related merchant example to obtain the pre-labeled transaction characteristics after the characteristic interference operation corresponding to the second-level related merchant example.

The method comprises the steps of randomly changing transaction information in pre-labeled transaction characteristics of a first merchant example to meet a non-preset index according to a target confidence coefficient through an encoder of an initial transaction characteristic extraction model to obtain pre-labeled transaction characteristics after feature interference operation corresponding to the first merchant example, namely randomly discarding real transaction information of the first merchant example according to the target confidence coefficient, namely, randomly introducing noise into the pre-labeled transaction characteristics of the first merchant example according to the target confidence coefficient. By introducing noise into the pre-labeled transaction features, the generalization capability of the first example features of the first merchant example is improved, namely the generalization capability of the transaction feature extraction model is improved.

D3 Performing correlation detection on the pre-labeled transaction features after the feature interference operation corresponding to the first merchant example, the pre-labeled transaction features after the feature interference operation corresponding to the second-level related merchant example, and the pre-labeled transaction features after the feature interference operation corresponding to the first-level related merchant example to obtain first example features of the first merchant example.

In the application, the server may perform first-level correlation detection on the pre-labeled transaction characteristics after the characteristic interference operation corresponding to the first merchant example and the pre-labeled transaction characteristics after the characteristic interference operation corresponding to the first-level related merchant example to obtain first correlation detection characteristics; and performing second-level correlation detection on the pre-labeled transaction characteristics after the characteristic interference operation corresponding to the first merchant example, the pre-labeled transaction characteristics after the characteristic interference operation corresponding to the first-level related merchant example and the pre-labeled transaction characteristics after the characteristic interference operation corresponding to the second-level related merchant example to obtain second correlation detection characteristics, and determining the first example characteristics of the first merchant example based on the first correlation detection characteristics and the second correlation detection characteristics.

It can be understood that the first-level correlation detection here has a similar meaning to the above-mentioned feature correlation selection operation of the first level, that is, the first-level correlation detection is to mine the implicit transaction features of the first merchant example based on the pre-labeled transaction features after the feature interference operation corresponding to the first merchant example and the pre-labeled transaction features after the feature interference operation corresponding to the first-level relevant merchant example. Similarly, the second-level correlation detection has a similar meaning to the second-level feature correlation selection operation, that is, the second-level correlation detection is to mine the implicit transaction features of the first business instance based on the pre-labeled transaction features after the feature interference operation corresponding to the first business instance, the pre-labeled transaction features after the feature interference operation corresponding to the first-level relevant business instance, and the pre-labeled transaction features after the feature interference operation corresponding to the second-level relevant business instance.

D4 Performing an iteration on the initial transaction feature extraction model based on the pre-labeled transaction features of the first merchant instance and the first instance features to obtain the transaction feature extraction model.

In the application, a euclidean distance between the first example feature and the pre-labeled transaction feature of the first merchant example may be obtained, and a transaction feature extraction loss value of the initial transaction feature extraction model is determined based on the euclidean distance, where the transaction feature extraction loss value is used to characterize the extraction accuracy of the transaction feature of the initial transaction feature extraction model. That is to say, the larger the euclidean distance is, the larger the difference between the first example feature and the pre-labeled transaction feature of the first merchant example is, that is, the higher the training cost value is, that is, the lower the accuracy of extracting the transaction feature of the initial transaction feature extraction model is; conversely, the smaller the euclidean distance is, the smaller the difference between the first example feature and the pre-labeled transaction feature of the first merchant example is, that is, the smaller the loss value of the transaction feature extraction is, that is, the higher the accuracy of the extraction of the transaction feature of the initial transaction feature extraction model is. Therefore, iteration can be performed on the initial transaction feature extraction model based on the transaction feature extraction loss value to obtain the transaction feature extraction model, and the accuracy of the transaction feature extraction model can be improved.

Optionally, the server may perform an iteration on the initial transaction feature extraction model based on the pre-labeled transaction features of the first merchant example and the first example features to obtain the transaction feature extraction model by any one of the following two examples:

example one: and performing iteration on the initial transaction feature extraction model based on the pre-labeled transaction features of the first merchant example and the first example features to obtain an iterated initial transaction feature extraction model, counting the iteration number of the initial transaction feature extraction model, and determining the iterated initial transaction feature extraction model as the transaction feature extraction model if the iteration number is greater than a threshold value.

In an example one, the server may obtain a euclidean distance between the pre-labeled transaction feature of the first merchant example and the first example feature, determine a transaction feature extraction loss value of the initial transaction feature extraction model based on the euclidean distance, and perform iteration on the initial transaction feature extraction model based on the transaction feature extraction loss value to obtain an iterated initial transaction feature extraction model. Here, performing iteration on the initial transaction feature extraction model based on the transaction feature extraction loss value may refer to performing iteration on a model variable value of the initial transaction feature extraction model, for example, determining an iteration step size based on the transaction feature extraction loss value, and performing iteration on the model variable value of the initial transaction feature extraction model based on the iteration step size, where the transaction feature extraction loss value is positively correlated with the iteration step size. Further, the number of iteration rounds of the initial transaction feature extraction model is recorded, and if the number of the iteration rounds is larger than a round threshold value, the initial transaction feature extraction model after iteration is determined as the transaction feature extraction model. If the iteration round number is less than or equal to the round threshold, executing a step of determining a verified merchant example group based on a fourth number of target undetermined related merchant examples respectively corresponding to the first number of transaction information and the second phase Guan Shanghu atlas in example two; by limiting the number of iteration rounds of the initial transaction feature extraction model, the resource waste caused by excessive number of iteration rounds of the initial transaction feature extraction model is avoided.

Example two: performing iteration on the initial transaction feature extraction model based on the pre-labeled transaction features of the first merchant example and the first example features to obtain an iterated initial transaction feature extraction model, and determining a verified merchant example group based on a fourth number of target undetermined related merchant examples and the second-phase Guan Shanghu maps corresponding to the first number of transaction information respectively; determining a training progress of the iterated initial transaction feature extraction model based on the validated merchant exemplar set; and determining the transaction characteristic extraction model based on the training progress and the initial transaction characteristic extraction model after iteration.

In example two, a euclidean distance between the pre-labeled transaction feature of the first merchant example and the first example feature may be obtained, a transaction feature extraction loss value of the initial transaction feature extraction model is determined based on the euclidean distance, and iteration is performed on the initial transaction feature extraction model to obtain an iterated initial transaction feature extraction model. Then, a target undetermined related merchant example is randomly selected from a fourth number of target undetermined related merchant examples respectively corresponding to the first number of transaction information to serve as a verified merchant example, the verified merchant example is different from the first merchant example, a first-level verification related merchant example with a first-level related merchant label to the verified merchant example and a second-level verification related merchant example with a second-level related merchant label to the verified merchant example are determined from the second related merchant map, and the verified merchant example, the first-level verification related merchant example and the second-level verification related merchant example are determined to serve as a verified merchant example group. Then, a training schedule of the iterated initial transaction feature extraction model may be determined based on the verified merchant exemplar set, where the training schedule is used to characterize whether the transaction feature extraction capability of the iterated initial transaction feature extraction model is optimal. Accordingly, the server may determine the transaction feature extraction model based on the training progress and the iterated initial transaction feature extraction model. The training progress of the iterated initial transaction feature extraction model is determined based on the verified merchant example group, namely verification of the transaction feature extraction accuracy of the iterated initial transaction feature extraction model based on the verified merchant example group is achieved, the transaction feature extraction accuracy of the transaction feature extraction model is improved, and the learning and induction capability of the transaction feature extraction model is improved.

Optionally, the verified merchant example group includes a verified merchant example, a first-level verified related merchant example, and a second-level verified related merchant example; the example of the verified merchant is different from the first merchant example, a first-level related merchant label is provided between the example of the verified merchant and the example of the first-level verified related merchant, and a second-level related merchant label is provided between the example of the verified merchant and the example of the second-level verified related merchant; the determining the training progress of the iterated initial transaction feature extraction model based on the verified merchant example set includes: and performing feature interference operation on the pre-labeled transaction features of the verification merchant example, the first-level verification-related merchant example and the second-level verification-related merchant example respectively through the iterated initial transaction feature extraction model to obtain the pre-labeled transaction features respectively processed correspondingly by the verification merchant example, the first-level verification-related merchant example and the second-level verification-related merchant example. And performing correlation detection on the pre-labeled transaction characteristics after the characteristic interference operation corresponding to the verification merchant example, the pre-labeled transaction characteristics after the characteristic interference operation corresponding to the primary verification-related merchant example and the pre-labeled transaction characteristics after the characteristic interference operation corresponding to the secondary verification-related merchant example through the iterated initial transaction characteristic extraction model to obtain first example characteristics of the verification merchant example. Determining a training cost value of the iterated initial transaction feature extraction model based on the first example features of the verified merchant example and the pre-labeled transaction features of the verified merchant example; and determining the training progress of the initial transaction feature extraction model after iteration based on the training cost value.

Performing characteristic interference operation on the pre-labeled transaction characteristics of the verification merchant paradigm through the initial transaction characteristic extraction model after iteration to obtain pre-labeled transaction characteristics after the characteristic interference operation corresponding to the verification merchant paradigm; performing characteristic interference operation on the pre-labeled transaction characteristics of the first-stage verification related merchant example to obtain pre-labeled transaction characteristics after the characteristic interference operation corresponding to the first-stage verification related merchant example; and executing characteristic interference operation on the pre-labeled transaction characteristics of the secondary verification related merchant example to obtain the pre-labeled transaction characteristics after the characteristic interference operation corresponding to the secondary verification related merchant example. Further, correlation detection is performed on the pre-labeled transaction characteristics after the characteristic interference operation corresponding to the verification merchant paradigm, the pre-labeled transaction characteristics after the characteristic interference operation corresponding to the primary verification-related merchant paradigm, and the pre-labeled transaction characteristics after the characteristic interference operation corresponding to the secondary verification-related merchant paradigm, so as to obtain the first paradigm characteristic of the verification merchant paradigm. Determining a training cost value of the iterated initial transaction feature extraction model based on the first example features of the verified merchant example and the pre-labeled transaction features of the verified merchant example; if the training cost value is smaller than a cost threshold value, determining that the training progress of the iterated initial transaction feature extraction model reaches the standard; and if the training cost value is greater than or equal to the cost threshold value, determining that the training progress of the iterated initial transaction feature extraction model meets the standard and does not meet the standard. By performing verification on the transaction feature extraction accuracy of the initial transaction feature extraction model after iteration based on the verification merchant paradigm, the transaction feature extraction accuracy of the transaction feature extraction model is improved, and the learning and induction capability of the transaction feature extraction model is improved.

Optionally, the first example feature of the verified merchant example and the pre-labeled transaction feature of the verified merchant example each include a transaction feature vector corresponding to a first amount of transaction information; the determining the training cost value of the iterative initial transaction feature extraction model based on the first example feature of the verified merchant example and the pre-labeled transaction feature of the verified merchant example includes: inputting the target confidence coefficient and a transaction feature vector corresponding to the Nth transaction information in the pre-labeled transaction features of the verification merchant example into an extraction cost evaluation function to obtain a feature extraction replacement value corresponding to the Nth transaction information; the target confidence degree is the confidence degree that the transaction feature vector corresponding to the Nth transaction information in the first example feature of the verification merchant example meets the preset index; n is a positive integer less than or equal to a first number; performing a homogenization operation on the feature extraction cost value corresponding to the Nth transaction information to obtain a feature extraction cost value after the homogenization operation corresponding to the Nth transaction information; and performing weighted summation on the feature extraction substitution values after the homogenization operation corresponding to the first quantity of transaction information respectively to obtain the training substitution value of the initial transaction feature extraction model after iteration. By calculating the training cost value corresponding to each transaction information, the accuracy of the transaction feature extraction model is improved, and by executing the homogenization operation on the training cost value corresponding to each transaction information, the problem that the accuracy of the transaction feature extraction model is low due to uneven distribution of the transaction information is avoided.

Optionally, the performing a normalization operation on the feature extraction cost value corresponding to the nth transaction information to obtain a feature extraction replacement value after the normalization operation corresponding to the nth transaction information includes: obtaining a third number of merchants of the merchant paradigm that transaction eigenvectors corresponding to the Nth transaction information in the verification merchant paradigm group meet preset indexes; and generating a homogenization coefficient based on the number of the third merchants, and executing homogenization operation on the feature extraction cost value corresponding to the Nth transaction information by adopting the homogenization coefficient to obtain a feature extraction cost value after the homogenization operation corresponding to the Nth transaction information. The number of the merchants meeting the preset index based on the transaction feature vector is determined to determine the homogenization coefficient, so that the situation that the number of the merchants with the transaction information is too large, the distribution of the merchants with the transaction information and the merchants without the transaction information is unbalanced, and the accuracy of a transaction feature extraction model is low is avoided.

Optionally, the determining the transaction feature extraction model based on the training progress and the iterated initial transaction feature extraction model includes: if the training progress of the iterated initial transaction feature extraction model does not reach the standard, iterating the iterated initial transaction feature extraction model based on the pre-labeled transaction feature of the first merchant example and the iterated first example feature to obtain the transaction feature extraction model, wherein the iterated first example feature is obtained by performing correlation detection on the pre-labeled transaction feature after the feature interference operation corresponding to the first merchant example and the pre-labeled transaction feature after the feature interference operation corresponding to the second-level related merchant example and the pre-labeled transaction feature after the feature interference operation corresponding to the first-level related merchant example through the iterated initial transaction feature extraction model; and if the training progress of the initial transaction feature extraction model after iteration reaches the standard, determining the initial transaction feature extraction model after iteration as the transaction feature extraction model.

If the training progress of the initial transaction feature extraction model after iteration does not reach the standard, it is indicated that the training cost value of the initial transaction feature extraction model after iteration does not reach the minimum, that is, it is indicated that the feature extraction accuracy of the initial transaction feature extraction model after iteration does not reach the optimum, therefore, the server may continue to iterate the initial transaction feature extraction model after iteration based on the pre-labeled transaction feature of the first merchant example and the first example feature after iteration until the training progress of the initial transaction feature extraction model after iteration reaches the standard, and obtain the transaction feature extraction model. If the training progress of the initial transaction feature extraction model after iteration reaches the standard, it is indicated that the training cost value of the initial transaction feature extraction model after iteration reaches the minimum, that is, the feature extraction accuracy of the initial transaction feature extraction model after iteration reaches the optimum, so that the server can determine the initial transaction feature extraction model after iteration as the transaction feature extraction model.

D5 Respectively obtaining rough transaction characteristics of the merchant to be recommended, the first-level related merchant and the second-level related merchant aiming at the commodity transaction data; and a first-stage related merchant label is arranged between the merchant to be recommended and the first-stage related merchant, and a second-stage related merchant label is arranged between the merchant to be recommended and the second-stage related merchant.

D6 The rough transaction characteristics of the merchant to be recommended and the rough transaction characteristics of the merchant related to the first level are subjected to a first-level characteristic correlation selection operation through a transaction characteristic extraction model, so that the first-level transaction characteristics of the merchant to be recommended are obtained.

D7 The rough transaction characteristics of the merchant to be recommended, the rough transaction characteristics of the second-level relevant merchant and the rough transaction characteristics of the first-level relevant merchant are subjected to second-level feature correlation selection operation through a transaction characteristic extraction model, so that the second-level transaction characteristics of the merchant to be recommended are obtained.

D8 Based on the primary transaction characteristics and the secondary transaction characteristics, generating target transaction characteristics of the merchant to be recommended, and recommending user data for the merchant to be recommended based on the target transaction characteristics.

In the embodiment of the application, rough transaction characteristics of the merchant to be recommended, the primary related merchant and the secondary related merchant can be respectively obtained, and the rough transaction characteristics are used for representing merchant attribute information and explicit transaction information of the merchant. Further, the rough transaction characteristics of the to-be-recommended merchant and the rough transaction characteristics of the first-level related merchant can be subjected to first-level feature correlation selection operation to obtain first-level transaction characteristics, and the rough transaction characteristics of the to-be-recommended merchant, the rough transaction characteristics of the first-level related merchant and the rough transaction characteristics of the second-level related merchant are subjected to second-level feature correlation selection operation to obtain second-level transaction characteristics. The implicit transaction information of the merchant to be recommended can be mined through the feature correlation selection operation of the first level and the feature correlation selection operation of the second level, that is, the primary transaction feature and the secondary transaction feature can represent basic merchant information and explicit transaction information of the merchant to be recommended and can also represent implicit transaction information of the merchant to be recommended. The target transaction characteristics of the to-be-recommended merchants are generated based on the primary transaction characteristics and the secondary transaction characteristics, and the target transaction characteristics can represent rich transaction information of the to-be-recommended merchants, so that the user data recommendation accuracy can be improved and accurate recommendation of the user data can be realized by recommending the user data for the to-be-recommended merchants based on the target transaction characteristics.

The embodiment of the present invention provides a server 100, where the server 100 includes a processor and a non-volatile memory storing computer instructions, and when the computer instructions are executed by the processor, the server 100 executes the aforementioned artificial intelligence-based user data recommendation method. As shown in fig. 2, fig. 2 is a block diagram of a server 100 according to an embodiment of the present invention. The server 100 includes a memory 111, a processor 112, and a communication unit 113. To facilitate the transfer or interaction of data, the elements of the memory 111, the processor 112 and the communication unit 113 are electrically connected to each other, directly or indirectly.

The foregoing description, for purpose of explanation, has been described with reference to specific embodiments. However, the illustrative discussions above are not intended to be exhaustive or to limit the disclosure to the precise forms disclosed. Many modifications and variations are possible in light of the above teaching. The embodiments were chosen and described in order to best explain the principles of the disclosure and its practical applications, to thereby enable others skilled in the art to best utilize the disclosure and various embodiments with various modifications as are suited to the particular use contemplated. The foregoing description, for purpose of explanation, has been described with reference to specific embodiments. However, the illustrative discussions above are not intended to be exhaustive or to limit the disclosure to the precise forms disclosed. Many modifications and variations are possible in light of the above teaching. The embodiments were chosen and described in order to best explain the principles of the disclosure and its practical applications, to thereby enable others skilled in the art to best utilize the disclosure and various embodiments with various modifications as are suited to the particular use contemplated.

Claims

1. A user data recommendation method based on artificial intelligence is applied to a server, and the method comprises the following steps:

respectively acquiring rough transaction characteristics of the merchant to be recommended, the first-level related merchant and the second-level related merchant aiming at the commodity transaction data; a first-level related merchant label is arranged between the merchant to be recommended and the first-level related merchant, and a second-level related merchant label is arranged between the merchant to be recommended and the second-level related merchant;

2. The artificial intelligence based user data recommendation method according to claim 1, wherein the obtaining of the rough transaction characteristics of the merchant to be recommended, the first-level related merchant, and the second-level related merchant for the commodity transaction data respectively comprises:

acquiring a first related merchant map corresponding to a merchant to be recommended, wherein the first related merchant map comprises a first map element used for representing rough transaction characteristics of the merchant to be recommended, a second map element used for representing rough transaction characteristics of the merchant to be recommended, and a correlation inference line obtained by performing correlation inference on the first map element and the second map element; the merchant to be recommended and the pending related merchant have related merchant labels;

according to a correlation inference path taking the first map element as an inference base point, sampling and inquiring second map elements in the first related merchant map one by one to obtain a first-stage related merchant having a first-stage related merchant label with the merchant to be recommended and a second-stage related merchant having a second-stage related merchant label with the merchant to be recommended;

rough transaction characteristics of the merchant to be recommended, the first-stage related merchant and the second-stage related merchant for commodity transaction data are respectively obtained from the first related merchant map;

the rough transaction characteristics comprise transaction characteristic vectors corresponding to a first quantity of transaction information; the step of performing sampling query on second map elements in the first related merchant map one by one according to a correlation inference path taking the first map element as an inference base point to obtain a first-level related merchant having a first-level related merchant label with the merchant to be recommended and a second-level related merchant having a second-level related merchant label with the merchant to be recommended includes:

according to a correlation reasoning path which takes the first map element as a reasoning base point, sampling and inquiring second map elements in the first correlation merchant map one by one to obtain a second number of first to-be-recommended correlation merchants having first-stage correlation merchant labels with the to-be-recommended merchants and a third number of second to-be-recommended correlation merchants having second-stage correlation merchant labels with the to-be-recommended merchants;

acquiring the number of first merchants of the first to-be-related merchants, wherein the number of the first merchants of the first to-be-related merchants does not meet a preset index by the transaction feature vectors corresponding to the first number of transaction information of the second to-be-related merchants; obtaining the number of second commercial tenants of second undetermined related commercial tenants, in which the transaction feature vectors corresponding to the first number of transaction information do not meet preset indexes, in the third number of second undetermined related commercial tenants;

if the number of the first merchants is smaller than a first calibration number, determining the second number of first to-be-recommended relevant merchants as first-level relevant merchants with first-level relevant merchant labels with the to-be-recommended merchants;

and if the number of the second merchants is smaller than a second calibrated number, determining the third number of second pending related merchants as the first-level related merchants with the first-level related merchant labels of the to-be-recommended merchants.

3. The artificial intelligence-based user data recommendation method according to claim 1, wherein the number of the first-level relevant merchants is greater than or equal to two, and the performing a first-level feature correlation selection operation on the rough transaction features of the to-be-recommended merchants and the rough transaction features of the first-level relevant merchants to obtain the first-level transaction features of the to-be-recommended merchants comprises:

performing feature selection operation on rough transaction features of at least two primary relevant merchants through a first-level relevance feature selection branch of a transaction feature extraction model to obtain first relevant features;

determining a first correlation parameter for at least two of the primary correlated merchants based on the first correlation feature;

determining a first-level transaction characteristic of the merchant to be recommended based on the first correlation parameter and the rough transaction characteristic of the merchant to be recommended;

the performing a second-level feature correlation selection operation on the rough transaction features of the to-be-recommended merchant, the rough transaction features of the second-level relevant merchant and the rough transaction features of the first-level relevant merchant to obtain the second-level transaction features of the to-be-recommended merchant includes:

performing feature selection operation on the rough transaction features of the to-be-recommended merchants and the rough transaction features of the second-level relevant merchants through a second-level relevant feature selection branch of the transaction feature extraction model to obtain second relevant features;

determining a second correlation parameter between the to-be-recommended merchant and the secondary correlation merchant based on the second correlation characteristic;

determining secondary transaction characteristics of the merchant to be recommended based on the second correlation parameters and the rough transaction characteristics of the primary relevant merchant;

the number of the second-level related merchants is greater than or equal to two, and the generating of the target transaction characteristics of the to-be-recommended merchants based on the first-level transaction characteristics and the second-level transaction characteristics comprises the following steps:

performing feature selection operation on the secondary transaction features respectively corresponding to at least two secondary related merchants through a target correlation feature selection branch of a transaction feature extraction model to obtain third correlation features;

and performing characteristic fusion operation on the first-level transaction characteristic and the third relevant characteristic to obtain a target transaction characteristic of the merchant to be recommended.

4. The artificial intelligence based user data recommendation method according to claim 1, wherein the determining a target user type of the merchant to be recommended based on the target transaction characteristics, and recommending user data to the merchant to be recommended according to the target user type includes:

acquiring a plurality of feature characterizations of a target transaction feature and a user type matrix of the target transaction feature, wherein the feature characterizations of the target transaction feature are obtained by traversing and retrieving the feature characterizations of the target transaction feature and feature characterizations in a feature characterization library, and the user type matrix of the target transaction feature is used for indicating the user type probability distribution of the target transaction feature;

determining an inefficiency characteristic representation corresponding to the user type probability distribution of the target transaction characteristic based on the user type matrix and the inefficiency characteristic set of the target transaction characteristic, wherein the inefficiency characteristic representation corresponding to the user type probability distribution of the target transaction characteristic comprises a characteristic representation with influence on the user type probability distribution of the target transaction characteristic lower than a threshold value, and the inefficiency characteristic set comprises a degree of matching between the user type probability distribution of the target transaction characteristic and the inefficiency characteristic representation corresponding to the user type probability distribution of the target transaction characteristic;

removing inefficient characteristic representations corresponding to the user type probability distribution of the target transaction characteristics from the plurality of characteristic representations of the target transaction characteristics to obtain target characteristic representations of the target transaction characteristics;

determining a target user type based on a target feature characterization of the target transaction feature and a user type set, wherein the user type set comprises a plurality of user types;

the user type set comprises the plurality of user types and the matching degree of a target feature representation corresponding to each user type in the plurality of user types, and the target feature representation corresponding to each user type is obtained by removing an inefficient feature representation corresponding to the user type probability distribution of each user type from the plurality of feature representations corresponding to each user type;

before the determining, based on the user type matrix and the set of inefficiency features of the target transaction feature, an inefficiency feature characterization corresponding to the user type probability distribution of the target transaction feature, the method further includes:

obtaining a plurality of feature representations of each sample transaction feature in a sample transaction feature set, a user type matrix of each sample transaction feature and a first sample transaction feature set, wherein the plurality of feature representations of each sample transaction feature are obtained by performing traversal retrieval on the plurality of feature representations of each sample transaction feature and feature representations in a feature representation library, the user type matrix of each sample transaction feature is used for indicating user type probability distribution of each sample transaction feature, and the first sample transaction feature set comprises sample transaction features of the same cluster in the sample transaction feature set which are subjected to clustering processing;

determining an indicator of association between a plurality of user type probability distributions of the sample transaction feature set including a user type probability distribution of the each sample transaction feature and a plurality of feature characterizations of the sample transaction feature set including a plurality of feature characterizations of the each sample transaction feature based on the plurality of feature characterizations of the each sample transaction feature, the user type matrix of the each sample transaction feature, and the first sample transaction feature set;

generating the inefficient feature set based on an association index between a plurality of user type probability distributions of the sample transaction feature set and a plurality of feature characterizations of the sample transaction feature set;

the determining an association indicator between a plurality of user type probability distributions of the sample transaction feature set and a plurality of feature characterizations of the sample transaction feature set based on the plurality of feature characterizations of each sample transaction feature, the user type matrix of each sample transaction feature, and the first sample transaction feature set comprises:

determining an association indicator between a first user type probability distribution and a first feature characterization based on a support degree that a first condition is satisfied, a support degree that a second condition is satisfied, and a support degree that the first condition is satisfied at the same time as the second condition, wherein a plurality of user type probability distributions of the sample transaction feature set include the first user type probability distribution, a plurality of feature characterizations of the sample transaction feature set include the first feature characterization, the first condition represents that a plurality of feature characterizations of a first sample transaction feature in the sample transaction feature set and a plurality of feature characterizations of a second sample transaction feature in the sample transaction feature set each include the first feature characterization, the user type probability distribution of the first sample transaction feature is the first user type probability distribution, and the second condition represents that the first sample transaction feature and the second sample transaction feature belong to the same cluster;

determining a target user type based on the target feature characterization of the target transaction feature and the set of user types, including:

determining a matching index between a target feature representation of the target transaction feature and a target feature representation of a user type in the set of user types;

and determining at least one user type with a matching index larger than a preset matching threshold value and the target characteristic representation of the target transaction characteristic as the target user type.

5. The artificial intelligence based user data recommendation method of claim 3, further comprising:

respectively acquiring pre-labeled transaction characteristics of a first merchant example, a first-level related merchant example and a second-level related merchant example aiming at commodity transaction data; the first business example and the first-level related business example have a first-level related business label therebetween, and the first business example and the second-level related business example have a second-level related business label therebetween;

performing feature interference operation on pre-labeled transaction features respectively corresponding to the first merchant example, the first-level related merchant example and the second-level related merchant example through an initial transaction feature extraction model to obtain pre-labeled transaction features respectively corresponding to the first merchant example, the first-level related merchant example and the second-level related merchant example after the feature interference operation;

performing correlation detection on the pre-labeled transaction characteristics after the characteristic interference operation corresponding to the first merchant example, the pre-labeled transaction characteristics after the characteristic interference operation corresponding to the second-level related merchant example, and the pre-labeled transaction characteristics after the characteristic interference operation corresponding to the first-level related merchant example to obtain first example characteristics of the first merchant example;

and performing iteration on the initial transaction feature extraction model based on the pre-labeled transaction features of the first merchant example and the first example features to obtain the transaction feature extraction model.

6. The artificial intelligence based user data recommendation method according to claim 5, wherein the performing, by an initial transaction feature extraction model, a feature interference operation on the pre-labeled transaction features respectively corresponding to the first business instance, the first-level related business instance, and the second-level related business instance to obtain pre-labeled transaction features respectively corresponding to the first business instance, the first-level related business instance, and the second-level related business instance after the feature interference operation comprises:

randomly changing a transaction characteristic vector in the pre-labeled transaction characteristics of the first merchant example through an initial transaction characteristic extraction model to obtain pre-labeled transaction characteristics after characteristic interference operation corresponding to the first merchant example;

randomly changing transaction feature vectors in the pre-labeled transaction features of the first-level related merchant example to obtain pre-labeled transaction features after feature interference operation corresponding to the first-level related merchant example;

randomly changing transaction feature vectors in the pre-labeled transaction features of the secondary related merchant examples to obtain the pre-labeled transaction features after feature interference operation corresponding to the secondary related merchant examples;

the obtaining of the pre-labeled transaction characteristics of the first merchant example, the first-level related merchant example and the second-level related merchant example with respect to the commodity transaction data respectively includes:

obtaining a second phase Guan Shanghu map; the second related merchant map comprises map elements used for representing pre-labeled transaction characteristics of the pending related merchant examples and correlation reasoning lines obtained by performing correlation reasoning on the map elements corresponding to the pending related merchant examples with related merchant labels, and the pre-labeled transaction characteristics comprise transaction characteristic vectors corresponding to the first amount of transaction information;

according to the correlation inference path in the second related merchant map, performing sampling query on map elements in the second related merchant map one by one to obtain a fourth number of target undetermined related merchant examples corresponding to the first number of transaction information respectively; a target undetermined related merchant example corresponding to one transaction information is an undetermined related merchant example in which a transaction feature vector corresponding to the transaction information meets a preset index in the second related merchant map;

determining a first commercial tenant example, a first-level related commercial tenant example and a second-level related commercial tenant example based on a fourth number of target undetermined related commercial tenant examples and the second-level Guan Shanghu atlases corresponding to the first number of transaction information respectively;

obtaining pre-labeled transaction characteristics of the first merchant example, the first-level relevant merchant example and the second-level relevant merchant example respectively aiming at commodity transaction data from the second relevant merchant map;

the determining a first commercial tenant example, a first-level related commercial tenant example and a second-level related commercial tenant example based on a fourth number of target pending related commercial tenant examples and the second-level Guan Shanghu atlases respectively corresponding to the first number of transaction information includes:

determining a first-stage undetermined related merchant example with a first-stage related merchant label from the second related merchant map; the target undetermined related merchant examples belong to a fourth number of target undetermined related merchant examples respectively corresponding to the first number of transaction information;

determining a second-stage undetermined related merchant example with a second-stage related merchant label from the second related merchant map;

taking the target undetermined related business example as a first business example, and sampling and inquiring a fifth number of primary undetermined related business examples from the primary undetermined related business example to be taken as the primary related business example;

and sampling and inquiring a sixth number of secondary undetermined related business examples from the secondary undetermined related business examples to serve as the secondary related business examples.

7. The artificial intelligence based user data recommendation method of claim 6, wherein the performing an iteration on the initial transaction feature extraction model based on the pre-labeled transaction features of the first merchant example and the first example features to obtain the transaction feature extraction model comprises:

performing iteration on the initial transaction feature extraction model based on the pre-labeled transaction features of the first merchant example and the first example features to obtain an iterated initial transaction feature extraction model;

determining a verification merchant example group based on a fourth number of target undetermined related merchant examples corresponding to the first number of transaction information respectively and the second phase Guan Shanghu atlas;

determining a training progress of the iterated initial transaction feature extraction model based on the validated merchant exemplar set;

determining the transaction feature extraction model based on the training progress and the iterated initial transaction feature extraction model;

the verification merchant example group comprises a verification merchant example, a first-level verification related merchant example and a second-level verification related merchant example; the example of the verified merchant is different from the first merchant example, a first-level related merchant label is provided between the example of the verified merchant and the example of the first-level verified related merchant, and a second-level related merchant label is provided between the example of the verified merchant and the example of the second-level verified related merchant;

the determining a training progress of the iterated initial transaction feature extraction model based on the validated merchant exemplar set includes:

performing feature interference operation on the pre-labeled transaction features of the verification merchant example, the first-level verification-related merchant example and the second-level verification-related merchant example respectively through the iterated initial transaction feature extraction model to obtain pre-labeled transaction features after the feature interference operation is respectively corresponding to the verification merchant example, the first-level verification-related merchant example and the second-level verification-related merchant example;

performing correlation detection on the pre-labeled transaction characteristics after the characteristic interference operation corresponding to the verification merchant example, the pre-labeled transaction characteristics after the characteristic interference operation corresponding to the primary verification-related merchant example, and the pre-labeled transaction characteristics after the characteristic interference operation corresponding to the secondary verification-related merchant example to obtain first example characteristics of the verification merchant example;

determining a training cost value of the iterated initial transaction feature extraction model based on the first example features of the verified merchant example and the pre-labeled transaction features of the verified merchant example;

and determining the training progress of the initial transaction feature extraction model after iteration based on the training cost value.

8. The artificial intelligence based user data recommendation method according to claim 7, wherein the first example feature of the verified merchant example and the pre-labeled transaction feature of the verified merchant example each include a transaction feature vector corresponding to a first amount of transaction information;

determining a training cost value of the iterated initial transaction feature extraction model based on the first example features of the verified merchant example and the pre-labeled transaction features of the verified merchant example, including:

inputting the target confidence coefficient and a transaction feature vector corresponding to the Nth transaction information in the pre-labeled transaction features of the verification merchant example into an extraction cost evaluation function to obtain a feature extraction replacement value corresponding to the Nth transaction information; the target confidence degree is the confidence degree that the transaction feature vector corresponding to the Nth transaction information in the first example feature of the verification merchant example meets the preset index; n is a positive integer less than or equal to a first number;

performing a homogenization operation on the feature extraction cost value corresponding to the Nth transaction information to obtain a feature extraction cost value after the homogenization operation corresponding to the Nth transaction information;

performing weighted summation on the feature extraction cost values after the homogenization operation corresponding to the first number of transaction information respectively to obtain the training cost value of the initial transaction feature extraction model after iteration;

the performing a normalization operation on the feature extraction cost value corresponding to the nth transaction information to obtain a feature extraction cost value after the normalization operation corresponding to the nth transaction information includes:

obtaining a third number of merchants of the merchant paradigm that transaction eigenvectors corresponding to the Nth transaction information in the verification merchant paradigm group meet preset indexes;

and generating a homogenization coefficient based on the number of the third merchants, and executing homogenization operation on the feature extraction cost value corresponding to the Nth transaction information by adopting the homogenization coefficient to obtain a feature extraction cost value after the homogenization operation corresponding to the Nth transaction information.

9. The artificial intelligence based user data recommendation method of claim 7, wherein the determining the transaction feature extraction model based on the training progress and the iterated initial transaction feature extraction model comprises:

if the training progress of the iterated initial transaction feature extraction model does not reach the standard, continuously iterating the iterated initial transaction feature extraction model based on the pre-annotated transaction feature of the first merchant example and the iterated first example feature to obtain the transaction feature extraction model, wherein the iterated first example feature is obtained by performing correlation detection on the pre-annotated transaction feature after the feature interference operation corresponding to the first merchant example, the pre-annotated transaction feature after the feature interference operation corresponding to the second-level related merchant example, and the pre-annotated transaction feature after the feature interference operation corresponding to the first-level related merchant example through the iterated initial transaction feature extraction model;

and if the training progress of the initial transaction feature extraction model after iteration reaches the standard, determining the initial transaction feature extraction model after iteration as the transaction feature extraction model.

10. A server system, comprising a server configured to perform the method of any one of claims 1-9.