Merchant push method and system based on merchant information and user information

Abstract

The present invention relates to a merchant push method and system based on merchant information and user information. The method includes: forming merchant tags based on merchant information, constructing a merchant stratification model using a clustering model, and stratifying merchants using the merchant stratification model based on the merchant tags; obtaining user preference tags and user tolerance tags based on user usage information of a certain application, and using a regression model to obtain user preference and user tolerance based on the user preference tags and user tolerance tags, thereby obtaining a user comprehensive tolerance score for the application; constructing a multi-objective fusion push model based on the merchant stratification, user comprehensive tolerance score, and user interests; screening merchants based on the multi-objective fusion push model based on real-time user information and pushing them to the user, and judging the authenticity of the merchant based on the user's feedback on the merchant. According to the present invention, merchants can be accurately pushed to users and the authenticity of the merchant data can be further verified.

Description

Technical Field

[0001] The present invention relates to computer technology for processing merchant information and user information, and in particular to a merchant push method and system based on merchant information and user information. Background Art

[0002] With the increasing popularity of the Internet, more and more users are accessing a wide range of merchant information online. However, the current problem is that there are a large number of merchants, but few effective ones. How can we better tap into the value of existing merchants and more accurately push merchant information to users for marketing and expansion?

[0003] In addition, on the other hand, the authenticity of merchant data is not high, and we also hope to propose a verification method that can verify the authenticity of merchant data.

[0004] Furthermore, in the current prior art, the merchant model for processing merchant information and the user model for processing user information are generally isolated from each other, and there is no correlation or overlap between the two. Summary of the Invention

[0005] In view of the above problems, the present invention aims to provide a merchant push method and system based on merchant information and user information, which can accurately push merchants to users.

[0006] Furthermore, the present invention also aims to provide a merchant push method and system based on merchant information and user information that can verify the authenticity of merchant data.

[0007] A merchant push method based on merchant information and user information according to one aspect of the present invention is characterized by comprising:

[0008] A merchant hierarchical clustering step, forming merchant tags based on merchant information, setting N levels for merchants, using a clustering model to construct a merchant hierarchical model, and implementing N-level merchant stratification for merchants based on the merchant tags using the merchant hierarchical model, where N is a natural number greater than 1;

[0009] A user fault tolerance scoring step comprises obtaining a user preference tag and a user tolerance tag based on the user's application usage information, obtaining a user preference and a user tolerance using a regression model based on the user preference and user tolerance tags, and obtaining a user comprehensive fault tolerance score for the application based on the user preference and user tolerance; and

[0010] a multi-objective fusion model construction step, forming a multi-objective fusion push model based on the merchant stratification obtained in the merchant stratification clustering step, the user comprehensive fault tolerance score obtained in the user fault tolerance scoring step, and user interests; and

[0011] The merchant list push step selects a merchant list based on the multi-objective fusion push model constructed in the multi-objective fusion model construction step according to the user's real-time information and pushes the merchant list to the user.

[0012] Optionally, after the merchant list recommendation step, the method further includes:

[0013] The user feedback verification step determines the authenticity of the merchant based on the user's feedback on the pushed merchant.

[0014] Optionally, the merchant hierarchical clustering step includes:

[0015] A merchant tag formation sub-step is to obtain a merchant basic tag by processing the merchant information, and to construct a merchant portrait based on the merchant basic tag to form a merchant tag;

[0016] A merchant stratification sub-step of setting N levels for merchants according to their credibility, constructing a merchant stratification model using a clustering model, and implementing N levels of merchant stratification using the merchant stratification model based on the merchant tags; and

[0017] In the merchant rating sub-step, merchants at different levels are given real ratings to obtain the merchant authenticity rating S m .

[0018] Optionally, the merchant tag forming sub-step includes:

[0019] Clean and process merchant information to obtain merchant basic tags;

[0020] Classify merchants into active merchants and inactive merchants based on historical transaction information contained in merchant information;

[0021] For active merchants, a merchant profile including merchant attribute profile and transaction profile is constructed; for inactive merchants, a merchant profile including merchant attribute profile is constructed; and

[0022] A merchant tag is formed based on the merchant portrait.

[0023] Optionally, the user error tolerance scoring step includes the following sub-steps:

[0024] The preference tag forming sub-step forms a user preference tag for the application based on the user's operation behavior information of the application;

[0025] a tolerance label forming sub-step, forming a user tolerance label for the application based on the operation failure information of the user who failed to operate the application;

[0026] a model analysis sub-step, using a regression model to construct a user preference scoring model and a tolerance scoring model, respectively; calculating the user preference based on the user preference label and the user preference scoring model; and calculating the user tolerance based on the user tolerance label and the user tolerance scoring model; and

[0027] The comprehensive fault tolerance scoring sub-step obtains the user's comprehensive fault tolerance score for the application based on the user preference and the user tolerance.

[0028] Optionally, the multi-objective fusion model building step includes the following sub-steps:

[0029] In the vector construction sub-step, a merchant vector is obtained based on the merchant tag using the embedding method, and a user vector is constructed based on the user's historical behavior;

[0030] A merchant similarity calculation sub-step, defining a recommendation ratio for each of N levels of merchants, and calculating merchant similarity for each of the N levels based on the merchant vectors;

[0031] The user interest calculation sub-step uses the user purchase rate as the target, uses a deep neural network to train the user multivariate interest model, obtains the user interest vector, and calculates the user interest score S based on the user interest vector. v ;

[0032] a sampling and recalling sub-step, sampling and recalling merchants that the user is interested in based on the recommendation ratio and the user interest vector; and

[0033] The multi-objective fusion push model construction sub-step is combined with the user interest score S v and the merchant authenticity score S m , score (u，m) =f(s v , s m |u) is used as the target training sorting model to obtain the multi-target fusion push model.

[0034] Optionally, in the user interest calculation sub-step, the merchant similarity is calculated according to the following formula:

[0035]

[0036]

[0037] Among them, m is the merchant vector, and the output of the formula is the similarity value between merchants.

[0038] Optionally, in the user interest calculation sub-step, a double-layer capsule network is used as a deep neural network.

[0039] Optionally, the merchant list pushing step includes the following sub-steps:

[0040] A matching sub-step, screening out merchants to be pushed that match the user's real-time location information based on the multi-objective fusion model according to the user's real-time information; and

[0041] Push sub-step: push the merchant to be pushed to the user through the specified communication method;

[0042] Optionally, the user feedback verification step includes:

[0043] In the user feedback sub-step, the user performs a prescribed operation on the merchant to provide feedback to the merchant; and

[0044] In the counting sub-step, the proportion of users' feedback on the merchant is counted. When the feedback proportion is greater than a threshold, the merchant is judged to be a real merchant; otherwise, the merchant is judged to be a fake merchant.

[0045] A merchant push device based on merchant information and user information according to one aspect of the present invention is characterized by comprising:

[0046] A merchant stratification and clustering module, configured to form merchant tags based on merchant information, set N levels for merchants, construct a merchant stratification model using a clustering model, and implement N-level merchant stratification for merchants based on the merchant tags using the merchant stratification model, where N is a natural number greater than 1;

[0047] A user fault tolerance scoring module is configured to obtain a user preference tag and a user tolerance tag based on information about the user's use of the application, obtain a user preference and a user tolerance using a regression model based on the user preference and user tolerance tags, and obtain a comprehensive user fault tolerance score for the application based on the user preference and user tolerance; and

[0048] A multi-objective fusion model construction module is used to construct a multi-objective fusion push model based on the merchant stratification obtained by the merchant stratification clustering module, the user comprehensive fault tolerance score obtained by the user fault tolerance scoring module, and user interests; and

[0049] The merchant list push module is used to filter out a merchant list based on the user's real-time information and the multi-objective fusion push model constructed in the multi-objective fusion model construction module and push it to the user.

[0050] Optionally, the method further includes the following steps after the merchant list recommendation module:

[0051] The user feedback verification module is used to determine the authenticity of merchants based on user feedback on the pushed merchants.

[0052] Optionally, the merchant hierarchical clustering module includes:

[0053] A merchant tag formation submodule is used to obtain a merchant basic tag by processing merchant information, and to construct a merchant portrait based on the merchant basic tag to form a merchant tag;

[0054] A merchant stratification submodule, configured to set N levels of merchants according to their credibility, construct a merchant stratification model using a clustering model, and implement N levels of merchant stratification using the merchant stratification model based on the merchant tags; and

[0055] The merchant rating submodule is used to assign real ratings to merchants at different levels and obtain the merchant authenticity rating S m .

[0056] Optionally, the merchant label formation submodule is used to clean and process merchant information to obtain merchant basic labels, divide merchants into active merchants and inactive merchants according to historical transaction information contained in the merchant information, construct merchant portraits including merchant attribute portraits and transaction portraits for active merchants, and construct merchant portraits including merchant attribute portraits for inactive merchants, and form merchant labels based on the merchant portraits.

[0057] Optionally, the user fault tolerance scoring module includes the following submodules:

[0058] The preference tag forming submodule is used to form a user preference tag for the application based on the user's operation behavior information of the application;

[0059] A tolerance label forming submodule is used to form a user tolerance label for the application based on the operation failure information of the user who failed to operate the application;

[0060] a model analysis submodule, configured to respectively construct a user preference scoring model and a tolerance scoring model using a regression model, calculate the user preference based on the user preference label and the user preference scoring model, and calculate the user tolerance based on the user tolerance label and the user tolerance scoring model; and

[0061] The comprehensive fault tolerance scoring submodule is used to obtain the user's comprehensive fault tolerance score for the application based on the user preference and the user tolerance.

[0062] Optionally, the multi-objective fusion model building module includes the following submodules:

[0063] A vector construction submodule, configured to obtain a merchant vector using an embedding device according to the merchant tag, and to construct a user vector according to the user's historical behavior;

[0064] A merchant similarity calculation submodule is used to define recommendation ratios for merchants in N levels, and calculate merchant similarity for each of the N levels based on the merchant vectors;

[0065] The user interest calculation submodule uses a deep neural network to train a user multivariate interest model with the user purchase rate as the target, obtains the user interest vector, and calculates the user interest score S based on the user interest vector. v ;

[0066] A sampling and recalling submodule, configured to sample and recall merchants that the user is interested in based on the recommendation ratio and the user interest vector; and

[0067] The multi-objective fusion push model construction submodule is used to combine the user interest score S v and the merchant authenticity score S m , score (u，m) =f(s v , s m |u) is used as the target training sorting model to obtain the multi-target fusion push model.

[0068] Optionally, in the user interest calculation submodule, the merchant similarity is calculated according to the following formula:

[0069]

[0070]

[0071] Among them, m is the merchant vector, and the output of the formula is the similarity value between merchants.

[0072] Optionally, in the user interest calculation submodule, a double-layer capsule network is used as a deep neural network.

[0073] Optionally, the merchant list push module includes the following submodules:

[0074] A matching submodule, configured to screen out merchants to be pushed that match the user's real-time location information based on the multi-objective fusion model according to the user's real-time information; and

[0075] The push submodule is used to push the merchants to the users through the specified communication method;

[0076] Optionally, the user feedback verification module includes:

[0077] A user feedback submodule, configured to provide feedback to merchants based on the user performing prescribed operations on the received merchant information; and

[0078] The statistics submodule is used to count the user's feedback ratio for the merchant. When the feedback ratio is greater than a threshold, the merchant is judged to be a real merchant, otherwise the merchant is judged to be a fake merchant.

[0079] A computer-readable medium according to one aspect of the present invention stores a computer program thereon, wherein when the computer program is executed by a processor, the merchant push method based on household information and user information is implemented.

[0080] A computer device according to one aspect of the present invention comprises: a memory; a processor; and a computer program stored in the memory and executable on the processor, wherein the computer program is characterized in that the execution of the computer program enables the processor to implement the merchant push method based on merchant information and user information when executing the computer program. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Figure 1 This is a flow chart showing an overview of the merchant push method based on merchant information and user information of the present invention.

[0082] Figure 2 It is a structural block diagram showing the merchant push system based on merchant information and user information of the present invention. DETAILED DESCRIPTION

[0083] The following describes some of the various embodiments of the present invention, which are intended to provide a basic understanding of the present invention, but are not intended to identify the key or decisive elements of the present invention or to limit the scope of protection.

[0084] For the purpose of brevity and illustration, the principles of the present invention are described herein primarily with reference to exemplary embodiments thereof. However, those skilled in the art will readily recognize that the same principles are equally applicable to all types of merchant push methods and systems based on merchant information and user information, and that these same principles may be implemented therein, and that any such variations do not depart from the true spirit and scope of this patent application.

[0085] Moreover, in the following description, reference is made to the accompanying drawings, which illustrate specific exemplary embodiments. Electrical, mechanical, logical, and structural changes may be made to these embodiments without departing from the spirit and scope of the present invention. In addition, although a feature of the present invention is disclosed in conjunction with only one of several embodiments, it may be desirable and / or advantageous to combine this feature with one or more other features of other embodiments as may be desired and / or advantageous for any given or identifiable function. Therefore, the following description should not be regarded in a limiting sense, and the scope of the present invention is defined by the appended claims and their equivalents.

[0086] Terms such as “having” and “including” indicate that in addition to the units (modules) and steps directly and clearly stated in the specification and claims, the technical solution of the present invention does not exclude the situation where it has other units (modules) and steps that are not directly or clearly stated.

[0087] First, the main technical concept of the merchant push method based on merchant information and user information of the present invention is described.

[0088] Figure 1 This is a flow chart showing an overview of the merchant push method based on merchant information and user information of the present invention.

[0089] like Figure 1 As shown, the merchant push method implemented based on merchant information and user information of the present invention mainly includes the following steps:

[0090] Merchant hierarchical clustering step S100: forming merchant tags based on merchant information, setting N levels for merchants, using a clustering model to construct a merchant hierarchical model, and implementing N-level merchant stratification for merchants based on the merchant tags using the merchant hierarchical model, where N is a natural number greater than 1;

[0091] User fault tolerance scoring step S200: obtaining a user preference tag and a user tolerance tag based on the user's usage information of a certain application, using a regression model to obtain a user preference and a user tolerance based on the user preference and user tolerance tags, and obtaining a user comprehensive fault tolerance score of the application based on the user preference and the user tolerance; and

[0092] Multi-objective fusion model construction step S300: constructing a multi-objective fusion push model based on the merchant stratification obtained in the merchant stratification clustering step, the user comprehensive fault tolerance score obtained in the user fault tolerance scoring step, and user interests; and

[0093] Merchant list push step S400: pushing the merchant list to the user based on the user's real-time information and the multi-objective fusion push model obtained in the multi-objective fusion model construction step; and

[0094] User feedback verification step S500 (optional step): judging the authenticity of merchants based on user feedback on the merchants in the pushed merchant list.

[0095] Among them, according to the merchant hierarchical clustering step S100, the user fault tolerance scoring step S200, the multi-objective fusion model construction step S300 and the merchant list push step S400, the merchant push method based on merchant information and user information of the present invention can be fully constituted. According to this merchant push method, merchants can be actively and accurately pushed to users based on merchant information and user information.

[0096] In existing technology, merchant verification primarily relies on on-site inspections by operators. These operators conduct field visits, provide sample labeled data, and then use big data analysis and other methods to classify and determine merchant authenticity. However, this method can be inefficient when dealing with a large number of merchants and lacks a reliable basis for identifying long-tail merchants.

[0097] Therefore, based on the above invention, the present invention further proposes a method for verifying the authenticity of merchants based on user feedback, that is, by further having a user feedback verification step S500, the merchant push method based on merchant information and user information of the present invention can not only accurately and actively push merchants to users, but also use user feedback to further verify the authenticity of merchant data.

[0098] The merchant hierarchical clustering step S100 includes:

[0099] A merchant tag formation sub-step is to obtain a merchant basic tag by processing the merchant information, and to construct a merchant portrait based on the merchant basic tag to form a merchant tag;

[0100] A merchant stratification sub-step of setting N levels for merchants according to their credibility, constructing a merchant stratification model using a clustering model, and implementing N levels of merchant stratification using the merchant stratification model based on the merchant tags; and

[0101] In the merchant rating sub-step, merchants at different levels are given real ratings to obtain the merchant authenticity rating S m .

[0102] Here, as an example, the specific contents of the merchant hierarchical clustering step S100 include:

[0103] Step S101: extracting and processing basic data such as the merchant's network access information, business registration information, and historical transaction information;

[0104] Step S102: Constructing a merchant profile, including the merchant's region, registered capital, legal person information, average transaction amount, number of transactions, and other items to form a merchant tag;

[0105] Step S103: Classify merchants into multiple levels according to their credibility (i.e., the authenticity of the merchants). The following description takes the case of classifying merchants into three levels as an example, i.e., classifying merchants into three levels: authentic merchants, suspicious merchants, and risky merchants; and

[0106] Step S104: using a clustering model (such as Kmeans, GMM, etc.) to build a merchant hierarchical model, clustering all merchants according to merchant tags, and clustering all merchants into three categories; and

[0107] Step S105: Assign authenticity scores S to merchants at different levels m ={s r , s q , s no_r}, where S r 、S q 、S no_r Represent the scores of real merchants, suspicious merchants and risky merchants respectively.

[0108] Next, the user fault tolerance scoring step S200 mainly includes the following sub-steps:

[0109] The preference tag forming sub-step forms a user preference tag for the application based on the user's operation behavior information of the application;

[0110] a tolerance label forming sub-step, forming a user tolerance label for the application based on the operation failure information of the user who failed to operate the application;

[0111] a model analysis sub-step, using a regression model to construct a user preference scoring model and a tolerance scoring model, respectively; calculating the user preference based on the user preference label and the user preference scoring model; and calculating the user tolerance based on the user tolerance label and the user tolerance scoring model; and

[0112] The comprehensive fault tolerance scoring sub-step obtains the user's comprehensive fault tolerance score for the application based on the user preference and the user tolerance.

[0113] Here, as an example, the specific contents of the user fault tolerance scoring model construction step S200 are listed as follows:

[0114] Step S201: extracting and processing basic data such as user basic data, behavior information, account information, application usage, authenticity feedback participation, etc.;

[0115] Step S202: Build a user profile, including the user's age, gender, frequent activity areas, activity level, check-in frequency, number of coupons received, number of failed transactions, etc., as well as scenario characteristics such as number of failed transactions, number of recommended merchants, number of historical user feedback, number of rewards received, etc., to form a user preference and tolerance label;

[0116] Step S203: Constructing a user rating model, specifically including: training a tolerance value s based on user tolerance information such as user loyalty to the application, number of failed transactions, and number of historical feedback. tol , train the preference value s based on the user's preference information such as the number of rewards the user has received in the application, app usage preferences, etc. like ;

[0117] Step S204: Based on the user's preference value s for the app like and tolerance value s tol Get the user's comprehensive evaluation value S for the app u =f(s like , s tol );as well as

[0118] Step S205: Customize the hyperparameter threshold α and set {S u The user set of ≥α} is selected as the users to be pushed to form a user pool.

[0119] Next, the multi-objective fusion model construction step S300 includes the following sub-steps:

[0120] A vector construction sub-step, using an embedding method to obtain a merchant vector based on the merchant tag, and constructing a user vector based on the user's historical behavior;

[0121] A merchant similarity calculation sub-step, defining a recommendation ratio for each of N levels of merchants, and calculating merchant similarity for each of the N levels based on the merchant vectors;

[0122] The user interest calculation sub-step uses the user purchase rate as the target, uses a deep neural network to train the user multivariate interest model, obtains the user interest vector, and calculates the user interest score S based on the user interest vector. v ;

[0123] a sampling and recalling sub-step, sampling and recalling merchants that the user is interested in based on the recommendation ratio and the user interest vector; and

[0124] The multi-objective fusion push model construction sub-step is combined with the user interest score S v and the merchant authenticity score S m , score (u，m )=f(s v , s m |u) is used as the target training sorting model to obtain the multi-target fusion push model.

[0125] Here, as an example, the specific content of the multi-objective fusion model construction step S300 includes:

[0126] Step S301: Obtain the merchant's initial vector based on the merchant tag, and construct a user vector based on the user's historical behavior, where the user's historical behavior is obtained from the user's transaction data;

[0127] Step S302: Define the recommendation ratios of three types of merchants, rato = {r1, r2, r3}, and calculate merchant similarity based on merchant tags and user behavior, where user behavior includes, for example, merchants that the user has browsed, clicked, and consumed within a period of time;

[0128] Step S303: Construct a double-layer capsule network and use a multi-source routing algorithm to extract user interests based on the user's behavior sequence to obtain the user's K-dimensional interest vector V u , taking the purchase rate CVR as the goal, a deep neural network is used to train the user multivariate interest model, and the user interest score s is obtained according to the user multivariate interest model v , where the user’s behavior is composed of the merchants that the user has consumed, and the user’s behavior sequence is the merchant sequence that the user has consumed;

[0129] Step S304: According to user interest v u ∈V u Using vector matching methods, we sample and recall merchants that users are interested in based on the stratified merchant recommendation ratio (rato); and

[0130] Step S305: Combine user interest score s v and merchant authenticity ratings m , score (u，m) =f(s v , s m |u) is used as the target training ranking model to obtain a multi-target fusion push model, and the final output of the multi-target fusion push model is the user's comprehensive rating of a merchant.

[0131] The merchant list push step S400 includes the following sub-steps:

[0132] A matching sub-step, screening out merchants to be pushed that match the user's real-time location information based on the multi-objective fusion model according to the user's real-time information; and

[0133] Push sub-step: push the merchant to be pushed to the user through the specified communication method;

[0134] The user feedback verification step S500 includes:

[0135] In the user feedback sub-step, the user performs prescribed operations on the merchant to provide feedback to the merchant; and

[0136] In the counting sub-step, the proportion of users' feedback on the merchant is counted. When the feedback proportion is greater than a threshold, the merchant is judged to be a real merchant; otherwise, the merchant is judged to be a fake merchant.

[0137] Here, as an example, in the merchant list push step S400, when the user arrives at a certain location, the user's intention is judged through the user's real-time information, and the merchant list to be pushed is obtained. It is replaced and sorted according to the hierarchical recommendation ratio, and the TopN is selected and pushed to the user through SMS, in-site messages, etc.

[0138] Next, as an example, the user feedback verification step S500 specifically includes:

[0139] Step S501: When the user arrives at the recommended merchant and fails to find the merchant or the merchant's actual address does not match the recommended address, the user can upload relevant credentials to the backend by taking a photo or other means;

[0140] Step S502: The backend regularly reviews credentials and calculates the user feedback ratio. When the feedback ratio is greater than a threshold, the merchant is considered to be a non-authentic merchant; and

[0141] Step S503: After the review is passed, rewards are issued to the user account in the form of coupons, red envelopes or points.

[0142] Here, user feedback is regarded as the real feedback of merchants, which gradually forms merchant seeds to facilitate further iterative optimization of subsequent models.

[0143] Figure 2 It is a structural block diagram showing the merchant push system based on merchant information and user information of the present invention.

[0144] like Figure 2 As shown, the merchant push system based on merchant information and user information of the present invention includes:

[0145] The merchant stratification and clustering module 100 is used to form merchant tags based on merchant information, set N levels for merchants according to their credibility, use a clustering model to build a merchant stratification model, and implement N-level merchant stratification based on the merchant tags using the merchant stratification model, where N is a natural number greater than 1;

[0146] The user fault tolerance scoring module 200 is used to obtain user preference tags and user tolerance tags based on the user's usage information of a certain application, use a regression model to obtain user preference and user tolerance based on the user preference and user tolerance tags, and obtain a comprehensive user fault tolerance score for the application based on the user preference and user tolerance;

[0147] A multi-objective fusion model module 300 is used to construct a multi-objective fusion push model based on the merchant stratification obtained by the merchant stratification clustering module, the user comprehensive fault tolerance score obtained by the user fault tolerance score module, and user interests;

[0148] The merchant list push module 400 is used to filter out a merchant list based on the user's real-time information and the multi-target fusion push model and push it to the user.

[0149] Furthermore, optionally, the merchant push system based on merchant information and user information of the present invention may further include:

[0150] The user feedback verification module 500 is used to determine the authenticity of the merchant based on the user's feedback on the pushed merchant.

[0151] The merchant hierarchical clustering module 100 includes:

[0152] The merchant tag forming submodule 110 is used to obtain a merchant basic tag by cleaning and processing the merchant information, and to construct a merchant portrait based on the merchant basic tag to form a merchant tag;

[0153] A merchant stratification submodule 120 is configured to set N levels of merchants according to their credibility, construct a merchant stratification model using a clustering model, and implement N levels of merchant stratification using the merchant stratification model based on the merchant tags; and

[0154] The merchant rating submodule 130 is used to assign authenticity ratings to merchants at different levels to obtain the merchant authenticity rating S m .

[0155] Among them, in the merchant label formation submodule 110, the merchant information is cleaned and processed to obtain the merchant basic label, and the merchants are divided into active merchants and inactive merchants according to the historical transaction information contained in the merchant information. For active merchants, a merchant portrait including a merchant attribute portrait and a transaction portrait is constructed, and for inactive merchants, a merchant portrait including a merchant attribute portrait is constructed, and a merchant label is formed according to the merchant portrait.

[0156] Among them, in the merchant stratification submodule 120, for active merchants, a knowledge graph is constructed, and after the link prediction system is completed, a community algorithm is used to calculate merchant clusters to achieve merchant stratification. For inactive merchants, a merchant stratification model is used according to the merchant attribute portrait to achieve merchant stratification.

[0157] The user fault tolerance scoring module 200 includes:

[0158] The preference tag forming submodule 210 is used to form a user preference tag for a certain application based on the user's operation behavior information of the application;

[0159] The tolerance label forming submodule 220 is used to form a user tolerance label for the application based on the operation failure information when the user fails to operate the application;

[0160] A model analysis submodule 230 is configured to construct a user preference scoring model and a tolerance scoring model using a regression model, calculate the user preference based on the user preference tag and the user preference scoring model, and calculate the user tolerance based on the user tolerance tag and the user tolerance scoring model;

[0161] A comprehensive fault tolerance scoring submodule 240 is configured to obtain a comprehensive fault tolerance score of the application from the user based on the user preference and the user tolerance; and

[0162] The user pool forming submodule 250 is configured to screen users based on the comprehensive fault tolerance score to form a user pool.

[0163] The multi-objective fusion model module 300 includes:

[0164] A vector construction submodule 310 is configured to obtain a merchant initial vector using an Embedding system based on the merchant tag and to construct a user vector based on the user's historical behavior;

[0165] The merchant similarity calculation submodule 320 is used to define the recommendation ratio for each of the N levels of merchants, and calculate the merchant similarity for each of the N levels based on the merchant tags and user historical behavior;

[0166] The user interest calculation submodule 330 is used to train the user multivariate interest model using a deep neural network with the user purchase rate as the target, obtain the user interest vector, and calculate the user interest score S based on the user interest vector. v ;

[0167] A sampling and recalling submodule 340 is configured to sample and recall merchants that the user is interested in based on the recommendation ratio and the user interest vector; and

[0168] Multi-objective fusion push model 350 is used to combine the user interest score S v And all

[0169] Merchant authenticity score S m , score (u，m) =f(s v , s m |u) is used as the target training sorting model to obtain the multi-target fusion push model.

[0170] In the user interest calculation submodule 330, the merchant similarity is calculated according to the following formula:

[0171]

[0172]

[0173] In the user interest calculation submodule 330 , a double-layer capsule network is used as a deep neural network.

[0174] The merchant list push module 400 includes:

[0175] A matching submodule 410 is configured to filter out merchants to be pushed that match the user's real-time location information based on the multi-objective fusion model according to the user's real-time information; and

[0176] The push submodule 420 is used to push the merchants to be pushed to the user through a prescribed communication method.

[0177] The user feedback verification module 500 includes:

[0178] The user feedback submodule 510 is used for the user to perform prescribed operations on the received merchant to provide feedback to the merchant; and

[0179] The statistics submodule 520 is used to count the user's feedback ratio for the merchant. When the feedback ratio is greater than a threshold, the merchant is judged to be a real merchant, otherwise the merchant is judged to be a fake merchant.

[0180] As described above, the merchant push method of the present invention, which is based on merchant information and user information, can firstly characterize merchant portrait characteristics based on the merchant's network access information, industrial and commercial data, historical transactions, etc., and use the clustering model to construct a merchant hierarchical clustering model. In this invention, the advantages of the clustering model, such as simplicity, efficiency, and rapid convergence, are utilized to achieve merchant stratification in a simple, efficient, and accurate manner. Furthermore, by stratifying merchants (for example, dividing them into three categories: real merchants, suspicious merchants, and fake merchants) and defining the push ratio for each type of merchant, it is possible to achieve precise push.

[0181] Secondly, the present invention can also use a regression model to construct a user tolerance rating model based on the user's activity level for a certain application, historical transaction behavior, and participation in feedback activities, and obtain the user's preference and tolerance for a certain application through the user tolerance rating model. The regression model is a predictive modeling technique that studies the relationship between the dependent variable (target) and the independent variable (predictor). This technique is commonly used for predictive analysis, time series models, and discovering causal relationships between variables. The regression model can indicate the significant relationship between independent and dependent variables, and it can also indicate the influence of multiple independent variables on a dependent variable. In the present invention, by utilizing the above advantages of the regression model, the user's preference and tolerance can be obtained more objectively, thereby enabling subsequent accurate merchant push.

[0182] Finally, a multi-target fusion push model is constructed by combining merchant stratification, user fault tolerance ratings, and user interests. This multi-target fusion push model comprehensively considers merchant stratification, user fault tolerance ratings, and user interests. Deep neural networks are used to train a user multivariate interest model during the construction of the multi-target fusion push model. This invention effectively leverages the following advantages of deep neural network training to achieve merchant recommendations: powerful nonlinear fitting capabilities, the ability to approximate any nonlinear function with arbitrary precision, strong feature extraction capabilities, and strong fault tolerance.

[0183] The above describes the merchant push method based on household information and user information and the system thereof of the present invention. The following cites a specific embodiment to illustrate the merchant push method based on household information and user information of the present invention.

[0184] In this embodiment, the SMS contact method is used as an application (APP) to list the The merchant push method based on household information and user information of the present invention is used to User pushes merchants, when users open It can be used as a verification certificate for merchant authenticity feedback based on GPS positioning.

[0185] 1. Merchant hierarchical clustering steps

[0186] When a merchant joins the network, it is necessary to fill in the merchant network information. At the same time, based on the merchant's unified social credit code, merchant name and other information, the merchant's industrial and commercial registration information can be queried. At the same time, some merchants have historical transaction data, and a large number of long-tail merchants have no transaction behavior.

[0187] The steps of constructing the merchant hierarchical clustering model in this embodiment mainly include:

[0188] (1) Clean and process the merchant's network access information, business registration information and other basic data to extract merchant tags;

[0189] (2) Based on merchant tags, select tags according to scene characteristics to build merchant attribute portraits, such as region, GPS location, industry, store address, etc.;

[0190] (3) Merchants are divided into active merchants and inactive merchants based on their historical transaction records. Active merchants are given transaction profiles, including their profitability, growth potential, and operational stability;

[0191] (4) Thus, the profile of an active merchant includes both attribute profile and transaction profile, while the profile of an inactive merchant is its attribute profile;

[0192] (5) Merchants are divided into three levels: trusted merchants, suspicious merchants, and risky merchants. Trusted merchants are considered to be authentic merchants, risky merchants are considered to be highly likely to be non-authentic merchants, and suspicious merchants are those that cannot be completely determined to be authentic.

[0193] (6) Build a merchant knowledge graph for active merchants to achieve merchant stratification;

[0194] (7) For long-tail merchants, the Gaussian hierarchical clustering method is used to cluster merchants according to their attribute portraits;

[0195] (8) Active merchants and long-tail merchants are stratified separately and then merged to obtain the stratification results of all merchants (here, active merchants and long-tail merchants are distinguished because of their different characteristics, so different stratification methods can be used, but they are finally merged because merchants are divided into three categories in total. No matter what stratification method is used, they will eventually belong to these three categories). The authenticity score S is assigned to each merchant layer. m :

[0196] s m ={s r , s q , s no_r}

[0197] Among them, S r 、S q 、S no_rRepresent the scores of real merchants, suspicious merchants and risky merchants respectively.

[0198] 2. User Fault Tolerance Scoring Steps

[0199] Users are using When recording the user's app usage, behavior trajectory, and basic user data such as user clicks, browsing, and purchases, the user's understanding of the app is obtained through analysis of these basic user data. Preference and tolerance.

[0200] The specific contents of the user fault tolerance scoring steps are as follows:

[0201] (1) Clean and process basic user data such as user behavior information, account red envelope information, activity participation, push feedback participation, etc., and extract the user's Behavioral tags, build user preference portraits, and describe users' preferences;

[0202] (2) According to the user Failed transactions (such as payment failure, repayment failure, coupon expiration, etc.), app reuse after failure, number of rewards received, etc., characterize users' app’s tolerance;

[0203] (3) Use regression models to construct user preference scoring models and tolerance scoring models respectively to obtain user App preference S like and tolerance S tol , where the regression model is a predictive modeling technique that can express the relationship between the dependent variable (target) and the independent variable (predictor);

[0204] (4) Combine preference and tolerance to obtain the user's Overall user comprehensive fault tolerance score S fault-tol :

[0205] s fault-tol =ω·s like +(1-ω)·s tol

[0206] Where ω is a number between 0 and 1, which is used to indicate which side the combined value of preference and tolerance is more inclined to. (5) Define the threshold α according to:

[0207]

[0208] That is, if the user's comprehensive fault tolerance score S fault-to is greater than or equal to the threshold α, then user u iBelongs to user pool S u , otherwise user u i Does not belong to user pool S u In this way, according to the above formula, the user u to be pushed is selected i Form a user pool S u .

[0209] 3. Steps to build a multi-objective fusion model

[0210] After obtaining the user pool to be pushed and the layered merchants, a multi-objective fusion model is constructed based on user preferences / real-time information, etc. The goal here is to guide users to consume and also to obtain positive feedback from users on the authenticity of the merchants.

[0211] The specific contents of the multi-objective fusion model construction steps are as follows:

[0212] (1) The merchant vector is constructed based on the merchant tag using the Embedding method, and the user vector is constructed based on the user's historical behavior sequence.

[0213] (2) Combining vector similarity and the swing method, the similarity between merchants is calculated according to the following formula, and the similar merchant lists are obtained according to the level to which the merchants belong. The formula is:

[0214]

[0215]

[0216] In the above formula, m is the merchant vector, and the output of the formula is the similarity value between merchants.

[0217] Among them, vector similarity and swing method are existing methods for calculating merchant similarity. Vector similarity is calculated from the user vector itself, and the swing method is calculated from the situation where two users consume at the same merchant. The two methods are combined as the final output similarity value.

[0218] (3) With the purchase rate as the target, a double-layer capsule network is constructed to extract user interests. The input of the double-layer capsule network is the user's historical behavior sequence, and the output is the user's interest vector. The user's multivariate interest model is trained to obtain the user's interest vector v n :

[0219]

[0220] Among them, V u Represents all user interest vectors calculated by the model.

[0221] Recall the list of merchants to be pushed based on the user's interest vector and calculate the user's interest score Sv .

[0222] Here, let's take an example to illustrate. If a user's interest is [supermarkets, restaurants], we use these two interest vectors to match (i.e., recall) the merchants with the highest user preference scores from the merchant list to form a push list. The user interest score here is the user's interest score for a merchant, which is obtained by multiplying the user interest vector by the merchant score.

[0223] (4) Define the recommendation ratio of the three types of merchants, and count the number of merchants in each layer in the list to be pushed. According to the number of missing and excess merchants in each layer, formulate a replacement strategy and replace them according to the merchant similarity list.

[0224] This is because the recommended merchants need to account for a certain proportion of each category, so it is best to count the number of merchants in each layer. When the number of merchants in a certain category is small, the most similar merchants are replaced by calculating the merchant vector similarity;

[0225] (5) Combined with user interest score S v and merchant authenticity score S m As the training target, a deep neural network is used to train the sorting model to obtain a multi-target fusion model.

[0226] Among them, the "Embedding" appearing in (1) is literally translated into "embedding" in Chinese and is often translated as "vectorization" or "vector mapping". Formally speaking, Embedding is to "represent" an object with a low-dimensional dense vector. The object here can be a word (Word2Vec), an item (Item2Vec), or a node in a network relationship (Graph Embedding). The word "representation" means that the Embedding vector can express certain characteristics of the corresponding object, and the distance between vectors reflects the similarity between objects. Moreover, since Embedding is a way to convert discrete variables into continuous vector representations, it can find the nearest neighbors in the embedding space. Therefore, in the present invention, it can be well used to implement merchant recommendations based on merchant tags.

[0227] Here, an example is given to illustrate the user multivariate interest model in (3). For example, the user multivariate interest model can be constructed by splicing the following vectors to output the user's interest vector group (also known as "all user interest vectors"): the user's basic features, the user's interest vector obtained from the historical behavior sequence through the capsule network (that is, the user's intention choice), the current user's trigger action (provided by the user's behavior), the target recommended merchant, and the contextual features (obtained from data statistics). Finally, these vectors are spliced ​​together to output the user's interest vector group.

[0228] In addition, in the present invention, in order to improve the push efficiency, it is preferred to train a multi-objective fusion model in advance, thereby being able to quickly implement the following merchant list push steps.

[0229] 4. Merchant list push steps

[0230] According to the user's real-time information, a merchant list is screened out based on the multi-objective fusion push model obtained in the multi-objective fusion model construction step and pushed to the user.

[0231] Specifically, when users use For example, real-time user information, such as the current time, location, and clicks, is returned. Merchant recommendations are then pushed based on this information, guiding the user to the designated merchant for purchase. Depending on the characteristics of different applications, the real-time user information that can be returned can also vary adaptively.

[0232] (1) Obtain the user's real-time information, determine the user's current intention, select the user's interest vector and recall the list of merchants to be pushed, calculate and replace the merchant stratification ratio, and use a multi-objective fusion push model to sort the merchant list;

[0233] (2) The sorted merchant list is filtered according to the geographical location, and the top N merchants within the specified distance are selected and pushed to users through SMS, in-site messages, etc.

[0234] 5. User feedback verification steps

[0235] The user feedback verification steps mainly include the following:

[0236] (1) Collect the user's behavior data on the merchant being pushed. If the user makes a transaction at a designated merchant, calculate the distance between the user's GPS and the merchant's location. If the distance is within the error range, the merchant is considered to be a real merchant.

[0237] (2) If the user does not find the merchant, or the merchant's actual location does not match the displayed location, guide the user to upload it to the backend by taking a photo or other means;

[0238] (3) The backend regularly reviews user feedback and collects statistics on the merchant's user feedback. When the number and proportion of inauthentic feedback exceeds a specified threshold, the merchant is considered to be an inauthentic merchant;

[0239] (4) Optionally, a reward mechanism can be developed to reward users who provide real and effective feedback by using points, red envelopes, coupons, etc. account.

[0240] As described above, the merchant push method of the present invention, which is implemented based on merchant information and user information, can first characterize merchant portrait features based on the merchant's network access information, industrial and commercial data, historical transactions, etc., and construct a merchant hierarchical clustering model. By stratifying merchants (for example, dividing merchants into three categories: real merchants, suspicious merchants, and fake merchants) and defining the push ratio of each type of merchant, all types of merchants, such as real merchants, suspicious merchants, and fake merchants, can have the opportunity to be recommended to users.

[0241] Secondly, it is also possible to build a user tolerance rating model based on the user's activity on a certain application, historical transaction behavior and participation in feedback activities, and obtain the user's preference and tolerance for a certain application through the user tolerance rating model. Based on the user's preference and tolerance, it is possible to select user groups suitable for authentic merchant feedback push, thereby further achieving accurate push.

[0242] Finally, a multi-objective fusion push model is constructed by combining merchant classification, user tolerance score and user interest. Merchants are pushed through SMS, in-site messages, etc., and the authenticity of merchants is verified through user feedback on the merchants.

[0243] A computer-readable medium according to one aspect of the present invention stores a computer program thereon, wherein when the computer program is executed by a processor, the merchant push method based on household information and user information is implemented.

[0244] A computer device according to one aspect of the present invention comprises: a memory; a processor; and a computer program stored on the memory and executable on the processor, wherein the execution of the computer program enables the processor to implement the above-mentioned merchant push method based on merchant information and user information when executing the computer program. The above examples mainly illustrate the merchant push method and system thereof based on merchant information and user information of the present invention. Although only some of the specific embodiments of the present invention are described, it should be understood by those skilled in the art that the present invention can be implemented in many other forms without departing from its subject matter and scope. Therefore, the examples and embodiments shown are to be regarded as illustrative rather than restrictive, and the present invention may cover various modifications and substitutions without departing from the spirit and scope of the present invention as defined in the appended claims.

Claims

1. A merchant push method based on merchant information and user information, characterized in that: include: A merchant hierarchical clustering step, forming merchant tags based on merchant information, setting N levels for merchants, using a clustering model to construct a merchant hierarchical model, and implementing N-level merchant stratification for merchants based on the merchant tags using the merchant hierarchical model, where N is a natural number greater than 1; A user fault tolerance scoring step comprises obtaining a user preference tag and a user tolerance tag based on the user's application usage information, obtaining a user preference and a user tolerance using a regression model based on the user preference and user tolerance tags, and obtaining a user comprehensive fault tolerance score for the application based on the user preference and user tolerance; and a multi-objective fusion model construction step, forming a multi-objective fusion push model based on the merchant stratification obtained in the merchant stratification clustering step, the user comprehensive fault tolerance score obtained in the user fault tolerance scoring step, and user interests; and The merchant list push step selects the merchant list based on the user's real-time information and the multi-objective fusion push model constructed in the multi-objective fusion model construction step and pushes it to the user. The merchant hierarchical clustering step includes: A merchant tag formation sub-step is to obtain a merchant basic tag by processing the merchant information, and to construct a merchant portrait based on the merchant basic tag to form a merchant tag; A merchant stratification sub-step of setting N levels for merchants according to their credibility, constructing a merchant stratification model using a clustering model, and implementing N levels of merchant stratification using the merchant stratification model based on the merchant tags; and In the merchant rating sub-step, merchants at different levels are given real ratings to obtain the merchant authenticity rating S m , The user fault tolerance scoring step includes the following sub-steps: The preference tag forming sub-step forms a user preference tag for the application based on the user's operation behavior information of the application; a tolerance label forming sub-step, forming a user tolerance label for the application based on the operation failure information of the user who failed to operate the application; a model analysis sub-step, using a regression model to construct a user preference scoring model and a tolerance scoring model, respectively; calculating the user preference based on the user preference label and the user preference scoring model; and calculating the user tolerance based on the user tolerance label and the user tolerance scoring model; and The comprehensive fault tolerance scoring sub-step obtains the user's comprehensive fault tolerance score for the application based on the user preference and the user tolerance.

2. The merchant push method based on merchant information and user information according to claim 1, characterized in that: After the merchant list recommendation step, the method further includes: The user feedback verification step determines the authenticity of the merchant based on the user's feedback on the pushed merchant.

3. The merchant push method based on merchant information and user information according to claim 1, characterized in that: The merchant tag forming sub-step includes: Clean and process merchant information to obtain merchant basic tags; Classify merchants into active merchants and inactive merchants based on historical transaction information contained in merchant information; For active merchants, a merchant profile including merchant attribute profile and transaction profile is constructed; for inactive merchants, a merchant profile including merchant attribute profile is constructed; and A merchant tag is formed based on the merchant portrait.

4. The merchant push method based on merchant information and user information according to claim 1, characterized in that: The multi-objective fusion model construction step includes the following sub-steps: In the vector construction sub-step, a merchant vector is obtained based on the merchant tag using the embedding method, and a user vector is constructed based on the user's historical behavior; A merchant similarity calculation sub-step, defining a recommendation ratio for each of N levels of merchants, and calculating merchant similarity for each of the N levels based on the merchant vectors; The user interest calculation sub-step uses the user purchase rate as the target, uses a deep neural network to train the user multivariate interest model, obtains the user interest vector, and calculates the user interest score S based on the user interest vector. v ; a sampling and recalling sub-step, sampling and recalling merchants that the user is interested in based on the recommendation ratio and the user interest vector; as well as The multi-objective fusion push model construction sub-step is combined with the user interest score S v and the merchant authenticity score S m , score (u,m) =f(s v ,s m |u) is used as the target training sorting model to obtain the multi-target fusion push model.

5. The merchant push method based on merchant information and user information as claimed in claim 4, characterized in that: In the user interest calculation sub-step, a double-layer capsule network is used as a deep neural network.

6. The merchant push method based on merchant information and user information according to claim 4, characterized in that: The merchant list push step includes the following sub-steps: A matching sub-step, screening out merchants to be pushed that match the user's real-time location information based on the multi-objective fusion model according to the user's real-time information; and The push sub-step pushes the merchant to be pushed to the user through the prescribed communication method.

7. The merchant push method based on merchant information and user information according to claim 6, characterized in that: The user feedback verification step includes: In the user feedback sub-step, the user performs a prescribed operation on the merchant to provide feedback to the merchant; and In the counting sub-step, the proportion of users' feedback on the merchant is counted. When the feedback proportion is greater than a threshold, the merchant is judged to be a real merchant; otherwise, the merchant is judged to be a fake merchant.

8. A merchant push device based on merchant information and user information, characterized in that: include: A merchant stratification and clustering module, configured to form merchant tags based on merchant information, set N levels for merchants, construct a merchant stratification model using a clustering model, and implement N-level merchant stratification for merchants based on the merchant tags using the merchant stratification model, where N is a natural number greater than 1; A user fault tolerance scoring module is used to obtain user preference tags and user tolerance tags based on user application usage information, use a regression model to obtain user preference and user tolerance based on the user preference and user tolerance tags, and obtain a comprehensive user fault tolerance score of the application based on the user preference and user tolerance; A multi-objective fusion model construction module is used to construct a multi-objective fusion push model based on the merchant stratification obtained by the merchant stratification clustering module, the user comprehensive fault tolerance score obtained by the user fault tolerance scoring module, and user interests; and The merchant list push module is used to filter out the merchant list based on the user's real-time information and the multi-objective fusion push model constructed in the multi-objective fusion model construction module and push it to the user. The merchant hierarchical clustering module includes: A merchant tag formation submodule is used to obtain a merchant basic tag by processing merchant information, and to construct a merchant portrait based on the merchant basic tag to form a merchant tag; A merchant stratification submodule, configured to set N levels of merchants according to their credibility, construct a merchant stratification model using a clustering model, and implement N levels of merchant stratification using the merchant stratification model based on the merchant tags; and The merchant rating submodule is used to assign real ratings to merchants at different levels and obtain the merchant authenticity rating S m , The user fault tolerance scoring module includes the following submodules: The preference tag forming submodule is used to form a user preference tag for the application based on the user's operation behavior information of the application; A tolerance label forming submodule is used to form a user tolerance label for the application based on the operation failure information of the user who failed to operate the application; a model analysis submodule, configured to respectively construct a user preference scoring model and a tolerance scoring model using a regression model, calculate the user preference based on the user preference label and the user preference scoring model, and calculate the user tolerance based on the user tolerance label and the user tolerance scoring model; and The comprehensive fault tolerance scoring submodule is used to obtain the user's comprehensive fault tolerance score for the application based on the user preference and the user tolerance.

9. The merchant push device based on merchant information and user information according to claim 8, characterized in that: After the business list recommendation module, the following is further included: The user feedback verification module is used to determine the authenticity of merchants based on user feedback on the pushed merchants.

10. The merchant push device based on merchant information and user information according to claim 8, characterized in that: The merchant tag formation submodule is used to clean and process merchant information to obtain merchant basic tags, divide merchants into active merchants and inactive merchants according to the historical transaction information contained in the merchant information, construct merchant portraits including merchant attribute portraits and transaction portraits for active merchants, and construct merchant portraits including merchant attribute portraits for inactive merchants, and form merchant tags based on the merchant portraits.

11. The merchant push device based on merchant information and user information according to claim 8, characterized in that: The multi-objective fusion model construction module includes the following submodules: A vector construction submodule, configured to obtain a merchant vector using an embedding device according to the merchant tag, and to construct a user vector according to the user's historical behavior; A merchant similarity calculation submodule is used to define recommendation ratios for merchants in N levels, and calculate merchant similarity for each of the N levels based on the merchant vectors; The user interest calculation submodule uses a deep neural network to train a user multivariate interest model with the user purchase rate as the target, obtains the user interest vector, and calculates the user interest score S based on the user interest vector. v ; A sampling and recalling submodule, configured to sample and recall merchants that the user is interested in based on the recommendation ratio and the user interest vector; as well as The multi-objective fusion push model construction submodule is used to combine the user interest score S v and the merchant authenticity score S m , score (u,m) =f(s v ,s m |u) is used as the target training sorting model to obtain the multi-target fusion push model.

12. The merchant push device based on merchant information and user information according to claim 11, characterized in that: In the user interest calculation submodule, a double-layer capsule network is used as a deep neural network.

13. The merchant push device based on merchant information and user information according to claim 11, characterized in that: The merchant list push module includes the following submodules: A matching submodule, configured to screen out merchants to be pushed that match the user's real-time location information based on the multi-objective fusion model according to the user's real-time information; as well as The push submodule is used to push the merchants to be pushed to users through the specified communication method.

14. The merchant push device based on merchant information and user information according to claim 13, characterized in that: The user feedback verification module includes: A user feedback submodule, configured to provide feedback to merchants based on the user performing prescribed operations on the received merchant information; and The statistics submodule is used to count the user's feedback ratio for the merchant. When the feedback ratio is greater than a threshold, the merchant is judged to be a real merchant, otherwise the merchant is judged to be a fake merchant.

15. A computer-readable medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the merchant push method based on household information and user information as described in any one of claims 1 to 7 is implemented.

16. A computer device comprising: Memory; processor; and a computer program stored on the memory and executable on the processor, characterized in that The running of the computer program enables the processor to implement the merchant push method based on merchant information and user information as described in any one of claims 1 to 7 when executing the computer program.

Images