Coupon recommendation method, system equipment and storage medium

Abstract

The invention discloses a coupon recommendation method and system equipment and a storage medium, and the method comprises the steps: coding the GPS coordinates of a user and a merchant through employing a user-user social network, a user-coupon exchange behavior graph, and a user-merchant transaction graph, and taking the coded GPS coordinates as the input information of a recommendation model; and extracting user preference features from the user-user social network and the user-merchant transaction atlas by using a graph neural network, and extracting hidden features of coupons from the user-coupon exchange behavior atlas by using the graph neural network. Preference features hidden by the user in the transaction data can be better utilized; the cold start problem of the new user and the new coupon is well solved by combining the knowledge graph and the transaction information; meanwhile, the GPS coordinates of the user and the merchant are converted into Geohash codes, then the Geohash is subjected to embedding to serve as graph neural network model input, and the Geohash is aggregated into representation of the geographic dimension of the user.

Description

Technical Field

[0001] The present invention belongs to the field of graph neural networks, and in particular to graph neural networks, recommendation systems, and knowledge graph technologies, and more particularly to a coupon recommendation method, system equipment, and storage medium. Background Art

[0002] The existing coupon recommendation system mainly considers the interaction between users and coupons, and does not add additional transaction information and geographic information between users and merchants as supplementary information; the existing coupon recommendation does not fully integrate user geographic information and user preference interaction information, and mainly relies on setting business rules for push.

[0003] Therefore, how to propose a coupon recommendation system that can additionally add transaction information and geographic information of users and merchants as supplementary information, and fully combine user geographic information and user preference interaction information as supplementary technical solutions, is a technical problem that technical personnel in this field urgently need to solve. Summary of the invention

[0004] In order to overcome the deficiencies of the prior art, the purpose of the present invention is to provide a coupon recommendation method, system device and storage medium, which solves the problem that the user preference representation in the prior art is not rich enough and the user geographic location information is not deeply utilized.

[0005] To achieve the above objectives, in a first aspect, the present invention provides a coupon recommendation method, comprising:

[0006] Build user social networks, user-coupon graphs, and user-merchant graphs;

[0007] Use the map service to query the GPS location of the merchant's specific address and calculate the representation of the Geohash code;

[0008] Through the user social network, user-coupon graph, and user-merchant graph, we can get three representations of users, namely h soci 、h uc 、h mech , and the coupons are shown c ;

[0009] Get the final score s of user u for coupon c *c , define the loss function and use the RMSproo optimizer for model training.

[0010] Furthermore, the construction of the user social network, the user-coupon graph, and the user-merchant graph includes:

[0011] Obtain customer contact data through the customer acquisition system, activity system, and marketing system to build a user social network, denoted as Gsoci { i |u j >};

[0012] in, i |u j > represents u i and u j There are social connections;

[0013] Get user redemption events from user activity redemption history and build a user-coupon graph, denoted as G coup { i |c i >};

[0014] in, i |c i > represents u i Redeemed the voucher i ;

[0015] Obtain the user-merchant graph from user transaction data, denoted as G mech { i |m i >};

[0016] in i |m i > represents u i and m i There is trading activity.

[0017] Furthermore, for the user-merchant graph G mech , query the GPS location of the merchant's specific address through the map service, and convert the user GPS and merchant GPS into 5-bit Geohash codes, represented by ge u andge m .

[0018] Furthermore, the three representations of users are learned through the user social network, user-coupon graph, and user-merchant graph, which are h soci 、h uc 、h mech , and the coupons are shown c ,include:

[0019] In the user social network, the input feature representation e of user u is u , u in G soci The set of adjacent users in is called nu, where the feature corresponding to the i-th adjacent user is expressed as After the graph neural network forward aggregation calculation, h soci , the calculation formula is:

[0020] ​​​​​​h soci =σ(W·Agg(e nu )+b);

[0021] Agg(e nu ) introduces the attention mechanism, and the specific formula of attention weight a is as follows:

[0022] a i =σ(w·(e nu +e u )+b);

[0023]

[0024] Agg(e nu )=e nu a;

[0025] In the user-coupon graph, the user u and coupon c are represented by graph neural aggregation, denoted as h uc and h c ;

[0026] In G coup In the example, the inputs of user u and the adjacent coupon set are e uc and e c , where the i-th adjacent coupon input of user u is recorded as

[0027] In the user-merchant graph G mech In the above example, we obtain the corresponding feature representation for user u, denoted as h mech .

[0028] Furthermore, the Geohash codes of user u and neighboring merchant set m are represented as e geo ;

[0029] Take the first 5 digits of Geohash and initialize the vector with dimension d as the initialization vector of Geohash encoding, denoted as e *geo ∈R k,d ;

[0030] e *geo As input, it passes through a 2-layer MLP (Multi-Layer Perceptron) network to obtain e geo , the specific formula is as follows:

[0031] e geo =σ2(W2σ1(W1·e *geo +b1)+b2)

[0032] User-Merchant Graph G mech The input feature of user u in is denoted as e*mechu , and the input features of the neighboring merchants are recorded as e *mech , the feature representation h of user u is calculated by the following method mech :

[0033] cat is a splicing operation,

[0034]

[0035] a i =σ(w·(e mech +e mechu )+b);

[0036]

[0037] Agg(e mech u )=e mech u a;

[0038] h mech =σ(W·Agg(e mechu )+b).

[0039] Further, according to the three representations of users obtained through the user social network, the user-coupon map, and the user-merchant map, they are h soci 、h uc 、h mech , and the coupons are shown c ,

[0040] The final representation of user u is h = cat (h soci ,h uc ,h mech ), cat is the splicing operation;

[0041] h and h c After concatenation, the result is input into a 2-layer MLP network to obtain the score of user u for coupon c:

[0042] s *c =σ2(W2σ1(W1·cat(h,h c )+b1)+b2)

[0043] Through the loss function Use RMSproo optimizer for training;

[0044] Among them, s c Indicates the actual redemption status of coupon c by user u, which is 1 if redeemed and 0 if not.

[0045] In a second aspect, the present invention further provides a coupon recommendation system, comprising: a data construction unit, a first operation unit, a second operation unit, and a model training unit;

[0046] Data construction unit: used to build user social networks, user-coupon graphs, and user-merchant graphs;

[0047] The first computing unit is used to query the GPS location of the specific address of the merchant by using the map service, and calculate the representation of the Geohash code;

[0048] The second computing unit is used to obtain three representations of users through user social networks, user-coupon graphs, and user-merchant graphs, respectively: h soci 、h uc 、h mech , and the coupons are shown c ;

[0049] Model training unit: used to obtain the final score s of user u for coupon c *c , define the loss function and use the RMSproo optimizer for model training.

[0050] In a third aspect, the present invention provides a computing device, comprising:

[0051] A memory for storing program instructions;

[0052] The processor is used to call the program instructions stored in the memory, and execute the above-mentioned coupon recommendation method in whole or in part according to the obtained program instructions.

[0053] In a fourth aspect, the present invention provides a computer-readable storage medium, including computer-readable instructions. When a computer reads and executes the computer-readable instructions, the above-mentioned coupon recommendation method is fully or partially implemented. The readable medium may be multiple, and the multiple readable media can operate independently of each other.

[0054] In a fifth aspect, the present invention provides a computer program product, which includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the above-mentioned coupon recommendation method is fully or partially executed.

[0055] Compared with the existing technology, this solution uses user-user social networks, user-coupon redemption behavior graphs, user-merchant transaction graphs, and encodes the GPS coordinates of users and merchants as input information for the recommendation model, and uses graph neural networks to extract user preference features from user-user social networks and user-merchant transaction graphs, and then uses graph neural networks to extract hidden features of coupons from user-coupon redemption behavior graphs. By learning the structural information and node information in the network through graph neural networks, users and coupons are fully characterized, so that the recommendation model has a more efficient improvement in accuracy. By adding user-merchant transaction graph information, not only the user's transaction preference characteristics are considered, but also the geographical dimension characteristics are added. In the new user operation scenario, due to the lack of new user behavior data, the new user information is aggregated and characterized through transaction information and social network graphs, so as to transmit transaction preferences between users, which solves the cold start problem to a certain extent. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the embodiments of the present disclosure or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0057] Figure 1 A schematic diagram of a flow chart of an embodiment of a coupon recommendation method provided by the present invention;

[0058] Figure 2 A structural block diagram of an embodiment of the coupon recommendation system provided by the present invention;

[0059] Figure 3 The present invention provides a schematic structural diagram of an electronic device. DETAILED DESCRIPTION

[0060] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be interpreted as limiting the present application.

[0061] It will be understood by those skilled in the art that, unless otherwise stated, the singular forms "a", "an" and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present application refers to the presence of features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. The term "and / or" used herein includes all or any unit and all combinations of one or more associated listed items.

[0062] In order to make the objectives, technical solutions and advantages of the present application more clear, the specific implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.

[0063] The core of the present invention is to provide a coupon recommendation method. Figure 1 The figure is a flow chart of the coupon recommendation method provided in this embodiment. In order to better understand the method, Figure 1 The technical solution of this embodiment is described in detail:

[0064] Step S1: construct user social network, user-coupon graph, user-merchant graph;

[0065] Step S2: Query the GPS location of the specific address of the merchant using the map service, and calculate the representation of the Geohash code;

[0066] Step S3: Obtain three representations of the user through the user social network, user-coupon graph, and user-merchant graph, which are h soci 、h uc 、h mech , and the coupons are shown c ;

[0067] Step S4: Obtain the final score s of user u for coupon c *c , define the loss function and use the RMSproo optimizer for model training.

[0068] According to one aspect of the present invention, a method for characterizing users and coupons using user social networks, user-coupon graphs, user-merchant graphs, and graph neural network technology is provided, comprising:

[0069] Build user social networks, user-coupon graphs, and user-merchant graphs;

[0070] Obtain customer contact data from the customer acquisition system, activity system, and marketing system to build social relationships (such as parents, colleagues, friends, etc.), denoted as G soci { i |u j ​>}, where i |u j > represents u i and u j There is a social relationship; obtain the user redemption event from the user activity redemption history to build a user-coupon graph, denoted as G coup { i |c i >}, where i |c i > represents u i Redeemed the voucher i ; Obtain the user-merchant graph from user transaction data, denoted as G mech { i |m i >}, where i |m i > represents u i and m i There is trading behavior;

[0071] For the user-merchant graph G mech , use the map service to query the GPS location of the merchant's specific address, and then convert the user GPS and merchant GPS into 5-digit Geohash codes, represented by geo u andge m The following uses the user social network, user-coupon graph, and user-merchant graph through the graph neural network to learn the user representation h:

[0072] First, in the user social network, the following method is used to obtain the user u in the social network G. soci The representation in is denoted as h soci ;

[0073] The input feature representation e of user u u , u in G soci The set of adjacent users in is called nu, where the feature corresponding to the i-th adjacent user is expressed as

[0074] Then, h is obtained by forward aggregation calculation of the graph neural network. soci , the calculation formula is as follows:

[0075] h soci =σ(W·Agg(e nu )+b)

[0076] Agg(e nu ) introduces the attention mechanism, and the specific formula of attention weight a is as follows:

[0077] a i =σ(w·(e nu ​​​​​+e u )+b)

[0078]

[0079] Agg(e nu )=e nu ·a

[0080] Then, in the user-coupon graph, the user u and coupon c can be represented by graph neural aggregation, denoted as h uc and h c ;

[0081] In G coup In the example, we can see that the inputs of user u and the adjacent coupon set are e uc and e c , where the i-th adjacent coupon input of user u is recorded as

[0082] In G soci The polymerization process is the same as that in uc and h c ;

[0083] In the user-merchant graph G mech In the above example, we only need to obtain the corresponding feature representation for user u, denoted as h mech ;

[0084] In order to add the feature information of geographic dimension, we need to first characterize the Geohash codes of user u and neighboring merchant set m, denoted as e geo , the specific steps are as follows:

[0085] Because they are the user's neighboring merchants, the first 5 digits of the Geohash code are the same, so the first 5 digits of the Geohash are taken;

[0086] Then initialize the vector of dimension d as the initialization vector of geohash encoding, denoted as e *geo ∈R k,d ;

[0087] e *geo As input, it passes through a 2-layer MLP (Multi-Layer Perceptron) network to obtain e geo , the specific formula is as follows:

[0088] e geo =σ2(W2σ1(W1·e *geo +b1)+b2)

[0089] User-Merchant Graph G mech The input feature of user u in is denoted as e *mech u, and the input features of the neighboring merchants are recorded as e *mech , the feature representation h of user u can be learned by calculating mech :

[0090] cat is the splicing operation

[0091]

[0092] a i =σ(w·(e mech +e mechu )+b)

[0093]

[0094] Agg(e mechu )=e mechu ·a

[0095] h mech =σ(W·Agg(e mechu )+b)

[0096] So far, three representations of users have been learned through user social networks, user-coupon graphs, and user-merchant graphs, which are h soci 、h uc 、h mech , and the coupons are shown c ;

[0097] The final representation of user u is h = cat (h soci ,h uc ,h mech ), cat is the splicing operation;

[0098] h and h c After concatenation, the results are input into a 2-layer MLP network to obtain the final score of user u for coupon c:

[0099] s *c =σ2(W2σ1(W1·cat(h,h c )+b1)+b2)

[0100] The final loss function defined as follows is trained using the RMSproo optimizer:

[0101]

[0102] Among them, s c Indicates the actual redemption status of coupon c by user u, which is 1 if redeemed and 0 if not.

[0103] For ease of understanding, a specific example is provided below to further explain the present application.

[0104] In the credit card industry, there are constantly new customers. In order to increase the activity and transaction volume of new customers, it is necessary to motivate customers through activities. Generally, cash red envelopes and coupons are used to incentivize new customers. In order to recommend accurate coupons to new customers, the present invention invents a coupon recommendation method based on a graph neural model that integrates transaction preferences and geographic information. Compared with the defects of the prior art, the method enriches the user's representation by adding transaction graphs and Geohash encoding information, thereby improving the accuracy of the recommendation model.

[0105] The coupon recommendation system will recommend the coupons that new customers are most interested in from the SKU library, and then remind new customers to redeem them on the APP via text messages.

[0106] In order to achieve the above object, the scheme of the present invention is as follows:

[0107] Sort out the list of coupons that are currently on the shelves as a candidate set for recommendation.

[0108] Obtain customer contact data from the customer acquisition system, activity system, and marketing system. The data generally includes ID number, relationship type, and basic customer information (name, gender, age, occupation, address, education, etc.), and build social relationships (such as parents, colleagues, friends, etc.), denoted as G soci { i |u j >}, where i |u j > represents u i and u j There is a social relationship; obtain user redemption events from the user activity redemption history, including information such as coupon name, type, brand, and redemption time, and build a user-coupon graph, denoted as G coup { i |c i >}, where i |c i > represents u i Redeemed the voucher i ; Obtain the transaction behavior between users and merchants from user transaction data, including merchant name, transaction amount, transaction time, etc., and build a user-merchant graph, denoted as G mech { i |m i >}, where i |m i > represents u i and m i There is trading behavior; ​​​​​​

[0109] One of the key points of this embodiment is to use the Geohash codes of users and merchants to characterize users. The specific steps are as follows:

[0110] For the user-merchant graph G mech In the process, Geohash attributes are added to the user entity and the merchant entity. The map service is used to query the GPS location of the merchant's specific address, and then the user GPS and merchant GPS are converted into 5-digit Geohash codes. Because they are the neighboring merchants of the user, the first 5 digits are the same, which is represented as geohash. u .

[0111] Then initialize the vector of dimension d as the initialization vector of geohash encoding, denoted as e *geo ∈R k,d ;

[0112] e *geo After passing through a 2-layer MLP network as input, we get e geo , the specific formula is as follows:

[0113] e geo =σ2(W2σ1(W1·e *geo +b1)+b2)

[0114] On the other hand, this embodiment utilizes three types of graph data, namely, user social network, user-coupon graph, and user-merchant graph, and extracts hidden representation forms of users and coupons, namely h and h respectively, by performing operations such as aggregation and nonlinear transformation on the graphs through graph neural networks. c .

[0115] h and h c The calculation process is as follows:

[0116] First, in the user social network, the following method is used to obtain the user u in the social network G. soci The representation in is denoted as h soci ;

[0117] The input feature representation e of user u u , u in G soci The set of adjacent users in is called nu, where the feature corresponding to the i-th adjacent user is expressed as

[0118] Then, h is obtained by forward aggregation calculation of the graph neural network. soci , the calculation formula is as follows:

[0119] h soci =σ(W·Agg(e nu )+b)

[0120] Agg(enu ) introduces the attention mechanism, and the specific formula of attention weight a is as follows:

[0121] a i =σ(w·(e nu +e u )+b)

[0122]

[0123] Agg(e nu )=e nu ·a

[0124] Then, in the user-coupon graph, the user u and coupon c can be represented by graph neural aggregation, denoted as h uc and h c ;

[0125] In G coup In the example, we can see that the inputs of user u and the adjacent coupon set are e uc and e c , where the i-th adjacent coupon input of user u is recorded as

[0126] In G soci The same polymerization process was used to obtain h uc and h c ;

[0127] In the user-merchant graph G mech In the above example, we only need to obtain the corresponding feature representation for user u, denoted as h mech ;

[0128] The input feature of user u is denoted as e *mechu , and the input feature of the adjacent merchant is recorded as e *mech , the feature representation h of user u can be learned by calculating mech :

[0129]

[0130] a i =σ(w·(e mech +e mechu )+b)

[0131]

[0132] Agg(e mechu )=e mechu ·a

[0133] h mech =σ(W·Agg(e mech )+b)

[0134] So far, three representations of users have been learned through user social networks, user-coupon graphs, and user-merchant graphs, which are h soci 、h uc 、h mech , and the coupons are shown c ;

[0135] The final representation of user u is h = cat (h soci ,h uc ,h mech ), cat is the splicing operation;

[0136] h and h c After concatenation, the results are input into a 2-layer MLP network to obtain the final score of user u for coupon c:

[0137] s *c =σ2(W2σ1(W1·cat(h,h c )+b1)+b2)

[0138] The final loss function defined as follows is trained using the RMSproo optimizer:

[0139]

[0140] Among them, s c Indicates the actual redemption status of coupon c by user u, which is 1 if redeemed and 0 if not.

[0141] Through the above method, historical redemption records are selected as positive samples. Since negative samples do not exist in reality, the coupons that users did not redeem when redeeming coupons are used as negative samples to construct a training data set.

[0142] The training data set is divided into training set, validation set, and test set in a ratio of 7:2:1.

[0143] By using the method of the present invention, the training set is trained to obtain a model file.

[0144] The model input includes the input features of user u in the user social network, user-coupon graph, user-merchant graph, and his one-hop adjacent nodes, as well as the input features of the coupons redeemed and unredeemed by user u.

[0145] The final output is the score of user u for the coupon. The score ranges from 0 to 1. The closer to 1, the more likely the user is to be interested in the coupon, and vice versa.

[0146] By using user-user social networks, user-coupon redemption behavior graphs, and user-merchant transaction graphs, the GPS coordinates of users and merchants are encoded as input information for the recommendation model, and the graph neural network is used to extract user preference features from the user-user social network and user-merchant transaction graphs, and the hidden features of coupons are extracted from the user-coupon redemption behavior graphs using the graph neural network. By learning the structural information and node information in the network through the graph neural network, users and coupons are fully characterized, so that the recommendation model has a more efficient improvement in accuracy. By adding user-merchant transaction graph information, not only the user's transaction preference characteristics are considered, but also the geographical dimension characteristics are added. In the new user operation scenario, due to the lack of new user behavior data, the new user information is aggregated and characterized through transaction information and social network graphs, so as to transmit transaction preferences between users, which solves the cold start problem to a certain extent.

[0147] In one embodiment, the present invention provides a coupon recommendation system. Figure 2 The figure shows a structural block diagram of a coupon recommendation system provided in this embodiment, wherein the system comprises: a data construction unit 10, a first operation unit 20, a second operation unit 30, and a model training unit 40;

[0148] Data construction unit 10: used to construct user social network, user-coupon map, user-merchant map;

[0149] The first computing unit 20 is used to query the GPS location of the specific address of the merchant by using a map service, and calculate the representation of the Geohash code;

[0150] The second computing unit 30 is used to obtain three representations of the user through the user social network, the user-coupon map, and the user-merchant map, respectively: soci 、h uc 、h mech , and the coupons are shown c ;

[0151] Model training unit 40: used to obtain the final score s of user u for coupon c *c , define the loss function and use the RMSproo optimizer for model training.

[0152] In the scheme of this embodiment, by using user-user social network, user-coupon redemption behavior graph, user-merchant transaction graph, the GPS coordinates of users and merchants are encoded as input information of the recommendation model, and the graph neural network is used to extract user preference features from the user-user social network and user-merchant transaction graph, and then the hidden features of the coupon are extracted from the user-coupon redemption behavior graph by using the graph neural network. By learning the structural information and node information in the network through the graph neural network, the users and coupons are fully characterized, so that the recommendation model has a more efficient improvement in accuracy. By adding user-merchant transaction graph information, not only the user's transaction preference characteristics are considered, but also the geographical dimension characteristics are added. In the new user operation scenario, due to the lack of new user behavior data, the new user information is aggregated and characterized through transaction information and social network graphs, so as to transmit transaction preferences between users, which solves the cold start problem to a certain extent.

[0153] The coupon recommendation system of this embodiment is used to implement the aforementioned coupon recommendation method. Therefore, the specific implementation method of the coupon recommendation system can be seen in the specific implementation method part of the coupon recommendation method in the previous text. For example, the data construction unit 10 is used to implement step S1 in the above-mentioned coupon recommendation method, the first operation unit 20 is used to implement step S2 in the above-mentioned coupon recommendation method, the second operation unit 30 is used to implement step S3 in the above-mentioned coupon recommendation method, and the model training unit 40 is used to implement step S4 in the above-mentioned coupon recommendation method, etc. Therefore, its specific implementation method can refer to the description of the corresponding various parts of the embodiments. The system provided by the present invention is attached to the method provided by the present invention, and its specific effects can refer to the above-mentioned method, which will not be repeated here.

[0154] In one embodiment, a computing device is provided, comprising:

[0155] A memory for storing program instructions;

[0156] A processor is used to implement the steps of any of the above-mentioned coupon recommendation methods when executing the computer program.

[0157] As an example, Figure 3 A schematic diagram of the structure of an electronic device applicable to the embodiment of the present application is shown. Figure 3 As shown, the electronic device 2000 includes: a processor 2001 and a memory 2003. The processor 2001 and the memory 2003 are connected, such as through a bus 2002. Optionally, the electronic device 2000 may also include a transceiver 2004. It should be noted that in actual applications, the transceiver 2004 is not limited to one, and the structure of the electronic device 2000 does not constitute a limitation on the embodiments of the present application.

[0158] The processor 2001 is applied in the embodiment of the present application to implement the method shown in the above method embodiment. The transceiver 2004 may include a receiver and a transmitter. The transceiver 2004 is applied in the embodiment of the present application to implement the function of the electronic device of the embodiment of the present application communicating with other devices when executed.

[0159] Processor 2001 may be a CPU (Central Processing Unit), a general purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array) or other programmable logic devices, transistor logic devices, hardware components or any combination thereof. It may implement or execute various exemplary logic blocks, modules and circuits described in conjunction with the disclosure of this application. Processor 2001 may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0160] The bus 2002 may include a path to transmit information between the above components. The bus 2002 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus. The bus 2002 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 3 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0161] The memory 2003 can be a ROM (Read Only Memory) or other types of static storage devices that can store static information and instructions, a RAM (Random Access Memory) or other types of dynamic storage devices that can store information and instructions, or an EEPROM (Electrically Erasable Programmable Read Only Memory), a CD-ROM (Compact Disc Read Only Memory) or other optical disk storage, optical disk storage (including compressed optical disk, laser disk, optical disk, digital versatile disk, Blu-ray disk, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited to these.

[0162] Optionally, the memory 2003 is used to store application code for executing the solution of the present application, and the execution is controlled by the processor 2001. The processor 2001 is used to execute the application code stored in the memory 2003 to implement the investment and financing project recommendation method provided in any embodiment of the present application.

[0163] The electronic device provided in the embodiment of the present application is applicable to any embodiment of the above method and will not be described in detail here.

[0164] A coupon recommendation method provided by the present invention has obvious advantages and beneficial effects compared with the prior art. It uses user-user social networks, user-coupon redemption behavior graphs, user-merchant transaction graphs, and encodes the GPS coordinates of users and merchants as input information of the recommendation model, and uses graph neural networks to extract user preference features from user-user social networks and user-merchant transaction graphs, and then uses graph neural networks to extract hidden features of coupons from user-coupon redemption behavior graphs. The structural information and node information in the network are learned by graph neural networks, and users and coupons are fully characterized, so that the recommendation model has a more efficient improvement accuracy. By adding user-merchant transaction graph information, not only the user's transaction preference characteristics are considered, but also the geographical dimension characteristics are added. In the new user business scenario, due to the lack of new user behavior data, the new user information is aggregated and characterized through transaction information and social network graphs, so as to transmit transaction preferences between users, which solves the cold start problem to a certain extent.

[0165] In one embodiment, a computer-readable storage medium is provided, including computer-readable instructions. When a computer reads and executes the computer-readable instructions, the steps of the coupon recommendation method described above are implemented. The coupon recommendation method provided by the present invention has obvious advantages and beneficial effects compared with the prior art. By using user-user social networks, user-coupon redemption behavior graphs, and user-merchant transaction graphs, the GPS coordinates of users and merchants are encoded as input information of the recommendation model, and the user preference features are extracted from the user-user social network and the user-merchant transaction graph using a graph neural network, and the hidden features of the coupon are extracted from the user-coupon redemption behavior graph using a graph neural network. The structural information and node information in the network are learned by the graph neural network, and the user and coupon are fully characterized, so that the recommendation model has a more efficient improvement accuracy. By adding user-merchant transaction graph information, not only the user's transaction preference characteristics are considered, but also the geographical dimension characteristics are added. In the new user business scenario, due to the lack of new user behavior data, the new user information is aggregated and characterized through transaction information and social network graphs, so as to transmit transaction preferences between users, which solves the cold start problem to a certain extent.

[0166] Through the description of the above implementation methods, technical personnel in the relevant field can understand that for the convenience and simplicity of description, only the division of the above-mentioned functional modules is used as an example. In actual applications, the above-mentioned functions can be assigned to different functional modules as needed, that is, the internal structure of the specific system can be divided into different functional modules to complete all or part of the functions described above.

[0167] Those of ordinary skill in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0168] In several embodiments covered by the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the division of modules is only a logical function division, and there may be other division methods in actual implementation, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or modules, or it can be an electrical, mechanical or other form of connection.

[0169] The modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules, that is, they may be located in one place or distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the embodiments of the present application.

[0170] In addition, each functional module in each embodiment of the present application can be implemented in the form of hardware or in the form of software functional modules. If these functional modules are implemented in the form of software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present application is essentially or part of the prior art that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a computer program product, which includes one or more computer instructions. When loading and executing computer program instructions on a computer, the process or function according to the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer program product is stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, a computer instruction can be transmitted from a website site, a computer, a server or a data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or a data center that includes one or more available media integrations. The available medium may be a magnetic medium (eg, a floppy disk, a hard disk, a magnetic tape), an optical medium (eg, a DVD), or a semiconductor medium (eg, a solid state disk (SSD)).

[0171] The above contents are merely examples and explanations of the structure of the present invention. The technicians in this technical field may make various modifications or additions to the specific embodiments described or replace them in a similar manner. As long as they do not deviate from the structure of the invention or exceed the scope defined by the claims, they should all fall within the protection scope of the present invention.

Claims

1. A coupon recommendation method, characterized in that: include: Build user social networks, user-coupon graphs, and user-merchant graphs; Use the map service to query the GPS location of the merchant's specific address and calculate the representation of the Geohash code; Through the user social network, user-coupon graph, and user-merchant graph, we can get three representations of users, namely h soci 、h uc 、h mech , and the coupons are shown c ; Get the final score s of user u for coupon c *c , define the loss function and use the RMSproo optimizer for model training.

2. The coupon recommendation method according to claim 1, characterized in that: The construction of the user social network, the user-coupon graph, and the user-merchant graph includes: Obtain customer contact data through the customer acquisition system, activity system, and marketing system to build a user social network, denoted as G soci { i |u j >};​ in, i |u j > represents u i and u j There are social connections;​ Get user redemption events from user activity redemption history and build a user-coupon graph, denoted as G coup { i |c i >};​ in, i |c i > represents u i Redeemed the voucher i ;​ Obtain the user-merchant graph from user transaction data, denoted as G mech { i |m i >};​ in i |m i > represents u i and m i There is trading activity.​ 3. The coupon recommendation method according to claim 2, characterized in that: For the user-merchant graph G mech , query the GPS location of the merchant's specific address through the map service, and convert the user GPS and merchant GPS into 5-bit Geohash codes, represented by geo u andge m .

4. The coupon recommendation method according to claim 1, characterized in that: The three representations of users are learned through user social network, user-coupon graph, and user-merchant graph, respectively: soci 、h uc 、h mech , and the coupons are shown c ,include: In the user social network, the input feature representation e of user u is u , u in G soci The set of adjacent users in is called nu, where the feature corresponding to the i-th adjacent user is expressed as After the graph neural network forward aggregation calculation, h soci , the calculation formula is: h soci =σ(W·Agg(e nu )+b); Agg(e nu ) introduces the attention mechanism, and the specific formula of attention weight a is as follows: am i σ(w·(e nu +e u )+b)! Add(e) nu )=and nu ·to; In the user-coupon graph, the user u and coupon c are represented by graph neural aggregation, denoted as h uc and h c ; In G coup In the example, the inputs of user u and the adjacent coupon set are e uc and e c , where the i-th adjacent coupon input of user u is recorded as In the user-merchant graph G mech In the above example, we obtain the corresponding feature representation for user u, denoted as h mech .

5. The coupon recommendation method according to any one of claims 1 to 4, characterized in that: The Geohash code of user u and neighboring merchant set m is represented as e geo ; Take the first 5 digits of Geohash and initialize the vector with dimension d as the initialization vector of Geohash encoding, denoted as e *geo ∈R k,d ; e *geo As input, it passes through a 2-layer MLP (Multi-Layer Perceptron) network to obtain e geo , the specific formula is as follows: e geo =σ2(W2σ1(W1·e *geo +b1)+b2) User-Merchant Graph G mech The input feature of user u in is denoted as e *mechu , and the input features of the neighboring merchants are recorded as e *mech , the feature representation h of user u is calculated by the following method mech : cat is a splicing operation, am i σ(w·(e mech +e mechu )+b)! Add(e) mech u )=and mech u ·to; h mech =σ(W·Agg(e mechu )+b).

6. The coupon recommendation method according to claim 5, characterized in that: According to the above, three representations of users are obtained through user social network, user-coupon graph, and user-merchant graph, respectively: h soci 、h uc 、h mech , and the coupons are shown c , The final representation of user u is h = cat (h soci ,h uc ,h mech ), cat is the splicing operation; h and h c After concatenation, the result is input into a 2-layer MLP network to obtain the score of user u for coupon c: s *c =σ2(W2σ1(W1·cat(h,h c )+b1)+b2) Through the loss function Use RMSproo optimizer for training; Among them, s c Indicates the actual redemption status of coupon c by user u, which is 1 if redeemed and 0 if not.

7. A coupon recommendation system, characterized in that: The system comprises: a data construction unit, a first operation unit, a second operation unit, and a model training unit; Data construction unit: used to build user social networks, user-coupon graphs, and user-merchant graphs; The first computing unit is used to query the GPS location of the specific address of the merchant by using the map service, and calculate the representation of the Geohash code; The second computing unit is used to obtain three representations of users through user social networks, user-coupon graphs, and user-merchant graphs, respectively: h soci 、h uc 、h mech , and the coupons are shown c ; Model training unit: used to obtain the final score s of user u for coupon c *c , define the loss function and use the RMSproo optimizer for model training.

8. A computing device, characterized in that include: A memory for storing program instructions; A processor is used to call the program instructions stored in the memory, and execute the method according to any one of claims 1 to 6 according to the obtained program instructions.

9. A computer-readable storage medium, characterized in that: It includes computer-readable instructions. When a computer reads and executes the computer-readable instructions, it implements the method according to any one of claims 1 to 6. There may be multiple readable media, and the multiple readable media can operate independently of each other.

10. A computer program product, characterized in that The computer program product comprises one or more computer instructions, and when the computer program instructions are loaded and executed on a computer, the method according to any one of claims 1 to 6 is fully or partially executed.

Images