Information recommendation method and device, computer device, and storage medium

Abstract

The application relates to an information recommendation method and device, computer equipment, a storage medium and a computer program product, which can be used in the field of artificial intelligence technology. The method comprises the following steps: determining, from a pre-constructed knowledge graph, associated information of historical behavior information of a to-be-recommended customer as to-be-recommended information; the knowledge graph comprises merchant information and activity information; determining, from the knowledge graph, a multi-layer associated information set corresponding to the to-be-recommended information; obtaining a prediction feature corresponding to each layer of the associated information set through a pre-trained multi-layer prediction model; performing fusion processing on the prediction feature corresponding to each layer of the associated information set to obtain customer features of the to-be-recommended customer; obtaining a recommendation probability of the to-be-recommended information according to the information features of the to-be-recommended information and the customer features of the to-be-recommended customer; and determining target recommended information corresponding to the to-be-recommended customer from the to-be-recommended information according to the recommendation probability. The method can improve the information recommendation accuracy.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to an information recommendation method, apparatus, computer device, storage medium, and computer program product. Background Technology

[0002] With the development of artificial intelligence technology, intelligent recommendation technology has emerged; for example, in financial scenarios, relevant information can be recommended to customers, such as merchant information and event information.

[0003] In traditional technologies, information recommendation is generally based on collaborative filtering, which mainly considers the similarity between customers and information. However, in sparse situations where customer behavior records are limited, the recommended information may not be accurate enough, resulting in a low information recommendation accuracy rate. Summary of the Invention

[0004] Therefore, it is necessary to provide an information recommendation method, apparatus, computer equipment, computer-readable storage medium, and computer program product that can improve the accuracy of information recommendation in order to address the above-mentioned technical problems.

[0005] Firstly, this application provides an information recommendation method. The method includes:

[0006] The relevant information of the historical behavior information of the customers to be recommended is determined from the pre-built knowledge graph and used as the recommendation information; the knowledge graph includes merchant information and activity information.

[0007] From the knowledge graph, a multi-layered set of related information corresponding to the information to be recommended is determined, and the predicted features corresponding to each layer of related information are obtained through a pre-trained multi-layered prediction model.

[0008] The predicted features corresponding to each layer of associated information set are fused to obtain the customer features of the customer to be recommended.

[0009] Based on the information characteristics of the information to be recommended and the customer characteristics of the customer to be recommended, the recommendation probability of the information to be recommended is obtained;

[0010] Based on the recommendation probability, target recommendation information corresponding to the customer to be recommended is determined from the information to be recommended.

[0011] In one embodiment, determining the multi-layered association information set corresponding to the information to be recommended from the knowledge graph, and obtaining the prediction features corresponding to each layer of association information set through a pre-trained multi-layered prediction model, includes:

[0012] From the knowledge graph, determine the set of hierarchical association information corresponding to the information to be recommended;

[0013] The information features of the information to be recommended and the information features of the associated information in the hierarchical association information set are input into a pre-trained multi-layer prediction model to obtain the prediction features corresponding to the hierarchical association information set.

[0014] The associated information in the hierarchical association information set is taken as the new information to be recommended, and the predicted features corresponding to the hierarchical association information set are taken as the information features of the information to be recommended. Then, the process jumps to the step of determining the hierarchical association information set corresponding to the information to be recommended from the knowledge graph, until the predicted features corresponding to the obtained hierarchical association information set are the predicted features corresponding to the Nth layer association information set; N is a positive integer greater than or equal to 2.

[0015] In one embodiment, before determining the association information of the historical behavior information of the customer to be recommended from the pre-built knowledge graph as the recommendation information, the method further includes:

[0016] Obtain historical customer behavior information;

[0017] Extract the merchant information and activity information that the customer has interacted with in the past from the historical behavior information;

[0018] Identify a first association between the merchant information and the activity information, a second association between the merchant information, and a third association between the activity information;

[0019] Based on the first association, the second association, and the third association, a knowledge graph is constructed as the pre-constructed knowledge graph.

[0020] In one embodiment, the pre-trained multilayer prediction model is trained in the following manner:

[0021] Extract the first and second information that the historical customers have interacted with sequentially from their historical behavior information;

[0022] The associated information of the first information is determined from the pre-constructed knowledge graph and used as sample information;

[0023] From the knowledge graph, a set of multi-layer sample association information corresponding to the sample information is determined. Through the multi-layer prediction model to be trained, the prediction features corresponding to each set of sample association information are obtained.

[0024] The predicted features corresponding to the sample association information set of each layer are fused to obtain the customer features of the historical customer.

[0025] Based on the information features of the sample information and the customer features of the historical customers, the recommendation probability of the sample information is obtained;

[0026] Based on the recommendation probability of the sample information, target information corresponding to the historical customer is determined from the sample information;

[0027] Based on the difference between the target information and the second information, the multi-layer prediction model to be trained is trained to obtain a trained multi-layer prediction model, which is used as the pre-trained multi-layer prediction model.

[0028] In one embodiment, determining a multi-layer sample association information set corresponding to the sample information from the knowledge graph, and obtaining the prediction features corresponding to each layer of sample association information set through a multi-layer prediction model to be trained, includes:

[0029] From the knowledge graph, determine the set of hierarchical sample association information corresponding to the sample information;

[0030] When the number of sample association information in the hierarchical sample association information set is less than or equal to a preset number, the information features of the sample information and the information features of the sample association information in the hierarchical sample association information set are input into the multi-layer prediction model to be trained to obtain the prediction features corresponding to the hierarchical sample association information set.

[0031] The sample association information in the hierarchical sample association information set is used as the new sample information, and the prediction feature corresponding to the hierarchical sample association information set is used as the information feature of the sample information. Then, the process jumps to the step of determining the hierarchical sample association information set corresponding to the sample information from the knowledge graph, until the prediction feature corresponding to the obtained hierarchical sample association information set is the prediction feature corresponding to the Nth layer sample association information set.

[0032] In one embodiment, the method further includes:

[0033] If the number of sample association information in the hierarchical sample association information set is greater than the preset number, the current customer characteristics of the historical customer are determined based on the predicted features corresponding to the obtained hierarchical sample association information set.

[0034] Based on the information features of the sample association information and the current customer features of the historical customers, the recommendation probability of the sample association information is obtained;

[0035] In the hierarchical sample association information set, the preset number of sample association information with the highest recommendation probability are retained to obtain a new hierarchical sample association information set;

[0036] The information features of the sample information and the information features of the sample association information in the new hierarchical sample association information set are input into the multi-layer prediction model to be trained to obtain the prediction features corresponding to the new hierarchical sample association information set.

[0037] In one embodiment, determining the target recommendation information corresponding to the customer to be recommended from the information to be recommended based on the recommendation probability includes:

[0038] From the information to be recommended, information with a recommendation probability greater than a preset probability is selected as the target recommendation information;

[0039] The method further includes:

[0040] The target recommendation information is stored in the recommendation list corresponding to the customer to be recommended;

[0041] According to the recommendation list, the target recommendation information is recommended to the customer to be recommended.

[0042] Secondly, this application also provides an information recommendation device. The device includes:

[0043] The information determination module is used to determine the related information of the historical behavior information of the customer to be recommended from the pre-built knowledge graph, and use it as the information to be recommended; the knowledge graph includes merchant information and activity information;

[0044] The feature determination module is used to determine the multi-layer association information set corresponding to the information to be recommended from the knowledge graph, and obtain the predicted features corresponding to each layer association information set through a pre-trained multi-layer prediction model.

[0045] The fusion processing module is used to fuse the predicted features corresponding to each layer of related information set to obtain the customer features of the customer to be recommended.

[0046] The probability determination module is used to obtain the recommendation probability of the information to be recommended based on the information features of the information to be recommended and the customer features of the customer to be recommended.

[0047] The information recommendation module is used to determine the target recommendation information corresponding to the customer to be recommended from the information to be recommended based on the recommendation probability.

[0048] Thirdly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to perform the following steps:

[0049] The relevant information of the historical behavior information of the customers to be recommended is determined from the pre-built knowledge graph and used as the recommendation information; the knowledge graph includes merchant information and activity information.

[0050] From the knowledge graph, a multi-layered set of related information corresponding to the information to be recommended is determined, and the predicted features corresponding to each layer of related information are obtained through a pre-trained multi-layered prediction model.

[0051] The predicted features corresponding to each layer of associated information set are fused to obtain the customer features of the customer to be recommended.

[0052] Based on the information characteristics of the information to be recommended and the customer characteristics of the customer to be recommended, the recommendation probability of the information to be recommended is obtained;

[0053] Based on the recommendation probability, target recommendation information corresponding to the customer to be recommended is determined from the information to be recommended.

[0054] Fourthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, performs the following steps:

[0055] The relevant information of the historical behavior information of the customers to be recommended is determined from the pre-built knowledge graph and used as the recommendation information; the knowledge graph includes merchant information and activity information.

[0056] From the knowledge graph, a multi-layered set of related information corresponding to the information to be recommended is determined, and the predicted features corresponding to each layer of related information are obtained through a pre-trained multi-layered prediction model.

[0057] The predicted features corresponding to each layer of associated information set are fused to obtain the customer features of the customer to be recommended.

[0058] Based on the information characteristics of the information to be recommended and the customer characteristics of the customer to be recommended, the recommendation probability of the information to be recommended is obtained;

[0059] Based on the recommendation probability, target recommendation information corresponding to the customer to be recommended is determined from the information to be recommended.

[0060] Fifthly, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, performs the following steps:

[0061] The relevant information of the historical behavior information of the customers to be recommended is determined from the pre-built knowledge graph and used as the recommendation information; the knowledge graph includes merchant information and activity information.

[0062] From the knowledge graph, a multi-layered set of related information corresponding to the information to be recommended is determined, and the predicted features corresponding to each layer of related information are obtained through a pre-trained multi-layered prediction model.

[0063] The predicted features corresponding to each layer of associated information set are fused to obtain the customer features of the customer to be recommended.

[0064] Based on the information characteristics of the information to be recommended and the customer characteristics of the customer to be recommended, the recommendation probability of the information to be recommended is obtained;

[0065] Based on the recommendation probability, target recommendation information corresponding to the customer to be recommended is determined from the information to be recommended.

[0066] The aforementioned information recommendation method, apparatus, computer equipment, storage medium, and computer program product first determine the related information of the historical behavior information of the customer to be recommended from a pre-constructed knowledge graph, which serves as the information to be recommended. The knowledge graph includes merchant information and activity information. Then, a multi-layered set of related information corresponding to the information to be recommended is determined from the knowledge graph. Through a pre-trained multi-layered prediction model, the prediction features corresponding to each layer of related information are obtained. Next, the prediction features corresponding to each layer of related information are fused to obtain the customer features of the customer to be recommended. Finally, based on the information features of the information to be recommended and the customer features of the customer to be recommended, the recommendation probability of the information to be recommended is obtained, and based on the recommendation probability, the target recommended information corresponding to the customer to be recommended is determined from the information to be recommended. In this way, when making information recommendations, a pre-constructed knowledge graph is first used to determine the information to be recommended to the customer and the multi-layered set of related information corresponding to the information to be recommended. Then, a multi-layered prediction model is used to output the predicted features corresponding to each layer of related information. These predicted features are then fused to obtain the customer features of the customer to be recommended. That is, by using the knowledge graph and the multi-layered prediction model to expand the historical behavior information of the customer to be recommended, the customer features of the customer to be recommended can be made denser, thereby enriching the customer-side features. This makes the recommendation probability obtained based on the information features of the information to be recommended and the customer features of the customer to be recommended more accurate, thus making the target recommended information determined based on the recommendation probability more accurate. In this way, the accuracy of information recommendation is improved, avoiding the defect of low information recommendation accuracy caused by sparse customer behavior. Attached Figure Description

[0067] Figure 1 This is a flowchart illustrating an information recommendation method in one embodiment;

[0068] Figure 2 This is a schematic diagram of multi-level prediction in a multi-level prediction model in one embodiment;

[0069] Figure 3 This is a flowchart illustrating the steps for obtaining the predicted features corresponding to each layer of associated information set in one embodiment.

[0070] Figure 4 This is a flowchart illustrating the training steps of a multi-layer prediction model in one embodiment.

[0071] Figure 5 This is a flowchart illustrating the information recommendation method in another embodiment;

[0072] Figure 6 This is a flowchart illustrating a knowledge graph-based marketing campaign recommendation method in one embodiment;

[0073] Figure 7This is a structural block diagram of an information recommendation device in one embodiment;

[0074] Figure 8 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0075] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0076] It should be noted that the information recommendation method, apparatus, computer equipment, storage medium, and computer program products provided in this application can be used in the field of fintech, such as accurately determining the target recommendation information for the customer to be recommended based on the customer's historical behavior information, such as activity information, merchant information, etc., thereby achieving high information recommendation accuracy; they can also be used in other related fields, such as in the field of artificial intelligence technology, using knowledge graphs and multi-layer prediction models to automatically determine the target recommendation information for the customer to be recommended, thereby achieving the purpose of intelligent information recommendation.

[0077] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data shall comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0078] In one embodiment, such as Figure 1 As shown, an information recommendation method is provided. This embodiment illustrates the application of this method to a server. It is understood that this method can also be applied to a terminal, or to a system including both a terminal and a server, and is implemented through interaction between the terminal and the server. The terminal can be, but is not limited to, various personal computers, laptops, smartphones, tablets, etc.; the server refers to a bank server, which can be implemented using a standalone server or a server cluster composed of multiple servers. In this embodiment, the method includes the following steps:

[0079] Step S101: Determine the related information of the historical behavior information of the customers to be recommended from the pre-built knowledge graph, and use it as the information to be recommended; the knowledge graph includes merchant information and activity information.

[0080] The knowledge graph includes multiple merchant and activity information entries, used to represent the relationships between merchant and activity information, between different merchant information entries, and between different activity entries. Furthermore, in the knowledge graph, both activity and merchant information are represented as nodes, and relationships are represented as edges. For example, assuming a relationship exists between activity information A1 and merchant information B1, then in the knowledge graph, there is a connecting edge between the node corresponding to activity information A1 and the node corresponding to merchant information B1. In practical scenarios, the pre-built knowledge graph refers to the customer consumption merchant knowledge graph.

[0081] It should be noted that the knowledge graph is built based on the historical behavioral information of multiple customers.

[0082] Merchant information, used to represent merchants, can refer to merchant name, merchant address, etc. Activity information, used to represent activities, can refer to activity name, activity address, activity details, etc. In practical scenarios, merchant information refers to merchant information, and activity information refers to marketing activity information.

[0083] Among them, "customers to be recommended" refers to authorized customers for whom recommendation information is currently required. The historical behavior information of customers to be recommended refers to their historical consumption and / or participation information; from the historical consumption information, we can learn about the merchants the customer has previously patronized (i.e., historical merchant information), and from the historical participation information, we can learn about the activities the customer has previously participated in (i.e., historical activity information).

[0084] It should be noted that the historical behavior information of the customers to be recommended falls into three categories: those that only include historical merchant information (such as which merchants they have visited), those that only include historical activity information (such as which activities they have participated in), and those that include both historical merchant information and historical activity information (such as visiting a certain merchant and participating in a certain activity of that merchant).

[0085] Among them, "related information" refers to information that is associated with the historical merchant information and / or historical activity information of the customer to be recommended. Specifically, it refers to information in the knowledge graph that has connection edges with the historical merchant information and / or historical activity information of the customer to be recommended, such as merchant information and activity information. "Information to be recommended" refers to information in the knowledge graph that has connection edges with the historical merchant information and / or historical activity information of the customer to be recommended.

[0086] Specifically, in response to an information recommendation request, the server retrieves the historical behavior information of the customer to be recommended from the database; it extracts historical consumption information and historical participation information from this information, identifying historical merchant information from the historical consumption information and historical activity information from the historical participation information. Next, the server retrieves a pre-built knowledge graph from the database and identifies information with connections to historical merchant information and / or historical activity information from the knowledge graph, using this as the association information for the customer's historical behavior information; finally, this association information is confirmed as the recommendation information for the customer.

[0087] For example, suppose the historical behavior information of the customer to be recommended includes historical activity information A2, and in the knowledge graph, the information connected to historical activity information A2 are activity information A3, activity information A4, merchant information B2, merchant information B3, and merchant information B4, then the information to be recommended refers to activity information A3, activity information A4, merchant information B2, merchant information B3, and merchant information B4.

[0088] Step S102: From the knowledge graph, determine the multi-layered association information set corresponding to the information to be recommended, and obtain the prediction features corresponding to each layer of association information set through a pre-trained multi-layered prediction model.

[0089] Each layer of related information includes multiple related information sets. (Reference) Figure 2 The first-level set of related information includes the related information of the information to be recommended in the knowledge graph (i.e., information in the knowledge graph associated with the information to be recommended), specifically, information in the knowledge graph that has a connection edge with the information to be recommended. The second-level set of related information includes the related information of the information included in the first-level set of related information in the knowledge graph, specifically, information in the knowledge graph that has a connection edge with the related information included in the first-level set of related information. The third-level set of related information includes the related information of the information included in the second-level set of related information in the knowledge graph, specifically, information in the knowledge graph that has a connection edge with the related information included in the second-level set of related information. And so on, resulting in multi-level sets of related information.

[0090] It should be noted that the number of layers in the associated information set can be a preset number or adjusted according to the actual situation.

[0091] The predicted features corresponding to each layer of related information set refer to the sub-customer features of the customers to be recommended, which are extended from each layer of related information set.

[0092] Among them, a pre-trained multi-layer prediction model refers to a prediction model that can output the prediction features corresponding to the set of related information at each layer, such as a multi-layer inference model.

[0093] Specifically, the server determines the related information of the information to be recommended from the knowledge graph, and combines this related information to obtain the first-level related information set. Then, it determines the related information of the related information in the first-level related information set from the knowledge graph, and combines them again to obtain the second-level related information set. Next, it determines the related information of the related information in the second-level related information set from the knowledge graph, and combines them again to obtain the third-level related information set. This process continues until multiple levels of related information sets corresponding to the information to be recommended are obtained. Simultaneously, the server inputs the related information from each level of related information set into a pre-trained multi-level prediction model, which then outputs the predicted features corresponding to each level of related information set.

[0094] Step S103: The predicted features corresponding to each layer of related information set are fused to obtain the customer features of the customer to be recommended.

[0095] Among them, customer features are used to characterize the customer information of the customer to be recommended. They can be represented by customer vectors, which are obtained by fusing the predicted features corresponding to each layer of related information sets.

[0096] It should be noted that by fusing the predicted features corresponding to each layer of related information sets, the sub-customer features of the customers to be recommended, which are extended from each layer of related information sets, can be spliced ​​together to obtain the customer features of the customers to be recommended. This achieves the goal of expanding the historical behavior information of the customers to be recommended when the historical behavior information of the customers to be recommended is sparse, making the customer features of the customers to be recommended more dense, thereby enriching the customer-side features.

[0097] Specifically, the server concatenates the predicted features corresponding to each layer of association information set to obtain concatenated features, and confirms the concatenated features as the customer features of the customers to be recommended. For example, if the predicted features corresponding to each layer of association information set are U1, U2, U3, ... Un, then the customer features of the customers to be recommended are U = U1 + U2 + U3 + ... + Un.

[0098] Step S104: Based on the information characteristics of the information to be recommended and the customer characteristics of the customer to be recommended, obtain the recommendation probability of the information to be recommended.

[0099] Among them, the information features of the information to be recommended are used to characterize the information content of the information to be recommended, and can be represented by information vectors.

[0100] The recommendation probability is used to measure the likelihood that a customer to be recommended will take action on the recommended information.

[0101] Specifically, the server performs feature extraction on the information to be recommended, obtaining its information features. These information features, along with the customer features of the customers to be recommended, are then input into a recommendation probability prediction model. The model processes these features to obtain the recommendation probability of the information. This recommendation probability prediction model is a model that calculates the recommendation probability of information based on its information features and the customer features.

[0102] Furthermore, the server can also calculate the feature similarity between the information features of the information to be recommended and the customer features of the customer to be recommended, as the similarity between the information to be recommended and the customer to be recommended; based on the similarity, query the correspondence between the similarity and the recommendation probability of the information (e.g., the higher the similarity, the greater the recommendation probability), and obtain the recommendation probability of the information to be recommended.

[0103] Step S105: Based on the recommendation probability, determine the target recommendation information corresponding to the customer to be recommended from the information to be recommended.

[0104] Among them, target recommendation information refers to information that meets the needs of the customers to be recommended, such as information about merchants that the customers to be recommended may consume, and information about activities that the customers to be recommended may participate in.

[0105] Specifically, the server selects a preset number (e.g., the top 5) of information with the highest recommendation probability from the information to be recommended, and uses them as target recommendation information corresponding to the customers to be recommended. This facilitates the subsequent recommendation of the target information to the customers to be recommended, thereby achieving accurate information recommendation.

[0106] In the information recommendation method provided in the above embodiments, the historical behavior information of the customer to be recommended is first determined from a pre-constructed knowledge graph as the information to be recommended. The knowledge graph includes merchant information and activity information. Then, a multi-layered set of related information corresponding to the information to be recommended is determined from the knowledge graph. Through a pre-trained multi-layered prediction model, the prediction features corresponding to each layer of related information are obtained. Next, the prediction features corresponding to each layer of related information are fused to obtain the customer features of the customer to be recommended. Finally, based on the information features of the information to be recommended and the customer features of the customer to be recommended, the recommendation probability of the information to be recommended is obtained. Based on the recommendation probability, the target recommendation information corresponding to the customer to be recommended is determined from the information to be recommended. In this way, when making information recommendations, a pre-constructed knowledge graph is first used to determine the information to be recommended to the customer and the multi-layered set of related information corresponding to the information to be recommended. Then, a multi-layered prediction model is used to output the predicted features corresponding to each layer of related information. These predicted features are then fused to obtain the customer features of the customer to be recommended. That is, by using the knowledge graph and the multi-layered prediction model to expand the historical behavior information of the customer to be recommended, the customer features of the customer to be recommended can be made denser, thereby enriching the customer-side features. This makes the recommendation probability obtained based on the information features of the information to be recommended and the customer features of the customer to be recommended more accurate, thus making the target recommended information determined based on the recommendation probability more accurate. In this way, the accuracy of information recommendation is improved, avoiding the defect of low information recommendation accuracy caused by sparse customer behavior.

[0107] In one embodiment, such as Figure 3 As shown, step S102 above, which determines the multi-layered association information set corresponding to the information to be recommended from the knowledge graph, and obtains the prediction features corresponding to each layer of association information set through a pre-trained multi-layered prediction model, specifically includes the following steps:

[0108] Step S301: Determine the set of hierarchical related information corresponding to the information to be recommended from the knowledge graph.

[0109] Step S302: Input the information features of the information to be recommended and the information features of the related information in the hierarchical association information set into the pre-trained multi-layer prediction model to obtain the prediction features corresponding to the hierarchical association information set.

[0110] Step S303: Take the associated information in the hierarchical association information set as the new information to be recommended, take the predicted features corresponding to the hierarchical association information set as the information features of the information to be recommended, and jump to step S301 until the predicted features corresponding to the obtained hierarchical association information set are the predicted features corresponding to the Nth layer association information set.

[0111] Where N is a positive integer greater than or equal to 2.

[0112] Among them, the hierarchical association information set refers to the first-level association information set, the second-level association information set, ... the Nth-level association information set.

[0113] Among them, the multi-level prediction model can predict the sub-behavioral information of the customer to be recommended based on the correlation between the information features of the information to be recommended and the information features of the related information in the hierarchical related information set, thereby outputting the sub-customer features of the customer to be recommended (represented by predicted features), which is beneficial for expanding the historical behavioral information of the customer to be recommended and making the customer features of the customer to be recommended more dense.

[0114] It should be noted that during the loop, the information to be recommended in step S301 will be replaced with the association information in the hierarchical association information set obtained in the previous round, and the information features of the information to be recommended in step S302 will be replaced with the predicted features corresponding to the hierarchical association information set obtained in the previous round.

[0115] Specifically, the server determines the related information of the information to be recommended from the knowledge graph; combines the related information of the information to be recommended to obtain the first-layer related information set; obtains the information features of the information to be recommended and the information features of the related information in the first-layer related information set, and inputs the information features of the information to be recommended and the information features of the related information in the first-layer related information set into a pre-trained multi-layer prediction model to obtain the prediction features corresponding to the first-layer related information set; for example, the server uses the multi-layer prediction model to calculate the feature similarity between the information features of the information to be recommended and the information features of the related information in the first-layer related information set, and uses the correspondence between feature similarity and weight to determine the weights corresponding to the information features of the related information in the first-layer related information set; using the weights corresponding to the information features of the related information in the first-layer related information set, the information features of the related information in the first-layer related information set are fused to obtain fused features, which are used as the prediction features corresponding to the first-layer related information set. Next, the server identifies the relationships between the first-layer relationships in the knowledge graph and combines them to obtain the second-layer relationship set. It then extracts the information features of the relationships in the second-layer relationship set and inputs the predicted features corresponding to the first-layer and second-layer relationships into a pre-trained multi-layer prediction model to obtain the predicted features corresponding to the second-layer relationship set. This process is repeated until the Nth-layer relationship set and its corresponding predicted features are obtained.

[0116] Furthermore, when the number of related information in the hierarchical related information set is greater than a preset number, the current customer characteristics of the customer to be recommended are determined based on the predicted features corresponding to the obtained hierarchical related information set; the recommendation probability of the related information in the hierarchical related information set is obtained based on the information features of the related information in the hierarchical related information set and the current customer characteristics of the customer to be recommended; a preset number (e.g., the top 5) of related information with the highest recommendation probabilities in the hierarchical related information set are retained to obtain a new hierarchical related information set; the information features of the information to be recommended and the information features of the related information in the new hierarchical related information set are input into a pre-trained multi-layer prediction model to obtain the predicted features corresponding to the new hierarchical related information set, which are used as the predicted features of the related information set of the corresponding layer.

[0117] In this embodiment, by determining the multi-layered association information set corresponding to the information to be recommended from the knowledge graph, and by using a pre-trained multi-layered prediction model, the predicted features corresponding to each layer of association information set are obtained. The predicted features corresponding to each layer of association information set can characterize the sub-customer features of the customer to be recommended. This is beneficial for expanding the historical behavior information of the customer to be recommended, making the expanded behavior information of the customer to be recommended more dense, thereby enriching the customer-side features and making the recommendation probability obtained based on the information features of the information to be recommended and the customer features of the customer to be recommended more accurate.

[0118] In one embodiment, before determining the related information of the historical behavior information of the customer to be recommended from the pre-built knowledge graph as the recommended information, step S101 above further includes the following: obtaining the historical behavior information of the historical customer; extracting the merchant information and activity information that the historical customer has interacted with from the historical behavior information; identifying the first association between the merchant information and the activity information, the second association between the merchant information, and the third association between the activity information; and constructing a knowledge graph based on the first association, the second association, and the third association as the pre-built knowledge graph.

[0119] Among them, the first association relationship is used to indicate that there is a relationship between merchant information and activity information, the second association relationship is used to indicate that there is a relationship between merchant information and merchant information, and the third association relationship is used to indicate that there is a relationship between activity information and activity information.

[0120] Specifically, the server retrieves historical customer behavior information from the database; from this historical behavior information, it extracts merchant information that the historical customer has consumed and activity information that the historical customer has participated in; it identifies the merchant information and activity information to obtain the first association between merchant information and activity information, the second association between merchant information and activity information, and the third association between activity information and activity information; using the merchant information and activity information as nodes, and utilizing the first, second, and third associations, it establishes connection edges between nodes corresponding to related merchant information and activity information, between nodes corresponding to related merchant information, and between nodes corresponding to related activity information, thereby obtaining a knowledge graph, which serves as a pre-constructed knowledge graph.

[0121] For example, the server retrieves a customer's historical spending data, extracts the merchants the customer has visited and the marketing activities they have participated in as nodes, and extracts relevant information such as region and amount as attributes of the nodes. At the same time, based on the relationship between different merchants and activities, it transforms the data into triplet data through embedding.<h,r,t> The data is stored; where h refers to the head node, t refers to the tail node, and r refers to the relationship between h and t. Finally, the server constructs a customer-merchant knowledge graph based on the triple data and maps it to a vector space; in the constructed knowledge graph, there can be relationships between merchants and activities, relationships between activities, and relationships between merchants.

[0122] In this embodiment, historical customer behavior information is obtained, and the first association between merchant information and activity information, the second association between merchant information, and the third association between activity information are extracted from the historical behavior information. These associations are then used to construct a knowledge graph, which is beneficial for subsequent use of the knowledge graph and multi-layer prediction models to expand the historical behavior information of the customers to be recommended, making the behavior information of the customers to be recommended more dense.

[0123] In one embodiment, such as Figure 4 As shown, the information recommendation method provided in this application also includes a training step of a pre-trained multi-layer prediction model, specifically including the following steps:

[0124] Step S401: Extract the first and second information that the historical customer has interacted with sequentially from the historical customer behavior information.

[0125] The first information refers to the merchant information or activity information that the customer has previously consumed from or participated in, while the second information refers to the merchant information or activity information that the customer has subsequently consumed from or participated in.

[0126] Step S402: Determine the related information of the first information from the pre-constructed knowledge graph, and use it as sample information.

[0127] Among them, the associated information of the first information refers to the information in the knowledge graph that is associated with the first information, specifically the information in the knowledge graph that has a connection edge with the first information.

[0128] Step S403: From the knowledge graph, determine the multi-layer sample association information set corresponding to the sample information, and obtain the prediction features corresponding to each layer sample association information set through the multi-layer prediction model to be trained.

[0129] Step S404: The predicted features corresponding to the sample association information set of each layer are fused to obtain the customer features of historical customers.

[0130] Step S405: Based on the information features of the sample information and the customer features of historical customers, obtain the recommendation probability of the sample information.

[0131] Step S406: Based on the recommendation probability of the sample information, determine the target information corresponding to the historical customers from the sample information.

[0132] Among them, target information refers to the information with the highest recommendation probability in the sample information.

[0133] Step S407: Based on the difference between the target information and the second information, train the multi-layer prediction model to be trained to obtain the trained multi-layer prediction model, which serves as the pre-trained multi-layer prediction model.

[0134] Specifically, the server retrieves historical customer behavior information from the database and extracts the first and second information that the historical customer has interacted with sequentially from this information. From a pre-built knowledge graph, it identifies information with connections to the first information, using these as the associated information of the first sample information, and confirms them as sample information. Then, the server determines a multi-layered set of sample association information corresponding to the sample information from the knowledge graph. Using a multi-layered prediction model to be trained, it obtains the predicted features corresponding to each layer of sample association information. The predicted features corresponding to each layer of sample association information are concatenated to obtain concatenated features, which serve as the customer features of the historical customer. Feature extraction is performed on the sample information to obtain its information features. The feature similarity between the information features of the sample information and the customer features of the historical customer is calculated as the similarity between the sample information and the historical customer, and the recommendation probability corresponding to the similarity is obtained as the recommendation probability of the sample information. Finally, the server selects the information with the highest recommendation probability from the sample information as the target information corresponding to historical customers. Based on the difference between the target information and the second information, the multi-layer prediction model to be trained is iteratively trained until the training termination condition is met, such as the loss value being less than a preset threshold. The multi-layer prediction model that meets the training termination condition is then taken as the trained multi-layer prediction model, thus obtaining the pre-trained multi-layer prediction model. For example, the server calculates the loss value based on the difference between the target information and the second information, combined with the loss function. If the loss value is greater than or equal to the preset threshold, the model parameters of the multi-layer prediction model to be trained are adjusted using the loss value, and steps S403 to S407 are repeated to train the adjusted multi-layer prediction model until the loss value obtained from the trained multi-layer prediction model is less than the preset threshold. Then, training stops, and the trained multi-layer prediction model is taken as the trained multi-layer prediction model.

[0135] For example, after the knowledge graph is constructed, the server builds a training sample set for a multi-layered inference model based on customers' historical consumption data, and then uses this training sample set to train the model. First, the historical consumption data of customers in the training sample set is randomly sampled to obtain corresponding merchant nodes and activity nodes. These merchant nodes and activity nodes are then used to expand into multiple adjacent nodes, such as nodes adjacent to merchant nodes or activity nodes in the knowledge graph. Each adjacent node is used as an input node for the model, and its vector representation is defined as v, where v∈R. d Then, the server performs multi-level reasoning based on the input nodes. Each reasoning iteration yields a new set of nodes for the next level. The triples in the node set are (h... i ,r i ,t i ), h i Represents the head node, r i To indicate a relation, t i The tail node is defined. The head node h is defined. i The corresponding head vector is H i Tail node t i The corresponding tail vector is T i H i ,T i ∈R d The relation matrix is ​​R. i R i ∈R d×d d represents the dimension. The relation matrix refers to the relationship matrix between the head node and other nodes in the knowledge graph. If the value of two nodes in the relation matrix is ​​1, it means that there is a relationship between these two nodes. Next, the server calculates the output vector of the first-level inference result using the following formula:

[0136]

[0137]

[0138] in, This represents the output vector of the first-level deduction result, used to assist in calculating the output vector of the second-level deduction result and the final client vector u; This indicates that v is represented by the relation matrix R. i The first-level node set is obtained; similarly, the second-level node set is obtained. By combining the first-level nodes Each node in the algorithm is used as the head node, and the results are calculated using the relation matrix. Specifically, the input nodes are used to obtain the first-level node set through the relation matrix, and this first-level node set is then used to obtain the second-level node set based on the relation matrix. In the second calculation, the output vector of the first-level calculation result is... As the input vector for the second round of calculation, it is used to replace v, and to utilize the second-level node set. Replace the first-level node set

[0139] Then, the server calculates the client vector u based on the output vectors of the calculation results from each layer. The specific calculation formula is as follows:

[0140]

[0141] in, This represents the output vector of the calculation result of the nth layer.

[0142] Next, the server uses a prediction function to calculate the customer vector u and the vector v of the input nodes (i.e., neighboring nodes) to predict the customer's potential consumption probability at merchant nodes or activity nodes. The formula for the prediction function is as follows:

[0143]

[0144]

[0145] Finally, the server uses the following loss function to train the multi-layer inference model to be trained, and obtains a high-precision multi-layer inference model through iteration:

[0146]

[0147] Where minL represents the loss value, y uv This indicates the actual interaction between the customer and the merchant or activity. I represents the predicted consumption probability. r R is a slice of the relation tensor in the knowledge graph, where V and E are the embedding matrices of entities in the knowledge graph, R is the embedding matrix of relations, and λ1 and λ2 represent different coefficients.

[0148] In this embodiment, historical customer behavior information is used in conjunction with a pre-built knowledge graph to iteratively train the multi-layer prediction model to be trained. This makes the predicted features output by the multi-layer prediction model after training more accurate, thereby making the customer features obtained based on the fusion of predicted features more accurate and further improving the accuracy of customer feature determination.

[0149] In one embodiment, step S403, which involves determining a multi-layer sample association information set corresponding to the sample information from the knowledge graph, and obtaining the prediction features corresponding to each layer of sample association information set through a multi-layer prediction model to be trained, specifically includes the following: determining a hierarchical sample association information set corresponding to the sample information from the knowledge graph; when the number of sample association information in the hierarchical sample association information set is less than or equal to a preset number, inputting the information features of the sample information and the information features of the sample association information in the hierarchical sample association information set into the multi-layer prediction model to be trained to obtain the prediction features corresponding to the hierarchical sample association information set; using the sample association information in the hierarchical sample association information set as new sample information, using the prediction features corresponding to the hierarchical sample association information set as the information features of the sample information, and then jumping to the step of determining the hierarchical sample association information set corresponding to the sample information from the knowledge graph, until the prediction features corresponding to the obtained hierarchical sample association information set are the prediction features corresponding to the Nth layer of sample association information set.

[0150] Among them, the set of hierarchical sample association information corresponding to the sample information consists of information in the knowledge graph that has connection edges with the sample information.

[0151] Among them, a preset number is used to limit the number of sample association information in the obtained hierarchical sample association information set, so as to avoid a large amount of sample association information in the hierarchical sample association information set, resulting in a large amount of data to be calculated.

[0152] Specifically, the server determines the association information of sample information from the knowledge graph; combines the association information of sample information to obtain the corresponding hierarchical sample association information set; determines whether the number of sample association information in the hierarchical sample association information set is less than or equal to a preset number; if so, it obtains the information features of the sample information and the information features of the sample association information in the hierarchical sample association information set, and inputs the information features of the sample information and the information features of the sample association information in the hierarchical sample association information set into the multi-level prediction model to be trained, to obtain the prediction features corresponding to the hierarchical sample association information set; for example, the server uses the multi-level prediction model to be trained to calculate the feature similarity between the information features of the sample information and the information features of the sample association information in the hierarchical association information set, and uses the correspondence between feature similarity and weight to determine the weight corresponding to the information features of the sample association information in the hierarchical sample association information set; using the weight corresponding to the information features of the sample association information in the hierarchical sample association information set, it performs fusion processing on the information features of the sample association information in the hierarchical sample association information set to obtain fused features, which are used as the prediction features corresponding to the hierarchical sample association information set, thereby completing one round of prediction. Next, the server uses the sample association information in the hierarchical sample association information set as the new sample information, and the prediction features corresponding to the hierarchical sample association information set as the information features of the sample information, and continues to execute the above process until the obtained hierarchical sample association information set is the Nth layer sample association information set, and the obtained hierarchical sample association information set is the prediction feature corresponding to the Nth layer sample association information set.

[0153] In this embodiment, a multi-layer sample association information set corresponding to the sample information is determined from the knowledge graph, and the prediction features corresponding to each layer sample association information set are obtained through the multi-layer prediction model to be trained. This is beneficial for the subsequent fusion processing of the prediction features corresponding to each layer sample association information set to obtain the customer features of historical customers. This achieves the purpose of expanding the sparse behavioral information of historical customers and making the customer features of historical customers more dense.

[0154] In one embodiment, after determining the hierarchical sample association information set corresponding to the sample information from the knowledge graph, the following steps are also included: if the number of sample association information in the hierarchical sample association information set is greater than a preset number, determine the current customer characteristics of the historical customer based on the prediction features corresponding to the obtained hierarchical sample association information set; obtain the recommendation probability of the sample association information based on the information features of the sample association information and the current customer characteristics of the historical customer; retain a preset number of sample association information with the highest recommendation probabilities in the hierarchical sample association information set to obtain a new hierarchical sample association information set; input the information features of the sample information and the information features of the sample association information in the new hierarchical sample association information set into the multi-layer prediction model to be trained to obtain the prediction features corresponding to the new hierarchical sample association information set.

[0155] The predicted features corresponding to the obtained hierarchical sample association information set refer to the predicted features corresponding to the sample association information sets of each layer that have already been predicted. For example, if the current prediction is for the third layer, then the predicted features corresponding to the obtained hierarchical sample association information set refer to the predicted features corresponding to the first layer sample association information set and the second layer sample association information set.

[0156] Specifically, when the number of sample association information in the hierarchical sample association information set (e.g., 7) is greater than a preset number (e.g., 5), the server concatenates the predicted features corresponding to the obtained hierarchical sample association information set to obtain the concatenated features, which serve as the current customer features of historical customers; it obtains the feature similarity between the information features of sample association information and the current customer features of historical customers, which serves as the similarity between sample association information and historical customers; based on the similarity, it queries the correspondence between similarity and recommendation probability to obtain the recommendation probability of sample association information; and it further refines the hierarchical sample association information set. In the associated information set, a predetermined number of sample associations with the highest recommendation probabilities are retained to obtain a new hierarchical sample association information set. For example, the server arranges the sample associations in the hierarchical sample association information set (I1, I2, I3, I4, I5, I6, I7) in descending order of recommendation probability, resulting in the arranged sample associations I7, I6, I5, I4, I3, I2, I1. Since the predetermined number is 5, the first 5 sample associations are retained, resulting in a new hierarchical sample association information set (I7, I6, I5, I4, I3). Finally, the server obtains the information features of the sample information and the information features of the sample associations in the new hierarchical sample association information set, and inputs these features into the multi-layer prediction model to be trained to obtain the prediction features corresponding to the new hierarchical sample association information set, which serve as the prediction features for the sample association information set of the corresponding layer.

[0157] For example, during training, since each layer of nodes can calculate multiple related nodes based on the relation matrix, the number of nodes may grow exponentially when there are many layers. Therefore, the following limitation is made on the number of nodes: a limit value L is set to limit the number of new nodes obtained in each prediction to prevent the amount of data to be calculated from growing exponentially; when the number of predicted nodes in a certain layer of inference is greater than the limit value L, the recommendation probabilities of the nodes obtained by the inference of that layer are sorted according to the result of the prediction function, and the L nodes with higher recommendation probabilities are retained.

[0158] In this embodiment, when the number of sample association information in the hierarchical sample association information set is greater than the preset number, the current customer characteristics of historical customers are determined according to the prediction features corresponding to the obtained hierarchical sample association information set. The hierarchical sample association information set is then updated using the prediction probability of the sample association information calculated based on the current customer characteristics. This helps to limit the number of sample association information in each layer of the sample association information set, effectively reducing the amount of data and avoiding the defect that a large amount of data to be calculated is caused by a large number of sample association information in the sample association information set.

[0159] In one embodiment, step S105, which determines the target recommendation information corresponding to the customer to be recommended from the information to be recommended based on the recommendation probability, specifically includes the following: filtering information with a recommendation probability greater than a preset probability from the information to be recommended, and using it as the target recommendation information. Further, after step S105, the following steps are also included: storing the target recommendation information in a recommendation list corresponding to the customer to be recommended; and recommending the target recommendation information to the customer to be recommended according to the recommendation list.

[0160] Specifically, the server filters information from the pending recommendation information that has a recommendation probability greater than a preset probability, and uses this information as target recommendation information. The target recommendation information is then stored sequentially in the recommendation list corresponding to the customers to be recommended, in descending order of recommendation probability. Next, the server responds to the recommendation requests from the customers to be recommended, sequentially recommending the target recommendation information from the recommendation list to the customers, thus achieving accurate information recommendation. Further, if there is new behavior information from the customers to be recommended, the server re-executes steps S101 to S105 based on this new behavior information to determine new target recommendation information corresponding to the customers to be recommended, thereby obtaining a new recommendation list corresponding to the customers to be recommended, and recommending the new target recommendation information to the customers to be recommended according to the new recommendation list. Further, when publishing new activity information, the server filters information related to merchants, regions, etc., involved in the new activity information from the recommendation list of the customers to be recommended, and recommends this information to the customers to be recommended. Furthermore, the server can also compile potential customer data for various information (such as merchant information or event information) based on each customer's recommendation list, and push it to the relevant business personnel to assist them in promoting the corresponding merchant information or event information.

[0161] For example, after obtaining the trained multi-layer prediction model, the server can apply the model to customized marketing campaign recommendations, mainly in the following scenarios:

[0162] ① Initially, the server will extract merchant and activity information from the customer's recent consumption and browsing data, use this as input to the model, and use a multi-layer inference model to predict which merchants the customer might consume from and which activities they might participate in, storing this information in a recommendation list. Based on the information in the recommendation list, the server will push relevant merchants' marketing activity information to the customer.

[0163] ② When a customer adds new consumption and browsing history, the server will clear the customer's recommendation list, use the new record as input to the model, re-infer a new customer recommendation list, and push marketing campaign information to the customer.

[0164] ③ When a new marketing campaign is launched, the server will obtain information such as the merchants and regions involved in the marketing campaign, and use this information to filter the data in the marketing campaign recommendation list, and push the marketing campaign information to customers based on the filtering results.

[0165] ④ Based on the marketing activity recommendation table, the server statistically analyzes the potential customer data of merchants / activities and generates reports to present to the business personnel responsible for marketing and promotion, thereby assisting the business personnel in promoting marketing activities and improving the efficiency and success rate of marketing cooperation between business personnel and merchants.

[0166] In this embodiment, based on the recommendation probability, target recommendation information corresponding to the customer to be recommended is determined from the information to be recommended, and the target recommendation information is recommended to the customer to be recommended. This helps to achieve accurate information recommendation and thus improves the accuracy of information recommendation.

[0167] In one embodiment, such as Figure 5 As shown, another information recommendation method is provided. Taking the application of this method to a server as an example, the steps include:

[0168] Step S501: Obtain historical customer behavior information; extract merchant information and activity information that the historical customer has interacted with from the historical behavior information.

[0169] Step S502: Identify the first association between merchant information and activity information, the second association between merchant information, and the third association between activity information.

[0170] Step S503: Based on the first association, the second association, and the third association, construct a knowledge graph as a pre-constructed knowledge graph.

[0171] Step S504: Determine the related information of the historical behavior information of the customer to be recommended from the pre-built knowledge graph, and use it as the recommendation information.

[0172] Step S505: Determine the set of hierarchical related information corresponding to the information to be recommended from the knowledge graph.

[0173] Step S506: Input the information features of the information to be recommended and the information features of the related information in the hierarchical association information set into the pre-trained multi-layer prediction model to obtain the prediction features corresponding to the hierarchical association information set.

[0174] Step S507: Take the associated information in the hierarchical association information set as the new information to be recommended, take the predicted features corresponding to the hierarchical association information set as the information features of the information to be recommended, and jump to step S505 until the predicted features corresponding to the obtained hierarchical association information set are the predicted features corresponding to the Nth layer association information set; N is a positive integer greater than or equal to 2.

[0175] Step S508: The predicted features corresponding to each layer of related information set are fused to obtain the customer features of the customer to be recommended.

[0176] Step S509: Based on the information characteristics of the information to be recommended and the customer characteristics of the customer to be recommended, obtain the recommendation probability of the information to be recommended.

[0177] Step S510: Select information with a recommendation probability greater than a preset probability from the information to be recommended, and use it as the target recommendation information.

[0178] Step S511: Store the target recommendation information in the recommendation list corresponding to the customer to be recommended; recommend the target recommendation information to the customer to be recommended according to the recommendation list.

[0179] The information recommendation method provided in the above embodiments expands the historical behavior information of the customers to be recommended through knowledge graphs and multi-layer prediction models. This makes the customer features of the customers to be recommended more dense, thereby enriching the customer-side features. This makes the recommendation probability obtained based on the information features of the information to be recommended and the customer features of the customers to be recommended more accurate, thus making the target recommendation information determined based on the recommendation probability more accurate. In this way, the information recommendation accuracy is improved, avoiding the defect of low information recommendation accuracy caused by sparse customer behavior.

[0180] In one embodiment, to more clearly illustrate the information recommendation method provided by the embodiments of this application, the following specific embodiment will be used to describe the information recommendation method in detail. In one embodiment, this application also provides a marketing campaign recommendation method based on knowledge graphs, such as... Figure 6 As shown, a marketing campaign knowledge graph is generated by using customers' historical consumption records. A multi-layered inference algorithm is used to assist the recommendation system in predicting customers' consumption tendencies. Based on the prediction results, marketing campaigns are accurately recommended to customers, which helps to provide customers with more accurate marketing campaign recommendations and to provide merchants with better marketing service support.

[0181] Specifically, it includes the following:

[0182] 1. Extract customers' historical consumption data to construct a knowledge graph of merchants and activities.

[0183] 2. Generate model training samples based on customers' historical consumption data.

[0184] 3. Randomly sample the sample records to obtain merchants and activity nodes, and obtain multiple associated nodes (i.e., nodes adjacent to the current node in the knowledge graph). Use a multi-layer inference model to predict the associated nodes, retain the nodes with higher prediction scores, and compare them with the actual merchants where customers make purchases in the sample database to achieve iterative training. For example, if customer A makes a purchase at store B and then later makes a purchase at store C, the multi-layer inference model will use store B as the source node during model training and expand it based on the knowledge graph to predict the next potential store for purchases. If the prediction result is C, it is correct; if the prediction result is not C, the model parameters of the multi-layer inference model will be adjusted.

[0185] 4. Apply the trained multi-layer inference model to the actual recommendation process. Use the customer's recent consumption and browsing data as model data to predict the merchants and activities that the customer may consume at. Mark the nodes and their attributes that meet the conditions and store them in the recommendation list.

[0186] 5. The recommendation system pushes marketing campaign information to the corresponding customers based on the records in the recommendation list, achieving precise marketing campaign recommendations.

[0187] 6. The recommendation system compiles potential customer data for merchants and activities, pushes it to relevant business personnel, assists them in promoting marketing activities, and helps them to cooperate with more merchants.

[0188] The above-mentioned knowledge graph-based marketing activity recommendation method can achieve the following technical effects: (1) It uses a knowledge graph to assist the recommendation system in recommending marketing activities, realizing personalized push of marketing activities. (2) It optimizes the sparsity problem of traditional recommendation methods, improves the accuracy of marketing activity recommendations, and can update the recommendation list according to the latest consumption and browsing information of customers. (3) It can assist business personnel in cooperating with more merchants to carry out marketing activities and provide merchants with better marketing services.

[0189] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0190] Based on the same inventive concept, this application also provides an information recommendation apparatus for implementing the information recommendation method described above. The solution provided by this apparatus is similar to the implementation scheme described in the above method; therefore, the specific limitations in one or more information recommendation apparatus embodiments provided below can be found in the limitations of the information recommendation method described above, and will not be repeated here.

[0191] In one embodiment, such as Figure 7 As shown, an information recommendation device is provided, including: an information determination module 710, a set determination module 720, a fusion processing module 730, a probability determination module 740, and an information recommendation module 750, wherein:

[0192] The information determination module 710 is used to determine the related information of the historical behavior information of the customer to be recommended from the pre-built knowledge graph, and use it as the information to be recommended; the knowledge graph includes merchant information and activity information.

[0193] The feature determination module 720 is used to determine the multi-layer association information set corresponding to the information to be recommended from the knowledge graph, and obtain the prediction features corresponding to each layer of association information set through a pre-trained multi-layer prediction model.

[0194] The fusion processing module 730 is used to fuse the predicted features corresponding to each layer of related information set to obtain the customer features of the customer to be recommended.

[0195] The probability determination module 740 is used to obtain the recommendation probability of the information to be recommended based on the information characteristics of the information to be recommended and the customer characteristics of the customers to be recommended.

[0196] The information recommendation module 750 is used to determine the target recommendation information corresponding to the customer to be recommended from the information to be recommended based on the recommendation probability.

[0197] In one embodiment, the feature determination module 720 is further configured to determine the hierarchical association information set corresponding to the information to be recommended from the knowledge graph; input the information features of the information to be recommended and the information features of the association information in the hierarchical association information set into a pre-trained multi-layer prediction model to obtain the prediction features corresponding to the hierarchical association information set; use the association information in the hierarchical association information set as the new information to be recommended, use the prediction features corresponding to the hierarchical association information set as the information features of the information to be recommended, and jump to the step of determining the hierarchical association information set corresponding to the information to be recommended from the knowledge graph, until the prediction features corresponding to the obtained hierarchical association information set are the prediction features corresponding to the Nth layer association information set; N is a positive integer greater than or equal to 2.

[0198] In one embodiment, the information recommendation device further includes a graph construction module, used to acquire historical customer behavior information; extract merchant information and activity information that the historical customer has interacted with from the historical behavior information; identify a first association between merchant information and activity information, a second association between merchant information, and a third association between activity information; and construct a knowledge graph based on the first association, the second association, and the third association, as a pre-constructed knowledge graph.

[0199] In one embodiment, the information recommendation device further includes a model training module, used to extract first information and second information that the historical customer has interacted with sequentially from the historical behavior information of the historical customer; determine the association information of the first information from a pre-constructed knowledge graph as sample information; determine a multi-layer sample association information set corresponding to the sample information from the knowledge graph, and obtain the prediction features corresponding to each layer sample association information set through a multi-layer prediction model to be trained; fuse the prediction features corresponding to each layer sample association information set to obtain the customer features of the historical customer; obtain the recommendation probability of the sample information based on the information features of the sample information and the customer features of the historical customer; determine the target information corresponding to the historical customer from the sample information based on the recommendation probability of the sample information; and train the multi-layer prediction model to be trained based on the difference between the target information and the second information to obtain the trained multi-layer prediction model as a pre-trained multi-layer prediction model.

[0200] In one embodiment, the model training module is further configured to determine the hierarchical sample association information set corresponding to the sample information from the knowledge graph; when the number of sample association information in the hierarchical sample association information set is less than or equal to a preset number, input the information features of the sample information and the information features of the sample association information in the hierarchical sample association information set into the multi-layer prediction model to be trained to obtain the prediction features corresponding to the hierarchical sample association information set; use the sample association information in the hierarchical sample association information set as new sample information, use the prediction features corresponding to the hierarchical sample association information set as the information features of the sample information, and jump to the step of determining the hierarchical sample association information set corresponding to the sample information from the knowledge graph, until the prediction features corresponding to the obtained hierarchical sample association information set are the prediction features corresponding to the Nth layer sample association information set.

[0201] In one embodiment, the model training module is further configured to: determine the current customer characteristics of historical customers based on the prediction features corresponding to the obtained hierarchical sample association information set when the number of sample association information in the hierarchical sample association information set is greater than a preset number; obtain the recommendation probability of sample association information based on the information features of sample association information and the current customer characteristics of historical customers; retain a preset number of sample association information with the highest recommendation probabilities in the hierarchical sample association information set to obtain a new hierarchical sample association information set; and input the information features of the sample information and the information features of the sample association information in the new hierarchical sample association information set into the multi-layer prediction model to be trained to obtain the prediction features corresponding to the new hierarchical sample association information set.

[0202] In one embodiment, the information recommendation module 750 is further configured to: filter out information with a recommendation probability greater than a preset probability from the information to be recommended, and use it as target recommendation information; store the target recommendation information in a recommendation list corresponding to the customer to be recommended; and recommend the target recommendation information to the customer to be recommended according to the recommendation list.

[0203] Each module in the aforementioned information recommendation device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0204] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 8As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media. The database stores knowledge graphs, historical behavior information, recommendation information, and other data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When executed by the processor, the computer program implements an information recommendation method.

[0205] Those skilled in the art will understand that Figure 8 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0206] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0207] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0208] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0209] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0210] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0211] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. An information recommendation method, characterized in that, The method includes: The relevant information of the historical behavior information of the customers to be recommended is determined from the pre-built knowledge graph and used as the recommendation information; the knowledge graph includes merchant information and activity information. From the knowledge graph, a multi-layered set of related information corresponding to the information to be recommended is determined. A pre-trained multi-layered prediction model is used to obtain the predicted features corresponding to each layer of related information. This includes: determining the hierarchical set of related information corresponding to the information to be recommended from the knowledge graph; inputting the information features of the information to be recommended and the information features of the related information in the hierarchical set of related information into the pre-trained multi-layered prediction model to obtain the predicted features corresponding to the hierarchical set of related information; using the related information in the hierarchical set of related information as new information to be recommended, using the predicted features corresponding to the hierarchical set of related information as the information features of the information to be recommended, and then returning to the step of determining the hierarchical set of related information corresponding to the information to be recommended from the knowledge graph, until the predicted features corresponding to the obtained hierarchical set of related information are the predicted features corresponding to the Nth layer of related information; N is a positive integer greater than or equal to 2. The predicted features corresponding to each layer of associated information set are fused to obtain the customer features of the customer to be recommended. Based on the information characteristics of the information to be recommended and the customer characteristics of the customer to be recommended, the recommendation probability of the information to be recommended is obtained; Based on the recommendation probability, target recommendation information corresponding to the customer to be recommended is determined from the information to be recommended.

2. The method according to claim 1, characterized in that, Before identifying the relevant information about the historical behavior of customers to be recommended from a pre-built knowledge graph, and using this information as the recommendation information, the following steps are also included: Obtain historical customer behavior information; Extract the merchant information and activity information that the customer has interacted with in the past from the historical behavior information; Identify a first association between the merchant information and the activity information, a second association between the merchant information, and a third association between the activity information; Based on the first association, the second association, and the third association, a knowledge graph is constructed as the pre-constructed knowledge graph.

3. The method according to claim 1, characterized in that, The pre-trained multilayer prediction model is trained in the following manner: Extract the first and second information that the historical customers have interacted with sequentially from their historical behavior information; The associated information of the first information is determined from the pre-constructed knowledge graph and used as sample information; From the knowledge graph, a set of multi-layer sample association information corresponding to the sample information is determined. Through the multi-layer prediction model to be trained, the prediction features corresponding to each set of sample association information are obtained. The predicted features corresponding to the sample association information set of each layer are fused to obtain the customer features of the historical customer. Based on the information features of the sample information and the customer features of the historical customers, the recommendation probability of the sample information is obtained; Based on the recommendation probability of the sample information, target information corresponding to the historical customer is determined from the sample information; Based on the difference between the target information and the second information, the multi-layer prediction model to be trained is trained to obtain a trained multi-layer prediction model, which is used as the pre-trained multi-layer prediction model.

4. The method according to claim 3, characterized in that, The step of determining a multi-layer sample association information set corresponding to the sample information from the knowledge graph, and obtaining the prediction features corresponding to each layer sample association information set through a multi-layer prediction model to be trained, includes: From the knowledge graph, determine the set of hierarchical sample association information corresponding to the sample information; When the number of sample association information in the hierarchical sample association information set is less than or equal to a preset number, the information features of the sample information and the information features of the sample association information in the hierarchical sample association information set are input into the multi-layer prediction model to be trained to obtain the prediction features corresponding to the hierarchical sample association information set. The sample association information in the hierarchical sample association information set is used as the new sample information, and the prediction feature corresponding to the hierarchical sample association information set is used as the information feature of the sample information. Then, the process jumps to the step of determining the hierarchical sample association information set corresponding to the sample information from the knowledge graph, until the prediction feature corresponding to the obtained hierarchical sample association information set is the prediction feature corresponding to the Nth layer sample association information set.

5. The method according to claim 4, characterized in that, The method further includes: If the number of sample association information in the hierarchical sample association information set is greater than the preset number, the current customer characteristics of the historical customer are determined based on the predicted features corresponding to the obtained hierarchical sample association information set. Based on the information features of the sample association information and the current customer features of the historical customers, the recommendation probability of the sample association information is obtained; In the hierarchical sample association information set, the preset number of sample association information with the highest recommendation probability are retained to obtain a new hierarchical sample association information set; The information features of the sample information and the information features of the sample association information in the new hierarchical sample association information set are input into the multi-layer prediction model to be trained to obtain the prediction features corresponding to the new hierarchical sample association information set.

6. The method according to any one of claims 1 to 5, characterized in that, The step of determining the target recommendation information corresponding to the customer to be recommended from the information to be recommended based on the recommendation probability includes: From the information to be recommended, information with a recommendation probability greater than a preset probability is selected as the target recommendation information; The method further includes: The target recommendation information is stored in the recommendation list corresponding to the customer to be recommended; According to the recommendation list, the target recommendation information is recommended to the customer to be recommended.

7. An information recommendation device, characterized in that, The device includes: The information determination module is used to determine the related information of the historical behavior information of the customer to be recommended from the pre-built knowledge graph, and use it as the information to be recommended; the knowledge graph includes merchant information and activity information; The feature determination module is used to determine the multi-layer association information set corresponding to the information to be recommended from the knowledge graph, and obtain the predicted features corresponding to each layer association information set through a pre-trained multi-layer prediction model. The feature determination module is further configured to: determine, from the knowledge graph, a set of hierarchical association information corresponding to the information to be recommended; input the information features of the information to be recommended and the information features of the association information in the set of hierarchical association information into a pre-trained multi-layer prediction model to obtain the prediction features corresponding to the set of hierarchical association information; use the association information in the set of hierarchical association information as new information to be recommended, use the prediction features corresponding to the set of hierarchical association information as the information features of the information to be recommended, and jump to the step of determining, from the knowledge graph, the set of hierarchical association information corresponding to the information to be recommended, until the prediction features corresponding to the obtained set of hierarchical association information are the prediction features corresponding to the Nth layer of association information; N is a positive integer greater than or equal to 2; The fusion processing module is used to fuse the predicted features corresponding to each layer of related information set to obtain the customer features of the customer to be recommended. The probability determination module is used to obtain the recommendation probability of the information to be recommended based on the information features of the information to be recommended and the customer features of the customer to be recommended. The information recommendation module is used to determine the target recommendation information corresponding to the customer to be recommended from the information to be recommended based on the recommendation probability.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images