Coupon data recommendation method and device, electronic equipment and computer readable medium

Abstract

This application discloses a coupon data recommendation method, apparatus, electronic device, and computer-readable medium, relating to the field of computer technology. The method includes: acquiring user behavior data, user attribute data, and coupon content data based on a coupon data recommendation request; extracting features from the user behavior data to generate temporal behavior features; invoking a clustering algorithm to determine a target clustering dimension based on the temporal behavior features and user attribute data, and generating atomic pseudo-labels based on the target clustering dimension; generating a knowledge graph based on the atomic pseudo-labels and coupon content data, and extracting high-order features from the knowledge graph; determining the central element corresponding to the coupon content data through a multi-head self-attention mechanism, and determining the target recommended coupon data based on the high-order features and the central element. This improves the accuracy and efficiency of coupon data recommendation.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a method, apparatus, electronic device, and computer-readable medium for recommending coupon data. Background Technology

[0002] Currently, collaborative filtering, matrix factorization, and deep learning are the mainstream technologies for data recommendation systems. However, with the continuous expansion of data scale and the increasing complexity of user needs, traditional data recommendation technologies face many challenges and problems when handling complex relationships and high-dimensional data. For example, recommendations for coupon data cannot adapt to rapidly changing user needs, exhibiting poor performance, low accuracy, and low efficiency. Summary of the Invention

[0003] In view of this, embodiments of this application provide a coupon data recommendation method, apparatus, electronic device, and computer-readable medium, which can solve the problems that existing coupon data recommendations cannot adapt to rapidly changing user needs, and have poor recommendation effects, low accuracy, and low efficiency.

[0004] To achieve the above objectives, according to one aspect of the embodiments of this application, a coupon data recommendation method is provided, comprising:

[0005] Based on the obtained coupon data, a recommendation request is made to obtain user behavior data, user attribute data, and coupon content data;

[0006] Feature extraction is performed on user behavior data to generate time-series behavioral features;

[0007] A clustering algorithm is invoked to determine the target clustering dimension based on temporal behavioral features and user attribute data, and then atomic pseudo-labels are generated based on the target clustering dimension;

[0008] Based on atomic pseudo-tags and coupon content data, a knowledge graph is generated, and high-order features are extracted from the knowledge graph.

[0009] The central element corresponding to the coupon content data is determined by a multi-head self-attention mechanism. Based on the high-order features and the central element, the target coupon data is determined for recommendation.

[0010] Optionally, a knowledge graph is generated, including:

[0011] Determine the coupon attributes corresponding to the coupon content data;

[0012] Generate a set of points based on atomic pseudo-tags and coupon attributes;

[0013] Generate an edge set based on the relationship between atomic pseudo-tags and coupon attributes;

[0014] A knowledge graph is generated based on the set of points and the set of edges.

[0015] Optionally, feature extraction is performed on user behavior data to generate time-series behavioral features, including:

[0016] Word embedding is performed on user behavior data to convert it into a continuous behavior sequence;

[0017] Perform the following training process:

[0018] The continuous action sequence is input into the encoder in the denoising autoencoder network to obtain the encoded continuous action sequence;

[0019] The encoded continuous action sequence is input into the decoder in the denoising autoencoder network to reconstruct the original continuous action sequence;

[0020] The encoder and decoder in the denoising autoencoder network are trained by minimizing the cross-entropy loss function to obtain the target denoising autoencoder network.

[0021] The trained target denoising autoencoder network is used to extract temporal behavioral features from user behavior data.

[0022] Optionally, a clustering algorithm is invoked to determine the target clustering dimension based on temporal behavioral features and user attribute data, including:

[0023] Determine the corresponding attribute tags based on user attribute data;

[0024] For each attribute label, determine the corresponding optional dimension, and perform clustering based on the corresponding temporal behavior features under each optional dimension to obtain each cluster.

[0025] If each cluster is non-empty, the corresponding optional dimension is determined as the target clustering dimension.

[0026] Optionally, atomic pseudo-labels are generated based on the target clustering dimension, including:

[0027] Atomic pseudo-labels are generated based on the definition data of the clusters corresponding to the target clustering dimension.

[0028] Optionally, a multi-head self-attention mechanism is used to determine the central element corresponding to the coupon content data, including:

[0029] The coupon content data is non-linearly transformed using a fully connected layer to obtain coupon content features;

[0030] The multi-head self-attention mechanism is invoked using time-series behavioral features and coupon content features as input parameters to obtain correlation data between time-series behavioral features and coupon content features;

[0031] Based on correlation data, predict the central element in the content features of coupons.

[0032] Optionally, based on higher-order features and central elements, target recommended coupon data is determined, including:

[0033] Based on higher-order features, identify nodes in the knowledge graph that are related to the node corresponding to the central element;

[0034] Calculate the similarity between the coupon content features corresponding to the relevant nodes and the central element, and determine the recommendation probability of the relevant nodes based on the similarity.

[0035] Based on the recommendation probability, determine the target recommended coupon data.

[0036] In addition, this application also provides a coupon data recommendation device, including:

[0037] The acquisition unit is configured to request user behavior data, user attribute data, and coupon content data based on the acquired coupon data.

[0038] The temporal behavior feature generation unit is configured to extract features from user behavior data to generate temporal behavior features;

[0039] The atomic pseudo-label generation unit is configured to call a clustering algorithm to determine the target clustering dimension based on temporal behavioral features and user attribute data, and then generate atomic pseudo-labels based on the target clustering dimension;

[0040] The high-order feature extraction unit is configured to generate a knowledge graph based on atomic pseudo-labels and coupon content data, and extract high-order features from the knowledge graph;

[0041] The target recommendation coupon data determination unit is configured to determine the central element corresponding to the coupon content data through a multi-head self-attention mechanism, and determine the target recommendation coupon data based on the high-order features and the central element.

[0042] Optionally, the higher-order feature extraction unit is further configured to:

[0043] Determine the coupon attributes corresponding to the coupon content data;

[0044] Generate a set of points based on atomic pseudo-tags and coupon attributes;

[0045] Generate an edge set based on the relationship between atomic pseudo-tags and coupon attributes;

[0046] A knowledge graph is generated based on the set of points and the set of edges.

[0047] Optionally, the temporal behavior feature generation unit is further configured to:

[0048] Word embedding is performed on user behavior data to convert it into a continuous behavior sequence;

[0049] Perform the following training process:

[0050] The continuous action sequence is input into the encoder in the denoising autoencoder network to obtain the encoded continuous action sequence;

[0051] The encoded continuous action sequence is input into the decoder in the denoising autoencoder network to reconstruct the original continuous action sequence;

[0052] The encoder and decoder in the denoising autoencoder network are trained by minimizing the cross-entropy loss function to obtain the target denoising autoencoder network.

[0053] The trained target denoising autoencoder network is used to extract temporal behavioral features from user behavior data.

[0054] Optionally, the atomic pseudo-tag generation unit is further configured to:

[0055] Determine the corresponding attribute tags based on user attribute data;

[0056] For each attribute label, determine the corresponding optional dimension, and perform clustering based on the corresponding temporal behavior features under each optional dimension to obtain each cluster.

[0057] If each cluster is non-empty, the corresponding optional dimension is determined as the target clustering dimension.

[0058] Optionally, the atomic pseudo-tag generation unit is further configured to:

[0059] Atomic pseudo-labels are generated based on the definition data of the clusters corresponding to the target clustering dimension.

[0060] Optionally, the target recommendation coupon data determination unit is further configured to:

[0061] The coupon content data is non-linearly transformed using a fully connected layer to obtain coupon content features;

[0062] The multi-head self-attention mechanism is invoked using time-series behavioral features and coupon content features as input parameters to obtain correlation data between time-series behavioral features and coupon content features;

[0063] Based on correlation data, predict the central element in the content features of coupons.

[0064] Optionally, the target recommendation coupon data determination unit is further configured to:

[0065] Based on higher-order features, identify nodes in the knowledge graph that are related to the node corresponding to the central element;

[0066] Calculate the similarity between the coupon content features corresponding to the relevant nodes and the central element, and determine the recommendation probability of the relevant nodes based on the similarity.

[0067] Based on the recommendation probability, determine the target recommended coupon data.

[0068] In addition, this application also provides an electronic device for recommending coupon data, including: one or more processors; and a storage device for storing one or more programs, which, when executed by one or more processors, enable the one or more processors to implement the coupon data recommendation method as described above.

[0069] In addition, this application also provides a computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements the coupon data recommendation method as described above.

[0070] To achieve the above objectives, according to another aspect of the embodiments of this application, a computer program product is provided.

[0071] A computer program product according to an embodiment of this application includes a computer program that, when executed by a processor, implements the coupon data recommendation method provided in an embodiment of this application.

[0072] One embodiment of the above invention has the following advantages or beneficial effects: This application obtains user behavior data, user attribute data, and coupon content data based on the obtained coupon data recommendation request; it extracts features from the user behavior data to generate time-series behavior features; it calls a clustering algorithm to determine the target clustering dimension based on the time-series behavior features and user attribute data, and then generates atomic pseudo-labels based on the target clustering dimension; it generates a knowledge graph based on the atomic pseudo-labels and coupon content data, and extracts high-order features from the knowledge graph; it determines the central element corresponding to the coupon content data through a multi-head self-attention mechanism, and determines the target recommended coupon data based on the high-order features and the central element. By generating atomic pseudo-labels from multiple attribute dimensions through a clustering algorithm, constructing a knowledge graph using the atomic pseudo-labels, and combining the high-order features and central elements of the knowledge graph, it accurately and efficiently predicts coupon data, improving the coupon data recommendation effect and increasing the accuracy and efficiency of coupon data recommendation.

[0073] The further effects of the aforementioned unconventional alternative methods will be explained below in conjunction with specific implementation methods. Attached Figure Description

[0074] The accompanying drawings are provided to better understand this application and do not constitute an undue limitation thereof. Wherein:

[0075] Figure 1 This is a schematic diagram of the main flow of a coupon data recommendation method according to an embodiment of this application;

[0076] Figure 2 This is a schematic diagram of the main flow of a coupon data recommendation method according to an embodiment of this application;

[0077] Figure 3 This is a schematic diagram of the overall framework of a knowledge graph-based intelligent coupon recommendation method according to an embodiment of this application;

[0078] Figure 4 This is a flowchart of the behavioral feature extraction module of a coupon data recommendation method according to an embodiment of this application;

[0079] Figure 5 This is a flowchart of the atomic pseudo-tag generation module of a coupon data recommendation method according to an embodiment of this application;

[0080] Figure 6 This is a diagram illustrating the effect of a clustering algorithm for segmenting user groups according to an embodiment of the coupon data recommendation method of this application;

[0081] Figure 7 This is a schematic diagram of the main units of the coupon data recommendation device according to an embodiment of this application;

[0082] Figure 8 This is an exemplary system architecture diagram to which embodiments of this application can be applied;

[0083] Figure 9 This is a schematic diagram of the structure of a computer system suitable for implementing terminal devices or servers in the embodiments of this application. Detailed Implementation

[0084] The following description, in conjunction with the accompanying drawings, illustrates exemplary embodiments of this application, including various details to aid understanding. These embodiments should be considered merely exemplary. Therefore, those skilled in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this application. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description. It should be noted that the acquisition, transmission, storage, use, and processing of data in the technical solutions of this application comply with relevant national laws and regulations. It should also be noted that certain software, components, models, and other existing industry solutions may be mentioned in the embodiments of this application. These should be considered exemplary, intended only to illustrate the feasibility of implementing the technical solutions of this application, and do not imply that the applicant has already used or necessarily used such solutions. The collection, gathering, updating, analysis, processing, use, transmission, and storage of user personal information involved in the technical solutions of this application all comply with relevant laws and regulations, are used for legal and reasonable purposes, do not violate public order and good morals, are not shared, disclosed, or sold outside of these legal uses, and are subject to supervision and management by regulatory authorities. Necessary measures should be taken to prevent unauthorized access to user personal information, safeguard user personal information security, cybersecurity, and national security, and ensure that those authorized to access personal information comply with relevant laws and regulations. Once this user personal information is no longer needed, risks should be minimized by restricting or even prohibiting data collection and / or deleting the data.

[0085] When used, including in certain relevant applications, data is deidentified to protect user privacy, for example by removing specific identifiers, controlling the amount or specificity of stored data, controlling how data is stored, and / or other methods.

[0086] Figure 1 This is a schematic diagram of the main flow of a coupon data recommendation method according to an embodiment of this application, as shown below. Figure 1 As shown, the coupon data recommendation method mainly includes the following steps S101-S105.

[0087] Step S101: Based on the obtained coupon data, a recommendation request is made to obtain user behavior data, user attribute data, and coupon content data.

[0088] In this embodiment, the executing entity of the coupon data recommendation method (e.g., a server) can receive coupon data recommendation requests via wired or wireless connections. Upon receiving a coupon data recommendation request, the executing entity can obtain the corresponding user behavior data, user attribute data, and coupon content data. For example, user behavior data may include user clickstream data, browsing behavior data, purchase behavior data, search behavior data, and other behavior data. Clickstream data captures user clicks on various elements on a page, recording the timestamp of the click event, user information, and clicked element information; browsing behavior data records user browsing behavior using front-end visible area capture technology, including browsing event timestamps, browsed page URLs, and dwell time information; purchase behavior data collects key nodes of order placement, payment, and refund in e-commerce applications, recording event time, user information, order information, and product information; search behavior data collects user search records, search frequency, and search timestamp information in e-commerce applications; other behavior data includes comment behavior, collection behavior, and follow behavior records. For example, user attribute data may include: basic information and tag information from the user's registration profile. Basic information may include name, email address, age, gender, and address. Tag information may include user product category preference tags, interest tags, personality tags, occupation tags, and custom tags. For example, coupon content data may include: coupon distribution record data and coupon-linked product content data. Coupon distribution record data may include coupon distribution ID, coupon distribution location, coupon distribution channel, coupon type, coupon price threshold, coupon discount price, number of times the user claimed the coupon, and number of times the user used the coupon. Coupon-linked product content data may include product SKU, product type, product price, product tags, and product description.

[0089] Step S102: Extract features from user behavior data to generate time-series behavior features.

[0090] Specifically, feature extraction is performed on user behavior data to generate temporal behavior features, including: word embedding of user behavior data to convert it into a continuous behavior sequence; and the following training process is performed: the continuous behavior sequence is input into the encoder of the denoising autoencoder network to obtain the encoded continuous behavior sequence; the encoded continuous behavior sequence is input into the decoder of the denoising autoencoder network to reconstruct the original continuous behavior sequence; the encoder and decoder of the denoising autoencoder network are trained by minimizing the cross-entropy loss function to train the target denoising autoencoder network; and temporal behavior features in user behavior data are extracted based on the trained target denoising autoencoder network.

[0091] For example, the executing entity can extract features from user behavior data by calling the behavior feature extraction module to generate time-series behavior features.

[0092] In some embodiments, the executing entity may invoke a behavior feature extraction module to extract features from user behavior data by constructing a denoising autoencoder to generate temporal behavior features. Specifically, the denoising autoencoder introduces noise when mapping user behavior data to a low-dimensional space, and then attempts to reconstruct the original noise-free data from the noisy input, thereby removing data noise and generating high-quality temporal behavior features B. t .

[0093] For example, for discrete user behavior data B, we first use an embedding layer to extract temporal features, that is, we use an embedding layer to transform it into a continuous vector representation B. t =Embedding(B), the purpose of which is to enable the denoising autoencoder to better learn the contextual relationships and temporal features between user behavior data. Then, the continuous behavior sequence B is... t The input is fed into the encoder of the denoising autoencoder network. The encoder uses a dilated 1D convolutional neural network to fuse user behavior context information. The dilated 1D convolutional neural network increases the receptive field by inserting spaces in the convolutional kernels without adding additional parameters. The encoded temporal behavior features are represented as B′. t :B′ t =F e (B t +z). Where z is noise, F e (·) represents a 1D dilated convolution. Next, the encoded temporal behavioral features B′ are... t The input is fed into the decoder, which attempts to reconstruct the original behavioral sequence B. t The decoder uses a convolutional neural network. Upsampling is used to restore the spatial resolution of the feature map to its original resolution. The convolutional neural network is used to extract features. The original sequence of actions that the decoder attempts to reconstruct is shown below: B t =F d (up(B′ t ), where up is upsampling, F d (·) represents a convolutional neural network. Based on the established denoising autoencoder's encoder and decoder, training is performed by minimizing the loss function, cross-entropy L. Among them, b′ t For B′ t The behavior at time step t, where p is the behavior data B. t The dimension enables the denoising autoencoder to learn an efficient representation of user behavior feature sequences (which, in this embodiment, can be referred to as user temporal features or temporal behavior features) B t Therefore, a well-trained denoising autoencoder is used to extract features from user behavior data in order to accurately generate temporal behavior features.

[0094] Step S103: Invoke the clustering algorithm to determine the target clustering dimension based on time-series behavioral features and user attribute data, and then generate atomic pseudo-labels based on the target clustering dimension.

[0095] The process involves identifying the possible dimensions for each user attribute data point, clustering based on temporal behavioral features along these dimensions, and then determining the dimensions that form clusters as target clustering dimensions. This facilitates the accurate generation of atomic pseudo-labels based on these target clustering dimensions. Atomic pseudo-labels provide more granular user segmentation, aiding in accurate coupon data recommendations. For example, the atomic pseudo-labels "Female 1," "Female 2," and "Female 3" represent the degree to which female users' purchasing behavior leans towards that of male users.

[0096] Step S104: Generate a knowledge graph based on the atomic pseudo-tags and coupon content data, and extract high-order features from the knowledge graph.

[0097] Specifically, generating a knowledge graph includes: determining the coupon attributes corresponding to the coupon content data; generating a set of points based on atomic pseudo-tags and coupon attributes; generating a set of edges based on the relationship between atomic pseudo-tags and coupon attributes; and generating a knowledge graph based on the set of points and the set of edges.

[0098] When generating the edge set based on the relationship between atomic pseudo-tags and coupon attributes, default edges, i.e., initial edges, can be used. After generating the knowledge graph based on the point set and edge set, higher-order features in the knowledge graph can be extracted based on a hypergraph neural network. That is, processing based on the hypergraph neural network yields more accurate relationships between points in the knowledge graph. The edges in the knowledge graph are optimized to make the connections between points more accurate.

[0099] Step S105: Determine the central element corresponding to the coupon content data through a multi-head self-attention mechanism, and determine the target recommended coupon data based on the high-order features and the central element.

[0100] Specifically, the central element corresponding to the coupon content data is determined through a multi-head self-attention mechanism, including: performing a non-linear transformation on the coupon content data through a fully connected layer to obtain coupon content features; calling the multi-head self-attention mechanism with temporal behavior features and coupon content features as input parameters to obtain correlation data between temporal behavior features and coupon content features; and predicting the central element in the coupon content features based on the correlation data.

[0101] In this embodiment, the central element can correspond to a coupon content feature. The executing entity can count the number of correlation data corresponding to each coupon content feature in the correlation data, and determine the coupon content feature corresponding to the largest number of correlation data as the central element, thereby improving the efficiency and accuracy of determining the central element.

[0102] In the feature representation space, the central element is the most representative element. The central element can be accurately identified using deep learning networks or probabilistic statistics.

[0103] Specifically, based on higher-order features and the central element, the target recommended coupon data is determined, including: determining the nodes related to the nodes corresponding to the central element in the knowledge graph based on higher-order features; calculating the similarity between the coupon content features corresponding to the related nodes and the central element, and determining the recommendation probability of the related nodes based on the similarity; and determining the target recommended coupon data based on the recommendation probability.

[0104] The central element can be a coupon or a coupon content feature corresponding to a coupon. The coupon content feature is related to the coupon attributes used to generate the knowledge graph. Higher-order features accurately characterize the relationships between nodes in the knowledge graph. The central element, as related data of the coupon, can correspond to a node in the knowledge graph. Through higher-order features, the nodes in the knowledge graph associated with the node corresponding to the central element can be accurately identified, and the cosine similarity between the coupon content feature of the associated node and the central element can be calculated. Based on the cosine similarity, the recommendation probability of the related nodes is calculated using a preset formula. The coupon data corresponding to the nodes in the knowledge graph associated with the node corresponding to the central element whose recommendation probability is higher than a preset probability threshold are determined as the target recommended coupon data.

[0105] This embodiment obtains user behavior data, user attribute data, and coupon content data based on the acquired coupon data recommendation request; it extracts features from the user behavior data to generate time-series behavior features; it calls a clustering algorithm to determine the target clustering dimension based on the time-series behavior features and user attribute data, and then generates atomic pseudo-labels based on the target clustering dimension; it generates a knowledge graph based on the atomic pseudo-labels and coupon content data, and extracts high-order features from the knowledge graph; it determines the central element corresponding to the coupon content data through a multi-head self-attention mechanism, and determines the target recommended coupon data based on the high-order features and the central element. By generating atomic pseudo-labels from multiple attribute dimensions through the clustering algorithm, constructing a knowledge graph using atomic pseudo-labels, and combining the high-order features and central elements of the knowledge graph, it accurately and efficiently predicts coupon data, improving the coupon data recommendation effect and increasing the accuracy and efficiency of coupon data recommendation.

[0106] Figure 2This is a schematic diagram of the main flow of a coupon data recommendation method according to an embodiment of this application, such as... Figure 2 As shown, the coupon data recommendation method mainly includes the following steps S201-S208.

[0107] Step S201: Based on the obtained coupon data, a recommendation request is made to obtain user behavior data, user attribute data, and coupon content data.

[0108] For example, user behavior data can include user clickstream data, browsing behavior data, purchase behavior data, search behavior data, and other behavior data. Clickstream data captures user clicks on various elements on a page, recording the timestamp of the click event, user information, and clicked element information. Browsing behavior data records user browsing behavior using front-end visible area capture technology, including browsing event timestamps, browsed page URLs, and dwell time information. Purchase behavior data collects key nodes of order placement, payment, and refund in e-commerce applications, recording event time, user information, order information, and product information. Search behavior data collects user search records, search frequency, and search timestamp information in e-commerce applications. Other behavior data includes comment behavior, favorite behavior, and follow behavior records. For example, user attribute data can include: basic information and tag information from user registration profiles. Basic information can include name, email, age, gender, and address. Tag information can include user product category preference tags, interest tags, personality tags, occupation tags, and custom tags. For example, coupon content data can include: coupon distribution record data and coupon-bundled product content data. The coupon distribution record data can include coupon distribution ID, coupon distribution location, coupon distribution channel, coupon type, coupon price threshold, coupon discount price, number of times users claimed the coupon, and number of times users used the coupon. The product content data associated with the coupon can include product SKU, product type, product price, product tags, and product description.

[0109] Step S202: Extract features from user behavior data to generate time-series behavior features.

[0110] The executing entity can input user behavior data into the feature extraction model to accurately and quickly obtain time-series behavioral features.

[0111] Step S203: Determine the corresponding attribute tags based on the user attribute data.

[0112] The executing entity extracts keywords from user attribute data and uses these extracted keywords as attribute labels for the corresponding user attribute data. For example, user attribute data can be city-tier level attribute data, including first-tier city attribute data, second-tier city attribute data, third-tier city attribute data, and fourth-tier city attribute data; the corresponding attribute labels could be "first-tier," "second-tier," "third-tier," and "fourth-tier." Similarly, user attribute data can be gender attribute data, including male and female data; the corresponding attribute labels could be "male" and "female."

[0113] Step S204: For each attribute label, determine the corresponding optional dimension, and perform clustering based on the corresponding temporal behavior features under each optional dimension to obtain each cluster.

[0114] For example, for the tag "female", the corresponding optional dimensions may include: purchase behavior bias dimension, purchase model dimension, shopping software usage type dimension, etc. This application embodiment does not specifically limit the various optional dimensions corresponding to the attribute tag.

[0115] After determining the optional dimensions corresponding to each attribute label, the executing entity can further cluster based on the temporal behavior features corresponding to each optional dimension to accurately obtain each cluster, which is used to quickly and accurately generate atomic pseudo-labels.

[0116] Step S205: If each cluster is non-empty, determine the corresponding optional dimension as the target clustering dimension.

[0117] After clustering based on the corresponding temporal behavior features under each optional dimension to obtain each cluster, it is determined whether each cluster is empty. If each cluster is empty, it means that clustering cannot be performed under the corresponding optional dimension, and that optional dimension cannot be used as the target clustering dimension. If each cluster is not empty, it means that clustering can be performed under the corresponding optional dimension, that is, there is a need for clustering. The obtained clusters can then be used as a data source for generating atomic pseudo-labels, which helps to improve the accuracy of coupon data recommendation based on atomic pseudo-labels.

[0118] Step S206: Generate atomic pseudo-labels based on the target clustering dimension.

[0119] Specifically, generating atomic pseudo-labels based on the target clustering dimension includes: generating atomic pseudo-labels based on the definition data of the clusters corresponding to the target clustering dimension.

[0120] For example, the executing entity can extract keywords from the definition data of the clusters corresponding to the target clustering dimension, and use these extracted keywords as atomic pseudo-tags to more accurately describe user characteristics. The definition data of the clusters corresponding to the target clustering dimension can be obtained based on user behavior data corresponding to the time-series behavioral features within those clusters. This improves the scientific rigor and accuracy of atomic pseudo-tag generation, and enhances the efficiency and accuracy of coupon data recommendation based on atomic pseudo-tags.

[0121] Step S207: Generate a knowledge graph based on the atomic pseudo-tags and coupon content data, and extract high-order features from the knowledge graph.

[0122] The atomic pseudo-tags and the coupon attributes corresponding to the coupon content data are used as a set of points, and the relationships between the default atomic pseudo-tags and coupon attributes are used as a set of edges to form triples. A knowledge graph is generated based on the triples, and high-order features in the knowledge graph are extracted through a hypergraph neural network.

[0123] Step S208: Determine the central element corresponding to the coupon content data through a multi-head self-attention mechanism, and determine the target recommended coupon data based on the high-order features and the central element.

[0124] A multi-head self-attention mechanism is used to capture the correlation between users and coupons, and the central element related to the coupon is determined based on this correlation. Higher-order features accurately characterize the relationships between nodes in the knowledge graph. The central element, as the relevant data of the coupon, can correspond to a node in the knowledge graph. Through higher-order features, the nodes in the knowledge graph associated with the node corresponding to the central element can be accurately identified, and the Euclidean distance between the coupon content features of the associated node and the central element is calculated. Based on the Euclidean distance, the recommendation probability of the related node is calculated using a preset formula. The coupon data corresponding to the nodes in the knowledge graph associated with the node corresponding to the central element whose recommendation probability is higher than a preset probability threshold is accurately determined as the target recommended coupon data.

[0125] Figure 3 This is a schematic diagram of the overall framework of a knowledge graph-based intelligent coupon recommendation method according to an embodiment of this application. This embodiment of the application uses a knowledge graph for intelligent coupon recommendation, enabling the system to generate high-quality atomic pseudo-labels in zero-shot learning and provide accurate coupon recommendation services through these atomic pseudo-labels.

[0126] This application's embodiments are based on knowledge graph-based intelligent coupon recommendations, involving: such as Figure 3The data acquisition module shown is used to collect data from the data source, perform preprocessing, and construct an initial dataset. The initial data mainly includes user behavior data, user attribute data, and coupon content data. The behavior feature extraction module extracts features from the original dataset, mines potential patterns from unlabeled data through self-supervised learning, and performs multi-scale feature fusion on user features to generate rich user feature representations and obtain a standard feature dataset. The atomic pseudo-label generation module inputs the extracted user behavior features into a clustering algorithm to generate user atomic pseudo-labels based on user attributes. Atomic pseudo-labels more accurately describe user features. The knowledge graph construction module uses atomic pseudo-labels and coupons as nodes, and the relationships between nodes as edges (obtained from initial default edges, which are subsequently optimized by the hypergraph processing module) to construct the graph. The hypergraph processing module uses a hypergraph neural network to perform multiple convolution operations on the knowledge graph to obtain higher-order features. The recommendation generation module uses a Transformer self-attention mechanism to capture the correlation between users and coupons and generates coupon recommendation probabilities by fusing the higher-order features obtained from the hypergraph neural network.

[0127] The data acquisition module collects user attribute data by retrieving basic information and tag information from user registration profiles. Basic information includes name, email, age, gender, and address. Tag information includes user product category preference tags, interest tags, personality tags, occupation tags, and custom tags; user data attributes are denoted as A. User behavior data collection includes user clickstream data, browsing behavior data, purchase behavior data, search behavior data, and other behavior data, denoted as B, and is time-series data. Clickstream data captures user clicks on various elements on the page, recording the timestamp of the click event, user information, and clicked element information. Browsing behavior data records user browsing behavior using front-end visible area capture technology, including browsing event timestamps, browsed page URLs, and dwell time information. Purchase behavior data is collected in e-commerce applications at key points such as order placement, payment, and refund, recording event time, user information, order information, and product information. Search behavior data is collected in e-commerce applications from user search records, search frequency, and search timestamp information. Other behavior data includes comment behavior, favorite behavior, and follow behavior records. Coupon content data includes coupon distribution records and coupon-linked product content data. Coupon distribution records include coupon distribution ID, distribution location, distribution channel, coupon type, coupon price threshold, coupon discount price, number of times users claimed the coupon, and number of times users used the coupon. Coupon-related product data includes product SKU, product type, product price, product tags, and product description. This coupon content data is denoted as C.

[0128] Figure 4This is a flowchart of the behavior feature extraction module of a coupon data recommendation method according to an embodiment of this application. The behavior feature extraction module extracts features by constructing a denoising autoencoder. The denoising autoencoder introduces noise when mapping user behavior data to a low-dimensional space, and then attempts to reconstruct the original noise-free data from the noisy input, thereby removing data noise and generating high-quality user behavior features B. t .

[0129] The behavioral feature extraction module works as follows: For discrete user behavior data B, the embedding layer is first used to extract temporal features, that is, the embedding layer is used to transform it into a continuous vector representation B. t =Embedding(B) aims to enable the denoising autoencoder to better learn the contextual relationships and temporal features between behavioral data.

[0130] Then, the continuous action sequence B t The input is fed into the encoder of the denoising autoencoder network. The encoder uses a dilated 1D convolutional neural network to fuse user behavior context information. The dilated 1D convolutional neural network increases the receptive field by inserting spaces in the convolutional kernels without adding additional parameters. The encoded temporal behavior features are represented as B′. t :B′ t =F e (B t +z). Where z is noise, F e (·) represents a 1D dilated convolution.

[0131] Next, the encoded temporal behavioral features B′ t The input is fed into the decoder, which attempts to reconstruct the original behavioral sequence B. t The decoder uses a convolutional neural network. Upsampling is used to restore the spatial resolution of the feature map to its original resolution. The convolutional neural network is used to extract features. The original sequence of actions that the decoder attempts to reconstruct is shown below:

[0132] B t =F d (up(B′ t ))

[0133] Where up represents upsampling, F d (·) represents a convolutional neural network.

[0134] Based on the established encoder and decoder of the denoising autoencoder, training is performed by minimizing the loss function cross-entropy L. Among them, b′ t For B′ t The behavior at time step t, where p is the behavior data B. tThe dimension of this allows the denoising autoencoder to learn an efficient representation of user behavior feature sequences, B. t (This can be referred to as user temporal features or temporal behavioral features). Therefore, a trained denoising autoencoder is used to extract features from user behavior data to accurately generate temporal behavioral features.

[0135] Figure 5 This is a flowchart of the atomic pseudo-tag generation module of a coupon data recommendation method according to an embodiment of this application. The atomic pseudo-tag generation module generates user behavior features B. t As input to the clustering algorithm, atomic pseudo-labels are generated for each attribute, with attributes as the dimension. For an attribute A... i By using attribute tags, a user group U can be divided into n categories. A user group can be represented as U = {U1, U2, ..., U...} n} can also be represented as Figure 5 In the context of {user group 1, user group 2, ..., user group n}, where n is attribute A i Category tags.

[0136] For user group U n They can be divided into K classes using clustering algorithms, and the user group can be represented as U. n ={U n，1 U n，2 , ..., U n，k}, or as Figure 5 As shown, the clustering is represented as {user group n_1, user group n_2, ..., user group n_k}, where k is an atomic pseudo-label, k = {1, 2, ..., K}. The clustering algorithm uses the k-means algorithm, which iteratively divides the original samples into K clusters. Initially, K points are randomly selected as the initial cluster centers, and then each point is assigned to the cluster represented by the nearest cluster center according to the nearest neighbor classification principle. Then, the center point of each cluster (e.g., y) is recalculated. i The clustering algorithm iterates until the cluster centers no longer change. The input to the clustering algorithm is the user population U. n Behavioral data B t，n To simplify the formula, In the following text, it will be represented as x i ,in User group U n Behavioral data B of instance X (a user) t，n，x In the i-th dimension (i.e., the more refined division direction), the cluster distance is calculated as follows:

[0137]

[0138] Where, dX，Y Let X be the clustering distance between instances X and Y, and Z be the dimension of the user behavior data features. i and y i Let X and Y be the coordinates of instances X and Y in the i-th dimension, respectively. Among these 1-Z dimensions, identify the dimensions that can cluster the classes and use them as the target clustering dimensions. Figure 6 This is a user group segmentation effect diagram based on a clustering algorithm of a coupon data recommendation method according to an embodiment of this application. When K=3, the user groups n_1, n_2, and n_3 obtained by the clustering algorithm for accurately generating atomic pseudo-labels can be as follows: Figure 6 As shown.

[0139] The atomic pseudo-label generation module generates an atomic pseudo-label for each attribute based on attribute labels and user temporal behavior (e.g., temporal behavior features). Therefore, atomic pseudo-labels can more accurately describe user characteristics. For example, for city tier-level attributes with labels "first-tier," "second-tier," "third-tier," and "fourth-tier," the purchasing power of "first-tier" users varies, with some "first-tier" users showing a greater preference for "second-tier" or "third-tier" spending. Similarly, for gender attributes with labels "male" and "female," female users' purchasing preferences tend to resemble those of male users. Clustering algorithms can further segment female users, generating atomic pseudo-labels "female 1," "female 2," and "female 3," representing the degree to which female users' purchasing behavior leans towards that of male users. Therefore, atomic pseudo-labels more accurately describe user characteristics.

[0140] The knowledge graph construction module uses the obtained atomic pseudo-tags and coupon attributes as a set of nodes V, and the relationships between atomic pseudo-tags and coupon attributes as a set of edges E, forming a triple T = {(v...} i e q v j )|v i v j ∈V, e q ∈E}, where v i Let v represent the i-th point. j Identify the j-th point, e q Let q represent the q-th edge.

[0141] The hypergraph processing module uses a hypergraph neural network to perform multi-layer convolutional operations based on the constructed knowledge graph to obtain high-order features. The module connects related nodes with edges and then extracts high-order features using the hypergraph convolutional neural network.

[0142]

[0143] Among them, H (l+1) H represents the node feature matrix of the (l+1)th layer.(l) Let be the node feature matrix of the l-th layer. H is the adjacency matrix of the hypergraph, representing the connection relationships between nodes and hyperedges. D v Let D be the degree matrix of the nodes. e Let W be the degree matrix of the hyperedge. (k) is the hypergraph weight matrix, where the user adjusts the weights relating nodes to hyperedges. ReLU is the activation function; after multiple convolutional operations using a hypergraph convolutional neural network, high-order features f are obtained.

[0144] f = Relu(H) (L) )

[0145] Where L represents the number of layers in the hypergraph convolutional neural network.

[0146] The high-order features obtained by the hypergraph processing module can describe the relationships between points in the hypergraph, and can be used to make accurate recommendations by finding the central element and elements similar to the central element.

[0147] Specifically, in the feature representation space, the central element is the most representative element. Central elements are typically identified using deep learning networks or probabilistic statistics. The selection of the central element has a decisive impact on the quality of the recommendation results because it serves as a reference point—the element most of interest to the user—determining the ranking and priority of the recommended objects. Once the central element is determined, the element with the highest similarity to it in the feature representation space can be found; that is, the point closest to the central element in the feature space. Feature space distance can be measured using Euclidean distance. Euclidean distance measures the straight-line distance between two vectors in a multidimensional space, thus reflecting their overall similarity. After calculating the distances of all elements to the central element, these elements can be sorted according to their distance. The closer the distance, the higher the similarity, and the better it meets the user's interests or needs. Therefore, the element closest to the central element in the recommendation list will be displayed first, and other elements will be arranged in order of increasing distance. The advantage of this method is that it can fully utilize the similarity information in the feature representation space, ensuring the relevance and accuracy of the recommendation results through precise distance calculation. This method not only quickly finds the most suitable recommendations for users in massive datasets, but also effectively prioritizes them, providing personalized and efficient recommendation services. Furthermore, this process demonstrates good scalability and robustness. Regardless of the size of the dataset or the complexity of the feature representation space, the system can effectively process and generate high-quality recommendation lists by calculating similarity and distance.

[0148] The Transformer deep learning algorithm (used for temporal prediction) excels in prediction tasks, especially in the field of Natural Language Processing (NLP). It is based on the Self-Attention mechanism, which effectively captures dependencies and multi-level representations in sequences.

[0149] The recommendation generation module uses Transformer to predict the center element and leverages features obtained from a hypergraph neural network to guide the prediction of coupon recommendation probabilities.

[0150] The recommendation generation module uses the Transformer's self-attention mechanism to capture the correlation between users and coupons. To further enhance the expressive power of coupon content features, the module first uses a fully connected layer to perform a non-linear transformation on the coupon content data C, capturing richer coupon content features C′:

[0151] C′=FC(C)

[0152] The recommendation generation module will use user behavior time-series data B t Using coupon content features C′ as input, a self-attention mechanism is used to capture the correlation between users and coupons. User behavior time-series data B... t Considering the key and value as input features, and the coupon content feature C′ as input features for the query, the attention in the Transformer can be represented as:

[0153]

[0154] Where softmax is the activation function, K = V = FC(B t ), Q = FC(C′), FC is a fully connected layer, d k Let be the dimension of the key vector. Multi-head attention is used to extract interaction information from different dimensions:

[0155] MultiHead(Q,K,V)=Concat(head1,...,head m , ..., head M W O

[0156] Among them, W OThe weight matrix is ​​a learnable matrix used to map the concatenated output back to the original dimensions. `Concat` is the concatenation operation. `M` is the number of heads. The output of the multi-head attention is fused with the original user behavior features (specifically, using residuals to recover lost features, making the features more complete), and then passed through a normalization layer to obtain the final feature U (a relatively complete feature). The final feature U is then added to the bias term to obtain the coupon recommendation probability value.

[0157] y t =softmax(WU+g)

[0158] Where W is the learnable matrix and g is the bias term. t Let y be an N-dimensional vector at time t, where each dimension corresponds to a coupon recommendation probability. For a coupon x, its probability of being recommended is y. t，x The higher-order feature of the node corresponding to this coupon in the hypergraph neural network is f. x Using coupon i as the central element, the probability of coupon x being recommended can be obtained by calculating the feature similarity between other nodes x and coupon i (where x represents the x-th coupon in the coupon ranking sequence and i represents the i-th coupon in the coupon ranking sequence) and applying a normalization function.

[0159]

[0160] Where D(·) is used to calculate the similarity between two features, and feature similarity can be calculated using Euclidean distance. X is the number of coupon types.

[0161] This application's embodiments generate atomic pseudo-labels from multiple attribute dimensions using a clustering algorithm, then construct a knowledge graph using these atomic pseudo-labels, and accurately guide coupon prediction.

[0162] Figure 7 This is a schematic diagram of the main units of a coupon data recommendation device according to an embodiment of this application. Figure 7 As shown, the coupon data recommendation device 700 includes an acquisition unit 701, a temporal behavior feature generation unit 702, an atomic pseudo-label generation unit 703, a high-order feature extraction unit 704, and a target recommended coupon data determination unit 705.

[0163] The acquisition unit 701 is configured to request user behavior data, user attribute data, and coupon content data based on the acquired coupon data.

[0164] The temporal behavior feature generation unit 702 is configured to extract features from user behavior data to generate temporal behavior features.

[0165] The atomic pseudo-label generation unit 703 is configured to call a clustering algorithm to determine the target clustering dimension based on temporal behavioral features and user attribute data, and then generate atomic pseudo-labels based on the target clustering dimension.

[0166] The high-order feature extraction unit 704 is configured to generate a knowledge graph based on atomic pseudo-labels and coupon content data, and extract high-order features from the knowledge graph.

[0167] The target recommendation coupon data determination unit 705 is configured to determine the central element corresponding to the coupon content data through a multi-head self-attention mechanism, and determine the target recommendation coupon data based on the high-order features and the central element.

[0168] In some embodiments, the high-order feature extraction unit 704 is further configured to: determine the coupon attributes corresponding to the coupon content data; generate a set of points based on the atomic pseudo-labels and the coupon attributes; generate a set of edges based on the relationship between the atomic pseudo-labels and the coupon attributes; and generate a knowledge graph based on the set of points and the set of edges.

[0169] In some embodiments, the temporal behavior feature generation unit 702 is further configured to: perform word embedding on user behavior data to convert it into a continuous behavior sequence; perform the following training process: input the continuous behavior sequence into the encoder in the denoising autoencoder network to obtain the encoded continuous behavior sequence; input the encoded continuous behavior sequence into the decoder in the denoising autoencoder network to reconstruct the original continuous behavior sequence; train the encoder and decoder in the denoising autoencoder network by minimizing the cross-entropy loss function to train the target denoising autoencoder network; and extract temporal behavior features from the user behavior data based on the trained target denoising autoencoder network.

[0170] In some embodiments, the atomic pseudo-label generation unit 703 is further configured to: determine the corresponding attribute label based on the user attribute data; for each attribute label, determine the corresponding optional dimension, and perform clustering based on the corresponding temporal behavior features under each optional dimension to obtain each cluster; if each cluster is non-empty, determine the corresponding optional dimension as the target clustering dimension.

[0171] In some embodiments, the atomic pseudo-label generation unit 703 is further configured to generate atomic pseudo-labels based on the definition data of the clusters corresponding to the target clustering dimension.

[0172] In some embodiments, the target recommendation coupon data determination unit 705 is further configured to: perform a nonlinear transformation on the coupon content data through a fully connected layer to obtain coupon content features; call a multi-head self-attention mechanism with the temporal behavior features and coupon content features as input parameters to obtain correlation data between the temporal behavior features and coupon content features; and predict the central element in the coupon content features based on the correlation data.

[0173] In some embodiments, the target recommendation coupon data determination unit 705 is further configured to: determine nodes related to the nodes corresponding to the central element in the knowledge graph based on higher-order features; calculate the similarity between the coupon content features corresponding to the related nodes and the central element; determine the recommendation probability of the related nodes based on the similarity; and determine the target recommendation coupon data based on the recommendation probability.

[0174] It should be noted that the coupon data recommendation method and coupon data recommendation device in this application are related in terms of specific implementation content, so the repeated content will not be described again.

[0175] Figure 8 An exemplary system architecture 800 is shown that can be applied to the coupon data recommendation method or coupon data recommendation apparatus of the embodiments of this application.

[0176] like Figure 8 As shown, system architecture 800 may include terminal devices 801, 802, and 803, a network 804, and a server 805. Network 804 serves as the medium for providing communication links between terminal devices 801, 802, and 803 and server 805. Network 804 may include various connection types, such as wired or wireless communication links or fiber optic cables, etc.

[0177] Users can use terminal devices 801, 802, and 803 to interact with server 805 via network 804 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 801, 802, and 803, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social media platform software, etc. (for example only).

[0178] Terminal devices 801, 802, and 803 can be various electronic devices with coupon data recommendation processing screens and support web browsing, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0179] Server 805 can be a server providing various services, such as a backend management server supporting user requests for coupon data recommendations submitted via terminal devices 801, 802, and 803 (this is just an example). The backend management server can obtain user behavior data, user attribute data, and coupon content data based on the received coupon data recommendation request; extract features from the user behavior data to generate time-series behavior features; call a clustering algorithm to determine the target clustering dimension based on the time-series behavior features and user attribute data, and then generate atomic pseudo-labels based on the target clustering dimension; generate a knowledge graph based on the atomic pseudo-labels and coupon content data, and extract high-order features from the knowledge graph; determine the central element corresponding to the coupon content data through a multi-head self-attention mechanism, and determine the target recommended coupon data based on the high-order features and the central element. By generating atomic pseudo-labels from multiple attribute dimensions through a clustering algorithm, constructing a knowledge graph using atomic pseudo-labels, and combining the high-order features and central element of the knowledge graph, coupon data can be predicted accurately and efficiently, improving the coupon data recommendation effect and increasing the accuracy and efficiency of coupon data recommendation.

[0180] It should be noted that the coupon data recommendation method provided in this application embodiment is generally executed by server 805, and correspondingly, the coupon data recommendation device is generally set in server 805.

[0181] It should be understood that Figure 8 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0182] The following is for reference. Figure 9 It shows a schematic diagram of the structure of a computer system 900 suitable for implementing a terminal device according to the embodiments of this application. Figure 9 The terminal device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0183] like Figure 9 As shown, the computer system 900 includes a central processing unit (CPU) 901, which can perform various appropriate actions and processes based on programs stored in read-only memory (ROM) 902 or programs loaded from storage section 908 into random access memory (RAM) 903. The RAM 903 also stores various programs and data required for the operation of the computer system 900. The CPU 901, ROM 902, and RAM 903 are interconnected via a bus 904. An input / output (I / O) interface 905 is also connected to the bus 904.

[0184] The following components are connected to I / O interface 905: an input section 906 including a keyboard, mouse, etc.; an output section 907 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 908 including a hard disk, etc.; and a communication section 909 including a network interface card such as a LAN card, modem, etc. The communication section 909 performs communication processing via a network such as the Internet. A drive 910 is also connected to I / O interface 905 as needed. A removable medium 911, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on drive 910 as needed so that computer programs read from it can be installed into storage section 908 as needed.

[0185] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 909, and / or installed from removable medium 911. When the computer program is executed by central processing unit (CPU) 901, it performs the functions defined above in the system of this application.

[0186] It should be noted that the computer-readable medium shown in this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. Computer-readable storage media can be, for example, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0187] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0188] The units described in the embodiments of this application can be implemented in software or hardware. The described units can also be housed in a processor; for example, a processor can be described as including an acquisition unit, a temporal behavior feature generation unit, an atomic pseudo-label generation unit, a high-order feature extraction unit, and a target recommendation coupon data determination unit. The names of these units do not necessarily constitute a limitation on the unit itself.

[0189] In another aspect, this application also provides a computer-readable medium, which may be included in the device described in the above embodiments; or it may exist independently and not assembled into the device. The computer-readable medium carries one or more programs that, when executed by the device, cause the device to obtain user behavior data, user attribute data, and coupon content data based on the obtained coupon data recommendation request; extract features from the user behavior data to generate temporal behavior features; invoke a clustering algorithm to determine the target clustering dimension based on the temporal behavior features and user attribute data, and then generate atomic pseudo-labels based on the target clustering dimension; generate a knowledge graph based on the atomic pseudo-labels and coupon content data, and extract high-order features from the knowledge graph; determine the central element corresponding to the coupon content data through a multi-head self-attention mechanism, and determine the target recommended coupon data based on the high-order features and the central element.

[0190] The computer program product of this application includes a computer program that, when executed by a processor, implements the coupon data recommendation method in the embodiments of this application.

[0191] According to the technical solution of this application embodiment, atomic pseudo-labels are generated from multiple attribute dimensions through clustering algorithms. A knowledge graph is constructed through atomic pseudo-labels, and the high-order features and central elements of the knowledge graph are combined to accurately and efficiently predict coupon data, thereby improving the coupon data recommendation effect and increasing the accuracy and efficiency of coupon data recommendation.

[0192] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can occur depending on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for recommending coupon data, characterized in that, include: Based on the obtained coupon data, a recommendation request is made to obtain user behavior data, user attribute data, and coupon content data; Feature extraction is performed on the user behavior data to generate time-series behavior features; A clustering algorithm is invoked to determine the target clustering dimension based on the temporal behavior features and the user attribute data, and then atomic pseudo-labels are generated according to the target clustering dimension; Based on the atomic pseudo-tags and the coupon content data, a knowledge graph is generated, and high-order features are extracted from the knowledge graph. The central element corresponding to the coupon content data is determined by a multi-head self-attention mechanism, and the target recommended coupon data is determined based on the higher-order features and the central element.

2. The method according to claim 1, characterized in that, The generation of the knowledge graph includes: Determine the coupon attributes corresponding to the coupon content data; Generate a point set based on the atomic pseudo-tags and the coupon attributes; Generate an edge set based on the relationship between the atomic pseudo-label and the coupon attribute; A knowledge graph is generated based on the set of points and the set of edges.

3. The method according to claim 1, characterized in that, The step of extracting features from the user behavior data to generate time-series behavior features includes: The user behavior data is word-embedded to convert it into a continuous behavior sequence; Perform the following training process: The continuous behavior sequence is input into the encoder in the denoising autoencoder network to obtain the encoded continuous behavior sequence; The encoded continuous action sequence is input into the decoder in the denoising autoencoder network to reconstruct the original continuous action sequence; The encoder and decoder in the denoising autoencoder network are trained by minimizing the cross-entropy loss function to obtain the target denoising autoencoder network. The trained target denoising autoencoder network extracts temporal behavioral features from the user behavior data.

4. The method according to claim 1, characterized in that, The invocation of the clustering algorithm to determine the target clustering dimension based on the temporal behavioral features and the user attribute data includes: Based on the user attribute data, determine the corresponding attribute tags; For each attribute label, determine the corresponding optional dimension, and perform clustering based on the corresponding temporal behavior features under each optional dimension to obtain each cluster. If all the clusters are non-empty, the corresponding optional dimension is determined as the target cluster dimension.

5. The method according to claim 4, characterized in that, The step of generating atomic pseudo-labels based on the target clustering dimension includes: Atomic pseudo-labels are generated based on the definition data of the clusters corresponding to the target clustering dimension.

6. The method according to claim 1, characterized in that, The process of determining the central element corresponding to the coupon content data through a multi-head self-attention mechanism includes: The coupon content data is non-linearly transformed using a fully connected layer to obtain coupon content features; The time-series behavioral features and the coupon content features are used as input parameters to invoke a multi-head self-attention mechanism to obtain correlation data between the time-series behavioral features and the coupon content features; Based on the correlation data, predict the central element in the coupon content features.

7. The method according to any one of claims 1 to 6, characterized in that, The process of determining the target recommended coupon data based on the higher-order features and the central element includes: Based on the higher-order features, determine the nodes in the knowledge graph that are related to the node corresponding to the central element; Calculate the similarity between the coupon content features corresponding to the relevant nodes and the central element, and determine the recommendation probability of the relevant nodes based on the similarity. Based on the recommendation probability, the target recommended coupon data is determined.

8. A coupon data recommendation device, characterized in that, include: The acquisition unit is configured to request user behavior data, user attribute data, and coupon content data based on the acquired coupon data. The temporal behavior feature generation unit is configured to extract features from the user behavior data to generate temporal behavior features; The atomic pseudo-label generation unit is configured to call a clustering algorithm to determine the target clustering dimension based on the temporal behavior features and the user attribute data, and then generate atomic pseudo-labels according to the target clustering dimension; The high-order feature extraction unit is configured to generate a knowledge graph based on the atomic pseudo-labels and the coupon content data, and extract high-order features from the knowledge graph; The target recommendation coupon data determination unit is configured to determine the central element corresponding to the coupon content data through a multi-head self-attention mechanism, and determine the target recommendation coupon data based on the higher-order features and the central element.

9. An electronic device for recommending coupon data, characterized in that, include: One or more processors; Storage device for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-7.

10. A computer-readable medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-7.

11. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-7.

Images