Data processing method and device, computer device and storage medium

Abstract

The application relates to a data processing method and device, computer equipment, a storage medium and a computer program product. First, the object attribute tree corresponding to a target object and the object original representation, the target object attribute tree corresponding to an object and the object original representation, and the basic data for data processing are acquired. Then, the information in the object attribute tree and the target object attribute tree is aggregated into attribute information, the target object attribute information is fused with the object original representation based on the object original representation to obtain an object fusion representation, the target object attribute information is fused with the object original representation based on the object original representation to obtain a target object fusion representation, and the target object and the target object are fused through double intentions. Finally, the data processing result of interest matching between the target object and the target object is obtained, and the accuracy of interest estimation is improved.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a data processing method, apparatus, computer equipment, and storage medium. Background Technology

[0002] With the development of computer technology and Artificial Intelligence (AI) technology, Machine Learning (ML) has emerged. Machine learning is a multidisciplinary field involving probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory, and many other disciplines. It specifically studies how computers can simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to endow computers with intelligence; its applications span all areas of artificial intelligence. For example, in the field of content recommendation, machine learning techniques can be used to predict the click-through rate of recommended content, thereby ensuring the effectiveness of content recommendations.

[0003] Current data processing for predicting click-through rates (CTR) of recommended content typically utilizes user interest models. However, while user interest models uncover users' interests in recommended content, they often overlook users' intentions regarding different attributes of the recommended content. Furthermore, the popularity of recommended content varies among user groups with different attributes. Therefore, using user interest models for interest-matching data processing results in relatively low accuracy. Summary of the Invention

[0004] Therefore, it is necessary to provide a data processing method, apparatus, computer equipment, computer-readable storage medium, and computer program product that can improve the accuracy of interest matching in response to the above-mentioned technical problems.

[0005] Firstly, this application provides a data processing prediction method. The method includes:

[0006] Obtain the object attribute tree that matches at least a portion of the attributes of the target object and the original object representation of the target object; and obtain the target attribute tree that matches at least a portion of the attributes of the target and the original target representation of the target.

[0007] Each of the object attribute trees is subjected to node information aggregation processing to obtain target object attribute information; each of the target object attribute trees is subjected to node information aggregation processing to obtain target object attribute information.

[0008] Based on the original representation of the target object, the attribute information of the target object is fused with the original representation of the object to obtain a fused representation of the object; based on the original representation of the object, the attribute information of the target object is fused with the original representation of the target object to obtain a fused representation of the target object.

[0009] Based on the object fusion representation, the target object fusion representation, the target object attribute information, and the target object attribute information, data matching processing is performed on the target object and the target object to obtain the data processing result of the target object on the target object.

[0010] Secondly, this application also provides a data processing apparatus. The apparatus includes:

[0011] The data acquisition module is used to acquire an object attribute tree that matches at least a portion of the attributes of the target object and the original object representation of the target object, and to acquire a target attribute tree that matches at least a portion of the attributes of the target object and the original target object representation of the target object.

[0012] The information aggregation module is used to perform node information aggregation processing on each of the object attribute trees to obtain target object attribute information, and to perform node information aggregation processing on each of the target object attribute trees to obtain target object attribute information.

[0013] The information fusion module is used to fuse the target object attribute information with the original object representation based on the original object representation to obtain a fused object representation, and to fuse the target object attribute information with the original object representation based on the original object representation to obtain a fused object representation.

[0014] The data processing module is used to perform data matching processing on the target object and the target object based on the object fusion representation, the target object fusion representation, the target object attribute information, and the target object attribute information, to obtain the data processing result of the target object on the target object.

[0015] 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:

[0016] Obtain the object attribute tree that matches at least a portion of the attributes of the target object and the original object representation of the target object; and obtain the target attribute tree that matches at least a portion of the attributes of the target and the original target representation of the target.

[0017] Each of the object attribute trees is subjected to node information aggregation processing to obtain target object attribute information; each of the target object attribute trees is subjected to node information aggregation processing to obtain target object attribute information.

[0018] Based on the original representation of the target object, the attribute information of the target object is fused with the original representation of the object to obtain a fused representation of the object; based on the original representation of the object, the attribute information of the target object is fused with the original representation of the target object to obtain a fused representation of the target object.

[0019] Based on the object fusion representation, the target object fusion representation, the target object attribute information, and the target object attribute information, data matching processing is performed on the target object and the target object to obtain the data processing result of the target object on the target object.

[0020] 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:

[0021] Obtain the object attribute tree that matches at least a portion of the attributes of the target object and the original object representation of the target object; and obtain the target attribute tree that matches at least a portion of the attributes of the target and the original target representation of the target.

[0022] Each of the object attribute trees is subjected to node information aggregation processing to obtain target object attribute information; each of the target object attribute trees is subjected to node information aggregation processing to obtain target object attribute information.

[0023] Based on the original representation of the target object, the attribute information of the target object is fused with the original representation of the object to obtain a fused representation of the object; based on the original representation of the object, the attribute information of the target object is fused with the original representation of the target object to obtain a fused representation of the target object.

[0024] Based on the object fusion representation, the target object fusion representation, the target object attribute information, and the target object attribute information, data matching processing is performed on the target object and the target object to obtain the data processing result of the target object on the target object.

[0025] 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:

[0026] Obtain the object attribute tree that matches at least a portion of the attributes of the target object and the original object representation of the target object; and obtain the target attribute tree that matches at least a portion of the attributes of the target and the original target representation of the target.

[0027] Each of the object attribute trees is subjected to node information aggregation processing to obtain target object attribute information; each of the target object attribute trees is subjected to node information aggregation processing to obtain target object attribute information.

[0028] Based on the original representation of the target object, the attribute information of the target object is fused with the original representation of the object to obtain a fused representation of the object; based on the original representation of the object, the attribute information of the target object is fused with the original representation of the target object to obtain a fused representation of the target object.

[0029] Based on the object fusion representation, the target object fusion representation, the target object attribute information, and the target object attribute information, data matching processing is performed on the target object and the target object to obtain the data processing result of the target object on the target object.

[0030] The aforementioned data processing methods, apparatus, computer equipment, storage media, and computer program products first acquire basic data for data processing, such as the object attribute tree and original representation of the target object, and the target object attribute tree and original representation of the target object. Then, through node information aggregation, the information in the object attribute tree and target object attribute tree is aggregated into attribute information. Next, based on the original representation of the target object, the target object attribute information is fused with the original object representation to obtain a fused object representation. Based on the original object representation, the target object attribute information is fused with the original target object representation to obtain a fused target object representation. This dual intention is used to fuse the target object and the target object. Finally, based on the obtained fused object representation, fused target object representation, target object attribute information, and target object attribute information, the data processing result of interest matching between the target object and the target object is obtained, improving the accuracy of interest prediction. Attached Figure Description

[0031] Figure 1 This is a diagram illustrating the application environment of a data processing method in one embodiment.

[0032] Figure 2 This is a flowchart illustrating a data processing method in one embodiment;

[0033] Figure 3 This is a schematic diagram of an object attribute tree in one embodiment;

[0034] Figure 4 This is a schematic diagram of the object attribute tree in one embodiment;

[0035] Figure 5 This is a schematic diagram of a dataset of the target objects in one embodiment;

[0036] Figure 6 This is a schematic diagram of the process for obtaining the object fusion representation in one embodiment;

[0037] Figure 7 This is an interaction diagram of a preset object target in one embodiment;

[0038] Figure 8 Here is a simplified flowchart of a data processing method in one embodiment;

[0039] Figure 9 This is a structural block diagram of a data processing device in one embodiment;

[0040] Figure 10 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0041] 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.

[0042] This application relates to the field of artificial intelligence, specifically the theories, methods, technologies, and application systems that utilize digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, artificial intelligence is a comprehensive technology within computer science that attempts to understand the essence of intelligence and produce a new type of intelligent machine capable of reacting in a manner similar to human intelligence. Artificial intelligence studies the design principles and implementation methods of various intelligent machines, enabling them to possess perception, reasoning, and decision-making capabilities. Artificial intelligence technology is a comprehensive discipline involving a wide range of fields, encompassing both hardware and software technologies. Fundamental artificial intelligence technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operating / interactive systems, and mechatronics. Artificial intelligence software technologies mainly include computer vision, speech processing, natural language processing, and machine learning / deep learning. This application specifically addresses machine learning technology within artificial intelligence.

[0043] In this article, it is important to understand the following terms:

[0044] Deep Neural Network (DNN): A neural network with multiple hidden layers, it is currently the most popular and effective click-through rate prediction model.

[0045] Recurrent Neural Network (RNN): A type of recurrent neural network that takes sequential data as input, recursively moves in the direction of the sequence, and all nodes (recurrent units) are connected in a chain-like manner.

[0046] Gated Recurrent Unit (GRU): A type of recurrent neural network designed to solve the vanishing gradient problem that occurs in standard recurrent neural networks.

[0047] Embedding: A method for converting discrete variables into continuous vector representations. Deep neural network models commonly convert discrete features into embedding vectors, which are then concatenated or otherwise processed before being used as the model's input layer.

[0048] Sum / MeanPooling: Sum pooling / mean pooling is used to sum or average all values ​​in a local receptive domain.

[0049] Click-Through Rate (CTR).

[0050] Conversion Rate (CVR).

[0051] The data processing method provided in this application embodiment can be applied to, for example... Figure 1In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104, or it can be located in the cloud or on another server. When staff on terminal 102 need to recommend content to individuals in the database, they need to determine the target of their data processing. Therefore, the data processing method of this application can be used to process the data between the target and the object, improving the recommendation success rate. First, the staff at terminal 102 can send the target object and the target item to server 104. Server 104 obtains the object attribute tree and the original object representation of the target object that match at least some of the attributes of the target object, and obtains the target item attribute tree and the original target item representation of the target item that match at least some of the attributes of the target item. For each object attribute tree, node information aggregation processing is performed to obtain the target object attribute information; for each target item attribute tree, node information aggregation processing is performed to obtain the target item attribute information. Based on the original target item representation, the target object attribute information and the original object representation are fused to obtain the object fused representation. Based on the original object representation, the target item attribute information and the original target item representation are fused to obtain the target item fused representation. Based on the object fused representation, the target item fused representation, the target object attribute information, and the target item attribute information, data matching processing is performed on the target object and the target item to obtain the data processing result of the target object for the target item. Then, terminal 102 can recommend the target item to the target object based on the data processing result. The terminal 102 can be, but is not limited to, various desktop computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, and smart in-vehicle systems. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted devices. The server 104 can be implemented using a standalone server or a server cluster consisting of multiple servers.

[0052] In one embodiment, such as Figure 2 As shown, a data processing method is provided, which can be applied to... Figure 1 Taking server 104 as an example, the following steps are included:

[0053] Step 201: Obtain the object attribute tree that matches at least a portion of the attributes of the target object and the original object representation of the target object, and obtain the target attribute tree that matches at least a portion of the attributes of the target object and the original target object representation of the target object.

[0054] In this context, the target object refers to the matching object specified by the data processing method of this application, specifically a user, such as a user of a video application or a game user. The subject matter refers to items, information, or other data that can be recommended to the target object. Specifically, the subject matter can be objects such as books or content information such as video advertisements. Whether to recommend the subject matter to the target object can be determined based on the data processing results of the target object. Data processing involves estimating the target object's level of interest in the subject matter, thereby recommending the subject matter that the target object is interested in, or recommending the subject matter to other interested objects. Attributes can be understood as tags carried by objects or subject matters. For objects, attributes may include gender, age, education level, region, and preferences. For subject matters, attributes may include subject matter number, manufacturer, industry, category, and brand. Attributes are specifically categorized according to the domain of the data processing application. The object attribute tree is a data structure built based on the dependencies between different attributes. Different object attribute trees can be constructed based on the dependencies between different attributes. For example, in a specific implementation, the target object is a specific user, and the object attributes specifically include gender, age, city, country, etc. The object attribute tree that can be constructed in this case can be referenced... Figure 3 As shown, there is a clear dependency between the user's city and country. Therefore, the user's city can be considered the parent node of the user, and the country can be considered the parent node of the user's city. This allows for the construction of three object attribute trees, each matching a subset of related object attributes. Similarly, in another embodiment, the target object is an advertisement. In this case, the target object attributes specifically include advertiser, industry, site, category, and brand. It can be identified that advertisers and industries have hierarchical relationships, and sites and categories also have hierarchical relationships. Based on these hierarchical relationships, three target object attribute trees can be constructed, each matching a subset of related target object attributes. The specific construction of the target object attribute trees can be found in [reference needed]. Figure 4 As shown, the original representation of an object is a vector representation obtained by embedding the target object. Embedding can transform a discrete variable into a continuous vector representation. Similarly, embedding can transform a target object into its corresponding continuous vector. Likewise, the object itself can be transformed into its corresponding continuous vector, i.e., the original representation of the object.

[0055] Specifically, when staff need to recommend items to a target object or recommend items to different objects, they can first match the object and the item to determine the data processing result of the object for the item. After confirming that the object is interested in the item, the item is then recommended to the user, improving the recommendation success rate. At this time, staff can specify several objects and several items to the server 104 through terminal 102 according to actual needs. For example, if a staff member needs to recommend several items to one object, they can specify one target object and multiple items to be recommended. If a staff member needs to recommend one item to multiple objects, they can specify multiple target objects and one item to be recommended. Similarly, a target user can specify one object and one item, determine the data processing result between them, or specify multiple objects and multiple items, determine the data processing result between each object and each item. After receiving the target object and item specified by terminal 102, server 104 can perform data processing accordingly, matching one target object and one item at a time. First, an object attribute tree matching at least some attributes of the target object is obtained. Simultaneously, the specified target object is embedded to obtain its original object representation. Then, a target object attribute tree matching at least some attributes of the target object is obtained, and the specified target object is embedded to obtain its original target object representation. In one embodiment, the data processing method of this application is specifically used for content recommendation in a video application. Here, the target object can specifically refer to a specific user of the video application, and the target objects are multiple pieces of content to be recommended. Recommendation involves identifying the parts of the content that the target object is most interested in from the multiple pieces of content to be recommended. First, the server obtains the object attribute tree corresponding to the video user, and simultaneously converts the specified user into its original object representation through embedding. Then, it selects one piece of content from the multiple pieces of content to be recommended as the target object for this matching, obtains the object attribute tree corresponding to the target object, and converts the target object into its original object representation through embedding.

[0056] Step 203: Perform node information aggregation processing on each object attribute tree to obtain target object attribute information, and perform node information aggregation processing on each target attribute tree to obtain target attribute information.

[0057] Node information aggregation refers to extracting information from different nodes in an object attribute tree or target attribute tree into the parent node. Aggregation can effectively improve the processing efficiency and accuracy of data processing. Target object attribute information refers to the attribute information obtained by combining an attribute with the attributes of its child nodes. Optimized target object attribute information can effectively display the attributes possessed by the target object. Similarly, optimized target attribute information can be obtained.

[0058] Specifically, existing feature interaction methods convert the attribute features of the target object into one-hot vectors, i.e., vectors with exactly one element equal to 1 and all other elements equal to 0. These features are then input into a machine learning model for data processing. However, the applicant discovered that these attributes have relationships and dependencies. For example, in the advertising field, an advertisement belongs to an advertiser, and an advertiser corresponds to an industry. Therefore, simply using sparse vectors to represent each attribute feature is insufficient. To improve the accuracy of data processing, the implicit structural information should be taken into account. Therefore, the solution of this application mines the implicit structural information between attributes by building an attribute tree. Thus, after obtaining the object attribute tree that matches at least some attributes of the target object, and the target attribute tree that matches at least some attributes of the target object, in order to optimize the attribute representation, the node information of the object attribute tree and the target attribute tree can be aggregated by aggregation, gathering the node information into higher-level nodes, realizing the fusion of related attributes in the same object attribute tree, and simultaneously realizing the fusion of related attributes in the same target attribute tree.

[0059] Step 205: Based on the original representation of the target object, the target object attribute information is fused with the original representation of the object to obtain the object fused representation.

[0060] Object fusion representation specifically characterizes an object's intention towards different target attribute information, while target fusion representation characterizes a target's intention towards different target object attribute information. Object fusion representation can show the popularity of different target attributes for an object, while target fusion representation can show the popularity of different target object attributes for a target.

[0061] Specifically, existing data processing methods often focus on the interest of research subjects in different objects, but neglect users' intentions regarding different object attributes. For example, a user clicks on an ad because they like that type of ad. Secondly, the popularity of an ad varies among different target groups. For instance, young women might prefer clicking on beauty ads, while young men tend to click on game ads. Therefore, it is necessary to mine and model the intentions for different attributes to enhance the interpretability of the model. That is, this application can fuse the optimized object attribute information with the original object representation to mine users' intentions for different object attributes, thus obtaining a fused object representation. Similarly, the optimized target object attribute information can be fused with the original object representation to mine users' intentions for different object attributes, thus obtaining a fused object representation. In a specific embodiment, the solution of this application is used to identify the interest of a target object in different types of advertisements. This can be achieved by aggregating node information to obtain the attribute information of different types of objects corresponding to the advertisements, and the attribute information of different types of target objects corresponding to the target object. Then, for each type of object attribute or target object attribute information, an attention mechanism can be used to assign a corresponding weight. The original representation of the object can then be fused with the target object attribute information of each type to obtain a fused representation of the object's intention towards each object attribute information. Similarly, the original representation of the object can be fused with the target object attribute information of each type to obtain a fused representation of the object's intention towards each object attribute information.

[0062] Step 207: Based on the object fusion representation, the target object fusion representation, the target object attribute information, and the target object attribute information, perform data matching processing on the target object and the target object to obtain the data processing result of the target object on the target object.

[0063] The purpose of data matching processing is specifically to find objects of interest to a target object, or to match objects of interest to a target object. The data processing result can be represented by the matching degree between the target object and the object; the higher the matching degree, the greater the target object's interest in the object.

[0064] Specifically, the solution of this application is applied to the input layer of a machine learning model. By optimizing the data of the input data processing model, more accurate matching is achieved. Therefore, after obtaining optimized object representations, target identifiers, and attribute representations such as object fusion representation, target object fusion representation, target object attribute information, and target object attribute information, the target object's interest and the target object's interest can be matched based on a preset machine learning model to obtain the corresponding data processing results. In one embodiment, the method of this application is specifically used to push video advertisements to a target object based on the data processing results. Therefore, the target object is a video advertisement. At this time, the target object can be processed with multiple candidate video advertisements to obtain the matching degree of the target object for different video advertisements, and finally, several video advertisements with high matching degrees are pushed to the target object. In another embodiment, the method of this application is used to push target video advertisements to users based on the data processing results. At this time, the target video advertisement can be processed with multiple candidate users to obtain the matching degree of the target video advertisement for different users, and finally, the target video advertisement is pushed to several users with high matching degrees.

[0065] The aforementioned data processing method first acquires the object attribute tree and original representation of the target object, as well as the target object attribute tree and original representation of the target object, as basic data for data processing. Then, through node information aggregation, the information in the object attribute tree and target object attribute tree is aggregated into attribute information. Next, based on the original representation of the target object, the target object attribute information is fused with the original object representation to obtain a fused object representation. Based on the original object representation, the target object attribute information is fused with the original target object representation to obtain a fused target object representation. This dual intention is used to fuse the target object and the target object. Finally, based on the obtained fused object representation, fused target object representation, target object attribute information, and target object attribute information, the data processing result of interest matching between the target object and the target object is obtained, improving the accuracy of interest prediction.

[0066] In one embodiment, step 203 includes: identifying the hierarchical relationship between each node in each object attribute tree; identifying the child nodes corresponding to each node in the object attribute tree based on the hierarchical relationship between each node; and aggregating the node information of the child nodes in each object attribute tree into the corresponding parent node of the child node to obtain the target object attribute information corresponding to each object attribute tree.

[0067] The object attribute tree is composed of attributes and objects. Objects are the leaf nodes in the attribute tree, while attributes occupy parent nodes at different levels based on their dependencies. Therefore, when aggregating node information, it is necessary to first identify the hierarchical relationships between different nodes before aggregating lower-level node information into higher-level nodes to ensure the effectiveness of the aggregation. After aggregating the node information of child nodes into parent nodes, the target object attribute information corresponding to the object attribute tree can be obtained through the node information of multiple parent nodes.

[0068] Specifically, when aggregating information from nodes in an object attribute tree, graph convolution is typically used. However, different attributes possess different semantic information; for example, the semantics and data distribution of the attributes "advertiser" and "product price" are completely different. Using traditional graph convolution for aggregation may introduce additional noise when aggregating information from different neighboring nodes. Therefore, this application proposes a bottom-up aggregation strategy to fuse node information in the object attribute tree. When learning the attribute information of a parent node in the object attribute tree, the information of its corresponding lower-level child nodes can be aggregated from bottom to top into the parent node, resulting in an aggregated parent node attribute representation. Then, the attribute representation of the parent node can be further aggregated into a higher-level parent node. Finally, the target object attribute information corresponding to each object attribute tree is obtained through the aggregated attribute representation of the parent node. The combined target object attribute information from all object attribute trees constitutes the attribute information corresponding to the target object. Similarly, for the aggregation of node information in the target item attribute tree, the node information in the target item attribute tree can be aggregated from the bottom up from the lower-level nodes to the corresponding upper-level parent nodes. Finally, by combining all the target item attribute information corresponding to the target item attribute trees, the attribute information corresponding to the target item is obtained. In one embodiment, the solution of this application is used to implement user data processing for advertisements, where the target item is specifically an advertisement, and its target item attribute tree can be specifically referred to... Figure 4 As shown. When node information aggregation is required, for the first target attribute tree, the information of the three ads under the advertiser can be aggregated into the parent advertiser to obtain two attribute information about the advertiser. Then, the attribute information of the two advertisers can be aggregated into the industry to obtain one attribute information about the industry. Similarly, the attribute information of two sites, the attribute information of a category, and the attribute information of a brand can be obtained. Combining these attribute information yields the target attribute information corresponding to the ad. In this embodiment, bottom-up node information aggregation can effectively reduce noise in the attribute aggregation process, thereby ensuring the accuracy of data processing.

[0069] In one embodiment, aggregating the node information of child nodes in each object attribute tree to the corresponding parent node of the child node to obtain the target object attribute information corresponding to each object attribute tree includes: calculating the Hadamard product between the parent node and the child node to obtain a first aggregation result; concatenating the parent node and the child node into vectors to obtain a second aggregation result; and obtaining the target object attribute information corresponding to each object attribute tree based on the first aggregation result and the second aggregation result.

[0070] The Hadamard product, or element-wise product, is symbolized by ⊙. It refers to the "product at the element level," where two vectors are multiplied element by element to form a new vector with the same size as the original vectors. The symbol for vector concatenation is ⊕, which refers to joining two vectors into a single vector.

[0071] Specifically, the solution in this application uses two interactive modeling methods to aggregate information from child nodes into parent nodes, thereby obtaining an optimized attribute representation of the parent node, and ultimately obtaining the target object attribute information corresponding to the object attribute tree. In one embodiment, aggregation can be implemented using the following aggregation function.

[0072]

[0073] Among them, e H Indicates the parent node, Represents the set of child nodes. Indicates child nodes The information in the middle is aggregated into the parent node e. h The attribute representation of the parent node obtained afterwards. This is the first aggregation result. The result represents the second aggregation. σ is the activation function, specifically the sigmoid activation function (S-shaped growth curve) used here. W1, W2, and W3 are parameter matrices, randomly initialized, and can be updated with the gradient during the training of the aggregation function. In this embodiment, the node information of child nodes is aggregated through Hadamard product and vector concatenation, which effectively ensures the validity of the node information aggregation process and thus guarantees the accuracy of data processing.

[0074] In one embodiment, obtaining an object attribute tree that matches at least a portion of the attributes of a target object includes: obtaining an object dataset; traversing the object dataset based on a preset object attribute tree structure to construct an object attribute tree.

[0075] The preset object attribute tree is specifically selected and constructed by staff based on the target objects and attributes required for matching. The object dataset contains several object data sets, and each object data set contains specific numerical values ​​for various object attributes. Similarly, the target attribute tree can also be constructed based on the target dataset, which contains multiple target data sets, each of which contains specific numerical values ​​for various target attributes.

[0076] Specifically, when data processing is required, server 104 needs to pre-construct an object attribute tree that matches at least some attributes of the target object, and also needs to construct a target attribute tree that matches at least some attributes of the object. Staff can determine the objects and targets to be matched according to actual data processing needs, then specify the object attributes and target attributes to be matched, and construct the corresponding object dataset and target dataset based on the specified object attributes and target attributes. Simultaneously, staff can refer to the relationships between object attributes and target attributes to construct corresponding preset attribute trees. For example, in the advertising field, the target attributes of advertising specifically include advertiser, industry, site, category, and brand. The target dataset can be specifically referenced... Figure 5 As shown, by summarizing the dependencies between attributes, the target attribute tree can be obtained. The first target attribute tree is: Advertisement-Advertiser-Industry; the second is: Advertisement-Site-Category; and the third is: Advertisement-Brand. After obtaining these three preset target attribute trees, the target dataset is traversed to populate the target attribute trees with the data from the dataset, resulting in the final usable target attribute tree. Similarly, the object attribute tree can be constructed by referring to the target attribute tree. In this embodiment, by traversing the dataset, both the object attribute tree and the target attribute tree can be effectively constructed, thereby effectively utilizing the dependencies between attributes and ensuring the accuracy of data processing.

[0077] In one embodiment, such as Figure 6 As shown, step 205 includes:

[0078] Step 601: Based on the original representation of the target object, obtain the attribute weights corresponding to various target object attributes. The attribute weights corresponding to various target object attributes are determined through an attention mechanism.

[0079] Step 603: Based on the attribute weights corresponding to various target object attributes, the target object attribute information is fused with the original object representation to obtain the object attribute fusion result.

[0080] Step 605: Based on the object attribute fusion result, perform convolution processing on the preset object target interaction graph through graph convolution to obtain the object representations corresponding to each order of neighbor nodes in the preset object target interaction graph.

[0081] Step 607: Perform sum pooling on the object representations corresponding to each level of neighbor nodes to obtain the object fusion representation.

[0082] The attribute weights corresponding to the target object attributes are weight parameters dynamically learned through an attention mechanism, used to characterize the popularity of the target object among users with different attributes. The attention mechanism is a data processing method in machine learning, widely used in various types of machine learning tasks such as natural language processing, image recognition, and speech recognition. It is mainly used to select the most critical information for the current task objective from a large amount of information. In this application, the weight allocation is achieved through an attention mechanism. The preset object-target interaction graph is established using a dataset of interactions between objects and targets. For example, if the dataset is an advertising dataset, the daily samples include object features, advertising features, and whether a click occurred. If an object clicks an advertisement, an edge is connected between the user and the advertisement, ultimately forming a preset object-target interaction graph. An example of a preset object-target interaction graph can be found by referring to [reference needed]. Figure 7 As shown in the diagram, the objects include u1, u2, u3, u4, and u5, while the targets include i1, i2, i3, i4, and i5. The lines connecting them represent their interaction relationships. For example, regarding advertisements, if user u1 clicks on both advertisements i1 and i2, then u1 is connected to both i1 and i2. Neighbor nodes refer to the points connected to the current node. A first-order neighbor node is a node connected via an edge, and a second-order neighbor node is a first-order node connected via another edge, and so on. For example, for... Figure 7 In this application, u1's first-order neighbors are i1 and i2, and its second-order neighbors are u2 and u3. Graph convolution is a deep learning method used to extract features from graph data. In this application, graph convolution is used to generate object representations corresponding to each order of neighbors. Sum pooling is used to sum and pool the object representations of each order of neighbors to obtain the final object representation.

[0083] Specifically, in the fusion processing of target object attribute information and the original representation of the object, the present application first needs to determine the attribute weights corresponding to various object attributes through the original representation of the target object. After combining the various target object attributes with their corresponding weights, they are fused into the original representation of the object to obtain the corresponding object attribute fusion result. After obtaining the object attribute fusion result, the object representations under each order of neighbor nodes in the graph network are obtained through graph convolution, thereby fusing the interaction information between the object and the target object into the user representation. In one embodiment, when fusing the target object attribute information and the original representation of the object, α(v,a) can be used to represent the popularity of target object v for different object attributes a. The calculation formula for determining the attribute weights through the attention mechanism is as follows:

[0084]

[0085] in, For the original representation of the object v, e a This is an embedded representation of attribute 'a' of the target object. u It is a collection constructed from all the attributes of the target objects. The formula for calculating the result of fusing all the target object attribute information with the original object representation is as follows:

[0086]

[0087] After determining the object attribute fusion result using the above formula, the obtained result can be... The data is fed into the graph network to obtain the object representations of subsequent neighbor nodes at each order in the object interaction graph of the preset object target. The specific object representations of each order neighbor node can be calculated using the following formula:

[0088]

[0089] Where, N u Let represent the neighbor nodes of the target object u. After determining each neighbor node of the target object u using the above formula, when considering the information of L-order neighbor nodes, we can obtain the representations of L objects. Finally, we perform comprehensive pooling to obtain the final optimized object representation. The specific calculation formula is as follows:

[0090]

[0091] In this embodiment, the popularity of the target object with different attributes is determined by the attention mechanism, and the attribute weight is determined accordingly. This allows the attributes of each object to be integrated into the original representation of the object. Furthermore, the object representation is optimized by methods such as graph convolution, which can effectively ensure the effectiveness of the object representation and thus improve the accuracy of data processing.

[0092] In one embodiment, step 205 further includes: obtaining attribute weights corresponding to various target attributes based on the original object representation, wherein the attribute weights corresponding to various target attributes are determined through an attention mechanism; fusing the target attribute information with the original target representation based on the attribute weights corresponding to various target attributes to obtain a target attribute fusion result; performing convolution processing on a preset object-target interaction graph through graph convolution based on the target attribute fusion result to obtain target representations corresponding to neighbor nodes of each order in the preset object-target interaction graph; and performing sum pooling processing on the target representations corresponding to neighbor nodes of each order to obtain a target fusion representation.

[0093] The attribute weights corresponding to the target attributes are weight parameters dynamically learned through an attention mechanism, used to represent users' opinions on different target attributes. Attention mechanisms are a data processing method in machine learning, widely used in various types of machine learning tasks such as natural language processing, image recognition, and speech recognition. They are primarily used to select information that is more crucial to the current task objective from a large amount of information.

[0094] Specifically, referring to the object attribute fusion process, the solution of this application, when fusing the target attribute information and the original representation of the target, firstly needs to determine the attribute weights corresponding to various target attributes through the original representation of the object. After combining the various target attributes with their corresponding weights, they are fused into the original representation of the target to obtain the corresponding target attribute fusion result. After obtaining the target attribute fusion result, the target representations under each order of neighbor nodes in the graph network are obtained through graph convolution, thereby fusing the interaction information between the object and the target into the target representation. In one embodiment, when fusing the target attribute information and the original representation of the target, α(u,a) can be used to represent the intention of object u towards different target attributes a. The calculation formula for determining the attribute weights through the attention mechanism is as follows:

[0095]

[0096] in, e is the primitive representation of object u. a This is an embedded representation of attribute 'a' of the target object. v It is a collection constructed from all the attributes of the target objects. The formula for calculating the object attribute fusion result by fusing all the target object attribute information with the original representation of the target objects is:

[0097]

[0098] After determining the object attribute fusion result using the above formula, the obtained result can be... The data is fed into the graph network to obtain the object representations of subsequent neighbor nodes at each order in the object interaction graph of the preset object target. The specific object representations of each order neighbor node can be calculated using the following formula:

[0099]

[0100] Where, N v This represents the neighbor nodes of the target object v. After determining each neighbor node of the target object v using the above formula, when considering the information of L-order neighbor nodes, we can obtain the representations of L objects. Finally, we perform comprehensive pooling to obtain the final optimized object representation. The specific calculation formula is as follows:

[0101]

[0102] In this embodiment, the attention mechanism is used to determine the object's intention for different target attributes, thereby determining the attribute weights. This allows each target attribute to be integrated into the original target representation, and the target representation is optimized through graph convolution and other methods. This effectively ensures the validity of the target representation and improves the accuracy of data processing.

[0103] In one embodiment, step 207 includes: inputting the object fusion representation, the target object fusion representation, the target object attribute information, and the target object attribute information into a preset feature interaction model to perform data matching processing on the target object and the target object, and obtain the data processing result of the target object on the target object.

[0104] Specifically, the scheme of this application is applied to the input layer of a machine learning model to optimize the data in the output preset feature interaction model. The object fusion representation, target object fusion representation, target object attribute information, and target object attribute information obtained in this application's scheme are all represented in the form of embedding vectors. After obtaining these data, the object fusion representation, target object fusion representation, target object attribute information, and target object attribute information can be combined to construct a set of input data. This input data is then input into the preset feature interaction model, where feature extraction, feature matching, and feature cross-processing are performed to ultimately obtain the data processing results between the target object and the target object. Specifically, the preset feature interaction model used in this application can be a DeepFM (Deep Factorization Machine) model, and the scheme of this application is specifically used to recommend video advertisements to users of video websites. Therefore, the object is the user, and the target object is the video advertisement. After obtaining the user fusion representation, video advertisement fusion representation, user attribute information, and video attribute information, these four pieces of information can be grouped into a single set of data, and then the obtained data is input into the already trained DeepFM model. In the DeepFM model, these data exhibit explicit interactions within the FM structure and implicit interactions within the DNN structure. The two are then summed, and the model output is the predicted matching value between the user and the video ad. The server uses this matching value to determine whether the user and video ad in the input data set match. Finally, based on the data processing results for each set of data, video ads are recommended to the user. In this embodiment, by using a preset feature interaction model combined with four input layer data—object fusion representation, target object fusion representation, target object attribute information, and target object attribute information—obtained in this application, the accuracy of data processing can be effectively improved.

[0105] This application also provides an application scenario in which the above-described data processing method is applied. Specifically, the data processing method is applied in this scenario as follows:

[0106] When video website operators need to push video ads to users, in order to improve the success rate of ad delivery, they can pre-determine the interests of target users. At this time, the data processing method described in this application can be used to determine the data processing results between the target user and multiple video ads to be recommended, determine the matching degree between the user and the video ads, and then push video ads with a high matching degree to the target user to complete the video ad recommendation process. The overall flowchart can be found here. Figure 8As shown, firstly, the server used for data processing can establish the target item attribute tree corresponding to the video advertisement and the object attribute tree corresponding to the target user. Simultaneously, embedding encoding is performed on the target user and the video advertisement to obtain the original object representation of the target user and the original target item representation of the video advertisement. In this process, the object attribute tree can be constructed by traversing the object dataset corresponding to the target user, while the target item attribute tree is constructed by traversing the target item dataset of the video advertisement. After obtaining the object attribute tree and the target item attribute tree, node information aggregation processing can be performed on each object attribute tree to obtain the target object attribute information, and node information aggregation processing can be performed on each target item attribute tree to obtain the target item attribute information. Specifically, the aggregation process can adopt a bottom-up aggregation approach, that is, through Hadamard product and vector concatenation, based on the dependencies between attributes, the node information in the lower-level nodes is aggregated into the parent node of the upper-level node to obtain the optimized attribute representation. Then, the obtained object attribute information needs to be fused with the original object representation to optimize the object representation. This process needs to consider the popularity of the target item for objects with different object attributes. First, based on the original representation of the target object, the attribute weights corresponding to various target object attributes are obtained, and the attribute weights corresponding to various target object attributes are determined through an attention mechanism. Based on the attribute weights corresponding to various target object attributes, the target object attribute information is fused with the original object representation to obtain the object attribute fusion result. Based on the object attribute fusion result, the preset object-target interaction graph is convolved through graph convolution to obtain the object representations corresponding to each order of neighbor nodes in the preset object-target interaction graph. The object representations corresponding to each order of neighbor nodes are then summed and pooled to obtain the object fusion representation. Similarly, for optimizing the original target representation, we can first obtain the attribute weights corresponding to various target attributes based on the original object representation. These attribute weights are determined through an attention mechanism. Based on these attribute weights, the target attribute information is fused with the original target representation to obtain the target attribute fusion result. Based on the target attribute fusion result, a graph convolution process is applied to the preset object-target interaction graph to obtain the target representations corresponding to each order of neighbor nodes in the preset object-target interaction graph. The target representations corresponding to each order of neighbor nodes are then summed and pooled to obtain the target fusion representation. Finally, the four pieces of information—the object fusion representation, the target fusion representation, the target object attribute information, and the target attribute information—are input into a preset feature interaction model to perform data matching processing on the target object and the target, obtaining the data processing result of the target object on the target. By processing the data of target users and multiple video ads to be recommended, the recommendation score of each video ad can be obtained. Then, based on the video ads with higher recommendation scores, the video ad recommendation process is achieved by recommending the video ads to the target users.

[0107] It should be understood that although the steps in the flowcharts of the above embodiments 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 above embodiments 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.

[0108] Based on the same inventive concept, this application also provides a data processing apparatus for implementing the data processing 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 data processing apparatus embodiments provided below can be found in the limitations of the data processing method described above, and will not be repeated here.

[0109] In one embodiment, such as Figure 9 As shown, a data processing apparatus is provided, comprising:

[0110] The data acquisition module 902 is used to acquire an object attribute tree that matches at least a portion of the attributes of the target object and the original object representation of the target object, and to acquire a target attribute tree that matches at least a portion of the attributes of the target object and the original target object representation of the target object.

[0111] The information aggregation module 904 is used to perform node information aggregation processing on each object attribute tree to obtain target object attribute information, and to perform node information aggregation processing on each target attribute tree to obtain target attribute information.

[0112] The information fusion module 906 is used to fuse the target object attribute information with the original object representation based on the original object representation to obtain the object fusion representation.

[0113] The data processing module 908 is used to perform data matching processing on the target object and the target object based on the object fusion representation, the target object fusion representation, the target object attribute information, and the target object attribute information, so as to obtain the data processing result of the target object on the target object.

[0114] In one embodiment, the information aggregation module 904 is specifically used to: identify the hierarchical relationship between each node in each object attribute tree; identify the child nodes corresponding to each node in the object attribute tree based on the hierarchical relationship between each node; and aggregate the node information of the child nodes in each object attribute tree to the corresponding parent node of the child node to obtain the target object attribute information corresponding to each object attribute tree.

[0115] In one embodiment, the information aggregation module 904 is further configured to: calculate the Hadamard product between the parent node corresponding to the child node and the child node to obtain a first aggregation result; concatenate the parent node corresponding to the child node and the child node into vectors to obtain a second aggregation result; and obtain the target object attribute information corresponding to each object attribute tree based on the first aggregation result and the second aggregation result.

[0116] In one embodiment, the data acquisition module 902 is specifically used to: acquire an object dataset; traverse the object dataset based on a preset object attribute tree structure, and construct an object attribute tree.

[0117] In one embodiment, the information fusion module 906 is specifically used for: obtaining attribute weights corresponding to various target object attributes based on the original representation of the target object, wherein the attribute weights corresponding to various target object attributes are determined through an attention mechanism; fusing the target object attribute information with the original object representation based on the attribute weights corresponding to various target object attributes to obtain an object attribute fusion result; performing convolution processing on a preset object-target interaction graph through graph convolution based on the object attribute fusion result to obtain object representations corresponding to neighbor nodes of each order in the preset object-target interaction graph; and performing sum pooling processing on the object representations corresponding to neighbor nodes of each order to obtain an object fusion representation.

[0118] In one embodiment, the information fusion module 906 is specifically used for: obtaining attribute weights corresponding to various target attributes based on the original representation of the object, wherein the attribute weights corresponding to various target attributes are determined through an attention mechanism; fusing the target attribute information with the original representation of the target based on the attribute weights corresponding to various target attributes to obtain a target attribute fusion result; performing convolution processing on a preset object-target interaction graph through graph convolution based on the target attribute fusion result to obtain the target representations corresponding to each order of neighbor nodes in the preset object-target interaction graph; and performing sum pooling processing on the target representations corresponding to each order of neighbor nodes to obtain the target fusion representation.

[0119] In one embodiment, the data processing module 908 is specifically used to: input the object fusion representation, the target object fusion representation, the target object attribute information, and the target object attribute information into a preset feature interaction model, so as to perform data matching processing on the target object and the target object, and obtain the data processing result of the target object on the target object.

[0120] Each module in the aforementioned data processing 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.

[0121] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 10 As shown, this computer device includes a processor, memory, input / output interfaces (I / O), 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 operation of the operating system and computer programs stored in the non-volatile storage media. The database stores data related to the data processing process. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements a data processing module method.

[0122] Those skilled in the art will understand that Figure 10 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.

[0123] 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.

[0124] In one embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0125] In one embodiment, a computer program product or computer program is provided, the computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and executes the computer instructions, causing the computer device to perform the steps in the above method embodiments.

[0126] 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.

[0127] 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. When executed, the computer program 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.

[0128] 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.

[0129] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of this 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. A data processing method, characterized in that, The method includes: Obtain an object attribute tree that matches at least a portion of the attributes of the target object and the original object representation of the target object, and obtain a target attribute tree that matches at least a portion of the attributes of the target object and the original object representation of the target object, wherein the original object representation of the target object is a vector representation obtained by converting the target object through an embedding method, and the original object representation of the target object is a vector representation obtained by converting the target object through an embedding method. Each of the object attribute trees is subjected to node information aggregation processing to obtain target object attribute information; each of the target object attribute trees is subjected to node information aggregation processing to obtain target object attribute information. Based on the original representation of the target object, the attribute information of the target object is fused with the original representation of the object to obtain a fused representation of the object; based on the original representation of the object, the attribute information of the target object is fused with the original representation of the target object to obtain a fused representation of the target object. Based on the object fusion representation, the target object fusion representation, the target object attribute information, and the target object attribute information, data matching processing is performed on the target object and the target object to obtain the data processing result of the target object on the target object.

2. The method according to claim 1, characterized in that, The step of aggregating node information for each of the object attribute trees to obtain the target object attribute information includes: Identify the hierarchical relationships between nodes in each of the object attribute trees; Identify the child nodes corresponding to each node in the object attribute tree based on the hierarchical relationship of each node; The node information of the child nodes in each of the object attribute trees is aggregated into the corresponding parent node of the child node to obtain the target object attribute information corresponding to each of the object attribute trees.

3. The method according to claim 2, characterized in that, The step of aggregating the node information of child nodes in each of the object attribute trees into the corresponding parent node of the child node to obtain the target object attribute information corresponding to each of the object attribute trees includes: Calculate the Hadamard product between the parent node corresponding to the child node and the child node to obtain the first aggregation result; The vectors of the parent node corresponding to the child node are concatenated with the child node to obtain the second aggregation result. Based on the first aggregation result and the second aggregation result, the target object attribute information corresponding to each of the object attribute trees is obtained.

4. The method according to claim 1, characterized in that, The process of obtaining the object attribute tree that matches at least a portion of the attributes of the target object includes: Retrieve object dataset; The object attribute tree is constructed by traversing the object dataset based on a preset object attribute tree structure.

5. The method according to any one of claims 1 to 4, characterized in that, The step of fusing the target object attribute information with the original object representation based on the original object representation to obtain the object fused representation includes: Based on the original representation of the target object, the attribute weights corresponding to various target object attributes are obtained, and the attribute weights corresponding to various target object attributes are determined through an attention mechanism. Based on the attribute weights corresponding to various target object attributes, the target object attribute information is fused with the original representation of the object to obtain the object attribute fusion result; Based on the object attribute fusion result, the preset object target interaction graph is convolved by graph convolution to obtain the object representations corresponding to each order of neighbor nodes in the preset object target interaction graph. The object representations corresponding to the neighbor nodes of each order are subjected to sum pooling to obtain the object fusion representation.

6. The method according to any one of claims 1 to 4, characterized in that, The step of fusing the target object attribute information with the original target object representation based on the original object representation to obtain the target object fused representation includes: Based on the original representation of the object, the attribute weights corresponding to various target attributes are obtained, and the attribute weights corresponding to various target attributes are determined through an attention mechanism. Based on the attribute weights corresponding to various target attributes, the target attribute information is fused with the original representation of the target to obtain the target attribute fusion result; Based on the target attribute fusion result, the target interaction graph of the preset object is convolved by graph convolution to obtain the target representation of each order of neighbor nodes in the target interaction graph of the preset object. The target representations corresponding to the neighbor nodes of each order are subjected to sum pooling to obtain the target fusion representation.

7. The method according to claim 1, characterized in that, The step of performing data matching processing on the target object and the target object based on the object fusion representation, the target object fusion representation, the target object attribute information, and the target object attribute information to obtain the data processing result of the target object on the target object includes: The object fusion representation, the target object fusion representation, the target object attribute information, and the target object attribute information are input into a preset feature interaction model to perform data matching processing on the target object and the target object, thereby obtaining the data processing result of the target object on the target object.

8. A data processing apparatus, characterized in that, The device includes: The data acquisition module is used to acquire an object attribute tree that matches at least a portion of the attributes of the target object and the original object representation of the target object, and to acquire a target attribute tree that matches at least a portion of the attributes of the target object and the original object representation of the target object. The original object representation of the target object is a vector representation obtained by converting the target object through an embedding method, and the original object representation of the target object is a vector representation obtained by converting the target object through an embedding method. The information aggregation module is used to perform node information aggregation processing on each of the object attribute trees to obtain target object attribute information, and to perform node information aggregation processing on each of the target object attribute trees to obtain target object attribute information. The information fusion module is used to fuse the target object attribute information with the original object representation based on the original object representation to obtain a fused object representation, and to fuse the target object attribute information with the original object representation based on the original object representation to obtain a fused object representation. The data processing module is used to perform data matching processing on the target object and the target object based on the object fusion representation, the target object fusion representation, the target object attribute information, and the target object attribute information, to obtain the data processing result of the target object on the target object.

9. 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 7.

10. 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 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 steps of the method according to any one of claims 1 to 7.

Images