Data processing method, apparatus, device, medium, and product

Abstract

Embodiments of the present application disclose a data processing method, device, equipment, medium and product, the method comprising: obtaining object information of a first type of object and object information of a second type of object; calculating a feature vector of the first type of object according to the object information of the first type of object, the object information of the second type of object, and a coupling relationship between the first type of object and the second type of object; generating a target recommended object corresponding to each second type of object according to a selection condition of the second type of object by the first type of object; obtaining a feature vector of the target recommended object, and determining a second type of object recommended to the target recommended object based on the feature vector of the target recommended object and the feature vector of the first type of object. The technical solution of the embodiments of the present application can improve the accuracy of recommended information.

Description

Technical Field

[0001] This application relates to the fields of computer and communication technology, and more specifically, to a data processing method, a data processing apparatus, an electronic device, a computer-readable storage medium, and a computer program product. Background Technology

[0002] In the field of information recommendation, a common practice is to promote information indiscriminately across the entire domain or to select a specific domain for promotion in a coarse-grained manner. This approach results in low accuracy of information recommendation. Therefore, how to improve the accuracy of information recommendation has become an urgent technical problem to be solved. Summary of the Invention

[0003] To address the aforementioned technical problems, embodiments of this application provide a data processing method, a data processing apparatus, an electronic device, a computer-readable storage medium, and a computer program product, which can improve the accuracy of recommendation information.

[0004] Other features and advantages of this application will become apparent from the following detailed description, or may be learned in part from practice of this application.

[0005] According to one aspect of the embodiments of this application, a data processing method is provided, comprising: acquiring object information of a first type of object and object information of a second type of object; calculating a feature vector of the first type of object based on the object information of the first type of object, the object information of the second type of object, and the coupling relationship between the first type of object and the second type of object; generating target recommended objects corresponding to each second type of object based on the selection status of the second type of object by the first type of object; acquiring the feature vector of the target recommended object; and determining the second type of object recommended to the target recommended object based on the feature vector of the target recommended object and the feature vector of the first type of object.

[0006] According to one aspect of the embodiments of this application, a data processing apparatus is also provided, the apparatus comprising: an acquisition module, configured to acquire object information of a first type of object and object information of a second type of object; a calculation module, configured to calculate a feature vector of the first type of object based on the object information of the first type of object, the object information of the second type of object, and the coupling relationship between the first type of object and the second type of object; a generation module, configured to generate target recommended objects corresponding to each second type of object based on the selection status of the second type of object by the first type of object; and a determination module, configured to acquire the feature vector of the target recommended object, and determine the second type of object recommended to the target recommended object based on the feature vector of the target recommended object and the feature vector of the first type of object.

[0007] In one embodiment of this application, the calculation module is further configured to obtain a first feature vector generation model, the first feature vector generation model being used to learn the coupling relationship between the first type of object and the second type of object; and to input the object information of the first type of object and the object information of the second type of object into the first feature vector generation model to obtain the feature vector of the first type of object.

[0008] In one embodiment of this application, the first feature vector generation model includes a first type of object feature extraction network, a second type of object feature extraction network, and a feature synthesis module for integrating the outputs of the first type of object feature extraction network and the second type of object feature extraction network. The calculation module is further used to input the object information of the first type of object into the first type of object feature extraction network, input the object information of the second type of object into the second type of object feature extraction network, and use the feature vector output by the first type of object feature extraction network as the feature vector of the first type of object.

[0009] In one embodiment of this application, the apparatus includes a first training module. The first training module is configured to construct a first sample pair and a second sample pair based on the selection of a second type of sample object by a first type of sample object. The first sample pair includes sample object information of the first type of sample object and sample object information of the second type of sample object selected by the first type of sample object. The second sample pair includes sample object information of the first type of sample object and sample object information of the second type of sample object not selected by the first type of sample object. The first sample pair is input as a positive sample and the second sample pair as a negative sample into a first feature vector generation model. The loss value of the first feature vector generation model is calculated based on the sample object similarity output by the first feature vector generation model for the positive sample and the negative sample. The model parameters of the first feature vector generation model are adjusted based on the loss value of the first feature vector generation model to train the first feature vector generation model.

[0010] In one embodiment of this application, after inputting the first sample pair as a positive sample and the second sample pair as a negative sample into the first feature vector generation model, the first training module is further configured to obtain the first type of sample information features corresponding to the sample object information of the first type of sample object through the first feature vector generation model, generate the sample sparse vector of the first type of sample object based on the first type of sample information features; divide the set of values ​​corresponding to the first type of sample information features into multiple first intervals and multiple second intervals, wherein the values ​​in the second intervals are different from the values ​​in the first intervals; map the sample sparse vector of the first type of sample object to each first interval and second interval to obtain the sample feature vector of the first type of sample object; and calculate the sample object similarity for the positive sample and the negative sample based on the sample feature vector of the first type of sample object, the sample feature vector of the second type of sample object selected by the first type of sample object, and the sample feature vector of the second type of sample object not selected by the first type of sample object.

[0011] In one embodiment of this application, the first training module is further configured to perform a remainder operation on the set of values ​​corresponding to the first type of sample information features to divide them into multiple first intervals; and to perform a quotient operation on the set of values ​​corresponding to the first type of sample information features to divide them into multiple second intervals.

[0012] In one embodiment of this application, the training module is further configured to create feature vector matrices for the first interval and the second interval respectively, wherein the feature vector matrix of the first interval and the feature vector matrix of the second interval are different; map the sparse vectors of the first class of sample objects to the feature vector matrix of each first interval to obtain the feature vector of each first interval; map the sparse vectors of the first class of sample objects to the feature vector matrix of each second interval to obtain the feature vector of each second interval; and generate the sample feature vector of the first class of sample objects based on the feature vector of each first interval and the feature vector of each second interval.

[0013] In one embodiment of this application, after the first training module inputs the first sample pair as a positive sample and the second sample pair as a negative sample into the first feature vector generation model, it is further configured to obtain the second type of sample information features corresponding to the sample object information of the second type of sample object selected by the first type of sample object through the first feature vector generation model; generate the sample sparse vector of the second type of sample object selected by the first type of sample object according to the second type of sample information features; divide the set of values ​​corresponding to the second type of sample information features into multiple third intervals and multiple fourth intervals; wherein, the values ​​in the third intervals are different from the values ​​in the fourth intervals, and the sample sparse vector of the second type of sample object selected by the first type of sample object is mapped to each third interval and fourth interval to obtain the sample feature vector of the second type of sample object selected by the first type of sample object.

[0014] In one embodiment of this application, the generation module is further configured to: select the first type of object of the second type of object as the target recommended object corresponding to the second type of object; or select the first type of object of the second type of object as the seed object, obtain other first type of objects similar to the seed object, and use the obtained other first type of objects as the target recommended object corresponding to the second type of object; or select the first type of object of the second type of object as the seed object, obtain other first type of objects similar to the seed object, and use the obtained other first type of objects and the seed object as the target recommended object corresponding to the second type of object.

[0015] In one embodiment of this application, the generation module is further configured to calculate the attribute correlation between the seed object and the other first-type objects; and to obtain other first-type objects similar to the seed object based on the attribute correlation between the seed object and the other first-type objects.

[0016] In one embodiment of this application, the determining module is further configured to generate an initial feature vector of the target recommendation object based on the feature vector of the first type of object; obtain a second feature vector generation model, which is obtained by training the sample feature vector of the first type of sample object and the sample feature vector of the sample recommendation object; and input the initial feature vector of the target recommendation object and the feature vector of the first type of object into the second feature vector generation model to obtain the feature vector of the target recommendation object.

[0017] In one embodiment of this application, the apparatus further includes a second training module, which is used to obtain the sample feature vector of the first type of sample object and the sample recommendation object; randomly initialize the sample feature vector of the sample recommendation object; and use the cross-entropy function as a loss function to train the second feature vector generation model based on the sample feature vector of the first type of sample object and the sample feature vector of the sample recommendation object.

[0018] In one embodiment of this application, the determining module is further configured to determine the similarity between the target recommended object and the first type of object based on the feature vector of the target recommended object and the feature vector of the first type of object; if the similarity between the target recommended object and the first type of object is higher than a preset threshold, then a second type of object recommended to the target recommended object is determined.

[0019] In one embodiment of this application, the determining module is further configured to obtain a global feature vector of the first type of object, wherein the global feature vector of the first type of object is used to describe the features of the global information of the first type of object; determine a local feature vector of the first type of object based on the feature vector of the target recommendation object and the feature vector of the first type of object; the local feature vector of the first type of object is used to describe the features of the local information of the first type of object related to the target recommendation object. The similarity between the target recommendation object and the first type of object is determined based on the similarity between the feature vector of the target recommendation object and the global feature vector of the first type of object, and the similarity between the feature vector of the target recommendation object and the local feature vector of the first type of object.

[0020] According to one aspect of the embodiments of this application, an electronic device is provided, including one or more processors; and a storage device for storing one or more computer programs, which, when executed by the one or more processors, cause the electronic device to implement the data processing method as described above.

[0021] According to one aspect of the present application, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor of an electronic device, causes the electronic device to perform the data processing method as described above.

[0022] According to one aspect of the present application, a computer program product is provided, including a computer program stored in a computer-readable storage medium, wherein a processor of an electronic device reads from the computer-readable storage medium and executes the computer program, causing the electronic device to perform the data processing method as described above.

[0023] In the technical solution provided by the embodiments of this application, the feature vector of the first type of object is calculated based on the object information of the first type of object, the object information of the second type of object, and the coupling relationship between the first type of object and the second type of object. That is, it not only relies on the object information of the first type of object, but also combines the object information of the second type of object and the coupling relationship between the first type of object and the second type of object to calculate the feature vector of the first type of object from multiple aspects, so that the feature vector of the first type of object can more accurately represent the first type of object. Then, by the selection of the second type of object by the first type of object, the target recommended object corresponding to each second type of object is accurately obtained. Then, based on the feature vector of the target recommended object and the feature vector of the first type of object, the second type of object recommended to the target recommended object is determined, ensuring that the second type of object recommended to the target recommended object is consistent with the target recommended object, and realizing accurate recommendation for the target recommended object.

[0024] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0025] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. It is obvious that the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort. In the drawings:

[0026] Figure 1 This is a schematic diagram of one implementation environment involved in this application;

[0027] Figure 2 This is a schematic diagram of one implementation environment involved in this application;

[0028] Figure 3 This is a flowchart illustrating a data processing method in an exemplary embodiment of this application;

[0029] Figure 4 This is a flowchart illustrating another data processing method as shown in an exemplary embodiment of this application;

[0030] Figure 5 This is a flowchart illustrating another data processing method as shown in an exemplary embodiment of this application;

[0031] Figure 6 This is a flowchart illustrating another data processing method as shown in an exemplary embodiment of this application;

[0032] Figure 7 This is a flowchart illustrating another data processing method as shown in an exemplary embodiment of this application;

[0033] Figure 8 This is a flowchart illustrating another data processing method as shown in an exemplary embodiment of this application;

[0034] Figure 9 This is a flowchart illustrating another data processing method as shown in an exemplary embodiment of this application;

[0035] Figure 10 This is a flowchart illustrating another data processing method as shown in an exemplary embodiment of this application;

[0036] Figure 11 This is a flowchart illustrating another data processing method as shown in an exemplary embodiment of this application;

[0037] Figure 12 This is a flowchart illustrating another data processing method as shown in an exemplary embodiment of this application;

[0038] Figure 13 This is a flowchart illustrating a data processing method in another exemplary embodiment of this application;

[0039] Figure 14 This is a flowchart illustrating another data processing method as shown in another exemplary embodiment of this application;

[0040] Figure 15 This is a flowchart illustrating another data processing method as shown in another exemplary embodiment of this application;

[0041] Figure 16 This is a flowchart illustrating another data processing method as shown in another exemplary embodiment of this application;

[0042] Figure 17 This is a flowchart illustrating another data processing method as shown in another exemplary embodiment of this application;

[0043] Figure 18 This is a flowchart illustrating another data processing method as shown in another exemplary embodiment of this application;

[0044] Figure 19 This is a schematic diagram illustrating the structure of the first feature vector generation model, as shown in yet another exemplary embodiment of this application.

[0045] Figure 20 This is a flowchart illustrating another data processing method as shown in another exemplary embodiment of this application;

[0046] Figure 21 This is a flowchart illustrating another data processing method as shown in another exemplary embodiment of this application;

[0047] Figure 22This is a structural block diagram of a data processing apparatus illustrated in an exemplary embodiment of this application;

[0048] Figure 23 A schematic diagram of the structure of a computer system suitable for implementing the electronic device of the present application is shown. Detailed Implementation

[0049] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0050] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0051] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0052] It should also be noted that "multiple" as mentioned in this application refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0053] The technical solutions of this application embodiment can be applied to the field of cloud technology. Before introducing the technical solutions of this application embodiment, cloud technology will be briefly introduced. Cloud technology refers to a hosting technology that unifies a series of resources such as hardware, software, and networks within a wide area network or local area network to realize the computation, storage, processing, and sharing of data.

[0054] Cloud computing refers to the delivery and usage model of IT infrastructure, meaning obtaining necessary resources in an on-demand and easily scalable manner through a network. In a broader sense, cloud computing also refers to the delivery and usage model of services, meaning obtaining necessary services in an on-demand and easily scalable manner through a network. These services can be IT and software related, internet-related, or other services. Cloud computing is a product of the development and integration of traditional computer and network technologies such as grid computing, distributed computing, parallel computing, utility computing, network storage technologies, virtualization, and load balancing.

[0055] Among these applications, cloud technology includes big data. Big data refers to data sets that cannot be captured, managed, and processed within a certain timeframe using conventional software tools. It represents massive, rapidly growing, and diverse information assets that require new processing models to achieve stronger decision-making, insightful discovery, and process optimization capabilities. With the advent of the cloud era, big data has attracted increasing attention. Big data requires specialized technologies to effectively process large amounts of data within a tolerable timeframe. Technologies suitable for big data include massively parallel processing databases, data mining, distributed file systems, distributed databases, cloud computing platforms, the internet, and scalable storage systems.

[0056] Based on cloud technology, the object information of the first type of object and the object information of the second type of object in this application embodiment can be obtained from massively parallel processing databases, distributed file systems, distributed databases, cloud computing platforms, the Internet and scalable storage systems.

[0057] It should be noted that the technical solutions of the embodiments of this application can also be applied to the field of blockchain technology. Blockchain technology is a novel distributed infrastructure and computing method that uses a block-chain data structure to verify and store data, uses distributed node consensus algorithms to generate and update data, uses cryptography to ensure the security of data transmission and access, and uses smart contracts composed of automated script code to program and manipulate data. Blockchain refers to a decentralized infrastructure with distributed storage characteristics. Specifically, it is a data structure that organizes data blocks in chronological order using a linked list-like manner. It can securely store data with sequential relationships that can be verified within the system, and uses cryptography to ensure that the data is immutable and unforgeable. Simply put, blockchain is a decentralized distributed ledger, where each chain is equivalent to an independent ledger.

[0058] Specifically, the implementation environment involved in this application embodiment can be a distributed system formed by multiple nodes (any form of computing device in the network, such as servers or terminals) connected through network communication. Taking a blockchain system as an example, the distributed system is formed by multiple nodes (any form of computing device in the network, such as servers or terminals), and the nodes form a peer-to-peer (P2P) network. The P2P protocol is an application layer protocol running on top of the Transmission Control Protocol (TCP). In the distributed system, any machine, such as a server or terminal, can join and become a node. A node includes a hardware layer, a middleware layer, an operating system layer, and an application layer.

[0059] The functions of each node in a blockchain system include at least the following: 1) Routing: A basic function of nodes used to support communication between nodes. In addition to routing, nodes may also have the following functions: 2) Application: Deployed in the blockchain to implement specific business needs, recording data related to the implementation of these functions to form record data. This record data carries a digital signature to indicate the source of the task data. The record data is then sent to other nodes in the blockchain system, whereby other nodes add the record data to a temporary block upon successful verification of its source and integrity.

[0060] Based on blockchain technology, in the embodiments of this application, a server node can generate recorded data from a trained first feature vector generation model and / or a second feature vector generation model. The recorded data carries a digital signature to indicate the source of the data, and the recorded data is sent to other nodes in the blockchain system. When other nodes, such as other servers or mobile terminals, successfully verify the source and integrity of the recorded data, they can obtain the first feature vector generation model and / or the second feature vector generation model. Then, other servers can complete the data processing in the embodiments of this application. Multiple servers and terminals can form a blockchain system, and the servers and terminals are nodes on the blockchain system.

[0061] The technical solutions of this application embodiment can also be applied to cloud technology and blockchain technology. The technical solutions of this application embodiment are described in detail below:

[0062] Please see Figure 1 , Figure 1 This is a schematic diagram of an implementation environment related to this application. The implementation environment includes a terminal 10 and a server 20, and the terminal 10 and the server 20 communicate with each other via a wired or wireless network.

[0063] In this context, terminal 10 can be the initiator of the data processing request, that is, to send a data processing request to server 20. Terminal 10 can obtain object information of the first type of object locally on the terminal and send it to server 20.

[0064] After receiving the data processing request, server 20 obtains object information of the second type of object from large-scale parallel processing databases with big data applications, distributed file systems, distributed databases, cloud computing platforms, the Internet, and scalable storage systems. Based on the object information of the first and second types of objects, and the coupling relationship between them, server 20 calculates the feature vector of the first type of object. Server 20 can also determine the selection status of the second type of object by the first type of object based on the coupling relationship, and then generate target recommended objects corresponding to each second type of object. Afterwards, it obtains the feature vector of the target recommended object, and based on the feature vectors of the target recommended object and the first type of object, determines the second type of object to be recommended to the target recommended object. Finally, it can send the second type of object to terminal 10 so that terminal 10 can display the second type of object to the first type of object.

[0065] Please see Figure 2 , Figure 2This is a schematic diagram of another implementation environment involved in this application. This embodiment environment includes a terminal 210, a server 220, and a blockchain network 230. The terminal 210 and the server 220 communicate with each other through a wired or wireless network, and the server 220 and the blockchain network 230 communicate with each other through a wired or wireless network.

[0066] In the blockchain network 230, node 101 is used to store the first feature vector generation model and the second feature vector generation model.

[0067] Terminal 210 can be the initiator of a data processing request, that is, to send a data processing request to server 220. Terminal 210 can obtain object information of the first type of object locally on the terminal and send it to server 220.

[0068] After receiving the data processing request, server 220 can obtain object information of the second type of object in the Internet network, and obtain the first feature vector generation model and the second feature vector generation model from the blockchain network. The object information of the first type of object and the object information of the second type of object are input into the first feature vector generation model to obtain the feature vector of the first object output by the first feature vector generation model based on the learned coupling relationship between the first type of object and the second type of object. Then, according to the selection of the second type of object by the first type of object, the target recommendation object corresponding to each second type of object is generated. Then, the feature vector of the target recommendation object is obtained through the second feature vector generation model. Finally, based on the feature vector of the target recommendation object and the feature vector of the first type of object, the second type of object recommended to the target recommendation object is determined.

[0069] Optionally, when the server inputs the object information of the first type of object and the object information of the second type of object into the first feature vector generation model, it can also obtain the coupling relationship between the first type of object and the second type of object through the feature vector generation model, and send the coupling relationship between the first type of object and the second type of object to the blockchain network for on-chain processing.

[0070] The aforementioned terminals can be electronic devices such as smartphones, tablets, laptops, computers, smart voice interaction devices, smart home appliances, vehicle terminals, and aircraft; the server can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and intelligent platforms. This document does not impose any restrictions on these.

[0071] It should be noted that in the specific implementation of this application, the obtained object information involves object-related matters. When the embodiments of this application are applied to specific products or technologies, it is necessary to obtain the object's permission or consent, and the collection, use and processing of relevant information must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0072] The following details the various implementation details of the technical solutions in the embodiments of this application:

[0073] like Figure 3 As shown, Figure 3 This is a flowchart illustrating a data processing method according to an embodiment of this application, which can be applied to... Figure 1 The implementation environment shown or Figure 2 In the implementation environment shown, this method can be executed by a terminal or a server, or by both a terminal and a server. In this embodiment, the method is described using the server as an example. The data processing method may include steps S310 to S340, which are detailed below:

[0074] S310. Obtain object information of the first type of object and object information of the second type of object.

[0075] In this application embodiment, the first type of object can be the user of the terminal. The object information of the first type of object is the usage information left by the user on the terminal through the use of the terminal, such as the user's basic attribute information, operation information, browsing information, etc. when using the terminal. The first type of object can also be an account on the corresponding application, such as a shopping website application or a content platform application. In this case, the object information of the first type of object is the information related to the account, such as the attribute information of the account when registering on the application.

[0076] In the embodiments of this application, the second type of object can be something that the first type of object can know, such as an advertisement, a product, or an article. The object information of the second type of object is information related to the product, such as the notification information corresponding to the advertisement, the product information of the product, or the content information described in the article.

[0077] In one example of this application embodiment, when the first type of object is the user of the terminal, the object information of the first type of object can be obtained through the log of the terminal corresponding to the first type of object. For example, if the user performs operation A and the terminal executes the instruction corresponding to operation A, the terminal log records the detection of the instruction corresponding to operation A and the execution result. Then, the user's operation A information can be obtained through the terminal log. When the first type of object is an account on the corresponding application, the attribute information required for registration when registering the account can be obtained through the application server corresponding to the application.

[0078] In one example of an embodiment of this application, the object information of the second type of object can be obtained by the publisher of the second type of object. For example, if a news article is published on a news website, the content information of the news article can be obtained from the news website.

[0079] It is worth noting that the first type of object in the embodiments of this application may include one first type of object or multiple first type of objects; similarly, the second type of object may include one second type of object or multiple second type of objects.

[0080] It should be noted that, in the specific embodiments of this application, the object information obtained involves information related to the object. When the embodiments of this application are applied to specific products or technologies, permission or consent from the object is required, and the collection, use and processing of relevant object information must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0081] S320. Calculate the feature vector of the first type of object based on the object information of the first type of object, the object information of the second type of object, and the coupling relationship between the first type of object and the second type of object.

[0082] The coupling relationship between the first type of object and the second type of object can be an interaction or mutual influence between them; alternatively, the coupling relationship can be one where the first type of object and the second type of object are unrelated. Optionally, this coupling relationship can be defined as the first type of object's choice of the second type of object.

[0083] The coupling relationship between the first type of object and the second type of object can be obtained in advance, such as by obtaining the object information of the first type of object and the object information of the second type of object.

[0084] After obtaining the object information of the first type of object, the object information of the second type of object, and the coupling relationship between the first and second type of objects, the feature vector of the first type of object can be calculated by influencing the object information of the first type of object through the coupling relationship between the first type of object and the object information of the second type of object. This feature vector is used to represent the feature information of the first type of object using a low-dimensional vector. Optionally, this feature vector is an embedding.

[0085] S330. Based on the selection of the second type of object by the first type of object, generate the target recommendation object corresponding to each second type of object.

[0086] The selection status of the second type of object by the first type of object refers to whether the second type of object is selected by the first type of object. For example, if the first type of object clicks to view the second type of object on the terminal, then the second type of object is selected by the first type of object; or if the first type of object purchases the second type of object on the terminal, then the second type of object is selected by the first type of object.

[0087] The target recommendation object corresponding to the second type of object refers to the object that may be interested in the second type of object. That is, the second type of object can be recommended to the target recommendation object. The target recommendation object can be the user of the terminal or the account of the application on the corresponding application.

[0088] In this embodiment, if a second type of object is selected by a first type of object, it means that the second type of object is being followed by the first type of object, indicating that the second type of object is popular. Based on whether the second type of object is selected by the first type of object, a target recommended object can be generated for each popular second type of object. Such target recommended objects include, but are not limited to, the first type of object, objects similar to the first type of object, and objects associated with the first type of object.

[0089] Optionally, the selection status of the second type of object by the first type of object can be obtained based on the coupling relationship between the first type of object and the second type of object; for example, if the coupling relationship between the first type of object and the second type of object is that the first type of object selects the second type of object, then the corresponding selection status is that the second type of object is selected by the first type of object.

[0090] S340. Obtain the feature vector of the target recommendation object, and based on the feature vector of the target recommendation object and the feature vector of the first type of object, determine the second type of object to be recommended to the target recommendation object.

[0091] In this embodiment, the feature vector of the target recommendation object is used to represent the feature information of the target recommendation object with a low-dimensional vector, and the feature vector of the target recommendation object can be obtained through the model.

[0092] As mentioned earlier, if the target recommended object is a first-class object, then the feature vector of the target recommended object is the feature vector of the first-class object. Since the first-class object has selected the second-class object, the second-class object can be directly recommended to the first-class object.

[0093] If the target recommendation object is an object similar to or related to the first type of object, then the second type of object to be recommended to the target recommendation object is determined based on the feature vector of the target recommendation object and the feature vector of the first type of object.

[0094] For example, the first type of object is object A, and the object information of the first type of object is A'. The second type of objects are objects 1 and 2, and the object information of the second type of objects is 1' and 2'. Assuming that there is a coupling relationship between object A and object 2, but no coupling relationship between object A and object 2, that is, object A selects object 1 and does not select object 2, then the feature vector of object A is calculated based on A', 1', 2', object A selecting object 1, and object A not selecting object 2. Since object A selected object 1, the target recommended object X corresponding to object 1 is generated. Then, based on the feature vector of the target recommended object X and A', it is determined whether to recommend object 1 to the target recommended object X.

[0095] In this embodiment, not only the object information of the first type of object is relied upon, but also the object information of the second type of object and the coupling relationship between the first and second type of objects are combined to calculate the feature vector of the first type of object from multiple aspects, so that the feature vector of the first type of object can more accurately represent the first type of object. Then, by observing the selection of the second type of object by the first type of object, the target recommended object corresponding to each second type of object is accurately obtained. Furthermore, based on the feature vector of the target recommended object and the feature vector of the first type of object, the second type of object recommended to the target recommended object is determined, ensuring that the second type of object recommended to the target recommended object is consistent with the target recommended object, and realizing accurate recommendation for the target recommended object.

[0096] In one embodiment of this application, another data processing method is provided, which can be applied to... Figure 1 The implementation environment shown or Figure 2 The implementation environment shown can be executed by a terminal or a server, or by both. In this embodiment, the method is described using the server as an example. Figure 4 As shown, the data processing method may include S410 to S430 and S330 to S340.

[0097] Steps S410 to S430 are described in detail below:

[0098] S410. Obtain object information of the first type of object and object information of the second type of object.

[0099] For information on the first type of object and the second type of object, please refer to [link to relevant documentation]. Figure 3 Step S310 shown in the diagram will not be described again here.

[0100] S420. Obtain the first feature vector generation model. The first feature vector generation model is used to learn the coupling relationship between the first type of object and the second type of object.

[0101] In this embodiment, the first feature vector generation model is pre-trained. It can be obtained from a blockchain network, a locally pre-trained first feature vector generation model, or a model server used to store various models. No further limitations are imposed here.

[0102] It should be noted that the first feature vector generation model is used to learn the coupling relationship between the first type of objects and the second type of objects. That is, the object information of the first type of objects and the object information of the second type of objects are input into the first feature vector generation model, and the first feature vector generation model can learn the coupling relationship between the first type of objects and the second type of objects based on the object information of the first type of objects and the object information of the second type of objects.

[0103] S430. Input the object information of the first type of object and the object information of the second type of object into the first feature vector generation model to obtain the feature vector of the first type of object.

[0104] The first feature vector generation model is obtained by inputting the object information of the first type of object and the object information of the second type of object into the first feature vector generation model. Since the first feature vector model can also learn the coupling relationship between the first type of object and the second type of object, that is, determine whether there is a coupling relationship between the first type of object and the second type of object based on the object information of the first type of object and the object information of the second type of object. If there is a coupling relationship, then further influence the object information of the first type of object based on the coupling relationship and the object information of the second type of object, that is, generate the feature vector of the first object by referring to the coupling relationship and the object information of the second type of object.

[0105] It should be noted that, Figure 4 For further details on steps S330 to S340 shown, please refer to [link to relevant documentation]. Figure 3 Steps S330 to S340 shown will not be repeated here.

[0106] In this embodiment of the application, the object information of the first type of object and the object information of the second type of object are input into the first feature vector generation model. The model learns the coupling relationship between the first type of object and the second type of object, thereby accurately and reliably obtaining the feature vector of the first type of object.

[0107] This application provides another data processing method that can be applied to... Figure 1 The implementation environment shown or Figure 2 The implementation environment shown can be executed by a terminal or a server, or by both. In this embodiment, the method is described using the server as an example. Figure 5 As shown, this data processing method is in Figure 4Based on S410~S430 and S330~S340 shown, Figure 4 The step S430 shown is extended to step S510.

[0108] in Figure 5 The first feature vector generation model in step S420 shown includes a first type of object feature extraction network, a second type of object feature extraction network, and a feature synthesis module for integrating the outputs of the first type of object feature extraction network and the second type of object feature extraction network; wherein the first type of object feature extraction network is used to output the feature vector of the first type of object, the second type of object feature extraction network is used to output the feature vector of the second type of object, and the feature synthesis module is used to integrate the feature vectors of the first type of object and the feature vectors of the second type of object, thereby learning the coupling relationship between the first type of object and the second type of object.

[0109] Step S510 is described in detail below:

[0110] S510. Input the object information of the first type of object into the first type of object feature extraction network, input the object information of the second type of object into the second type of object feature extraction network, and use the feature vector output by the first type of object feature extraction network as the feature vector of the first type of object. First feature vector generation model.

[0111] In this embodiment, the object information of the first type of object is input into the first type of object feature extraction network to obtain the initial feature vector of the first type of object, and the object information of the second type of object is input into the second type of object feature extraction network to obtain the initial feature vector of the second type of object. The feature synthesis module learns the coupling relationship between the first type of object and the second type of object based on the initial feature vector of the first type of object and the initial feature vector of the second type of object, and feeds back the coupling relationship to the first type of object feature extraction network. The first type of object feature extraction network adjusts the initial feature vector of the first type of object with reference to the coupling relationship between the first type of object and the second type of object output by the feature synthesis module and the initial feature vector of the second type of object, and then outputs the feature vector of the first type of object.

[0112] It should be noted that, Figure 5 For detailed information on steps S410-S420 and S330-S340 shown, please refer to [link to relevant documentation]. Figure 4 Steps S410 to S420 and S330 to S340 shown will not be repeated here.

[0113] The first feature vector generation model in this embodiment processes the object information of the first type of object and the object information of the second type of object through two network structures respectively, and by combining the outputs of the two networks, the feature vector of the first type of object output by the feature extraction network is accurate and reliable.

[0114] This application also provides another data processing method, which can be applied to... Figure 1 The implementation environment shown or Figure 2 The implementation environment shown can be executed by a terminal or a server, or by both. In this embodiment, the method is described using the server as an example. Figure 6 As shown, the data processing method also includes the training process of the first feature vector generation model. Figure 4 Steps S610 to S630 are added before S410 shown.

[0115] Steps S610 to S630 are described in detail below:

[0116] S610. Based on the selection of the second type of sample objects by the first type of sample objects, construct a first sample pair and a second sample pair. The first sample pair includes the sample object information of the first type of sample objects and the sample object information of the second type of sample objects selected by the first type of sample objects. The second sample pair includes the sample object information of the first type of sample objects and the sample object information of the second type of sample objects not selected by the first type of sample objects.

[0117] In this embodiment, the first feature vector generation model is trained using a first type of sample object and a second type of sample object. The selection of the first type of sample object relative to the second type of sample object can be either that the first type of sample object selects the second type of sample object or that the first type of sample object does not select the second type of sample object. Since the selection of the first type of sample object relative to the second type of sample object is known, this selection can be used as the label for the second type of sample object. For example, if the first type of sample object a selects the second type of sample object b, the labeled second type of sample object is object b(a-1); if the first type of sample object a does not select the second type of sample object b, the labeled second type of sample object is object b(a-0); if the first type of sample object a selects the second type of sample object c, the labeled second type of sample object is c(a-1); if the first type of sample object a does not select the second type of sample object d, the labeled second type of sample object is d(a-0).

[0118] Based on the selection of the second type of sample objects by the first type of sample objects, construct the first sample pair and the second sample pair. The first sample pair can be one pair or multiple pairs; the second sample pair can be one pair or multiple pairs. Continuing from the previous example, the first sample pair includes the object information of a and the sample information of b (a-1), and the object information of a and the sample information of c (a-1); the second sample pair includes the object information of a and the object information of d (a-0).

[0119] S620. Input the first sample pair as a positive sample and the second sample pair as a negative sample into the first feature vector generation model. Calculate the loss value of the first feature vector generation model based on the sample object similarity output by the first feature vector generation model for the positive and negative samples.

[0120] In this embodiment of the application, the first sample pair is used as a positive sample and the second sample pair is used as a negative sample and input into the first feature vector generation model to obtain the sample object similarity output by the first feature vector generation model for the positive sample and the negative sample, and then the loss value of the first feature vector generation model is calculated based on the sample object similarity.

[0121] The sample object similarity output for the positive and negative samples refers to the similarity between the sample feature vectors of the positive and negative samples. The lower the similarity, the less related the positive and negative samples are. Since the difference between the positive and negative samples lies only in whether the second type of sample object is selected, the less related the positive and negative samples are, the less different the second type of sample object selected by the first type of sample object is from the second type of sample object not selected by the first type of sample object.

[0122] S630. Adjust the model parameters of the first feature vector generation model according to the loss value of the first feature vector generation model, so as to train the first feature vector generation model.

[0123] The model parameters of the first feature vector generation model are adjusted based on the loss value of the first feature vector generation model until the first feature vector generation model converges. At this point, the model parameters of the first feature vector generation model are optimal, and the training of the first feature vector generation model is completed.

[0124] It should be noted that, Figure 6 For further details on steps S410-S430 and S330-S340 shown, please refer to [link to detailed information]. Figure 4 Steps S410 to S430 and S330 to S340 shown in the diagram will not be repeated here.

[0125] In this embodiment, the loss value of the first feature vector generation model is calculated by the sample object similarity between positive and negative samples, so that the model can fully learn the relationship between positive and negative samples during training, thus ensuring the reliability and accuracy of the training of the first feature vector generation model.

[0126] In one embodiment of this application, another data processing method is also provided, which can be applied to... Figure 1 The implementation environment shown or Figure 2 The implementation environment shown can be executed by a terminal or a server, or by both. In this embodiment, the method is described using the server as an example. Figure 7 As shown, this data processing method is in Figure 6 Steps S710 to S740 have been added between the steps within step S620 shown, as follows: Figure 7 As shown, the data processing method includes steps S6201, S710-S740, and S6202, wherein steps S6201 and S6202 are described in detail below. Figure 6 The step S620 shown.

[0127] Steps S710 to S740 are described in detail below:

[0128] S710. Obtain the first type of sample information features corresponding to the sample object information of the first type of sample object through the first feature vector generation model.

[0129] In this embodiment of the application, the sample object information of the first type of sample object is converted into the first type of sample information feature, that is, the sample object information is characterized by ID. For example, the types of sample object information of the first type of sample object include 1, 2, 3, and 4, and each type corresponds to ID1, ID2, ID3, and ID4. If the type of sample object information of the first type of sample object is type 3, then the corresponding first type of sample information feature is ID3.

[0130] Optionally, the sample object information of the first type of sample object is input into the first feature vector generation model, and the first type of sample information features generated by the first type of object feature extraction network of the first feature vector generation model are obtained.

[0131] S720. Generate a sparse vector of the first type of sample object based on the information features of the first type of sample, and divide the set of values ​​corresponding to the information features of the first type of sample into multiple first intervals and multiple second intervals, wherein the values ​​in the second intervals are different from the values ​​in the first intervals.

[0132] The sample object information is represented by an ID. Then, a sparse vector of the first type of sample object is generated based on the features of the first type of sample information. This sparse vector is the unique identifier of the sample object information. For example, if the sparse vector is a one-hot vector, then the one-hot feature vector (0,0,1,0) is obtained by one-hot encoding the first type of sample information feature ID3.

[0133] It should be understood that after converting the sample object information of the first type of sample objects into first type of sample information features, the sample information features have as many possible values ​​as there are types of sample object information. After obtaining the set of values ​​corresponding to the first type of sample information features, the set of values ​​corresponding to the first type of sample information features is divided into multiple first intervals and multiple second intervals. The values ​​in the second intervals are different from the values ​​in the first intervals. That is, the integration of the values ​​in the first intervals and the second intervals constitutes the set of values ​​corresponding to the first type of sample information features.

[0134] For example, if the set of values ​​corresponding to the information features of the first type of sample is {0,1,2,3,4}, and the first interval is {1,2}, then the second interval is {0,3,4}. This is a pair of first intervals and second intervals, and the first interval and the second interval are complementary intervals. If the first interval is {0,1,2}, then the second interval is {3,4}. This is another pair of first intervals and second intervals.

[0135] S730. Map the sparse vectors of the first type of sample objects to each first interval and second interval to obtain the sample feature vectors of the first type of sample objects.

[0136] The sparse vector of the first type of sample object is mapped to the interval sample feature vector of each pair of the first interval and the second interval. The sample feature vector of the first type of sample object is obtained through the interval sample feature vector of multiple intervals. For example, the one-hot vector is mapped to the embedding vector of the first interval and the embedding vector of the second interval. Then the sample feature vector of the first type of sample object is obtained through the embedding vector of multiple intervals.

[0137] S740. Calculate the sample object similarity for positive and negative samples based on the sample feature vector of the first type of sample object, the sample feature vector of the second type of sample object selected by the first type of sample object, and the sample feature vector of the second type of sample object not selected by the first type of sample object.

[0138] In this embodiment, the sample feature vector of a positive sample can be obtained based on the sample feature vector of the first type of sample object and the sample feature vector of the second type of sample object selected by the first type of sample object; the sample feature vector of a negative sample can be obtained based on the sample feature vector of the first type of sample object and the sample feature vector of the second type of sample object not selected by the second type of sample object; and then the sample object similarity for the positive sample and the negative sample is calculated based on the sample feature vector of the positive sample and the sample feature vector of the negative sample.

[0139] Optionally, the feature vector of the positive sample can be obtained by taking the inner product of the feature vector of the first type of sample object and the feature vector of the second type of sample object selected by the first type of sample object; the feature vector of the negative sample can be obtained by taking the inner product of the feature vector of the first type of sample object and the feature vector of the second type of sample object not selected by the second type of sample object; and then the similarity between the positive sample and the negative sample can be obtained by taking the dot product of the feature vector of the positive sample and the feature vector of the negative sample.

[0140] It should be noted that, Figure 7 For further details on steps S610, S630, S410-S430 and S330-S340 shown, please refer to [link to detailed description]. Figure 6 Steps S610, S630, S410-S430 and S330-S340 shown will not be repeated here.

[0141] This application embodiment accurately represents the first type of sample object by supplementing the interval learning of the sample feature vector of the first type of sample object, and ensures the accuracy of the sample object similarity when calculating the sample object similarity for positive and negative samples using the sample feature vector of the first type of sample object, the sample feature vector of the second type of sample object selected by the first type of sample object, and the sample feature vector of the second type of sample object not selected by the first type of sample object.

[0142] This application provides another data processing method, which can be applied to... Figure 1 The implementation environment shown or Figure 2 The implementation environment shown can be executed by a terminal or a server, or by both. In this embodiment, the method is described using the server as an example. The data processing method is as follows: Figure 8 As shown, this data processing method will Figure 7 Based on step S720 shown, step S720 is extended to steps S7201 and S810-S820. For step S7201, please refer to... Figure 7 The process of generating sample sparse vectors in step S720.

[0143] Steps S810 to S820 are described in detail below:

[0144] S810. Perform a remainder operation on the set of values ​​corresponding to the information features of the first type of sample to divide it into multiple first intervals.

[0145] S820. Perform quotient processing on the set of values ​​corresponding to the information features of the first type of sample to divide it into multiple second intervals.

[0146] It is understandable that the quotient and remainder are complementary. By performing remainder and quotient processing on the set of values ​​corresponding to the information features of the first type of samples, the set of values ​​corresponding to the information features of the first type of samples can be divided into multiple complementary intervals, that is, the first interval and the second interval are complementary.

[0147] For example, randomly initializing to obtain N embedding tables, where the first interval is P1 = {{h(x)mod m1}}x∈S; where S represents the set of all feature values; h(x): x∈S represents the hash value of the feature value, m i This represents the length of the i-th embedding table.

[0148] The second interval is Where M j The denominator of the modulo function for the j-th interval is... Where j = 2, ..., k.

[0149] It should be noted that, Figure 8 For further details regarding steps S610, S6201, S710, S730-S740, S6202, S630, S410-S430, and S330-S340 shown, please refer to [link to detailed information]. Figure 7 Steps S610, S6201, S710, S730~S740, S6202, S630, S410~S430, and S330~S340 shown in the diagram will not be repeated here.

[0150] This application provides another data processing method, which can be applied to... Figure 1 The implementation environment shown or Figure 2 The implementation environment shown can be executed by a terminal or a server, or by both. In this embodiment, the method is described using the server as an example. Figure 9 As shown, this data processing method is in Figure 8 Based on what is shown, Figure 8 The S730 shown is extended to steps S910 to S940. Steps S910 to S940 are described in detail below:

[0151] S910. Create eigenvector matrices for the first interval and the second interval respectively. The eigenvector matrices for the first interval and the second interval are different.

[0152] Create an eigenvector matrix for each first interval Where D j Let be the vector dimension of the j-th first interval.

[0153] Create an eigenvector matrix for each second interval Where D j Let be the vector dimension of the j-th second interval.

[0154] S920. Map the sparse vectors of the first type of sample objects to the feature vector matrix of each first interval to obtain the feature vector of each first interval.

[0155] For the onehot eigenvector e ε The j-th value is mapped to the eigenvector matrix W of the j-th first interval. j Generate the embedding vector x of the j-th first interval. j :

[0156] x j =W j *e j

[0157] S930. Map the sparse vectors of the first type of sample objects to the feature vector matrix of each second interval to obtain the feature vector of each second interval.

[0158] For the onehot eigenvector e ε The j-th value is mapped to the eigenvector matrix of the j-th second interval. Generate the embedding vector x of the j-th second interval. j :

[0159]

[0160] S940. Generate the sample feature vector of the first type of sample object based on the feature vector of each first interval and the feature vector of each second interval.

[0161] The embedding of feature ε can be represented by embedding x for multiple intervals. ε The calculation is as follows:

[0162]

[0163] Optionally, the feature vectors of each interval can be concatenated to obtain the sample feature vector of the first type of sample object.

[0164] Optionally, the feature vectors of each interval can be summed to obtain the sample feature vector of the first type of sample object.

[0165] Optionally, the feature vectors of each interval can be multiplied by the Hadamard product to obtain the sample feature vectors of the first type of sample objects.

[0166] It should be noted that, Figure 9 For detailed descriptions of steps S610, S6201, S710, S7201, S810~S830, S740, S6202, S630, S410~S430, and S330~S340, please refer to [link to documentation]. Figure 8 Steps S610, S6201, S710, S7201, S810~S830, S740, S6202, S630, S410~S430, and S330~S340 shown in the diagram will not be repeated here.

[0167] In this embodiment, each operation (quotient or remainder) divides the set of values ​​into multiple intervals. By combining the feature vectors corresponding to the quotient and remainder intervals, a unique feature vector can be generated for each sample object.

[0168] This application provides another data processing method, which can be applied to... Figure 1 The implementation environment shown or Figure 2 The implementation environment shown can be executed by a terminal or a server, or by both. In this embodiment, the method is described using the server as an example. Figure 10 As shown, this data processing method is in Figure 6 Steps S1010 to S1030 have been added between the steps in step S620 shown, as follows: Figure 10 As shown, the data processing method includes steps S6201, S1010 to S1030, and S6202. For steps S6201 and S6202, please refer to... Figure 6 The step S620 is shown. Steps S1010 to S1030 are described in detail below:

[0169] S1010. Obtain the second type of sample information features corresponding to the sample object information of the second type of sample object selected by the first type of sample object through the first feature vector generation model.

[0170] The process of obtaining the features of the second type of sample information corresponding to the sample object information of the second type of sample object selected by the first type of sample object is detailed in step S710, and will not be repeated here.

[0171] S1020. Generate a sparse vector of the second type of sample object selected by the first type of sample object based on the information features of the second type of sample, and divide the set of values ​​corresponding to the information features of the second type of sample into multiple third intervals and multiple fourth intervals; wherein the values ​​in the third interval are different from the values ​​in the fourth interval.

[0172] The process of generating the sparse vector of the sample of the second type of sample object selected by the first type of sample object is described in step S720, which will not be repeated here.

[0173] S1030, Map the sparse vectors of the second-class sample objects selected by the first-class sample objects to each third and fourth interval to obtain the sample feature vectors of the second-class sample objects selected by the first-class sample objects.

[0174] The process of obtaining the sample feature vector of the second type of sample object selected by the first type of sample object is described in step S730, and will not be repeated here.

[0175] It should be noted that, Figure 10 For further details on steps S610-S630, S410-S430, and S330-S340 shown, please refer to [link to detailed information]. Figure 6 Steps S610-S630, S410-S430 and S330-S340 shown will not be repeated here.

[0176] It is worth noting that this application provides another data processing method, which can be applied to... Figure 1 The implementation environment shown or Figure 2 The implementation environment shown can be executed by a terminal or a server, or by both. In this embodiment, the method is described using the server as an example. Figure 11 As shown, this data processing method is in Figure 3 In step S330 shown, step S330 is extended to step S1110. Step S1110 is described in detail below:

[0177] S1110. Select the first type of object that is selected as the second type of object as the target recommendation object corresponding to the second type of object.

[0178] Based on whether the second type of object was selected by the first type of object, the first type of object that selected the second type of object will be the target recommended object corresponding to the second type of object. That is, if the second type of object is selected, it is likely that the first type of object is interested in it, therefore the first type of object of the second type of object will be selected, and thus it can be recommended to the selected first type of object in the next recommendation. For example, if first type objects 1 and 3 click on second type object B, but first type object 2 does not click on the second type of object, then the target recommended objects corresponding to the second type of object will be first type objects 1 and 3.

[0179] It should be noted that, Figure 11 For detailed information on steps S310 to S320 and S340 shown, please refer to [link to relevant documentation]. Figure 3 Steps S310 to S320 and S340 shown in the diagram will not be described again here.

[0180] In this application embodiment, a data processing method is also provided, which can be applied to Figure 1 The implementation environment shown or Figure 2 The implementation environment shown can be executed by a terminal or a server, or by both. In this embodiment, the method is described using the server as an example. Figure 12 As shown, this data processing method is in Figure 3 In step S330 shown, step S330 is extended to S1210. Step S1210 is described in detail below:

[0181] S1210. Select the first type of object of the second type of object as the seed object, obtain other first type of objects similar to the seed object, and use the obtained other first type of objects as the target recommendation objects corresponding to the second type of object.

[0182] Based on whether the second type of object is selected by the first type of object, the first type of object that has selected the second type of object is used as the seed object. The seed object represents the group of objects that are interested in the second type of object.

[0183] Obtain other first-class objects similar to the seed object. These can be objects similar to all seed objects, such as those whose registration information includes Y, and whose registration information is similar to Y. Other first-class objects similar to the seed object can also be objects similar to some or individual seed objects, such as other first-class objects 9 that are only similar to seed object 1, and other first-class objects 7 that are similar to both seed object 2 and seed object 3.

[0184] Since the seed object has chosen the second type of object, other first-type objects similar to the seed object are more likely to choose the second type of object. Therefore, when other first-type objects similar to the seed object are obtained, all other first-type objects obtained are used as the target recommended objects corresponding to the second type of object.

[0185] It should be noted that, Figure 12 For detailed information on steps S310 to S320 and S340 shown, please refer to [link to relevant documentation]. Figure 3 Steps S310 to S320 and S340 shown in the diagram will not be described again here.

[0186] In this embodiment, a random number generation service and a seed source service are deployed through a generation service deployment request sent by the business party, so as to provide support for the subsequent implementation of corresponding random number generation based on the random number generation service and the seed source service.

[0187] This application provides another data processing method, which can be applied to... Figure 1 The implementation environment shown or Figure 2 The implementation environment shown can be executed by a terminal or a server, or by both. In this embodiment, the method is described using the server as an example. Figure 13 As shown, this data processing method is in Figure 3 In step S330 shown, step S330 is extended to S1310. Step S1310 is described in detail below:

[0188] S1310. Select the first type of object of the second type of object as the seed object, obtain other first type objects similar to the seed object, and use the obtained other first type objects and the seed object as the target recommendation objects corresponding to the second type of object.

[0189] After obtaining all other first-class objects similar to the seed object, all other first-class objects and the seed object are used as target recommendation objects.

[0190] It should be noted that, Figure 13 For detailed information on steps S310 to S320 and S340 shown, please refer to [link to relevant documentation]. Figure 3 Steps S310 to S320 and S340 shown in the diagram will not be described again here.

[0191] Based on the selection of the second type of object by the first type of object, the embodiments of this application generate target recommended objects in various ways, making the target recommended objects more comprehensive and meeting the needs of various scenarios.

[0192] This application provides another data processing method, which can be applied to... Figure 1 The implementation environment shown or Figure 2 The implementation environment shown can be executed by a terminal or a server, or by both. In this embodiment, the method is described using the server as an example. Figure 14 As shown, this data processing method is in Figure 12 Based on the S1210 shown, S1210 includes steps S1410 to S1440. Steps S1410 to S1440 are described in detail below:

[0193] S1410. Select the first type of object of the second type of object as the seed object.

[0194] S1420, Calculate the attribute correlation between the seed object and other first-class objects.

[0195] In this embodiment of the application, the attribute correlation between the seed object and other first-class objects refers to the correlation between the attribute information of the seed object and the attribute information of other first-class objects.

[0196] Optionally, attribute relevance includes at least one of interest relevance, operation relevance, and basic information relevance, where the basic information may be the basic registration information of the first type of object on the application.

[0197] In one example, attribute relevance can be characterized by the degree of overlap of attribute information. For example, the degree of overlap of attributes between the seed object and other first-class objects, such as interest overlap, can be calculated, and the degree of attribute overlap can be used as the attribute relevance between the seed object and other first-class objects.

[0198] In another example, attribute correlation can be characterized by the similarity of attribute information. For example, the attributes of a seed object can be converted into a seed attribute vector, and the attributes of other first-class objects can be converted into other first-class attribute vectors. Then, the similarity between the seed attribute vector and other first-class attribute vectors can be used as the attribute correlation between the seed object and other first-class objects.

[0199] S1430. Based on the attribute correlation between the seed object and other first-class objects, obtain other first-class objects similar to the seed object.

[0200] In this embodiment, other first-class objects similar to the seed object are obtained based on the degree of correlation between the attributes of the seed object and other first-class objects. In one example, if the attribute similarity between the seed object and other first-class objects exceeds a preset similarity threshold, then objects with attribute similarity exceeding the similarity threshold are considered as other first-class objects similar to the seed object.

[0201] In another example of this application's embodiments, if the attribute similarity between the seed object and other first-type objects exceeds a preset similarity threshold, then the objects with attribute similarity exceeding the similarity threshold are considered candidate objects. Furthermore, potential objects associated with the candidate objects are mined from the candidate objects, and the candidate objects and potential objects are considered as other first-type objects similar to the seed object. Specifically, the mining of potential objects from the candidate objects can be based on the industry of the candidate objects; for example, if the industry of the candidate objects is the training industry, then the potential objects could be students.

[0202] S1440. Use the other first-class objects obtained as the target recommendation objects corresponding to the second-class objects.

[0203] In this embodiment, the correlation between the attributes of the seed object and other first-class objects is used to make the identification of other first-class objects similar to the seed object reliable and comprehensive.

[0204] It should be noted that, Figure 14 For detailed information on steps S310 to S320 and S340 shown, please refer to [link to relevant documentation]. Figure 12 Steps S310 to S320 and S340 shown in the diagram will not be described again here.

[0205] Understandable, Figure 13 Please refer to step 1310 for the process of obtaining other first-class objects similar to the seed object. Figure 14 Steps S1420 and S1430. This application embodiment provides another data processing method, which can be applied to… Figure 1 The implementation environment shown or Figure 2 The implementation environment shown can be executed by a terminal or a server, or by both. In this embodiment, the method is described using the server as an example. Figure 15 As shown, this data processing method is in Figure 3 In step S340 shown, steps S1510 to S1530 and step S3401 are extended. Steps S1510 to S1530 and step S3401 are described in detail below:

[0206] S1510. Based on the feature vector of the first type of object, generate the initial feature vector of the target recommendation object.

[0207] In this embodiment of the application, if the target recommendation object is a first type of object, then the feature vector of the first type of object is used as the initial feature vector of the target recommendation object.

[0208] If the target recommendation object is a seed object and / or other first-class objects similar to the seed object, the initial feature vector of the target recommendation object can be obtained by random initialization.

[0209] S1520. Obtain the second feature vector generation model. The second feature vector generation model is obtained by training the sample feature vectors of the first type of sample objects and the sample feature vectors of the sample recommendation objects.

[0210] In this embodiment, a second feature vector generation model can be obtained from a blockchain network. This second feature vector generation model is pre-trained and can be obtained locally or from a model server used to store various models; no further limitation is made here. The second feature vector generation model is used to learn the coupling relationship between the feature vectors of the first type of object and the feature vectors of the target recommendation object, thereby obtaining the feature vector of the target recommendation object. The second feature vector generation model is trained using the sample feature vectors of the first type of sample object and the sample feature vectors of the sample recommendation object, where the sample feature vectors of the first type of sample object are obtained by the first feature vector generation model.

[0211] S1530. Input the initial feature vector of the target recommendation object and the feature vector of the first type of object into the second feature vector generation model to obtain the feature vector of the target recommendation object.

[0212] The initial feature vector of the target recommendation object and the feature vector of the first type of object are input into the second feature vector generation model. The second feature vector generation model learns the relationship between the first type of object and the target recommendation object. Then, based on the relationship between the first type of object and the target recommendation object and the feature vector of the first type of object, the initial feature vector is adjusted to obtain the feature vector of the target recommendation object.

[0213] Optionally, the second feature vector generation model includes a first-class object feature vector extraction network and a recommendation object feature vector extraction network, as well as a synthesis module for integrating the outputs of the first-class object feature vector extraction network and the second-class object feature vector extraction network; then, the initial feature vector of the target recommendation object is input into the target recommendation object feature vector extraction network, the feature vector of the first-class object is input into the first-class object feature vector extraction network, and the feature vector output by the recommendation object feature vector extraction network is used as the feature vector of the target recommendation object.

[0214] S3401. Based on the feature vector of the target recommendation object and the feature vector of the first type of object, determine the second type of object to be recommended to the target recommendation object.

[0215] It should be noted that, Figure 15For detailed information on steps S310 to S330 shown, please refer to [link to relevant documentation]. Figure 3 Steps S310 to S330 shown in the diagram will not be repeated here.

[0216] In this embodiment, the initial feature vector of the target recommendation object and the feature vector of the first type of object are input into the second feature vector generation model. The model learns the relationship between the first type of object and the target recommendation object, thereby accurately and reliably obtaining the feature vector of the target recommendation object.

[0217] This application provides another data processing method, which can be applied to... Figure 1 The implementation environment shown or Figure 2 The implementation environment shown can be executed by a terminal or a server, or by both. In this embodiment, the method is described using the server as an example. Figure 16 As shown, this data processing method is in Figure 15 The steps S1610 to S1630 are added before S1520 shown. Steps S1610 to S1630 constitute the training process of the second feature vector generation model, detailed below:

[0218] S1610. Obtain the sample feature vector of the first type of sample object, and the sample recommendation object.

[0219] In this embodiment, the object information of the first type of sample object can be input into the first feature vector generation model, and then obtained through the first feature vector generation model. The specific acquisition process can be found in [reference needed]. Figures 7-9 As shown, I will not go into detail here.

[0220] The sample recommendation object refers to the sample recommendation object corresponding to the second type of sample object, and the second type of sample object can be... Figure 7 The recommended sample objects corresponding to the second type of sample objects can be obtained based on the selection of the second type of sample objects by the first type of sample objects. For details, please refer to [link / reference / documentation]. Figures 11-13 I will not go into details here.

[0221] S1620. Randomly initialize to obtain the sample feature vector of the sample recommendation object.

[0222] In this embodiment of the application, the sample recommendation object is taken as a non-first type sample object as an example. After obtaining the sample recommendation object, the object information of the sample recommendation object can be converted into a one-hot vector, and then the sample feature vector of the sample recommendation object can be obtained by random initialization based on the one-hot vector.

[0223] S1630. Using the cross-entropy function as the loss function, train the second feature vector generation model based on the sample feature vectors of the first type of sample objects and the sample feature vectors of the sample recommendation objects.

[0224] The cross-entropy function is either sigmoid or softmax. The sample feature vectors of the first type of sample objects and the sample feature vectors of the sample recommendation objects are input into the second feature vector generation model. The training of the second feature vector generation model is completed when the loss function converges.

[0225] Optionally, when the second feature vector generation model includes a feature vector extraction network for the first type of object and a feature vector extraction network for the recommended object, two sets of cross-entropy functions are used as the loss functions for the feature vector extraction network for the first type of object and the feature vector extraction network for the target recommended object, respectively, to train the second feature vector generation model.

[0226] It should be noted that, Figure 16 For detailed information on steps S310-S330, S1510-S1530, and S3401 shown, please refer to [link to relevant documentation]. Figure 15 Steps S310-S330, S1510-S1530, and S3401 shown in the diagram will not be repeated here.

[0227] In this embodiment, the cross-entropy function is used as the loss function, and the second feature vector generation model is trained based on the sample feature vectors of the first type of sample objects and the sample feature vectors of the sample recommendation objects. This enables the second feature vector generation model to fully learn the relationship between the sample feature vectors of the first type of sample objects and the sample feature vectors of the sample recommendation objects, ensuring the accuracy of the trained second feature vector generation model.

[0228] This application provides another data processing method, which can be applied to... Figure 1 The implementation environment shown or Figure 2 The implementation environment shown can be executed by a terminal or a server, or by both. In this embodiment, the method is described using the server as an example. Figure 17 As shown, this data processing method is in Figure 3 In step S340 shown, step S340 is extended to S1710 to S1730. Steps S1710 to S1730 are described in detail below:

[0229] S1710. Obtain the feature vector of the target recommendation object.

[0230] The process of obtaining the feature vector of the target recommendation object can participate in Figure 15 As shown.

[0231] S1720. Based on the feature vector of the target recommendation object and the feature vector of the first type of object, determine the similarity between the target recommendation object and the first type of object.

[0232] In one example of this application embodiment, since the feature vector of an object is used to characterize the feature information of the object, the similarity between the feature vector of the target recommended object and the feature vector of the first type of object is used to characterize the similarity between the feature information of the target recommended object and the first type of object.

[0233] In another example, if the target recommendation object is another first-class object similar to the seed object, and the seed object is a first-class object that has selected a second-class object, the similarity between the target recommendation object and the seed object can be determined based on the feature vector of the target recommendation object and the feature vector of the selected seed object.

[0234] S1730. If the similarity between the target recommended object and the first type of object is higher than a preset threshold, then the second type of object recommended to the target recommended object is determined.

[0235] Understandably, the higher the similarity between the target recommendation object and the first type of object, the more likely the corresponding second type of object is to match the target recommendation object. Therefore, a similarity threshold can be preset. If the similarity between the target recommendation object and the first type of object is higher than the preset threshold, the second type of object to be recommended to the target recommendation object is determined, and then the determined second type of object is recommended to the target recommendation object. This similarity threshold can be flexibly adjusted according to actual needs; for example, a similarity threshold of 80%.

[0236] In one example, if the similarity between the target recommended object and the first type of object is lower than a preset threshold, the second type of object will be filtered out, that is, it will not be recommended to the target recommended object.

[0237] It should be noted that, Figure 17 For detailed information on steps S310 to S330 shown, please refer to [link to relevant documentation]. Figure 3 Steps S310 to S330 shown in the diagram will not be repeated here.

[0238] In this embodiment, the similarity between the target recommended object and the first type of object is determined by the feature vector of the target recommended object and the feature vector of the first type of object. Then, based on the similarity, the second type of object recommended to the target recommended object is determined, further ensuring that the recommended second type of object conforms to the target recommended object.

[0239] This application provides another data processing method, which can be applied to... Figure 1 The implementation environment shown or Figure 2The implementation environment shown can be executed by a terminal or a server, or by both. In this embodiment, the method is described using the server as an example. Figure 18 As shown, this data processing method is in Figure 17 In step S1720 shown, S1720 is extended to steps S1810 to S1830. Steps S1810 to S1830 are described in detail below:

[0240] S1810. Obtain the global feature vector of the first type of object. The global feature vector of the first type of object is used to describe the features of the global information of the first type of object.

[0241] In this embodiment of the application, a global feature vector of a first type of object is obtained, wherein the global feature vector is used to describe the features of the global information of the first type of object. Optionally, the feature vector of the first type of object output by the second feature vector generation model can be used as the global feature vector of the first type of object.

[0242] Optionally, the global feature vector of the first type of object is Where E g For the global embedding of the first type of object, E s W represents the embedding input into the second feature vector generation model. g Generate model parameters for the second feature vector.

[0243] S1820. Determine the local feature vector of the first type of object based on the feature vector of the target recommendation object and the feature vector of the first type of object; the local feature vector of the first type of object is used to describe the features of the local information of the first type of object related to the target recommendation object.

[0244] In this embodiment, the local feature vector of the first type of object is used to describe the features of the local information of the first type of object related to the target recommendation object. That is, the local information of the first type of object is related to the target recommendation object, and the local feature vector corresponding to the local information of the first type of object is determined based on the feature vector of the target recommendation object and the feature vector of the first type of object.

[0245] Optionally, the local feature vector of the first type of object is Where E l E is a local embedding representation of the first type of object. s E represents the embedding input into the second feature vector generation model. u This represents the embedding of the target recommended object output by the second feature vector generation model.

[0246] S1830. Based on the similarity between the feature vector of the target recommended object and the global feature vector of the first type of object, and the similarity between the feature vector of the target recommended object and the local feature vector of the first type of object, determine the similarity between the target recommended object and the first type of object.

[0247] In this embodiment, the global correlation between the target recommended object and the first type of object can be obtained based on the similarity between the feature vector of the target recommended object and the global feature vector of the first type of object; the local correlation between the target recommended object and the first type of object can be obtained based on the similarity between the feature vector of the target recommended object and the local feature vector of the first type of object; and then the similarity between the target recommended object and the first type of object can be determined based on the global correlation and the local correlation.

[0248] Optionally, the cosine similarity between the feature vector of the target recommended object and the global feature vector of the first type of object can be used as the global correlation, and the cosine similarity between the feature vector of the target recommended object and the local feature vector of the first type of object can be used as the local correlation.

[0249] Optionally, the sum of global relevance and local relevance can be used as the similarity between the target recommended object and the first type of object.

[0250] Optionally, weights can be set for global relevance and local relevance respectively, and the similarity between the target recommended object and the first type of object can be obtained by weighted summation of global relevance and local relevance.

[0251] It should be noted that, Figure 18 For detailed information on steps S310-S330, S1710, and S1730 shown, please refer to [link to relevant documentation]. Figure 17 Steps S310 to S330, S1710 and S1730 shown in the diagram will not be described again here.

[0252] In this embodiment, the similarity between the target recommended object and the first type of object is calculated by the global and local correlation between the first type of object and the target recommended object, thus ensuring the accuracy of the similarity from two dimensions.

[0253] For ease of understanding, the following is based on Figure 1 or Figure 2 The implementation environment shown below provides a detailed description of the data processing method provided in the embodiments of this application:

[0254] The data processing method includes the training process of the first feature vector generation model and the second feature vector generation model, which are described in detail below:

[0255] Among them, such as Figure 19As shown, the first feature vector generation model includes a first-type object feature extraction network, a second-type object feature extraction network, and a feature synthesis module for integrating the outputs of the first-type object feature extraction network and the second-type object feature extraction network; wherein the first-type object feature extraction network and the second-type object feature extraction network have the same structure: feature input layer, everage pooling layer, attention layer, and fully connected layer, respectively.

[0256] like Figure 20 As shown, the data processing method includes:

[0257] S2010. Input the sample object information of the first type of sample object and the object information of the second type of sample object into the first feature vector generation model for training, wherein the second type of sample object carries a label indicating whether it was selected by the first type of sample object.

[0258] In this embodiment of the application, there is a coupling relationship between the first type of sample object and the second type of sample object, that is, whether the second type of sample object is selected by the first type of sample object.

[0259] After inputting the sample object information of the first type of sample object and the object information of the second type of sample object into the first feature vector generation model, the following steps are included:

[0260] Step (1) inputs the sample object information of the first type of sample object into the first type of object feature extraction network. The feature input layer of the first type of object feature extraction network converts the sample object information of the first type of sample object into information features of the first type of sample object and generates a one-hot vector. For example, the sample object information of the first type of sample object is basic attributes, operation behavior, etc. The sample object information is sparsely represented as ID to obtain features. Define all values ​​S of a certain feature ε: {0,....n-1} S is the set of all ID values ​​of the feature, and n represents the size of the set; after one-hot encoding of feature ε, the one-hot feature vector e is obtained. ε e ε Let e ​​be the one-hot vector represented by feature ε. ε ∈R S , where R is the set of all one-hot eigenvectors.

[0261] Step (2) The equalization pooling layer of the first-class object feature extraction network divides the set of values ​​corresponding to the sample objects of the first class into multiple first intervals and obtains second intervals to supplement the first intervals. That is, by using two complementary functions (integer quotient and remainder functions), two separate embedding tables can be generated, and the embeddings are combined in a way that generates unique embeddings for each class in a certain way. For example, the set of values ​​corresponding to feature ε is divided into k intervals using quotient and remainder functions, and each equivalence class of a given partition is assigned a "bucket" that maps to the embedding vector. Therefore, each interval of each partition corresponds to an embedding table. Under the complementary intervals, after each embedding generated in each interval is combined through some operation, each index is mapped to a different embedding vector. For example, N embedding tables are randomly initialized, and m i Let represent the length of the i-th embedding table. Let S denote the set of all feature values; let h(x):x∈S represent the hash value of the feature; let M... j Let represent the denominator of the modulo function for the j-th interval; Where j = 2, ..., k; then the first interval P1 and the supplementary interval The calculation formula is as follows:

[0262] P1={{h(x)mod m1}}x∈S

[0263]

[0264] The second interval P2 and the supplementary interval The calculation formula is as follows:

[0265] P2={{h(x)mod m2}}x∈S

[0266]

[0267] The j-th interval P j and supplementary interval The calculation formula is as follows:

[0268] P j ={{h(x)mod m j x∈S

[0269]

[0270] Step (3) The attention layer of the first type of object feature extraction network maps the one-hot vector of the first type of sample object to the embedding vector of each interval to obtain the sample feature vector of the first type of sample object; for example, after concatenating the first interval and the second interval, j intervals are obtained, W′ j Let e ​​be the embedding table of the j-th interval; then for the one-hot eigenvector e ε Map it to the embedding table of the j-th interval to generate the embedding vector x of the j-th interval. j :

[0271] x′ j =W′ j *e i

[0272] The embedding of feature ε can be represented by embedding x for multiple intervals. ε The calculation is as follows:

[0273] x ε = f{x′1,x′2,...,x′ j}

[0274] Where f is a mapping function, which can be an MLP (multi-layer perceptrons) function.

[0275] Step (4) inputs the sample feature vectors of the first type of sample objects and the second type of sample objects into the feature synthesis module. The feature synthesis module learns the coupling relationship between the first type of sample objects and the second type of sample objects. For example, it learns the embeddings of the first type of sample objects and the embeddings of the second type of sample objects respectively. Then, it calculates the similarity between the first type of sample objects and the second type of sample objects using the dot product method. The similarity between the first type of sample objects and the second type of sample objects represents the coupling relationship between them. If the similarity is greater than a threshold, it means that the second type of sample object is selected by the first type of sample object. After learning the coupling relationship between the first type of sample objects and the second type of sample objects, the feature synthesis module adjusts the network parameters of the feature extraction network of the first type of object according to the sample feature vector of the second type of sample objects. The feature vector output by the fully connected layer of the feature extraction network of the first type of object is used as the feature vector of the first type of sample object.

[0276] The loss function for the first feature vector generation model is as follows:

[0277]

[0278] Where u represents the first type of sample object, i represents the second type of sample object, D represents the set of the first type of sample object and the second type of sample object pair, D=1 indicates that the first type of sample object selected the second type of sample object, D=0 indicates that the first type of sample object did not select the second type of sample object, K represents the set of negatively sampled samples that did not select the second type of sample object, and p represents the probability output by the first feature vector generation model.

[0279] The sample object information of the second type of sample object is input into the second type of sample object feature extraction network. The feature input layer of the second type of sample object feature extraction network converts the sample object information of the second type of sample object into the information features of the second type of sample object and generates a one-hot vector. The everage pooling layer of the second type of sample object feature extraction network divides the set of values ​​corresponding to the sample object of the second type of sample object into multiple third intervals and obtains a fourth interval to supplement the third intervals. The attention layer of the second type of sample object feature extraction network maps the one-hot vector of the second type of sample object to the embedding vector of each interval to obtain the sample feature vector of the second type of sample object. For the specific process, please refer to steps (1) to (4) above, which will not be repeated here.

[0280] S2020. For each second-class sample object, generate a seed object based on the first-class sample object it selected, and obtain sample recommendation objects similar to the seed object.

[0281] The target recommendation object is the recommendation object corresponding to the second type of sample object.

[0282] S2030. Use the sample feature vector of the first type of sample object output by the first feature vector generation model as the sample feature vector of the seed object, and randomly initialize it to obtain the sample feature vector of the sample recommendation object.

[0283] S2040. Input the sample feature vector of the seed object and the sample feature vector of the sample recommended object into the second feature vector generation model for training.

[0284] The second feature vector generation model includes a feature vector extraction network for the first type of object and a feature vector extraction network for the recommendation object. The feature vector extraction network for the first type of object includes a fully connected layer, a self-attention unit, and an attention unit. The fully connected layer fits the input, and the pRELU activation function introduces non-linearity. The self-attention unit calculates the weights, and average pooling is used to calculate the average. The feature vector extraction network for the first type of object takes the sample feature vector of the seed object as input, passes it through a spatial transformation layer (fully connected layer and PRELU activation function), then through a self-attention unit and an attention unit, and finally through average pooling to output the sample feature vector of the seed object. The feature vector extraction network for the recommendation object takes the embedding of the sample feature vector of the recommended object as input, passes it through a spatial transformation (using the same fully connected layer and PRELU activation function as the first type of object feature vector extraction network), and outputs the sample feature vector of the recommended object.

[0285] The second feature vector generation model also includes a synthesis module, which is used to synthesize the sample feature vectors of the seed objects output by the first type of object feature vector extraction network and the sample feature vectors of the sample recommended objects output by the recommendation object feature vector extraction network. It learns the relationship between the sample feature vectors of the seed objects and the sample feature vectors of the sample recommended objects, and then feeds it back to the recommendation object feature vector extraction network.

[0286] Multiple sets of cross-entropy are used as the loss function for generating the second feature vector.

[0287] In one example, after training the first and second feature vector generation models, the output is the model parameters of the first and second feature vector generation models, and the feature vector of the target recommended object. The model parameters can be loaded into memory for later use, and the feature vector of the target recommended object can be stored in a key-value (kv) cache.

[0288] like Figure 21 As shown, Figure 21 For based on Figure 20 The data processing methods for the model shown, i.e., the application scenarios of the model, include:

[0289] S2110, Construct the information characteristics of the first type of object and the information characteristics of the second type of object.

[0290] S2120. The information features of the first type of object and the information features of the second type of object generate the feature vector of the first type of object.

[0291] Steps S2110 and S2120 can be obtained through the first feature vector generation model. That is, the object information of the first type of object and the object of the second type of object are input into the first feature vector generation model, and the feature vector output by the fully connected layer of the feature extraction network of the first type of object in the first feature vector generation model is used as the feature vector of the first type of object.

[0292] S2130. Generate feature vectors of seed objects based on clustering.

[0293] Optionally, the first feature vector generation model also predicts whether the second type of object is selected by the first type of object. Based on whether the second type of object is selected by the first type of object, for each second type of object, the first type of objects that have selected the second type of object are aggregated. The first type of objects that have selected the second type of object are used as seed objects, and the feature vectors of each first type of object that has selected the second type of object in step S2120 are aggregated to obtain the feature vector of the seed object.

[0294] S2140. Generate the feature vector of the target recommended object based on the feature vector of the seed object.

[0295] The target recommendation object is other first-class objects similar to the seed object, and the initial feature vector of the target recommendation object is obtained by random initialization.

[0296] The feature vectors of the seed object and the initial feature vectors of the target recommended object are input into the second feature vector generation model. The feature vectors of the recommended object feature vectors extracted by the second feature vector generation model are used as the feature vectors of the target recommended object.

[0297] S2150. Calculate the similarity between the seed object and the target recommended object based on the feature vector of the seed object and the feature vector of the target recommended object.

[0298] Obtain the local feature vector of the seed object

[0299] Where E l For the local embedding of the seed object, E s E represents the embedding of the seed object. u W represents the embedding of the target recommendation object. l Here, T represents the model parameters, and T is the matrix transpose.

[0300] Get the global feature vector of the seed object

[0301] Ultimately, E l E g respectively with E u The similarity score between the target recommendation object and the seed object is obtained by calculating the weighted sum of cosines.

[0302] score u,s =α*cosine(E u E g )+β*cosine(e u E l )

[0303] Where α and β are weight values.

[0304] In one example, after obtaining the similarity between the seed object and the target recommended object, it can be determined whether to recommend the second type of object to the target recommended object based on their similarity. If the similarity between the seed object and the target recommended object is higher than a preset threshold, then the second type of object is recommended to the target recommended object.

[0305] In another example, the similarity between the seed object and the target recommended object can also be used as a feature vector to generate the first feature vector of the model or other models.

[0306] The apparatus embodiments described herein can be used to execute the data processing methods described above in the embodiments of this application. For details not disclosed in the apparatus embodiments of this application, please refer to the embodiments of the data processing methods described above in this application.

[0307] This application provides a data processing apparatus, such as... Figure 22 As shown, the data processing device can be configured in an execution node, and the device includes:

[0308] The acquisition module 2210 is used to acquire object information of the first type of object and object information of the second type of object;

[0309] The calculation module 2220 is used to calculate the feature vector of the first type of object based on the object information of the first type of object, the object information of the second type of object, and the coupling relationship between the first type of object and the second type of object;

[0310] The generation module 2230 is used to generate target recommendation objects corresponding to each second type of object based on the selection of the second type of object by the first type of object;

[0311] The determining module 2240 is used to obtain the feature vector of the target recommended object, and based on the feature vector of the target recommended object and the feature vector of the first type of object, determine the second type of object to be recommended to the target recommended object. In one embodiment of this application, based on the aforementioned scheme, the calculation module 2220 is further used to obtain a first feature vector generation model, which is used to learn the coupling relationship between the first type of object and the second type of object; the object information of the first type of object and the object information of the second type of object are input into the first feature vector generation model to obtain the feature vector of the first type of object.

[0312] In one embodiment of this application, based on the aforementioned scheme, the first feature vector generation model includes a first type of object feature extraction network, a second type of object feature extraction network, and a feature synthesis module for integrating the outputs of the first type of object feature extraction network and the second type of object feature extraction network. The calculation module 2220 is further used to input the object information of the first type of object into the first type of object feature extraction network, input the object information of the second type of object into the second type of object feature extraction network, and use the feature vector output by the first type of object feature extraction network as the feature vector of the first type of object.

[0313] In one embodiment of this application, based on the aforementioned scheme, the apparatus includes a first training module. The first training module is used to construct a first sample pair and a second sample pair according to the selection of a second type of sample object by a first type of sample object. The first sample pair includes sample object information of the first type of sample object and sample object information of the second type of sample object selected by the first type of sample object. The second sample pair includes sample object information of the first type of sample object and sample object information of the second type of sample object not selected by the first type of sample object. The first sample pair is input as a positive sample and the second sample pair as a negative sample into a first feature vector generation model. The loss value of the first feature vector generation model is calculated based on the sample object similarity output by the first feature vector generation model for the positive and negative samples. The model parameters of the first feature vector generation model are adjusted according to the loss value of the first feature vector generation model to train the first feature vector generation model.

[0314] In one embodiment of this application, based on the aforementioned scheme, after the first training module inputs the first sample pair as a positive sample and the second sample pair as a negative sample into the first feature vector generation model, it is further configured to obtain the first type of sample information features corresponding to the sample object information of the first type of sample object through the first feature vector generation model, generate the sample sparse vector of the first type of sample object according to the first type of sample information features; divide the set of values ​​corresponding to the first type of sample information features into multiple first intervals and multiple second intervals, wherein the values ​​in the second intervals are different from the values ​​in the first intervals; map the sample sparse vector of the first type of sample object to each first interval and second interval to obtain the sample feature vector of the first type of sample object; calculate the sample object similarity for positive and negative samples based on the sample feature vector of the first type of sample object, the sample feature vector of the second type of sample object selected by the first type of sample object, and the sample feature vector of the second type of sample object not selected by the first type of sample object.

[0315] In one embodiment of this application, based on the aforementioned scheme, the first training module is further configured to perform a modulo operation on the set of values ​​corresponding to the information features of the first type of samples to divide them into multiple first intervals; and to perform a quotient operation on the set of values ​​corresponding to the information features of the first type of samples to divide them into multiple second intervals. In another embodiment of this application, based on the aforementioned scheme, the first training module is further configured to create feature vector matrices for the first intervals and the second intervals respectively, wherein the feature vector matrices for the first intervals and the second intervals are different; to map the sparse vectors of the first type of sample objects to the feature vector matrix of each first interval to obtain the feature vector of each first interval; to map the sparse vectors of the first type of sample objects to the feature vector matrix of each second interval to obtain the feature vector of each second interval; and to generate the sample feature vector of the first type of sample objects based on the feature vectors of each first interval and each second interval.

[0316] In one embodiment of this application, based on the aforementioned scheme, after the first training module inputs the first sample pair as a positive sample and the second sample pair as a negative sample into the first feature vector generation model, it is further configured to obtain the second-class sample information features corresponding to the sample object information of the second-class sample object selected by the first-class sample object through the first feature vector generation model; generate the sample sparse vector of the second-class sample object selected by the first-class sample object according to the second-class sample information features; divide the set of values ​​corresponding to the second-class sample information features into multiple third intervals and multiple fourth intervals; wherein, the values ​​in the third interval are different from the values ​​in the fourth interval, and the sample sparse vector of the second-class sample object selected by the first-class sample object is mapped to each third interval and fourth interval to obtain the sample feature vector of the second-class sample object selected by the first-class sample object.

[0317] In one embodiment of this application, based on the foregoing scheme, the generation module 2230 is further configured to use the first type of object selected as the second type of object as the target recommended object corresponding to the second type of object; or use the first type of object selected as the second type of object as the seed object, obtain other first type of objects similar to the seed object, and use the obtained other first type of objects as the target recommended object corresponding to the second type of object; or use the first type of object selected as the second type of object as the seed object, obtain other first type of objects similar to the seed object, and use the obtained other first type of objects and the seed object as the target recommended object corresponding to the second type of object.

[0318] In one embodiment of this application, based on the aforementioned scheme, the generation module 2230 is further used to calculate the attribute correlation between the seed object and other first-class objects; and to obtain other first-class objects similar to the seed object based on the attribute correlation between the seed object and other first-class objects.

[0319] In one embodiment of this application, based on the foregoing scheme, the determining module 2240 is further configured to generate an initial feature vector of the target recommendation object based on the feature vector of the first type of object; obtain a second feature vector generation model, which is obtained by training the sample feature vector of the first type of sample object and the sample feature vector of the sample recommendation object; and input the initial feature vector of the target recommendation object and the feature vector of the first type of object into the second feature vector generation model to obtain the feature vector of the target recommendation object.

[0320] In one embodiment of this application, based on the foregoing scheme, the device further includes a second training module. The second training module is used to obtain the sample feature vector of the first type of sample object and the sample recommendation object; randomly initialize the sample feature vector of the sample recommendation object; use the cross-entropy function as the loss function, and train the second feature vector generation model based on the sample feature vector of the first type of sample object and the sample feature vector of the sample recommendation object.

[0321] In one embodiment of this application, based on the foregoing scheme, the determining module 2240 is further configured to determine the similarity between the target recommended object and the first type of object based on the feature vector of the target recommended object and the feature vector of the first type of object; if the similarity between the target recommended object and the first type of object is higher than a preset threshold, then a second type of object recommended to the target recommended object is determined. In one embodiment of this application, based on the foregoing scheme, the determining module 2240 is further configured to obtain the global feature vector of the first type of object, the global feature vector of the first type of object being used to describe the features of the global information of the first type of object; determine the local feature vector of the first type of object based on the feature vector of the target recommended object and the feature vector of the first type of object; the local feature vector of the first type of object being used to describe the features of the local information of the first type of object related to the target recommended object. The similarity between the target recommended object and the first type of object is determined based on the similarity between the feature vector of the target recommended object and the global feature vector of the first type of object, and the similarity between the feature vector of the target recommended object and the local feature vector of the first type of object. It should be noted that the apparatus provided in the above embodiments and the method provided in the above embodiments belong to the same concept, and the specific manner in which each module and unit performs its operation has been described in detail in the method embodiments, and will not be repeated here.

[0322] Embodiments of this application also provide an electronic device, including one or more processors and a storage device, wherein the storage device is used to store one or more computer programs, which, when executed by one or more processors, cause the electronic device to implement the data processing method described above.

[0323] Figure 23 A schematic diagram of the structure of a computer system suitable for implementing the electronic device of the present application is shown.

[0324] It should be noted that, Figure 23 The computer system 2300 of the electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application, wherein the electronic device may be a terminal or a server.

[0325] like Figure 23As shown, the computer system 2300 includes a central processing unit (CPU) 2301, which can perform various appropriate actions and processes, such as executing the methods described in the above embodiments, based on a program stored in read-only memory (ROM) 2302 or a program loaded from storage portion 2308 into random access memory (RAM) 2303. The RAM 2303 also stores various programs and data required for system operation. The CPU 2301, ROM 2302, and RAM 2303 are interconnected via a bus 2304. An input / output (I / O) interface 2305 is also connected to the bus 2304.

[0326] In some embodiments, the following components are connected to the I / O interface 2305: an input section 2306 including a keyboard, mouse, etc.; an output section 2307 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 2308 including a hard disk, etc.; and a communication section 2309 including a network interface card such as a LAN (Local Area Network) card, modem, etc. The communication section 2309 performs communication processing via a network such as the Internet. A drive 2310 is also connected to the I / O interface 2305 as needed. A removable medium 2311, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 2310 as needed so that computer programs read from it can be installed into the storage section 2308 as needed.

[0327] Specifically, according to embodiments of this application, the processes described above with reference to the flowcharts can be implemented as computer programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program including a computer program 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 2309, and / or installed from removable medium 2311. When the computer program is executed by processor (CPU) 2301, it performs various functions defined in the system of this application.

[0328] It should be noted that the computer-readable medium shown in the embodiments of this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory, flash memory, optical fiber, portable compact disc read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying a computer-readable computer program. 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 computer program contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wired, etc., or any suitable combination thereof.

[0329] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this application. Each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains 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 a computer program.

[0330] The units or modules described in the embodiments of this application can be implemented in software or hardware, and can also be located in a processor. The names of these units or modules do not necessarily limit the specific unit or module itself.

[0331] Another aspect of this application provides a computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the method described above. This computer-readable storage medium may be included in the electronic device described in the above embodiments, or it may exist independently and not assembled into the electronic device.

[0332] Another aspect of this application provides a computer program product comprising a computer program stored in a computer-readable storage medium. A processor of an electronic device reads the computer program from the computer-readable storage medium and executes the computer program, causing the electronic device to perform the methods described above in the various embodiments.

[0333] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to the embodiments of this application, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0334] Other embodiments of this application will readily conceive of by considering the specification and practicing the embodiments disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein.

[0335] The above content is merely a preferred exemplary embodiment of this application and is not intended to limit the implementation of this application. Those skilled in the art can easily make corresponding modifications or alterations based on the main concept and spirit of this application. Therefore, the scope of protection of this application should be determined by the scope of protection claimed in the claims.

Claims

1. A data processing method, characterized in that, Application servers, including: In response to a data processing request sent by the terminal, object information of a first type of object and object information of a second type of object are obtained. The second type of object is something known to the first type of object. The first type of object is the user of the terminal or an application account. The object information of the first type of object includes basic attribute information, operation information, and / or browsing information generated by the first type of object through the terminal. The second type of object is an advertisement, product, or article displayed on the terminal. The object information of the second type of object includes notification information corresponding to the advertisement, product information, or content information of the article. The object information of the first type of object is input into the first type of object feature extraction network of the first feature vector generation model to obtain the initial feature vector of the first type of object. The object information of the second type of object is input into the second type of object feature extraction network of the first feature vector generation model to obtain the initial feature vector of the second type of object. The initial feature vector of the first type of object and the initial feature vector of the second type of object are combined by the feature synthesis module of the first feature vector generation model to learn the coupling relationship between the first type of object and the second type of object. The coupling relationship between the first type of object and the second type of object is fed back to the feature extraction network of the first type of object, so that the feature extraction network of the first type of object can adjust the initial feature vector of the first type of object by referring to the coupling relationship between the first type of object and the second type of object and the initial feature vector of the second type of object, and obtain the feature vector of the first type of object. Based on the selection of the second type of object by the first type of object, a target recommended object corresponding to each second type of object is generated. The selection includes the first type of object viewing or purchasing the second type of object by clicking on the terminal. Obtain the feature vector of the target recommended object; based on the feature vector of the target recommended object and the feature vector of the first type of object, determine the second type of object to be recommended to the target recommended object; and send the determined second type of object to the terminal corresponding to the target recommended object for display. The first feature vector generation model is obtained through the following training steps: Based on the selection of the second type of sample objects by the first type of sample objects, a first sample pair and a second sample pair are constructed. The first sample pair includes the sample object information of the first type of sample objects and the sample object information of the second type of sample objects selected by the first type of sample objects. The second sample pair includes the sample object information of the first type of sample objects and the sample object information of the second type of sample objects not selected by the first type of sample objects. The first sample pair is used as a positive sample and the second sample pair is used as a negative sample. The first feature vector generation model is then input into the first feature vector generation model. The loss value of the first feature vector generation model is calculated based on the sample object similarity output by the first feature vector generation model for the positive sample and the negative sample. The model parameters of the first feature vector generation model are adjusted based on the loss value of the first feature vector generation model in order to train the first feature vector generation model.

2. The method according to claim 1, characterized in that, After inputting the first sample pair as positive samples and the second sample pair as negative samples into the first feature vector generation model, the method further includes: The first feature vector generation model is used to obtain the first type of sample information features corresponding to the sample object information of the first type of sample object; and the first type of sample sparse vector is generated based on the first type of sample information features. The set of values ​​corresponding to the information features of the first type of sample is divided into multiple first intervals and multiple second intervals, wherein the values ​​in the second intervals are different from the values ​​in the first intervals. Map the sparse vectors of the first type of sample objects to each first interval and second interval to obtain the sample feature vectors of the first type of sample objects; Based on the sample feature vectors of the first type of sample objects, the sample feature vectors of the second type of sample objects selected by the first type of sample objects, and the sample feature vectors of the second type of sample objects not selected by the first type of sample objects, the sample object similarity for the positive sample and the negative sample is calculated.

3. The method according to claim 2, characterized in that, The step of dividing the set of values ​​corresponding to the information features of the first type of sample into multiple first intervals and multiple second intervals includes: The set of values ​​corresponding to the information features of the first type of sample is moduloed to divide it into multiple first intervals; The set of values ​​corresponding to the information features of the first type of sample is divided into multiple second intervals by taking the quotient.

4. The method according to claim 3, characterized in that, The step of mapping the sparse vectors of the first type of sample objects to each first interval and second interval to obtain the sample feature vectors of the first type of sample objects includes: Create feature vector matrices for the first interval and the second interval respectively, wherein the feature vector matrix of the first interval is different from the feature vector matrix of the second interval; The sparse vectors of the first type of sample objects are mapped to the feature vector matrix of each first interval to obtain the feature vector of each first interval; The sparse vectors of the first type of sample objects are mapped to the feature vector matrix of each second interval to obtain the feature vector of each second interval; Based on the feature vectors of each first interval and each second interval, a sample feature vector of the first type of sample object is generated.

5. The method according to claim 1, characterized in that, After inputting the first sample pair as positive samples and the second sample pair as negative samples into the first feature vector generation model, the method further includes: The second type of sample information features corresponding to the sample object information of the second type of sample object selected by the first type of sample object are obtained through the first feature vector generation model; Generate a sparse vector of the second type of sample object selected by the first type of sample object based on the information features of the second type of sample; The set of values ​​corresponding to the information features of the second type of sample is divided into multiple third intervals and multiple fourth intervals; wherein the values ​​in the third intervals are different from the values ​​in the fourth intervals. The sparse vectors of the second-class sample objects selected by the first-class sample objects are mapped to each third and fourth interval to obtain the sample feature vectors of the second-class sample objects selected by the first-class sample objects.

6. The method according to claim 1, characterized in that, The step of generating target recommendation objects corresponding to each second type of object based on the selection of the second type of object by the first type of object includes: The first type of object selected from the second type of object will be used as the target recommendation object corresponding to the second type of object; or The first type of object selected from the second type of object is used as the seed object. Other first type of objects similar to the seed object are obtained, and these other first type of objects are used as the target recommendation objects corresponding to the second type of object; or The first type of object selected from the second type of object is used as the seed object. Other first type of objects similar to the seed object are obtained, and the obtained other first type of objects and the seed object are used as the target recommendation objects corresponding to the second type of object.

7. The method according to claim 6, characterized in that, The step of obtaining other first-type objects similar to the seed object includes: Calculate the attribute correlation between the seed object and the other first-class objects; Based on the attribute correlation between the seed object and the other first-class objects, other first-class objects similar to the seed object are obtained.

8. The method according to claim 1, characterized in that, The step of obtaining the feature vector of the target recommendation object includes: Based on the feature vectors of the first type of object, an initial feature vector of the target recommendation object is generated; A second feature vector generation model is obtained, which is trained using the sample feature vectors of the first type of sample objects and the sample feature vectors of the sample recommendation objects. The initial feature vector of the target recommendation object and the feature vector of the first type of object are input into the second feature vector generation model to obtain the feature vector of the target recommendation object.

9. The method according to claim 8, characterized in that, The second feature vector generation model is obtained through the following training steps: Obtain the sample feature vector of the first type of sample object, and the sample recommendation object; The sample feature vector of the recommended sample object is obtained by random initialization; Using the cross-entropy function as the loss function, the second feature vector generation model is trained based on the sample feature vectors of the first type of sample objects and the sample feature vectors of the sample recommendation objects.

10. The method according to any one of claims 1 to 9, characterized in that, The step of determining the second type of object to be recommended to the target object based on the feature vector of the target recommendation object and the feature vector of the first type of object includes: Based on the feature vector of the target recommendation object and the feature vector of the first type of object, the similarity between the target recommendation object and the first type of object is determined. If the similarity between the target recommended object and the first type of object is higher than a preset threshold, then the second type of object is determined to be recommended to the target recommended object.

11. The method according to claim 10, characterized in that, Determining the similarity between the target recommendation object and the first type of object based on the feature vector of the target recommendation object and the feature vector of the first type of object includes: Obtain the global feature vector of the first type of object; the global feature vector of the first type of object is used to describe the features of the global information of the first type of object. Based on the feature vector of the target recommendation object and the feature vector of the first type of object, a local feature vector of the first type of object is determined; the local feature vector of the first type of object is used to describe the features of local information of the first type of object related to the target recommendation object. The similarity between the target recommended object and the first type of object is determined based on the similarity between the feature vector of the target recommended object and the global feature vector of the first type of object, and the similarity between the feature vector of the target recommended object and the local feature vector of the first type of object.

12. A data processing apparatus, characterized in that, The device is deployed on a server and includes: The acquisition module is used to respond to a data processing request sent by the terminal to acquire object information of a first type of object and object information of a second type of object, wherein the second type of object is something known to the first type of object; wherein, the first type of object is the user of the terminal or an application account, and the object information of the first type of object includes basic attribute information, operation information and / or browsing information generated by the first type of object through the terminal; the second type of object is an advertisement, product or article displayed on the terminal, and the object information of the second type of object includes notification information corresponding to the advertisement, product information or content information of the article; The calculation module is used to input the object information of the first type of object into the first type of object feature extraction network of the first feature vector generation model to obtain the initial feature vector of the first type of object; input the object information of the second type of object into the second type of object feature extraction network of the first feature vector generation model to obtain the initial feature vector of the second type of object; and combine the initial feature vectors of the first type of object and the initial feature vectors of the second type of object through the feature synthesis module of the first feature vector generation model to learn the coupling relationship between the first type of object and the second type of object; and feed back the coupling relationship between the first type of object and the second type of object to the first type of object feature extraction network so that the first type of object feature extraction network can adjust the initial feature vector of the first type of object with reference to the coupling relationship between the first type of object and the second type of object and the initial feature vector of the second type of object to obtain the feature vector of the first type of object. The generation module is used to generate target recommended objects corresponding to each second type of object based on the selection of the second type of object by the first type of object. The selection includes the first type of object viewing or purchasing the second type of object by clicking on the terminal. The determination module is used to obtain the feature vector of the target recommendation object, determine the second type of object to be recommended to the target recommendation object based on the feature vector of the target recommendation object and the feature vector of the first type of object, and send the determined second type of object to the terminal corresponding to the target recommendation object for display; The device includes a first training module, which is used to construct a first sample pair and a second sample pair based on the selection of a second sample pair by a first type of sample object. The first sample pair includes sample object information of the first type of sample object and sample object information of the second type of sample object selected by the first type of sample object. The second sample pair includes sample object information of the first type of sample object and sample object information of the second type of sample object not selected by the first type of sample object. The first sample pair is input as a positive sample and the second sample pair is input as a negative sample into a first feature vector generation model. The loss value of the first feature vector generation model is calculated based on the sample object similarity output by the first feature vector generation model for the positive and negative samples. The model parameters of the first feature vector generation model are adjusted based on the loss value of the first feature vector generation model to train the first feature vector generation model.

13. An electronic device, characterized in that, include: One or more processors; A storage device for storing one or more computer programs that, when executed by one or more processors, cause the electronic device to perform the method of any one of claims 1 to 11.

14. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by the processor of the electronic device, causes the electronic device to perform the method described in any one of claims 1 to 11.

15. A computer program product, characterized in that, The computer program product includes a computer program stored in a computer-readable storage medium, wherein a processor of an electronic device reads from and executes the computer program, causing the electronic device to perform the method described in any one of claims 1 to 11.

Images