A task execution method, apparatus, device, and computer-readable storage medium

Abstract

This application provides a task execution method, apparatus, device, and computer-readable storage medium. The method includes: acquiring a target industry label for a target industry; extracting features from the target industry label to obtain industry label features; determining a target cluster identifier corresponding to the industry label features based on a preset cluster set, wherein the preset cluster set is a set composed of cluster information corresponding to industry features of multiple industries; determining a target industry feature corresponding to the target cluster identifier from the correspondence between preset features and cluster identifiers, wherein the correspondence between preset features and cluster identifiers is a set composed of the correspondence between each sub-industry feature in the industry features and a cluster identifier in the preset cluster set; and training a preset model corresponding to the target industry using the target industry features, so as to execute a prediction task using the trained preset model. This application embodiment can improve the efficiency of preset model training in the field of artificial intelligence.

Description

Technical Field

[0001] This application relates to information processing technology in the field of artificial intelligence, and more particularly to a task execution method, apparatus, device, and computer-readable storage medium. Background Technology

[0002] With the rapid development of artificial intelligence, network models have been widely used in various industries. Network models can improve the intelligence and efficiency of functional applications across these industries; therefore, training network models is a crucial part of the rapid development of various sectors.

[0003] Generally, training a model for a specific industry involves hard filtering user behavior information from various industries using industry-specific labels to obtain industry-specific user behavior data. This data is then used for feature extraction and clustering to obtain industry-specific features. Finally, a prediction model for that industry is trained using these features. However, this training process involves numerous steps, including filtering user behavior information from various industries and training the model. These steps require manual intervention, resulting in low efficiency in obtaining industry-specific features and consequently, low efficiency in training the prediction model. Summary of the Invention

[0004] This application provides a task execution method, apparatus, device, and computer-readable storage medium that can improve the efficiency of predictive model training.

[0005] The technical solution of this application embodiment is implemented as follows:

[0006] This application provides a task execution method, including:

[0007] Obtain the target industry label for the target industry, where the target industry label is the representation information of the target industry;

[0008] Extract the features of the target industry tags to obtain industry tag features;

[0009] Based on the cluster set, the target cluster identifier corresponding to the industry tag feature is determined. The cluster set is a collection of cluster information corresponding to the industry features of multiple industries. The cluster set is obtained by clustering the behavioral data of the multiple industries based on the corpus data.

[0010] From the correspondence between features and cluster identifiers, the target industry features corresponding to the target cluster identifiers are determined. The correspondence between features and cluster identifiers is a set consisting of the correspondence between each sub-industry feature in the industry features and a cluster identifier in the cluster set.

[0011] The correspondence between the features and the cluster identifiers is obtained by clustering the behavioral data of the multiple industries based on the corpus data.

[0012] Using the characteristics of the target industry, a prediction model corresponding to the target industry is trained, and the trained prediction model is used to perform prediction tasks.

[0013] This application also provides a task execution method, including:

[0014] Obtain the target user's target operation data and the seed operation data corresponding to the information to be recommended in the target industry, wherein the seed operation data is the operation data of the seed user of the information to be recommended;

[0015] Based on the above-trained prediction model, the target similarity between the target operation data and the seed operation data is predicted;

[0016] The target similarity and a similarity threshold are compared to determine the recommendation result for the information to be recommended.

[0017] Based on the recommendation results, recommendation processing is performed on the information to be recommended in order to complete the prediction task.

[0018] This application provides a first task execution device, including:

[0019] The tag acquisition module is used to acquire target industry tags for the target industry, wherein the target industry tags are representational information of the target industry.

[0020] The feature extraction module is used to extract the features of the target industry label to obtain industry label features;

[0021] The identifier determination module is used to determine the target cluster identifier corresponding to the industry tag feature based on the cluster set. The cluster set is a set of cluster information corresponding to the industry features of multiple industries. The cluster set is obtained by clustering the behavioral data of the multiple industries based on the corpus data.

[0022] The feature determination module is used to determine the target industry feature corresponding to the target cluster identifier from the correspondence between features and cluster identifiers. The correspondence between features and cluster identifiers is a set consisting of the correspondence between each sub-industry feature in the industry feature and a cluster identifier in the cluster set. The correspondence between features and cluster identifiers is obtained by clustering the behavioral data of the multiple industries based on the corpus data.

[0023] The model training module is used to train a prediction model corresponding to the target industry using the characteristics of the target industry, so as to perform prediction tasks using the trained prediction model.

[0024] In this embodiment of the application, the model training device further includes a feature extraction module, used to acquire corpus data; classify the corpus data using various industry labels to obtain the correspondence between labels and corpus; interleave the labels and corpus in the correspondence between labels and corpus to obtain negative samples; and train an initial feature extraction model using the negative samples and the correspondence between labels and corpus to obtain the feature extraction model.

[0025] In this embodiment of the application, the feature determination module is further configured to use the feature extraction model to extract the features of the target industry label and obtain the industry label features.

[0026] In this embodiment of the application, the model training device further includes a feature clustering module, used to acquire industry operation data; extract text information from the industry operation data to obtain industry key text; extract features from the industry key text to obtain industry features; and cluster the industry features according to the number of industries corresponding to the multiple industries to obtain the cluster set, and the correspondence between the features and the cluster identifiers.

[0027] In this embodiment of the application, the feature clustering module is further configured to: determine the number of clusters based on the number of industries corresponding to the plurality of industries, wherein the number of clusters is greater than the number of industries; cluster the industry features based on the number of clusters to obtain each cluster, which includes a cluster identifier, a cluster center feature, and a cluster feature; construct the cluster set by combining the cluster identifiers and cluster center features corresponding to each cluster, wherein each cluster information contains a cluster identifier and a cluster center feature; and determine the correspondence between the feature and the cluster identifier based on the cluster identifiers and cluster features corresponding to each cluster.

[0028] In this embodiment of the application, the feature clustering module is further configured to obtain target sub-industry features that match each sub-cluster feature in the cluster features of the current cluster from the industry features, wherein the current cluster is any one of the clusters; and determine the correspondence between the feature and the cluster identifier based on the target sub-industry feature and the current cluster; wherein the correspondence between the feature and the cluster identifier corresponds to each cluster; and the correspondence between the target sub-industry feature and the current cluster is a sub-correspondence in the correspondence between the feature and the cluster identifier.

[0029] In this embodiment, the identifier determination module is further configured to compare the current sub-industry label feature with each cluster center feature in the cluster set to obtain a current distance set, wherein the current sub-industry label feature is any sub-industry label feature in the industry label features; determine the cluster identifier corresponding to the smallest current distance in the current distance set as the current cluster identifier corresponding to the current sub-industry label feature, thereby obtaining each current cluster identifier corresponding to the industry label feature; and obtain the target cluster identifier based on each current cluster identifier.

[0030] In this embodiment of the application, the identifier determination module is further configured to integrate the current cluster identifiers to obtain the sub-target cluster identifiers and the occurrence frequencies corresponding to the sub-target cluster identifiers, wherein the sub-target cluster identifiers and the occurrence frequencies correspond one-to-one; sort the sub-target cluster identifiers according to the occurrence frequencies; select a preset number of sub-target cluster identifiers with the highest occurrence frequencies from the sorted sub-target cluster identifiers; and combine the preset number of sub-target cluster identifiers with the highest occurrence frequencies into the target cluster identifier.

[0031] In this embodiment of the application, the feature determination module is further configured to obtain the current sub-target industry feature corresponding to the current sub-target cluster identifier from the correspondence between the feature and the cluster identifier, thereby obtaining each current sub-target industry feature corresponding to the target cluster identifier, wherein the current sub-target cluster identifier is any sub-target cluster identifier among the target cluster identifiers; and determine each current sub-target industry feature as the target industry feature.

[0032] In this embodiment of the application, the feature extraction module is further configured to: utilize the initial feature extraction model to obtain target label features corresponding to target labels in target samples, wherein the target sample is the negative sample or the correspondence between the label and the corpus; utilize the initial feature extraction model to obtain target corpus features corresponding to target corpus in the target samples; obtain correlation information between the target label features and the target corpus features; and iteratively train the initial feature extraction model based on the difference between the correlation information and the preset annotation data corresponding to the target sample, and when a preset cutoff condition is met, determine the iteratively trained initial feature extraction model as the feature extraction model.

[0033] This application provides a second task execution device, including:

[0034] The information acquisition module is used to acquire the target operation data of the target user and the seed operation data corresponding to the information to be recommended in the target industry. The seed operation data is the operation data of the seed user of the information to be recommended.

[0035] The similarity module is used to predict the target similarity between the target operation data and the seed operation data based on the above-trained prediction model.

[0036] The result determination module is used to compare the target similarity with the similarity threshold to determine the recommendation result of the information to be recommended;

[0037] The task execution module is used to perform recommendation processing on the information to be recommended based on the recommendation results, so as to complete the execution of the prediction task.

[0038] This application provides a first task execution device, including:

[0039] The first memory is used to store executable instructions;

[0040] The first processor, when executing executable instructions stored in the first memory, implements the task execution method applied to the first task execution device.

[0041] This application provides a second task execution device, including:

[0042] The second memory is used to store executable instructions;

[0043] The second processor, when executing executable instructions stored in the second memory, implements the task execution method applied to the second task execution device.

[0044] This application provides a computer-readable storage medium storing executable instructions, which, when executed by a first processor, implement the task execution method provided in this application for a first task execution device; or, when executed by a second processor, implement the task execution method provided in this application for a second task execution device.

[0045] The embodiments of this application have at least the following beneficial effects: Since the target industry features used in the training process of the prediction model are obtained by determining the target cluster identifier from the set of cluster information corresponding to each industry feature through the target industry label, and then filtering the cluster identifier corresponding to each sub-industry feature through the target cluster identifier; that is, by calling the cluster set and filtering the correspondence between features and cluster identifiers, the target industry features of the target industry can be obtained. Therefore, when obtaining the specified industry features, the processing flow is simplified and the degree of manual intervention is reduced; thus, the efficiency of obtaining the target industry features is high, and the training efficiency of the prediction model is high when training the prediction model based on the target industry features. Attached Figure Description

[0046] Figure 1 This is an exemplary flowchart for obtaining specific industry characteristics;

[0047] Figure 2 This is an optional architecture diagram of the task execution system provided in the embodiments of this application;

[0048] Figure 3a This is provided by the embodiments of this application. Figure 2 A schematic diagram of the component structure of a server in the system;

[0049] Figure 3b This is provided by the embodiments of this application. Figure 2 A schematic diagram of the composition structure of another server in the system;

[0050] Figure 4 This is an optional flowchart illustrating the task execution method provided in the embodiments of this application;

[0051] Figure 5 This is another optional flowchart illustrating the task execution method provided in the embodiments of this application;

[0052] Figure 6 This is another optional flowchart illustrating the task execution method provided in the embodiments of this application;

[0053] Figure 7 This is another optional flowchart illustrating the task execution method provided in the embodiments of this application;

[0054] Figure 8 This is yet another optional flowchart illustrating the task execution method provided in the embodiments of this application;

[0055] Figure 9 This is another optional flowchart illustrating the task execution method provided in the embodiments of this application;

[0056] Figure 10This is an exemplary flowchart of obtaining target industry characteristics provided in an embodiment of this application;

[0057] Figure 11 This is another exemplary flowchart of obtaining target industry characteristics provided in the embodiments of this application. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0059] In the implementation of this application, the collection and processing of relevant data should strictly comply with the requirements of relevant laws and regulations, obtain the informed consent or separate consent of the personal information subject, and carry out subsequent data use and processing within the scope of laws and regulations and the authorization of the personal information subject.

[0060] In this application embodiment, data related to user behavior information is involved. When this application embodiment is applied to a specific product or technology, user permission or consent is required, and the collection, use and processing of related data must comply with relevant laws, regulations and standards.

[0061] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0062] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0063] Before providing a further detailed description of the embodiments of this application, the nouns and terms involved in the embodiments of this application will be explained, and the nouns and terms involved in the embodiments of this application shall be interpreted as follows.

[0064] 1) Industry tags: These are textual information used to distinguish different industries; for example, the industry tags for the gaming industry could be "Honor of Kings" and "Peacekeeper Elite"; another example is that the industry tags for the e-commerce industry could be "women's clothing" and "comfortable, fashionable and breathable trendy shoes".

[0065] 2) Industry characteristics: refers to the information used to train the prediction model for that industry, such as the vector representation information of information that users are interested in.

[0066] 3) 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 computing, storage, processing, and sharing of data.

[0067] 4) Artificial Intelligence (AI): refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0068] Generally, training a model for a specific industry involves hard filtering user behavior information from various industries using industry-specific labels to obtain user behavior information for that specific industry. Then, feature extraction and clustering are performed on this user behavior information to obtain industry-specific features. Finally, these industry-specific features are used to train a prediction model for that specific industry.

[0069] For example, see Figure 1 , Figure 1 This is an exemplary flowchart for obtaining specific industry characteristics; such as... Figure 1 As shown, user behavior information 1-1 and specified industry labels 1-2 are obtained from various industries. The user behavior information 1-1 from each industry is then filtered using the specified industry labels 1-2, with the filtering method being hard matching. After filtering, specified industry behavior information 1-3 is obtained. Here, a vector model 1-5 for obtaining vectors of user behavior information is also obtained by acquiring corpus data 1-4. Then, the vector of specified industry behavior information 1-3 is obtained using vector model 1-5, resulting in user interest vector 1-6. Next, to achieve data compression during training, user interest vector 1-6 is clustered, thus obtaining specified industry features 1-7 including user interest vectors and cluster identifiers.

[0070] It should be noted that each time based on Figure 1The illustrated process for obtaining specified industry features involves selecting specified industry behavior information from data sources (user behavior information from various industries) (processing from user behavior information 1-1 to obtaining specified industry behavior information 1-3), followed by a series of model training processes (processing from specified industry behavior information 1-3 to obtaining specified industry features 1-7). This involves numerous steps, and both selection and model training require manual intervention, resulting in significant time consumption and a high degree of human involvement. Consequently, the efficiency of obtaining specified industry features is low, leading to low efficiency in predictive model training. Furthermore, the selection process, which involves obtaining specified industry behavior information, cannot accurately distinguish between different industry behaviors. In other words, the obtained specified industry behaviors often contain information from other industries or irrelevant information. Therefore, when determining specified industry features based on specified industry behavior information and training a predictive model based on these features, the accuracy of the trained predictive model is poor.

[0071] It should also be noted that during the aforementioned screening and model training processes, on the one hand, during screening—that is, hard filtering of user behavior information from various industries using specified industry labels—different specified industry labels correspond to different parameters, and parameter modification requires manual intervention. On the other hand, during model training, the process of obtaining vectors of specified industry behavior information requires manual modification of input and output path parameters because different behavior information corresponds to different vectors. Furthermore, during model training, when clustering user interest vectors, different user interest vectors correspond to different clustering model parameters, thus requiring manual modification of input and output path parameters. Finally, after obtaining the clustering results, manual modification of input and output path parameters is also required to determine the cluster identifier corresponding to each user interest vector. In other words, each stage of obtaining specified industry features requires manual intervention to configure various path parameters.

[0072] Based on this, embodiments of this application provide a task execution method, apparatus, device, and computer-readable storage medium. By calling a cluster set and filtering the correspondence between features and cluster identifiers, the target industry features of the target industry can be obtained. The acquisition efficiency of target industry features is high, and the training efficiency of the prediction model is high when training the prediction model based on the target industry features. Furthermore, when filtering the correspondence between features and cluster identifiers, highly discriminative target clusters are used, resulting in high accuracy of the obtained target industry features. Consequently, the prediction accuracy of the prediction model is high when training the prediction model based on the target industry features.

[0073] The following describes exemplary applications of the task execution device provided in the embodiments of this application. The task execution device provided in the embodiments of this application can be implemented as various types of user terminals such as laptops, tablets, desktop computers, set-top boxes, and mobile devices (e.g., mobile phones, portable music players, personal digital assistants, dedicated messaging devices, portable gaming devices), or as a server. The following will describe exemplary applications when the device is implemented as a server.

[0074] See Figure 2 , Figure 2 This is an optional architecture diagram of the task execution system provided in the embodiments of this application; as shown... Figure 2 As shown, to support a task execution application, in the task execution system 100, a terminal 200 (terminals 200-1 and 200-2 are shown as examples) connects to a server 600 (a second task execution device) via a network 300. The network 300 can be a wide area network (WAN), a local area network (LAN), or a combination of both. Additionally, the task execution system 100 also includes a database 500 and a server 400 (a first task execution device). The database 500 stores a set of clusters and the correspondence between features and cluster identifiers, providing data support for model training to the server 400 through the stored set of clusters and the correspondence between features and cluster identifiers. The database 500 also stores the trained prediction model, providing data support for prediction tasks to the server 600 through the stored trained prediction model.

[0075] Terminal 200-1 is used to receive target operation data corresponding to the target user determined by the user's operation through graphical interface 200-11, and send the target operation data to server 600 through network 300; it is also used to receive recommendation information sent by server 600 through network 300 in response to the target operation data, and display it on graphical interface 200-11.

[0076] Terminal 200-2 is used to receive target operation data corresponding to the target user determined by the user's operation through the graphical interface 200-21, and send the target operation data to the server 600 through the network 300; it is also used to receive the recommendation information sent by the server 600 through the network 300 in response to the target operation data, and display it on the graphical interface 200-21.

[0077] Server 400 is used to obtain target industry labels for the target industry, which are representative information of the target industry; extract features from the target industry labels to obtain industry label features; obtain a set of clusters from database 500, and determine the target cluster identifier corresponding to the industry label features based on the cluster set, which is a set of cluster information corresponding to the industry features of multiple industries; obtain the correspondence between features and cluster identifiers from the database, and determine the target industry features corresponding to the target cluster identifiers from the correspondence between features and cluster identifiers, which is a set of correspondences between each sub-industry feature in the industry features and a cluster identifier in the cluster set; use the target industry features to train a prediction model corresponding to the target industry, and use the trained prediction model to perform prediction tasks.

[0078] Server 600 is used to receive the information to be predicted sent by terminal 200 through network 300, obtain the recommendation result for the target operation data using the trained prediction model obtained from database 500, and send the information to be recommended to terminal 200 through network 300 based on the recommendation result.

[0079] In some embodiments, servers 400 and 600 can be independent physical servers, server clusters or distributed systems composed of multiple physical servers, or cloud servers providing 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 artificial intelligence platforms. Furthermore, the functions corresponding to server 400 and server 600 can be integrated into a single server. Terminal 200 can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, etc., but is not limited to these. The terminal and server can be directly or indirectly connected via wired or wireless communication, which is not limited in this embodiment.

[0080] It should also be noted that the task execution method provided in this application embodiment can be applied to model training in any application scenario. For example, it can be applied to model training in a cloud platform to improve the generalization ability of the model in the cloud platform, thereby improving the quality of cloud services; in this case, cloud technology is required.

[0081] See Figure 3a , Figure 3a This is provided by the embodiments of this application. Figure 2 A schematic diagram of the structure of a server in the system. Figure 3aThe server 400 shown includes at least one first processor 410, a first memory 450, at least one first network interface 420, and a first user interface 430. The various components in the first server 400 are coupled together via a first bus system 440. It is understood that the first bus system 440 is used to implement communication between these components. In addition to a data bus, the first bus system 440 also includes a power bus, a control bus, and a status signal bus. However, for clarity, ... Figure 3a The various buses are all labeled as the first bus system 440.

[0082] The first processor 410 can be an integrated circuit chip with signal processing capabilities, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., wherein the general-purpose processor can be a microprocessor or any conventional processor, etc.

[0083] The first user interface 430 includes one or more first output devices 431 that enable the presentation of media content, including one or more speakers and / or one or more visual displays. The first user interface 430 also includes one or more first input devices 432, including user interface components that facilitate user input, such as a keyboard, mouse, microphone, touch screen display, camera, other input buttons and controls.

[0084] The first memory 450 may be removable, non-removable, or a combination thereof. Exemplary hardware devices include solid-state storage, hard disk drives, optical disk drives, etc. The first memory 450 may optionally include one or more storage devices physically located remote from the first processor 410.

[0085] The first memory 450 may include volatile memory or non-volatile memory, or both. The non-volatile memory may be read-only memory (ROM), and the volatile memory may be random access memory (RAM). The first memory 450 described in this application embodiment is intended to include any suitable type of memory.

[0086] In some embodiments, the first memory 450 is capable of storing data to support various operations, examples of which include programs, modules, and data structures or subsets or supersets thereof, as illustrated below.

[0087] The first operating system 451 includes system programs for handling various basic system services and performing hardware-related tasks, such as the framework layer, core library layer, and driver layer, for implementing various basic business functions and handling hardware-based tasks.

[0088] The first network communication module 452 is used to reach other computing devices via one or more (wired or wireless) first network interfaces 420, exemplary first network interfaces 420 including: Bluetooth, Wi-Fi, and Universal Serial Bus (USB), etc.

[0089] The first presentation module 453 is configured to enable the presentation of information (e.g., a user interface for operating peripheral devices and displaying content and information) via one or more first output devices 431 (e.g., a display screen, a speaker, etc.) associated with the first user interface 430.

[0090] The first input processing module 454 is configured to detect and translate one or more user inputs or interactions from one or more first input devices 432.

[0091] In some embodiments, the first task execution device provided in this application can be implemented in software. Figure 3a A first task execution device 455 stored in a first memory 450 is shown. It may be software in the form of programs and plug-ins, including the following software modules: a tag acquisition module 4551, a feature extraction module 4552, an identifier determination module 4553, a feature determination module 4554, a model training module 4555, a feature extraction module 4556, and a feature clustering module 4557. These modules are logically related and can therefore be arbitrarily combined or further split according to the functions they implement.

[0092] See Figure 3b , Figure 3b This is provided by the embodiments of this application. Figure 2 A schematic diagram of the structural composition of another server in the system. Figure 3b The server 600 shown includes at least one second processor 610, a second memory 650, at least one second network interface 620, and a second user interface 630. The various components in the second server 600 are coupled together via a second bus system 640. It is understood that the second bus system 640 is used to implement communication between these components. In addition to a data bus, the second bus system 640 also includes a power bus, a control bus, and a status signal bus. However, for clarity, in… Figure 3b The various buses are all labeled as the second bus system 640.

[0093] The second processor 610 can be an integrated circuit chip with signal processing capabilities, such as a general-purpose processor, a digital signal processor, or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., wherein the general-purpose processor can be a microprocessor or any conventional processor, etc.

[0094] The second user interface 630 includes one or more second output devices 431 that enable the presentation of media content, including one or more speakers and / or one or more visual displays. The second user interface 630 also includes one or more second input devices 632, including user interface components that facilitate user input, such as a keyboard, mouse, microphone, touch screen display, camera, other input buttons and controls.

[0095] The second memory 650 may be removable, non-removable, or a combination thereof. Exemplary hardware devices include solid-state storage, hard disk drives, optical disk drives, etc. The second memory 650 may optionally include one or more storage devices physically located remote from the second processor 610.

[0096] The second memory 650 may include volatile memory or non-volatile memory, or both. The non-volatile memory may be read-only memory, and the volatile memory may be random access memory. The second memory 650 described in the embodiments of this application is intended to include any suitable type of memory.

[0097] In some embodiments, the second memory 650 is capable of storing data to support various operations, examples of which include programs, modules, and data structures or subsets or supersets thereof, as illustrated below.

[0098] The second operating system 651 includes system programs for handling various basic system services and performing hardware-related tasks, such as the framework layer, core library layer, and driver layer, for implementing various basic business functions and handling hardware-based tasks.

[0099] The second network communication module 652 is used to reach other computing devices via one or more (wired or wireless) second network interfaces 620, exemplary second network interfaces 620 including: Bluetooth, wireless compatibility authentication, and Universal Serial Bus, etc.

[0100] The second presentation module 653 is configured to enable the presentation of information (e.g., a user interface for operating peripheral devices and displaying content and information) via one or more second output devices 631 (e.g., a display screen, a speaker, etc.) associated with the second user interface 630.

[0101] The second input processing module 654 is used to detect and translate one or more user inputs or interactions from one or more second input devices 632.

[0102] In some embodiments, the second task execution device provided in this application can be implemented in software. Figure 3b A second task execution device 655 stored in a second memory 650 is shown. It may be software in the form of programs and plug-ins, including the following software modules: information acquisition module 6551, similarity module 6552, result determination module 6553, and task execution module 6554. These modules are logical and can therefore be arbitrarily combined or further split according to the functions they implement.

[0103] The functions of each module will be explained below.

[0104] In other embodiments, the first task execution device and the second task execution device provided in this application can be implemented in hardware. As an example, the first task execution device and the second task execution device provided in this application can be processors in the form of hardware decoding processors, which are programmed to execute the task execution methods provided in this application. For example, the processor in the form of hardware decoding processors can be one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components.

[0105] The task execution method provided in this application will be described below with reference to exemplary applications and implementations of the server provided in the embodiments of this application.

[0106] See Figure 4 , Figure 4 This is an optional flowchart illustrating a task execution method provided in an embodiment of this application, which will be combined with... Figure 4 The steps shown are explained.

[0107] S401. Obtain the target industry label for the target industry.

[0108] In this embodiment of the application, when optimizing a model in a specified industry or training a model in a specified industry, the server also receives the target industry label of the target industry.

[0109] It should be noted that the target industry refers to a specific industry, such as the gaming industry or the e-commerce industry. The target industry label is the characteristic information of the target industry. For example, when the target industry is the gaming industry, the target industry labels are labels such as "Honor of Kings" and "Peacekeeper Elite". Furthermore, the target industry label includes at least one sub-target industry label. For example, "Honor of Kings" is a sub-target industry label, and "Peacekeeper Elite" is also a sub-target industry label. Each sub-target industry label in the target industry label can be in the form of text, images, or other forms of information. This application embodiment does not specifically limit this.

[0110] S402. Extract the features of the target industry label to obtain the industry label features.

[0111] In this embodiment of the application, after the server obtains the target industry label, in order to use the target industry label to obtain the characteristics of the target industry, the server performs a transformation process on the target industry label to obtain information that can characterize the target industry label and can be used for calculation; here, the server's transformation process on the target industry label is the process of extracting the characteristics of the target industry label, and the extracted characteristics are the industry label characteristics.

[0112] It is easy to see that the industry label features are the features of the target industry label; wherein, the features are information that can be used for calculation, which can be in vector form, matrix form, etc., and this application embodiment does not specifically limit them.

[0113] It should be noted that the process of the server extracting features from the target industry label can be described as the server using a vector model to obtain the vector representation corresponding to the target industry label. This vector model could be a trained DSSM (Deep Structured Semantic Model, also known as a dual-tower model, applied in text similarity matching scenarios), a trained FM (Factorization Machine) model, or other models used to obtain vector representations. Furthermore, when the target industry label includes at least one sub-target industry label, the industry label features correspondingly include at least one sub-industry label feature, and there is a one-to-one correspondence between the at least one sub-target industry label and the at least one sub-target industry label.

[0114] S403. Based on the cluster set, determine the target cluster identifier corresponding to the industry label features.

[0115] In this embodiment, the server stores a cluster set, or the server is able to obtain the cluster set. This cluster set is a collection of cluster information corresponding to industry characteristics of multiple industries. Therefore, the server selects the cluster identifier corresponding to the cluster to which each sub-industry label feature belongs from the cluster set, and then obtains the target cluster identifier corresponding to the industry label feature based on the cluster identifier corresponding to each sub-industry label feature. Here, the cluster identifier represents a cluster, and a cluster may include a cluster center and an identifier, etc. The cluster set is obtained by clustering behavioral data from multiple industries based on corpus data, where the industry characteristics of multiple industries are the characteristics of the behavioral data of multiple industries.

[0116] It should be noted that "industry characteristics" is a collective term for characteristics of multiple industries, including at least one sub-industry characteristic, and each sub-industry characteristic corresponds to an object (user). The target cluster identifier can be a set of all cluster identifiers corresponding to all sub-industry label characteristics, or it can be a set of at least one cluster identifier selected from the set of all cluster identifiers corresponding to all sub-industry label characteristics based on the distinguishing characteristics of the cluster identifiers. This application embodiment does not specifically limit this.

[0117] S404. From the correspondence between features and cluster identifiers, determine the target industry features corresponding to the target cluster identifier.

[0118] In this embodiment of the application, the server stores the correspondence between features and cluster identifiers, or the server is able to obtain the correspondence between features and cluster identifiers; the correspondence between features and cluster identifiers is a set consisting of the correspondence between each sub-industry feature in the industry features and a cluster identifier in the cluster set; therefore, the server uses the target cluster to filter the correspondence between features and cluster identifiers, and the obtained filtering result is the target industry feature corresponding to the target cluster.

[0119] It should be noted that each sub-industry feature in the industry features corresponds to a cluster identifier. This cluster identifier is used to compress the sample data during the training of the prediction model; that is, it determines the samples input to the prediction model based on the cluster identifier. The target industry features are the features of the target industry.

[0120] It should also be noted that the correspondence between features and cluster identifiers is obtained by clustering behavioral data from multiple industries based on corpus data; that is, the correspondence between features and cluster identifiers, as well as the cluster set, are all clustering results obtained by processing behavioral data from multiple industries based on corpus data to obtain industry features of multiple industries, and then clustering the industry features of multiple industries.

[0121] S405. Utilize the characteristics of the target industry to train a prediction model corresponding to the target industry, and then use the trained prediction model to perform prediction tasks.

[0122] In this embodiment of the application, after the server obtains the target industry characteristics, it can use the target industry characteristics to optimize or train the target industry model, that is, use the target industry characteristics to train the prediction model corresponding to the target industry.

[0123] It should be noted that the prediction model can be a training model for the target industry or an optimization model for the target industry; this application does not specifically limit this. Furthermore, the trained prediction model is used to perform prediction tasks, such as determining the relevance between information (or behavioral information) that a user is interested in and a certain advertisement.

[0124] Understandably, the target industry features used in the training process of the prediction model are obtained by determining the target cluster identifier from the set of cluster information corresponding to each industry feature through the target industry label, and then filtering the cluster identifiers corresponding to each sub-industry feature through the target cluster identifier. In other words, by calling the cluster set and filtering the correspondence between features and cluster identifiers, the target industry features can be obtained. Therefore, when obtaining specified industry features, the processing flow is simplified and the degree of manual intervention is reduced. As a result, the efficiency of obtaining target industry features is high, and consequently, the training efficiency of the prediction model is high when training the prediction model based on the target industry features.

[0125] See Figure 5 , Figure 5 This is another optional flowchart illustrating the task execution method provided in the embodiments of this application; as shown below. Figure 5 As shown in the embodiment of this application, S406-S409 are included before S402; that is, before the server extracts the features of the target industry tag and obtains the industry tag features, the task execution method also includes S406-S409. Each step is described below.

[0126] S406. Obtain corpus data.

[0127] It should be noted that the server obtains corpus data by acquiring information related to user actions on the internet. In other words, corpus data consists of text information that users encounter on the internet, such as product descriptions: "summer dresses, 2020 new women's clothing, Korean style slimming chiffon blouses, plus-size puff sleeves, mid-length floral waist-cinching dresses, and garden daisy dresses"; or a collection of keywords related to "Honor of Kings" in an article or a related text: "Most marksmen are countered by Dunshan and Miyamoto Musashi, the skill mechanics are there."

[0128] Here, the server can obtain corpus data through various channels, such as in various functional applications, on various websites, etc.

[0129] S407. Classify the corpus data using each industry tag to obtain the corresponding relationship between the tag and the corpus.

[0130] In the embodiment of the present application, each industry tag, that is, the tag of each industry, is stored in the server, or the server can obtain each industry tag; each tag in each industry tag is used to represent the operation intention of the user on the Internet, such as the object clicked, the information searched, etc. Thus, after the server obtains the corpus data, it classifies the corpus data using each industry tag to obtain the corpus corresponding to each tag in each industry; here, the corpus corresponding to each tag in each industry is the corresponding relationship between the tag and the corpus.

[0131] It should be noted that in the corresponding relationship between the tag and the corpus, since the tag and the corpus are related, the corresponding relationship between the tag and the corpus is a positive sample for training the initial feature extraction model, and the corresponding annotation information is, for example, 1.

[0132] S408. Interleave and combine the tags and corpora in the corresponding relationship between the tag and the corpus to obtain negative samples.

[0133] In the embodiment of the present application, the server interleaves and combines the tags and corpora in the corresponding relationship between the tag and the corpus, and thus obtains negative samples where the tag and the corpus are not related; it is easy to know that the negative samples are also sample pairs of the tag and the corpus, but the tags and corpora of the negative samples are not related. For example, when the tag in the negative sample is "Honor of Kings", the corpus in the negative sample is "Automobile sunshade, snow shield, front windshield sunscreen, heat insulation shade, car sun visor, sun shield, heat insulation board"; the corresponding annotation information of the negative sample is, for example, 0.

[0134] It should be noted that in addition to the methods of obtaining positive samples and negative samples described in S406 - S408, other methods can also be used to obtain positive samples and negative samples, and the embodiment of the present application does not make specific limitations on this.

[0135] S409. Use the negative samples and the corresponding relationship between the tag and the corpus to train the initial feature extraction model to obtain the feature extraction model.

[0136] In the embodiment of the present application, the server can obtain the model to be trained, that is, the initial feature extraction model. Therefore, after obtaining the positive samples and negative samples, it can perform iterative training on the initial feature extraction model using the positive samples and negative samples; when the iterative training is completed, the obtained trained model is the feature extraction model.

[0137] Here, the initial feature extraction model is, for example, DSSM or FM; in addition, when a label is input into the feature extraction model, it will output features representing that input label.

[0138] Accordingly, in the embodiments of this application, S402 can be implemented by S4021, wherein:

[0139] S4021. Using a feature extraction model, extract the features of the target industry label to obtain the industry label features.

[0140] It should be noted that, in addition to using the feature extraction model described above to obtain the features of the target industry label, the server can also use other methods to extract the features of the target industry label, and this application embodiment does not specifically limit this.

[0141] See Figure 6 , Figure 6 This is another optional flowchart illustrating the task execution method provided in the embodiments of this application; as shown Figure 6 As shown in the embodiment of this application, S403 is followed by S4010-S4013; that is, before the server determines the target cluster identifier corresponding to the industry tag feature based on the cluster set, the task execution method further includes S4010-S4013. Each step is described below.

[0142] S4010, Obtain industry operation data.

[0143] It should be noted that when the server obtains user behavior information in various industries, it also obtains behavioral operation data; that is, industry operation data is behavioral data of various industries.

[0144] S4011. Extract text information from industry operation data to obtain key industry text.

[0145] It should be noted that industry operational data includes data in various forms. Here, we will take text data as an example: the server extracts text information from the industry operational data, and the extracted text information is the key industry text.

[0146] S4012. Extract the features of key industry texts to obtain industry features.

[0147] In this embodiment of the application, after the server obtains the key industry text, in order to use the key industry text to obtain the characteristics of each industry, the key industry text is transformed to obtain information that can characterize the key industry text and can be used for calculation; here, the transformation of the key industry text by the server is the process of extracting the characteristics of the key industry text, and the extracted characteristics are the industry characteristics.

[0148] It should be noted that once the server obtains the feature extraction model, S4012 can be implemented through S40121, where:

[0149] S40121. Using a feature extraction model, extract the features of key industry texts to obtain industry features.

[0150] S4013. Based on the number of industries corresponding to multiple industries, cluster the industry features to obtain a set of clusters and the correspondence between features and cluster identifiers.

[0151] In this embodiment of the application, in order to compress the number of industry features and improve the training efficiency of the model, the server clusters the industry features based on the number of industries corresponding to multiple industries. Based on the clustering results, a set of clusters and the correspondence between features and cluster identifiers are obtained.

[0152] In this embodiment of the application, S4013 can be implemented through S40131-S40134; that is, the server clusters industry features based on the number of industries corresponding to multiple industries to obtain a set of clusters and the correspondence between features and cluster identifiers, including S40131-S40134. Each step is described below.

[0153] S40131. Determine the number of clusters based on the number of industries corresponding to multiple industries.

[0154] It should be noted that the number of clusters is greater than or equal to the number of industries corresponding to multiple industries; for example, when the number of industries is 5, the number of clusters can be 15 or 20.

[0155] S40132. Based on the number of clusters, cluster the industry characteristics to obtain each cluster, which includes a cluster identifier, a cluster center feature, and a cluster feature.

[0156] It should be noted that once the server determines the number of clusters, it also determines how many clusters the industry characteristics will be grouped into. Then, based on the number of clusters, the industry characteristics are clustered, resulting in each cluster containing a cluster identifier, a cluster center feature, and cluster features. Furthermore, the cluster identifier represents the cluster, the cluster center feature is the cluster center of the cluster, and the cluster features are the individual sub-industry characteristics belonging to that cluster.

[0157] S40133. Combine the cluster identifiers and cluster center features corresponding to each cluster to form a cluster set.

[0158] In this embodiment, the server combines the cluster identifier and cluster center feature of each cluster into a cluster information, thus obtaining the individual cluster information. The set of these individual cluster information is the cluster set. It is easy to see that each cluster information contains a cluster identifier and a cluster center feature.

[0159] S40134. Based on the cluster identifier and cluster feature corresponding to each cluster, determine the correspondence between the feature and the cluster identifier.

[0160] It should be noted that the cluster features of each cluster include multiple sub-industry features. Here, the server determines the cluster identifier corresponding to each sub-industry feature based on the cluster identifier and cluster features corresponding to each cluster, thus obtaining the correspondence between features and cluster identifiers.

[0161] In the embodiments of this application, S40134 can be implemented through S401341 and S401342; that is, the server determines the correspondence between the feature and the cluster identifier based on the cluster identifier and cluster feature corresponding to each cluster, including S401341 and S401342. The steps are described below.

[0162] S401341. From the industry features, obtain the target sub-industry features that match each sub-cluster feature in the cluster features of the current cluster.

[0163] It should be noted that the current cluster is any cluster among various clusters; each cluster feature belongs to the industry feature; when each sub-industry feature in the industry feature includes user information (such as user identifier) ​​and interest features, and the cluster feature includes at least one sub-cluster feature, each sub-cluster feature is an interest feature; thus, the server can determine a sub-industry feature that matches each sub-cluster feature from the industry feature, that is, the target sub-industry feature.

[0164] S401342. Based on the characteristics of the target sub-industry and the current cluster, determine the correspondence between the characteristics and the cluster identifier.

[0165] It should be noted that the correspondence between features and cluster identifiers corresponds to each individual cluster; the correspondence between target sub-industry features and the current cluster is a sub-correspondence within the correspondence between features and cluster identifiers.

[0166] See Figure 7 , Figure 7 This is another optional flowchart illustrating the task execution method provided in the embodiments of this application; as shown below. Figure 7As shown in the embodiment of this application, S403 can be implemented through S4031-S4033; that is, the server determines the target cluster identifier corresponding to the industry label feature based on the cluster set, including S4031-S4033. The steps are described below.

[0167] S4031. Compare the current sub-industry label features with the cluster center features in the cluster set to obtain the current distance set.

[0168] In this embodiment, the server iterates through each sub-industry label feature in the industry label features. For the currently iterated sub-industry label feature, i.e. the current sub-industry label feature, the server compares the current sub-industry label feature with each cluster center feature in the cluster set, thus obtaining the set of current distances between the current sub-industry label feature and each cluster center feature, i.e. the current distance set.

[0169] Here, the current sub-industry label feature is any sub-industry label feature among the industry label features.

[0170] S4032. Determine the cluster identifier corresponding to the minimum current distance in the current distance set from the cluster set as the current cluster identifier corresponding to the current sub-industry label feature, thereby obtaining each current cluster identifier corresponding to the industry label feature.

[0171] It should be noted that the server obtains the current distance set and selects the minimum current distance from the current distance set; it is easy to know that the cluster identifier corresponding to the minimum current distance is the cluster represented by the current cluster identifier, which is the cluster to which the current sub-industry label feature belongs; here, the server uses the current cluster identifier to indicate the cluster to which the current sub-industry label feature belongs.

[0172] S4033. Based on each current cluster identifier, obtain the target cluster identifier.

[0173] In this embodiment, the server can directly combine each current cluster identifier into a target cluster identifier, or it can select at least one current cluster identifier from each current cluster identifier to combine into a target cluster identifier. This embodiment does not specifically limit this.

[0174] In this embodiment of the application, S4033 can be implemented through S40331-S40334; that is, the server obtains the target cluster identifier based on each current cluster identifier, including S40331-S40334. The steps are described below.

[0175] S40331. Integrate the current cluster identifiers to obtain the identifiers of each sub-target cluster and the frequency of occurrence of each sub-target cluster identifier.

[0176] It should be noted that there are identical cluster identifiers among the current cluster identifiers; here, the server merges at least two identical current cluster identifiers into a sub-target cluster identifier, and at the same time, counts the frequency of occurrence of the sub-target cluster identifier.

[0177] Here, each sub-target cluster identifier corresponds one-to-one with each occurrence frequency.

[0178] S40332. Sort the identifiers of each sub-target cluster according to their frequency of occurrence.

[0179] It should be noted that if a sub-target cluster identifier appears frequently, it indicates that the sub-target cluster identifier has high coverage and distinguishability. Therefore, the server sorts the sub-target cluster identifiers according to their frequency of occurrence, and selects the target cluster identifiers with high coverage and distinguishability from the sorted sub-target cluster identifiers. Here, high coverage and distinguishability mean that the coverage is greater than the coverage threshold and the distinguishability is greater than the distinguishability threshold.

[0180] S40333. Select a preset number of sub-target cluster identifiers with the highest frequency from the sorted sub-target cluster identifiers.

[0181] In this embodiment of the application, the server is pre-set with a preset quantity ratio (e.g., 80%), or the server is able to obtain the preset quantity ratio, which is used to determine the preset quantity of the selected sub-target cluster identifiers.

[0182] S40334. Combine the preset number of sub-target cluster identifiers with the highest frequency of occurrence into a target cluster identifier.

[0183] It should be noted that after the server obtains a preset number of the most frequently occurring sub-target cluster identifiers, it combines the preset number of most frequently occurring sub-target cluster identifiers, and the resulting combination is the target cluster identifier; it is easy to see that the target cluster identifier includes the preset number of sub-target cluster identifiers.

[0184] based on Figure 7 See Figure 8 , Figure 8 This is yet another optional flowchart illustrating the task execution method provided in the embodiments of this application; as shown Figure 8 As shown in the embodiment of this application, S404 can be implemented through S4041 and S4042; that is, the server determines the target industry feature corresponding to the target cluster identifier from the correspondence between features and cluster identifiers, including S4041 and S4042. Each step is described below.

[0185] S4041. From the correspondence between features and cluster identifiers, obtain the current sub-target industry features corresponding to the current sub-target cluster identifier, thereby obtaining each current sub-target industry feature corresponding to the target cluster identifier.

[0186] It should be noted that, since the target cluster identifier includes multiple sub-target cluster identifiers, the server iterates through the target cluster identifiers. For the currently traversed sub-sub-target cluster identifier, i.e., the current sub-target cluster identifier, the server selects at least one sub-industry feature corresponding to the current sub-target cluster identifier from the correspondence between features and cluster identifiers. The selected at least one sub-industry feature corresponding to the current sub-target cluster identifier constitutes the current sub-target industry feature corresponding to the current sub-target cluster identifier. When the traversal of the target cluster identifiers is completed and the current sub-target industry features of each sub-target cluster identifier are obtained, the current sub-target industry features corresponding to the target cluster identifier are also obtained.

[0187] Here, the current sub-target cluster identifier is any sub-target cluster identifier among the target cluster identifiers.

[0188] S4042. Determine the industry characteristics of each current sub-target as the target industry characteristics.

[0189] In this embodiment of the application, after the server obtains the characteristics of each current sub-target industry, it combines the characteristics of each current sub-target industry, and the resulting combination is the target industry characteristic.

[0190] In this embodiment of the application, S409 can be implemented through S4091-S4094; that is, the server uses negative samples and the correspondence between labels and corpus to train an initial feature extraction model to obtain a feature extraction model, including S4091-S4094. Each step is described below.

[0191] S4091. Using the initial feature extraction model, obtain the target label features corresponding to the target label in the target sample.

[0192] It should be noted that when the server uses negative samples and the correspondence between labels and corpus to train the initial feature extraction model, for each sample, i.e. the target sample, the server inputs the label of the target sample, i.e. the target label, into the initial feature extraction model, and the feature of the target label is the target label feature.

[0193] Here, the target sample is either a negative sample or a correspondence between the label and the corpus.

[0194] S4092. Using the initial feature extraction model, obtain the target corpus features corresponding to the target corpus in the target sample.

[0195] Similarly, in this embodiment of the application, the server inputs the target sample's corpus, i.e., the target corpus, into the initial feature extraction model, and the features of the target corpus are the target corpus features.

[0196] S4093. Obtain the correlation information between the target label features and the target corpus features.

[0197] In this embodiment, after the server obtains the target label features and the target corpus features, it calculates the correlation between the target label features and the target corpus features, thus obtaining the correlation information. Here, when both the target label features and the target corpus features are in vector form, the correlation information can be obtained by calculating the inner product between the target label features and the target corpus features, and then using a mapping function (such as the "softmax" function) to transform the inner product result.

[0198] S4094. Based on the difference between the relevance information and the preset labeled data corresponding to the target sample, the initial feature extraction model is iteratively trained. When the preset cutoff condition is met, the initial feature extraction model after iterative training is determined as the feature extraction model.

[0199] In this embodiment, since the target sample corresponds to preset labeled data, such as 1 or 0, after obtaining the relevance information, the server calculates the difference between the relevance information and the preset labeled data using a preset calculation method. This preset calculation method may be a loss function or a set calculation formula. Because the difference between the relevance information and the preset labeled data characterizes the feature extraction performance of the initial feature extraction model to a certain extent, a larger difference indicates a worse feature extraction performance, the model parameters in the initial feature extraction model are iteratively adjusted based on this difference to achieve iterative training of the initial feature extraction model. Here, the iterative training ends when a preset cutoff condition is met during the iterative training process; at this point, the iteratively trained initial feature extraction model is the feature extraction model.

[0200] It should be noted that, from the description of S401 up to this point, the execution entity (server) described is the aforementioned server 400.

[0201] See Figure 9 , Figure 9 This is another optional flowchart illustrating the task execution method provided in the embodiments of this application; as shown Figure 9 As shown in the embodiment of this application, the server uses the characteristics of the target industry to train the prediction model corresponding to the target industry. After the prediction task is executed using the trained prediction model, the task execution method further includes S4014-S4017. Each step is described below.

[0202] S4014. Obtain the target user's target operation data and the seed operation data corresponding to the information to be recommended in the target industry. The seed operation data is the operation data of the seed users of the information to be recommended.

[0203] In this embodiment of the application, when it is determined whether to recommend information (such as advertisements, text and other multimedia information) in the target industry to the target user, the behavioral data of the target user is obtained, thus obtaining the target operation data.

[0204] It should be noted that there are seed users for the information to be recommended. These seed users are users who have performed conversion operations on the information to be recommended, or users who are highly likely to perform conversion operations on the information to be recommended. The server obtains the operation data of the seed users, which is the seed operation data. The server obtains the target operation data and the vertical axis operation data, and determines whether to recommend the information to the target user based on the seed operation data and the target operation data.

[0205] S4015. Based on the trained prediction model, predict the target similarity between the target operation data and the seed operation data.

[0206] In this embodiment, after obtaining the target operation data and seed operation data, the server can directly input the information to be predicted into the trained prediction model to perform prediction, or it can extract features from the target operation data and seed operation data separately, and input the extracted features into the trained prediction model to perform prediction. This embodiment does not specifically limit the specific method used. Here, the result output by the trained prediction model is the target similarity between the target operation data and the seed operation data.

[0207] S4016. Compare the target similarity and the similarity threshold to determine the recommendation result of the information to be recommended.

[0208] It should be noted that the server has a pre-set similarity threshold, or the server can obtain a similarity threshold. This similarity threshold is used to measure whether the target operation data and the seed operation data are similar. For example, when the target similarity is greater than the similarity threshold, it indicates that the target operation data and the seed operation data are similar; when the target similarity is less than or equal to the similarity threshold, it indicates that the target operation data and the seed operation data are dissimilar. It is easy to see that when the target operation data and the seed operation data are similar, the information to be recommended is recommended to the target user, and the target user is more likely to perform an action on the recommended information; conversely, when the target operation data and the seed operation data are dissimilar, the information to be recommended is recommended to the target user, and the target user is less likely to perform an action on the recommended information. Therefore, the server determines whether to recommend the information to the target user based on the comparison result of the target similarity and the similarity threshold.

[0209] S4017. Based on the recommendation results, perform recommendation processing on the information to be recommended in order to complete the prediction task.

[0210] In this embodiment, after obtaining the recommendation result, the server, since the recommendation result represents whether to recommend the information to the target user, sends the information to be recommended as the target user's recommendation information to the terminal based on the recommendation result, so as to display the information to be recommended on the terminal, thereby achieving accurate recommendation to the target user and improving recall; or, the server does not recommend the information to be recommended to the target user. Here, the execution of the prediction task described in S4014-S4017 can be online or offline, and this embodiment does not specifically limit it.

[0211] It should also be noted that the execution entity (server) described in S4014-S4017 is the aforementioned server 600.

[0212] It should be noted that when obtaining the characteristics of a target industry, if the following methods are used... Figure 1 The process shown requires multiple people to work together, and it takes 3-4 days to obtain the target industry characteristics. If the task execution method provided in this application is used to obtain the target industry characteristics, the time to obtain the target industry characteristics is 1 day, which is 80% shorter.

[0213] Understandably, since the cluster set and the correspondence between features and cluster identifiers are obtained through preprocessing, the information of each cluster corresponding to the industry features of multiple industries and the routine of each sub-industry feature corresponding to a cluster identifier are realized; thus, the target industry features of the target industry are obtained quickly.

[0214] The following will describe an exemplary application of the embodiments of this application in a real-world application scenario.

[0215] See Figure 10 , Figure 10 This is an exemplary flowchart illustrating the process of obtaining target industry characteristics, provided in an embodiment of this application; for example... Figure 10 As shown in the diagram, this exemplary process for obtaining target industry characteristics includes the following steps:

[0216] S1001. Obtain corpus data.

[0217] It should be noted that the corpus data consists of text information that users encounter on the Internet.

[0218] S1002. Use corpus data to obtain positive samples (the correspondence between labels and corpus data) and negative samples.

[0219] It should be noted that the format of positive and negative samples is <search term (tags in various industry labels), document (corpus), target field (preset labeled data)>. Here, the search term refers to a user's click or search intent on the internet. The document refers to the corpus corresponding to the search term. In positive samples, the search term is related to the document, meaning the document is the text information in the search results; in negative samples, the search term is not related to the document. In positive samples, the target field is 1, and in negative samples, the target field is 0.

[0220] Here, when obtaining positive and negative samples, the corpus data is classified using pre-collected industry tags for each industry. Tags and their corresponding corpora are extracted from the classification results to construct positive samples. Then, the search terms and documents from two different positive samples are interleaved to obtain negative samples. Alternatively, information that the user did not search for or click on can be used as documents and combined with search terms to construct negative samples.

[0221] S1003. Train a vector model (feature extraction model) using positive and negative samples.

[0222] It should be noted that the search terms (target labels) and documents (target corpus) from the positive or negative samples are input into the DSSM (initial feature extraction model). After multiple nonlinear transformations in the DSSM, the vectors corresponding to the search terms (target label features) and the vectors corresponding to the documents (target corpus features) are obtained respectively. By calculating the inner product of the vectors corresponding to the search terms and the documents, the relevance score of the search terms and documents in the sample (target sample) is obtained. Then, the "softmax" function is used to convert the relevance score into the probabilities of positive and negative samples (relevance information). Finally, the DSSM is iteratively trained based on the difference between the probabilities of the positive and negative samples and the target field. When the training is completed, the vector model is obtained.

[0223] S1004. Use a vector model to obtain user interest vectors (industry features) corresponding to users' positive behavior data (industry operation data).

[0224] It should be noted that positive user behavior data refers to behavioral information that users generate on the Internet, such as clicks, favorites, orders, and searches, which can express user intent. Text information is extracted from positive user behavior data, such as extracting user search terms and clicked article titles. The extracted text information (key industry text) is input into a vector model to obtain the corresponding vector, which is the user interest vector in the format <user identifier, vector>.

[0225] S1005. Cluster the user interest vectors to obtain the correspondence between cluster identifiers and cluster centers (cluster center features) (cluster set), as well as the user interest vectors with cluster identifiers (correspondence between features and cluster identifiers).

[0226] It's important to note that the cluster center is a vector; the clustering model can be any model that produces clustering results, such as the "K-means" clustering model, an unsupervised algorithm that, given a sample set, divides the set into K clusters based on the distance between samples, making points within a cluster as close together as possible while maximizing the distance between clusters. Furthermore, the correspondence between cluster labels and cluster centers includes n pairs of cluster labels and cluster centers, where n is set based on the number of industries. Both K and n are positive integers greater than 0. In addition, other clustering algorithms such as mean-shift clustering can also be used during clustering.

[0227] Here, S1001-S1005 can be encapsulated as an HTTP (HyperText Transfer Protocol) service being invoked. The HTTP request during the invocation carries text information representing the label of the specified industry, and the request result is the corresponding cluster identifier set (see S1007) and vector (see S1006).

[0228] S1006. Use a vector model to obtain the interest vector (target industry feature) of the specified industry (target industry) corresponding to the label data (target industry label).

[0229] S1007. The interest vector of the specified industry is compared one by one with each cluster center in the correspondence between cluster identifier and cluster center to determine the cluster identifier set.

[0230] Here, you specify the industry-specific tag data, such as the given tags "Honor of Kings" and "Peacekeeper Elite".

[0231] S1008. According to the preset rules, the set of cluster identifiers is filtered to obtain the filtered set of cluster identifiers (target cluster identifiers).

[0232] It should be noted that the preset rule can be: filter the 80% of the most frequently occurring cluster identifiers.

[0233] S1009. Using the filtered set of cluster identifiers, filter out the user interest vectors (target industry features) of the specified industry from the user interest vectors with cluster identifiers. This filtering can be achieved using a Spark task.

[0234] See Figure 11 , Figure 11 This is another exemplary flowchart illustrating the process of obtaining target industry characteristics provided in this application embodiment; such as Figure 11 As shown, to obtain the target industry features, the corresponding inputs include user positive behavior 11-1 (industry operation data), corpus 11-2 (corpus data), and industry labels 11-3 (target industry labels). User positive behavior 11-1 consists of positive behavior data generated by users on the internet, such as user clicks and order placements. Corpus 11-2 consists of text information encountered by users on the internet, such as an article read on a WeChat public account or product descriptions encountered by users on e-commerce platforms; corpus 11-2 is used to generate the positive and negative samples required for training vector model 11-4 (feature extraction model). Industry labels 11-3 are textual labels that distinguish a specific industry; for example, the industry labels for the gaming industry could be "Honor of Kings" and "Peacekeeper Elite"; the industry labels for the e-commerce industry could be "women's clothing" and "comfortable, fashionable, and breathable trendy shoes." It is easy to see that the user interest vector (target industry feature) for the specified industry is the output result, and industry labels 11-3 are the labels for the specified industry.

[0235] Here, the process of obtaining user interest vectors for a specific industry using user positive behavior 11-1, corpus 11-2, and industry tags 11-3 includes the following steps: First, training samples are used with corpus 11-2, and vector model 11-4 is trained using the trained samples. Additionally, text information is extracted from user positive behavior 11-1 and input into vector model 11-4 to obtain user interest vectors 11-5 (industry features across multiple industries). Second, user interest vectors 11-5 are input into clustering model 11-6 for clustering, resulting in cluster identifiers and cluster centers 11-7 (cluster sets). Then, the cluster identifiers and cluster centers 11-7 are concatenated to user interest vectors 11-5 through concatenation 11-8, resulting in user interest vectors and cluster identifiers 11-9 (the correspondence between features and cluster identifiers). Next, when obtaining user interest vectors for a specific industry, the industry label 11-3 (target industry label) is input into the vector model 11-4, and combined with the cluster identifier and cluster center 11-7 to obtain label cluster identifiers 11-10 (identifiers of each current cluster). According to rule 11-11 (the rule of selecting the sub-target cluster identifiers corresponding to the most frequently occurring number of occurrences), discriminative label cluster identifiers 11-12 (target cluster identifiers) are selected from label cluster identifiers 11-10. Finally, the discriminative label cluster identifiers 11-12 are used to filter the user interest vector and cluster identifier 11-9 11-13 to obtain the user interest vector 11-14 (target industry features) for the specified industry. Additionally, Figure 11 The process of handling the dashed box in the code can be used as an HTTP service request.

[0236] Understandably, on the one hand, the processed data is positive behavior data of users across the entire industry, not just user positive behavior data from a specific industry. Therefore, the processed data remains consistent regardless of changes in the specific industry, achieving data standardization and eliminating the need for manual modification of various parameters. On the other hand, the resulting set of filtered cluster identifiers possesses high discriminative power and coverage, leading to high accuracy in the filtered user interest vectors for the specific industry, thereby improving the training effect of the prediction model. Furthermore, the manual processing involved in determining the cluster identifier set and filtering out interest vectors for the specific industry is simple, thus improving the overall efficiency of obtaining interest vectors for the specific industry.

[0237] The following description continues to illustrate the exemplary structure of the first task execution device 455 provided in the embodiments of this application as a software module. In some embodiments, such as Figure 3a As shown, the software module stored in the first task execution device 455 of the first memory 450 may include:

[0238] The tag acquisition module 4551 is used to acquire target industry tags for the target industry, wherein the target industry tags are representational information of the target industry.

[0239] Feature extraction module 4552 is used to extract features of the target industry label to obtain industry label features;

[0240] The identifier determination module 4553 is used to determine the target cluster identifier corresponding to the industry tag feature based on the cluster set. The cluster set is a set of cluster information corresponding to the industry features of multiple industries. The cluster set is obtained by clustering the behavioral data of the multiple industries based on the corpus data.

[0241] The feature determination module 4554 is used to determine the target industry feature corresponding to the target cluster identifier from the correspondence between features and cluster identifiers. The correspondence between features and cluster identifiers is a set consisting of the correspondence between each sub-industry feature in the industry features and a cluster identifier in the cluster set. The correspondence between features and cluster identifiers is obtained by clustering the behavioral data of the multiple industries based on the corpus data.

[0242] The model training module 4555 is used to train a prediction model corresponding to the target industry using the characteristics of the target industry, so as to perform prediction tasks using the trained prediction model.

[0243] In this embodiment of the application, the model training device 455 further includes a feature extraction module 4556, used to acquire the corpus data; classify the corpus data using various industry labels to obtain the correspondence between labels and corpus; interleave the labels and corpus in the correspondence between labels and corpus to obtain negative samples; and train an initial feature extraction model using the negative samples and the correspondence between labels and corpus to obtain the feature extraction model.

[0244] In this embodiment of the application, the feature determination module 4554 is further configured to use the feature extraction model to extract the features of the target industry label and obtain the industry label features.

[0245] In this embodiment of the application, the model training device 455 further includes a feature clustering module 4557, which is used to acquire industry operation data, wherein the industry operation data is behavioral data of various industries; extract text information from the industry operation data to obtain industry key text; extract features from the industry key text to obtain industry features; and cluster the industry features according to the number of industries corresponding to the multiple industries to obtain the cluster set and the correspondence between the features and the cluster identifiers.

[0246] In this embodiment of the application, the feature clustering module 4557 is further configured to: determine the number of clusters based on the number of industries corresponding to the plurality of industries, wherein the number of clusters is greater than the number of industries; cluster the industry features based on the number of clusters to obtain each cluster, which includes a cluster identifier, a cluster center feature, and a cluster feature; construct the cluster set by combining the cluster identifiers and cluster center features corresponding to each cluster, wherein each cluster information contains a cluster identifier and a cluster center feature; and determine the correspondence between the feature and the cluster identifier based on the cluster identifiers and cluster features corresponding to each cluster.

[0247] In this embodiment of the application, the feature clustering module 4557 is further configured to obtain, from the industry features, a target sub-industry feature that matches each sub-cluster feature in the cluster features of the current cluster, wherein the current cluster is any one of the clusters; and determine the correspondence between the feature and the cluster identifier based on the target sub-industry feature and the current cluster; wherein the correspondence between the feature and the cluster identifier corresponds to each cluster; and the correspondence between the target sub-industry feature and the current cluster is a sub-correspondence in the correspondence between the feature and the cluster identifier.

[0248] In this embodiment of the application, the identifier determination module 4553 is further configured to compare the current sub-industry label feature with each cluster center feature in the cluster set to obtain a current distance set, wherein the current sub-industry label feature is any sub-industry label feature in the industry label features; determine the cluster identifier corresponding to the smallest current distance in the current distance set as the current cluster identifier corresponding to the current sub-industry label feature, thereby obtaining each current cluster identifier corresponding to the industry label feature; and obtain the target cluster identifier based on each current cluster identifier.

[0249] In this embodiment of the application, the identifier determination module 4553 is further configured to integrate the current cluster identifiers to obtain the sub-target cluster identifiers and the occurrence frequencies corresponding to the sub-target cluster identifiers, wherein the sub-target cluster identifiers and the occurrence frequencies correspond one-to-one; sort the sub-target cluster identifiers according to the occurrence frequencies; select a preset number of sub-target cluster identifiers with the highest occurrence frequencies from the sorted sub-target cluster identifiers; and combine the preset number of sub-target cluster identifiers with the highest occurrence frequencies into the target cluster identifier.

[0250] In this embodiment of the application, the feature determination module 4554 is further configured to obtain the current sub-target industry feature corresponding to the current sub-target cluster identifier from the correspondence between the feature and the cluster identifier, thereby obtaining each current sub-target industry feature corresponding to the target cluster identifier, wherein the current sub-target cluster identifier is any sub-target cluster identifier among the target cluster identifiers; and determine each current sub-target industry feature as the target industry feature.

[0251] In this embodiment of the application, the feature extraction module 4556 is further configured to: use the initial feature extraction model to obtain target label features corresponding to target labels in target samples, wherein the target sample is the negative sample or the correspondence between the label and the corpus; use the initial feature extraction model to obtain target corpus features corresponding to target corpus in the target samples; obtain correlation information between the target label features and the target corpus features; and perform iterative training on the initial feature extraction model based on the difference between the correlation information and the preset annotation data corresponding to the target sample. When a preset cutoff condition is met, the iteratively trained initial feature extraction model is determined as the feature extraction model.

[0252] The following description continues to illustrate the exemplary structure of the second task execution device 655 provided in the embodiments of this application as a software module. In some embodiments, such as Figure 3b As shown, the software module stored in the second task execution device 655 in the second memory 650 may include:

[0253] The information acquisition module 6551 is used to acquire the target operation data of the target user and the seed operation data corresponding to the information to be recommended in the target industry, wherein the seed operation data is the operation data of the seed user of the information to be recommended;

[0254] Similarity module 6552 is used to predict the target similarity between the target operation data and the seed operation data based on the preset model obtained after training by the first task execution device 455.

[0255] The result determination module 6553 is used to compare the target similarity and the similarity threshold to determine the recommendation result of the information to be recommended;

[0256] The task execution module 6554 is used to perform recommendation processing on the information to be recommended based on the recommendation results, so as to complete the execution of the prediction task.

[0257] This application provides a computer program product or computer program that includes computer instructions stored in a computer-readable storage medium.

[0258] This application provides a computer-readable storage medium storing executable instructions. When these executable instructions are executed by a first processor, the first processor will execute the task execution method provided in this application for a first task execution device, for example... Figure 4 The illustrated task execution method; or, when the executable instructions are executed by the second processor, the second processor will execute the task execution method applied to the second task execution device provided in the embodiments of this application, for example, such as... Figure 9 The task execution method is shown.

[0259] In some embodiments, the computer-readable storage medium may be a memory such as FRAM, ROM, PROM, EP ROM, EEPROM, flash memory, magnetic surface memory, optical disk, or CD-ROM; or it may be a variety of devices including one or any combination of the above-mentioned memories.

[0260] In some embodiments, executable instructions may take the form of a program, software, software module, script, or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a standalone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

[0261] As an example, executable instructions may, but do not necessarily, correspond to files in a file system. They may be stored as part of a file that holds other programs or data, for example, in one or more scripts in a Hyper Text Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple collaborating files (e.g., a file that stores one or more modules, subroutines, or code sections).

[0262] As an example, executable instructions can be deployed to execute on a single computing device, or on multiple computing devices located in one location, or on multiple computing devices distributed across multiple locations and interconnected via a communication network.

[0263] In summary, through the embodiments of this application, the target industry features used in the training process of the prediction model are obtained by determining the target cluster identifier from the set of cluster information corresponding to each industry feature through target industry labels, and then filtering the cluster identifiers corresponding to each sub-industry feature using the target cluster identifier. In other words, by calling the cluster set and filtering the correspondence between features and cluster identifiers, the target industry features can be obtained. Therefore, the process of obtaining specified industry features is simplified, and manual intervention is reduced. This results in high efficiency in obtaining target industry features, and consequently, high training efficiency of the prediction model based on these features. Furthermore, since the obtained target industry features have high coverage and high discriminative power, the accuracy of the prediction model training can be improved.

[0264] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, and improvements made within the spirit and scope of this application are included within the scope of protection of this application.

Claims

1. A task execution method, characterized in that, include: Obtain the target industry label for the target industry, where the target industry label is the representation information of the target industry; Extract the features of the target industry tags to obtain industry tag features; Based on the cluster set, the target cluster identifier corresponding to the industry tag feature is determined. The cluster set is a collection of cluster information corresponding to the industry features of multiple industries. The cluster set is obtained by clustering the behavioral data of the multiple industries based on the corpus data. From the correspondence between features and cluster identifiers, obtain the current sub-target industry features corresponding to the current sub-target cluster identifier, obtain each current sub-target industry feature corresponding to the target cluster identifier, wherein the current sub-target cluster identifier is any sub-target cluster identifier in the target cluster identifier, and determine each current sub-target industry feature as the target industry feature; The correspondence between the features and the cluster identifiers is a set consisting of the correspondence between each sub-industry feature in the industry features and a cluster identifier in the cluster set. The sub-industry features and objects have a one-to-one correspondence. The target industry features include multiple corresponding sub-industry features, and the corresponding sub-industry features are the current sub-target industry features. The correspondence between the features and the cluster identifiers is obtained by clustering the behavioral data of the multiple industries based on the corpus data; Using the characteristics of the target industry, a prediction model corresponding to the target industry is trained, and the trained prediction model is used to perform prediction tasks.

2. The method according to claim 1, characterized in that, Before extracting the features of the target industry label to obtain the industry label features, the method further includes: Obtain the corpus data; The corpus data is classified using industry tags to obtain the correspondence between tags and corpus data; The labels and corpora in the correspondence between the labels and corpora are interleaved to obtain negative samples; Using the negative samples and the correspondence between the labels and the corpus, an initial feature extraction model is trained to obtain the feature extraction model; The step of extracting features from the target industry label to obtain industry label features includes: Using the feature extraction model, the features of the target industry label are extracted to obtain the industry label features.

3. The method according to claim 1, characterized in that, Before determining the target cluster identifier corresponding to the industry label feature based on the cluster set, the method further includes: Acquire industry operational data, which is behavioral data from multiple industries; Text information is extracted from the industry operation data to obtain key industry text; Extract the features of the key text in the industry to obtain the industry features; Based on the number of industries corresponding to the multiple industries, the industry features are clustered to obtain the cluster set and the correspondence between the features and the cluster identifiers.

4. The method according to claim 3, characterized in that, The step of clustering the industry features based on the number of industries corresponding to the multiple industries to obtain the cluster set, and the correspondence between the features and cluster identifiers, includes: The number of clusters is determined based on the number of industries corresponding to the multiple industries, and the number of clusters is greater than the number of industries; Based on the number of clusters, the industry characteristics are clustered to obtain individual clusters, each including a cluster identifier, a cluster center feature, and a cluster feature. The cluster set is constructed by combining the cluster identifiers and cluster center features corresponding to each cluster, and each cluster information contains a cluster identifier and a cluster center feature. Based on the cluster identifier and cluster feature corresponding to each of the clusters, the correspondence between the feature and the cluster identifier is determined.

5. The method according to claim 4, characterized in that, The step of determining the correspondence between features and cluster identifiers based on the cluster identifiers and cluster features corresponding to each of the clusters includes: From the industry features, obtain the target sub-industry features that match each sub-cluster feature in the cluster features of the current cluster, where the current cluster is any one of the clusters; Based on the target sub-industry characteristics and the current cluster, determine the correspondence between the characteristics and the cluster identifier; The correspondence between the features and the cluster identifiers corresponds to each of the clusters; the correspondence between the target sub-industry features and the current cluster is a sub-correspondence in the correspondence between the features and the cluster identifiers.

6. The method according to any one of claims 1 to 5, characterized in that, The step of determining the target cluster identifier corresponding to the industry tag feature based on the cluster set includes: The current sub-industry label feature is compared with the cluster center feature in the cluster set to obtain the current distance set, wherein the current sub-industry label feature is any sub-industry label feature in the industry label features; The cluster identifier corresponding to the minimum current distance in the current distance set is determined as the current cluster identifier corresponding to the current sub-industry label feature, thereby obtaining each current cluster identifier corresponding to the industry label feature; Based on the current cluster identifiers, the target cluster identifier is obtained.

7. The method according to claim 6, characterized in that, The step of obtaining the target cluster identifier based on each current cluster identifier includes: The current cluster identifiers are integrated to obtain the sub-target cluster identifiers and the occurrence frequencies corresponding to the sub-target cluster identifiers. The sub-target cluster identifiers and the occurrence frequencies correspond one-to-one. The sub-target cluster identifiers are sorted according to their frequency of occurrence. From the sorted sub-target cluster identifiers, select a preset number of sub-target cluster identifiers that appear most frequently; The preset number of sub-target cluster identifiers with the highest frequency of occurrence are combined to form the target cluster identifier.

8. The method according to claim 2, characterized in that, The step of using the negative samples and the correspondence between the labels and the corpus to train an initial feature extraction model to obtain the feature extraction model includes: Using the initial feature extraction model, target label features corresponding to target labels in target samples are obtained, wherein the target sample is the negative sample or the correspondence between the label and the corpus; Using the initial feature extraction model, the target corpus features corresponding to the target corpus in the target sample are obtained; Obtain the correlation information between the target label features and the target corpus features; Based on the difference between the correlation information and the preset labeled data corresponding to the target sample, the initial feature extraction model is iteratively trained. When the preset cutoff condition is met, the initial feature extraction model after iterative training is determined as the feature extraction model.

9. A task execution method, characterized in that, include: Obtain the target user's target operation data and the seed operation data corresponding to the information to be recommended in the target industry, wherein the seed operation data is the operation data of the seed user of the information to be recommended; Based on the trained prediction model according to any one of claims 1 to 8, predict the target similarity between the target operation data and the seed operation data; The target similarity and a similarity threshold are compared to determine the recommendation result for the information to be recommended. Based on the recommendation results, recommendation processing is performed on the information to be recommended in order to complete the prediction task.

10. A first task execution device, characterized in that, include: The tag acquisition module is used to acquire target industry tags for the target industry, wherein the target industry tags are representational information of the target industry. The feature extraction module is used to extract the features of the target industry label to obtain industry label features; The identifier determination module is used to determine the target cluster identifier corresponding to the industry tag feature based on the cluster set. The cluster set is a set of cluster information corresponding to the industry features of multiple industries. The cluster set is obtained by clustering behavioral data of multiple industries based on corpus data. The feature determination module is used to obtain the current sub-target industry feature corresponding to the current sub-target cluster identifier from the correspondence between features and cluster identifiers, obtain each current sub-target industry feature corresponding to the target cluster identifier, wherein the current sub-target cluster identifier is any sub-target cluster identifier in the target cluster identifiers, and determine each current sub-target industry feature as the target industry feature; The correspondence between the features and the cluster identifiers is a set consisting of the correspondence between each sub-industry feature in the industry features and a cluster identifier in the cluster set. The sub-industry features and objects have a one-to-one correspondence. The target industry features include multiple corresponding sub-industry features, and the corresponding sub-industry features are the current sub-target industry features. The correspondence between the features and the cluster identifiers is obtained by clustering the behavioral data of the multiple industries based on the corpus data; The model training module is used to train a prediction model corresponding to the target industry using the characteristics of the target industry, so as to perform prediction tasks using the trained prediction model.

11. The apparatus according to claim 10, characterized in that, The device further includes: The feature extraction module is further configured to extract features of the target industry labels, and to obtain the corpus data before obtaining the industry label features; to classify the corpus data using each industry label to obtain the correspondence between labels and corpus; to interleave and combine the labels and corpus in the correspondence between labels and corpus to obtain negative samples; and to train an initial feature extraction model using the negative samples and the correspondence between labels and corpus to obtain the feature extraction model. The feature determination module is also used to extract the features of the target industry label using the feature extraction model, thereby obtaining the industry label features.

12. The apparatus according to claim 10, characterized in that, The device further includes: a feature clustering module; The feature clustering module is used to obtain industry operation data before determining the target cluster identifier corresponding to the industry label feature based on the cluster set. The industry operation data is the behavioral data of various industries. Text information is extracted from the industry operation data to obtain key industry text. Extract the features of the key text in the industry to obtain the industry features; Based on the number of industries corresponding to the multiple industries, the industry features are clustered to obtain the cluster set and the correspondence between the features and the cluster identifiers.

13. The apparatus according to claim 10, characterized in that, The device further includes: The identifier determination module is further configured to compare the current sub-industry label feature with the cluster center feature of each cluster in the cluster set to obtain the current distance set, wherein the current sub-industry label feature is any sub-industry label feature in the industry label features; The cluster identifier corresponding to the minimum current distance in the current distance set is determined as the current cluster identifier corresponding to the current sub-industry label feature, thereby obtaining each current cluster identifier corresponding to the industry label feature; Based on the current cluster identifiers, the target cluster identifier is obtained.

14. A second task execution device, characterized in that, include: The information acquisition module is used to acquire the target operation data of the target user and the seed operation data corresponding to the information to be recommended in the target industry. The seed operation data is the operation data of the seed user of the information to be recommended. A similarity module is used to predict the target similarity between the target operation data and the seed operation data based on the trained prediction model according to any one of claims 1 to 8. The result determination module is used to compare the target similarity with the similarity threshold to determine the recommendation result of the information to be recommended; The task execution module is used to perform recommendation processing on the information to be recommended based on the recommendation results, so as to complete the execution of the prediction task.

15. A first task execution device, characterized in that, include: The first memory is used to store executable instructions; A first processor, when executing executable instructions stored in the first memory, implements the method according to any one of claims 1 to 8.

16. A second task execution device, characterized in that, include: The second memory is used to store executable instructions; The second processor, when executing executable instructions stored in the second memory, implements the method of claim 9.

17. A computer-readable storage medium, characterized in that, The device stores executable instructions for implementing the method of any one of claims 1 to 8 when executed by a first processor; or for implementing the method of claim 9 when executed by a second processor.

18. A computer program product comprising computer-executable instructions or a computer program, characterized in that, When the computer-executable instructions or computer program are executed by a processor, they implement the method according to any one of claims 1 to 9.

Images