Recommended model construction method and apparatus, and computer-readable storage medium

Abstract

The application discloses a recommendation model construction method and device and a computer readable storage medium, wherein the method comprises the following steps: acquiring business data through multiple business channels, and processing the business data into data features; processing the data features corresponding to different business channels into data features with the same calculation caliber, and mapping to obtain view features; constructing a business template according to the view features, wherein the business template comprises a model template, an index template and a rule template; and constructing a recommendation model suitable for a target business scene according to a target business template selected in an activity configuration interface. The application aims to reduce the cost of constructing the recommendation model.

Description

Technical Field

[0001] This invention relates to the field of recommendation model construction technology, and in particular to a recommendation model construction method, apparatus, and computer-readable storage medium. Background Technology

[0002] Recommendation models are widely used in e-commerce, media, social networking, tourism, and many other fields. Different industries and scenarios employ different recommendation models, such as collaborative filtering, content-based recommendation, and deep learning-based recommendation. Within these technologies, different scenarios require different recommendation models, and most recommendation models are designed for specific scenarios and problems. This leads to the need for different algorithms and models across different scenarios, requiring the construction of separate recommendation models for different business scenarios, resulting in high development and maintenance costs for recommendation models.

[0003] The above content is only used to help understand the technical solution of the present invention and does not represent an admission that the above content is prior art. Summary of the Invention

[0004] The main objective of this invention is to provide a method, apparatus, and computer-readable storage medium for building recommendation models, aiming to reduce the cost of building recommendation models.

[0005] To achieve the above objectives, the present invention provides a method for constructing a recommendation model, which obtains business data through multiple business channels and processes the business data into data features;

[0006] Data features corresponding to different business channels are processed into data features with the same calculation caliber and mapped to obtain view features;

[0007] A business template is constructed based on the view features, and the business template includes a model template, an indicator template, and a rule template.

[0008] A recommendation model adapted to the target business scenario is built based on the target business template selected in the activity configuration interface.

[0009] Optionally, the step of constructing a recommendation model adapted to the target business scenario based on the target business template selected in the activity configuration interface includes:

[0010] Replace the corresponding data features with the view features used in the target business template;

[0011] The target business templates are combined according to preset combination rules to obtain the recommendation model.

[0012] Optionally, the target service template includes a target back-end rule template and a target recall rule template. After the step of replacing the corresponding data features with the view features used in the target service template, the method further includes:

[0013] The activity configuration interface receives the recall operator selected by the user for the target recall rule template, and the activity configuration interface also receives the back row operator selected by the user for the target back row rule template.

[0014] Configure the target recall rule template according to the recall operator, and configure the target back row rule template according to the back row operator.

[0015] Optionally, before the step of combining the target business templates according to preset combination rules to obtain the recommendation model, the method further includes:

[0016] Determine the selection order of the target rule template after the user selects it;

[0017] The selected order will be used as the execution order of the target back row rule template;

[0018] The execution order is used as one of the preset combination rules to execute the back row rules corresponding to the target back row rule template in a series on the sorting results.

[0019] Optionally, after the step of acquiring business data through multiple business channels and processing the business data into data features, the method further includes:

[0020] Based on the data type corresponding to the data feature, the data feature is stored in the corresponding feature library;

[0021] Based on the business requirements of different business scenarios, determine the business templates and feature libraries involved in the business scenarios;

[0022] Establish a template pool corresponding to the business scenario based on the relevant business templates.

[0023] Optionally, the step of constructing a recommendation model adapted to the target business scenario based on the target business template selected in the activity configuration interface includes:

[0024] When a user-triggered activity creation operation is received, the target business scenario corresponding to the activity creation operation is determined, and the template pool corresponding to the target business scenario is output.

[0025] In response to the user's selection action, the target business template selected by the user in the template pool is determined.

[0026] Optionally, the target business template includes a target model template. After the step of constructing a recommendation model adapted to the target business scenario based on the target business template selected in the activity configuration interface, the method further includes:

[0027] The activity configuration interface receives the field information selected by the user for the model template;

[0028] Training parameters are constructed based on the field information and the model template;

[0029] The recommendation model is trained based on the training parameters and the selected target feature library used for training.

[0030] Optionally, the step of constructing a business template based on the view features includes:

[0031] Define the metric template name, associated engine operator, and description information for each business scenario;

[0032] The business metric name, metric storage type, storage configuration, and metric calculation logic are defined based on the metric template.

[0033] In addition, to achieve the above objectives, the present invention also provides a recommendation model building apparatus, which includes a memory, a processor, and a recommendation model building program stored in the memory and executable on the processor. When the recommendation model building program is executed by the processor, it implements the steps of the recommendation model building method as described above.

[0034] In addition, to achieve the above objectives, the present invention also provides a computer-readable storage medium storing a recommendation model building program, which, when executed by a processor, implements the steps of the recommendation model building method as described above.

[0035] This invention proposes a method, apparatus, and computer-readable storage medium for building a recommendation model. First, business data is acquired through multiple business channels and processed into data features. The data features corresponding to different business channels are then processed into data features with the same computational scope, mapping them to view features. A business template is built based on the view features, including a model template, an indicator template, and a rule template. Finally, a recommendation model adapted to the target business scenario is built based on the target business template selected in the activity configuration interface. By mapping the processed data features from different business channels to view features, business data from different channels can be indexed based on these view features, achieving data universality. Building business templates using unified view features allows for arbitrary combinations of business templates. For recommendation models to be built in different business scenarios, the same business template can be used for identical structures, improving the reusability of business templates. Therefore, during the design phase of the recommendation model, the overhead of data acquisition and model building can be saved, thereby reducing the cost of building the recommendation model. Attached Figure Description

[0036] Figure 1 This is a schematic diagram of the terminal structure of the hardware operating environment involved in the embodiments of the present invention;

[0037] Figure 2 This is a flowchart illustrating an embodiment of the recommended model construction method of the present invention;

[0038] Figure 3 This is a flowchart illustrating another embodiment of the recommended model construction method of the present invention;

[0039] Figure 4 This is a schematic diagram of the recommendation model architecture involved in an embodiment of the present invention;

[0040] Figure 5 This is a schematic diagram illustrating the application of view features in an embodiment of the present invention;

[0041] Figure 6 The recommended model in this embodiment of the invention involves the following steps and processes;

[0042] Figure 7 The sorting process involved in the facility of this invention;

[0043] Figure 8 The downstream processes involved in the facility of this invention;

[0044] Figure 9 This is a data transmission flowchart related to an embodiment of the present invention.

[0045] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0046] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0047] Because of the limitations of related technologies, different algorithms and models are required for different scenarios, and recommendation models need to be built separately for different business scenarios, resulting in high development and maintenance costs for recommendation models.

[0048] To reduce the cost of building recommendation models, this invention proposes a method, apparatus, and computer-readable storage medium for building recommendation models. The main steps of the method include:

[0049] Business data is acquired through multiple business channels and processed into data features.

[0050] Data features corresponding to different business channels are processed into data features with the same calculation caliber and mapped to obtain view features;

[0051] A business template is constructed based on the view features, and the business template includes a model template, an indicator template, and a rule template.

[0052] A recommendation model adapted to the target business scenario is built based on the target business template selected in the activity configuration interface.

[0053] By mapping the processed data features of business data from different business channels into view features, business data from different business channels can be indexed based on these view features, thus achieving data universality. By constructing business templates using unified view features, business templates can be combined arbitrarily. For recommendation models to be built in different business scenarios, the same business template can be used for the same structure, improving the reusability of business templates. Therefore, in the design phase of recommendation models, the overhead of data acquisition and model building can be saved, thereby reducing the cost of building recommendation models.

[0054] The claims of this invention will be described in detail below with reference to the accompanying drawings.

[0055] like Figure 1 As shown, Figure 1 This is a schematic diagram of the terminal structure of the hardware operating environment involved in the embodiments of the present invention.

[0056] In this embodiment of the invention, the terminal can be a recommendation model building device.

[0057] like Figure 1As shown, the terminal may include: a processor 1001, such as a CPU, a memory 1003, and a communication bus 1002. The communication bus 1002 is used to enable communication between these components. The memory 1003 may be high-speed RAM or stable non-volatile memory, such as disk storage. Optionally, the memory 1003 may also be a storage device independent of the aforementioned processor 1001.

[0058] Those skilled in the art will understand that Figure 1 The terminal structure shown does not constitute a limitation on the terminal and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0059] like Figure 1 As shown, the memory 1003, which serves as a computer storage medium, may include an operating system and a recommendation model building program.

[0060] exist Figure 1 In the terminal shown, processor 1001 can be used to call the recommendation model building program stored in memory 1003 and perform the following operations:

[0061] Business data is acquired through multiple business channels and processed into data features.

[0062] Data features corresponding to different business channels are processed into data features with the same calculation caliber and mapped to obtain view features;

[0063] A business template is constructed based on the view features, and the business template includes a model template, an indicator template, and a rule template.

[0064] A recommendation model adapted to the target business scenario is built based on the target business template selected in the activity configuration interface.

[0065] Furthermore, the processor 1001 can call the recommendation model building program stored in the memory 1003 and also perform the following operations:

[0066] Replace the corresponding data features with the view features used in the target business template;

[0067] The target business templates are combined according to preset combination rules to obtain the recommendation model.

[0068] Furthermore, the processor 1001 can call the recommendation model building program stored in the memory 1003 and also perform the following operations:

[0069] The activity configuration interface receives the recall operator selected by the user for the target recall rule template, and the activity configuration interface also receives the back row operator selected by the user for the target back row rule template.

[0070] Configure the target recall rule template according to the recall operator, and configure the target back row rule template according to the back row operator.

[0071] Furthermore, the processor 1001 can call the recommendation model building program stored in the memory 1003 and also perform the following operations:

[0072] Determine the selection order of the target rule template after the user selects it;

[0073] The selected order will be used as the execution order of the target back row rule template;

[0074] The execution order is used as one of the preset combination rules to execute the back row rules corresponding to the target back row rule template in a series on the sorting results.

[0075] Furthermore, the processor 1001 can call the recommendation model building program stored in the memory 1003 and also perform the following operations:

[0076] Based on the data type corresponding to the data feature, the data feature is stored in the corresponding feature library;

[0077] Based on the business requirements of different business scenarios, determine the business templates and feature libraries involved in the business scenarios;

[0078] Establish a template pool corresponding to the business scenario based on the relevant business templates.

[0079] Furthermore, the processor 1001 can call the recommendation model building program stored in the memory 1003 and also perform the following operations:

[0080] When a user-triggered activity creation operation is received, the target business scenario corresponding to the activity creation operation is determined, and the template pool corresponding to the target business scenario is output.

[0081] In response to the user's selection action, the target business template selected by the user in the template pool is determined.

[0082] Furthermore, the processor 1001 can call the recommendation model building program stored in the memory 1003 and also perform the following operations:

[0083] The activity configuration interface receives the field information selected by the user for the model template;

[0084] Training parameters are constructed based on the field information and the model template;

[0085] The recommendation model is trained based on the training parameters and the selected target feature library used for training.

[0086] Furthermore, the processor 1001 can call the recommendation model building program stored in the memory 1003 and also perform the following operations:

[0087] Define the metric template name, associated engine operator, and description information for each business scenario;

[0088] The business metric name, metric storage type, storage configuration, and metric calculation logic are defined based on the metric template.

[0089] The following explanation, through specific exemplary solutions, clarifies the scope of protection claimed in the claims of this invention, so that those skilled in the art can better understand the scope of protection of the claims. It is understood that the following exemplary solutions do not limit the scope of protection of this invention, but are only used to explain this invention.

[0090] For example, refer to Figure 2 In one embodiment of the recommendation model construction method of the present invention, the recommendation model construction method includes the following steps:

[0091] Step S10: Obtain business data through multiple business channels and process the business data into data features;

[0092] Step S20: Process the data features corresponding to different business channels into data features with the same calculation caliber, and map them to obtain view features;

[0093] Step S30: Construct a business template based on the view features. The business template includes a model template, an indicator template, and a rule template.

[0094] Step S40: Build a recommendation model adapted to the target business scenario based on the target business template selected in the activity configuration interface.

[0095] In this embodiment, the recommendation model can recommend information that the recommender is interested in. This information can be item information, text information, image information, etc., and there are no restrictions here.

[0096] Reference Figure 4 The recommendation model building system includes a data management module, a template management module, and an activity management module.

[0097] The data management module manages the data needed to build the recommendation model. Data is the foundation of the entire recommendation process. It is responsible for collecting business data from different business scenarios and channels, processing features according to business needs, and obtaining data features. These features can provide valuable information for various business scenarios and can be used for model training and recommendation result generation. The data management module accesses business data from different channels. Data from channels with the same access rules is considered to be from the same channel. Data from different channels, after processing, can generate features with the same calculation criteria, mapped to a single view feature. (See reference...) Figure 5 By using a view feature, you can index different data features corresponding to different channels.

[0098] In this embodiment, the template management module can build and manage multiple business templates. View features are a management probability of different data features, which can index data features from different data channels. Using view features as the construction language, view features are used instead of data names used in the traditional construction process when building business templates. The constructed model templates can be freely combined. The template management module manages and saves multiple built model templates. Business templates come in various types, including model templates, rule templates, and indicator templates. Each template has its own characteristics and uses, but their basic function is to provide general, reusable components to reduce the development cost and time of recommendation models. At the same time, each template can be configured and adjusted as needed to meet different business requirements, thereby improving the quality and accuracy of personalized recommendation services.

[0099] The Activity Management module is responsible for creating, managing, and running recommendation activities in specific recommendation scenarios. Based on business needs, the module selects an appropriate scenario configuration scheme for instantiation and then builds a recommendation chain to achieve accurate recommendations. Users can create recommendation service businesses within the Activity Management module; each activity constitutes a recommendation service business. The module supports the creation, management, and operation of recommendation activities in specific recommendation scenarios, providing users with personalized and efficient recommendation activity services. When creating a recommendation service business, users need to determine the activity template, scheme, and performance indicators. They then trigger the corresponding activity creation operation through the configuration interface in the Activity Management module. In the activity configuration interface, users select the desired target business template, and a recommendation model adapted to the target business scenario is built based on the selected template.

[0100] Optionally, the step of constructing a recommendation model adapted to the target business scenario based on the target business template selected in the activity configuration interface includes:

[0101] Replace the corresponding data features with the view features used in the target business template;

[0102] The target business templates are combined according to preset combination rules to obtain the recommendation model.

[0103] In this embodiment, since the business template is constructed based on view features, the business module uses view features to describe the data features required by the corresponding steps of the module. Therefore, when constructing a recommendation model based on the business template, it is necessary to restore the view features in the business template to data features. After restoration, the target business template is combined according to the preset combination rules to obtain the recommendation model.

[0104] Optionally, the target service template includes a target back-end rule template and a target recall rule template. After the step of replacing the corresponding data features with the view features used in the target service template, the method further includes:

[0105] The activity configuration interface receives the recall operator selected by the user for the target recall rule template, and the activity configuration interface also receives the back row operator selected by the user for the target back row rule template.

[0106] Configure the target recall rule template according to the recall operator, and configure the target back row rule template according to the back row operator.

[0107] In this embodiment, the obtained recommendation model includes the following steps, referred to Figure 6 :

[0108] Recall: Items meeting the requirements are recalled to build a recall product pool. The business template management module provides various target recall rule templates. The target recall rule templates and their corresponding back-end operators can be configured in the activity configuration interface according to business needs. Recall consists of three parts: recall strategy, filtering rules, and fusion ranking. The candidate set generated by the recall strategy must include the item ID (itemId) and ranking weight (weight). Figure 7 and Figure 8 The images show the recall configuration and the fusion process, respectively.

[0109] Sorting: Select templates from the model templates that match the business metrics and model metrics (selected by default). After the solution is submitted, model training is automatically initiated based on the model templates, and the training results are monitored.

[0110] Back-row: The business template management module also provides various target back-row rule templates, allowing users to perform secondary intervention on the recommendation results, i.e., back-row, according to business needs. Rule types include material protection, material placement, PV weight adjustment, deduplication, fragmentation, grouping, category proportion, category top, etc. Target back-row rule templates and their corresponding back-row operators can be configured in the activity configuration interface according to business needs.

[0111] Optionally, before the step of combining the target business templates according to preset combination rules to obtain the recommendation model, the method further includes:

[0112] Determine the selection order of the target rule template after the user selects it;

[0113] The selected order will be used as the execution order of the target back row rule template;

[0114] The execution order is used as one of the preset combination rules to execute the target back row rule template in series on the sorting results.

[0115] In this embodiment, multiple business rules can be configured within the same activity configuration. Users can configure multiple target back-end rule templates, and the execution order of these templates affects the sorting results differently. Therefore, users need to adjust the priority of the target back-end rule templates. Users can select target back-end rules according to their needs, or they can use the system's default execution rules, i.e., use the selected order as the execution order of the target back-end rule templates.

[0116] Specifically, when selecting a target back-row rule template, a user can select one rule template at a time, meaning there is a selection order for the target back-row rule templates. This selection order serves as the execution order of the target back-row rule templates. When combining templates, the execution order of the target back-row rule templates is used as one of the combination rules for the combined templates, and the corresponding back-row rules of the target back-row rule templates are executed in sequence based on the sorting results.

[0117] Specifically, users can select a target back-ranking rule template from the template pool according to business needs. Based on this template, they can then perform secondary intervention on the ranking results. This secondary intervention may result in a re-ranking of the ranking results, leading to target recommendations that better meet user needs. This allows users to independently select target back-ranking rules, configure them according to their requirements, and then re-rank the results based on these rules. This enables users to expand and configure ranking rules according to actual needs, thereby improving the system's scalability and maintainability. It also provides the ability to monitor and adjust recommendation paths and parameters in real time, allowing for timely optimization and adjustment of recommendation activities, improving recommendation effectiveness and efficiency, and enhancing the versatility of the recommendation system.

[0118] In the technical solution disclosed in this embodiment, business data is first acquired through multiple business channels and processed into data features. The data features corresponding to different business channels are then processed into data features with the same calculation caliber, and mapped to obtain view features. A business template is constructed based on the view features, including a model template, an indicator template, and a rule template. A recommendation model adapted to the target business scenario is constructed based on the target business template selected in the activity configuration interface. By mapping the processed data features of business data from different business channels to view features, business data from different business channels can be indexed based on the view features, thus achieving data universality. Constructing business templates using unified view features allows for arbitrary combinations of business templates. For recommendation models to be constructed in different business scenarios, the same business template can be used for the same structure, improving the reusability of business templates. The introduction of concepts such as feature libraries and view features enables a single feature to be used in multiple business scenarios, avoiding resource waste caused by repeated calculations. This improves efficiency in data processing and enhances system performance.

[0119] Therefore, during the design phase of a recommendation model, the overhead of data acquisition and model building can be saved, thereby reducing the cost of building a recommendation model.

[0120] Optionally, refer to Figure 3 Based on any of the above embodiments, in another embodiment of the recommended model construction method of the present invention, after the step of obtaining business data through multiple business channels and processing the business data into data features, the method further includes:

[0121] Step S50: Store the data features into the corresponding feature library according to the data type corresponding to the data features;

[0122] Step S60: Based on the business requirements of different business scenarios, determine the business templates and feature libraries involved in the business scenarios;

[0123] Step S70: Establish a template pool corresponding to the business scenario based on the relevant business templates.

[0124] In this embodiment, the data management module manages the data needed to build the recommendation model. Data is the foundation of the entire recommendation process. It is responsible for collecting business data from different business scenarios and channels, processing features according to business needs, and obtaining data features. These data features can provide valuable information for various recommendation scenarios and can be used for model training and recommendation result generation. For each business scenario, at least user data, item data, and user behavior data must be accessed. (Refer to...) Figure 9 Data access from different business channels is generally achieved through message queues, with data transmission occurring incrementally or fully. Data flows to various locations, including online feature libraries, background libraries, and data warehouses. After data access, a specific system needs to be established before it can be used downstream. The main uses of this data include attribute filtering during recall, model building, subsequent rule usage, and business metric calculation. Feature libraries can be broadly categorized into two types: User Feature Libraries (including user attribute features, statistical features, asset features, interaction features, and advertising / product response features); and Material Feature Libraries (including material attribute features, statistical features, interaction features, text features, and image features). Based on the data type, data features are stored in the corresponding feature libraries for use in subsequent recall and training processes.

[0125] Reference Figure 4 The recommendation model building system also includes a business scenario management module. This module manages multiple business scenarios, which are abstractions of recommendation business domains, such as content recommendation, financial product recommendation, and advertising recommendation. Each business scenario has independent business meaning, metrics for evaluating business performance, and the templates and data required to implement the business. It is essentially a collection of templates and business data that can be flexibly combined according to the specific business needs of each scenario. After the recommendation template is instantiated, the system runs and manages the recommendation activities. Furthermore, after classifying business scenarios, the system supports preset scenario templates, providing developers with a convenient and quick scenario configuration solution by associating templates with business models.

[0126] Optionally, the step of constructing a recommendation model adapted to the target business scenario based on the target business template selected in the activity configuration interface includes:

[0127] When a user-triggered activity creation operation is received, the target business scenario corresponding to the activity creation operation is determined, and the template pool corresponding to the target business scenario is output.

[0128] In response to the user's selection action, the target business template selected by the user in the template pool is determined.

[0129] Based on multiple business scenarios, the recommendation model building system can provide more convenient and efficient configuration solutions. When receiving a user's activity creation operation, the system determines the target business scenario corresponding to the operation, thereby identifying the template pool within that scenario. This template pool can then be displayed in the activity management module's configuration interface, allowing users to select appropriate model modules, indicator modules, and / or rule templates from it. In response to the user's selection, the system determines the target business template and target feature library chosen by the user from the template pool. The target business module is used to compose the recommendation model.

[0130] Optionally, the target business template includes a target model template. After the step of constructing a recommendation model adapted to the target business scenario based on the target business template selected in the activity configuration interface, the method further includes:

[0131] The activity configuration interface receives the field information selected by the user for the model template;

[0132] Training parameters are constructed based on the field information and the model template;

[0133] The recommendation model is trained based on the training parameters and the selected target feature library used for training.

[0134] In this embodiment, the target business template includes a target model template, which is the basic framework of the recommendation model. The model template provides a templated model for different data features, algorithm models, and tuning parameters. The model template is used to sort the items to be recommended. Selecting a suitable model template as the target model template and training and adapting the model based on it can achieve efficient and accurate recommendations from the system. Specifically, different training field information is provided for different model templates. The field information consists of configuration parameters for debugging the model template, including training parameters during training and the structural parameters of the model itself. After the model template is selected as the target model template, the field information is displayed on the activity configuration interface for the user to select. The activity configuration interface receives the field information selected by the user for the model template and constructs training parameters based on the field information and the model template. The recommendation model is trained based on the training parameters and the selected target feature library used for training.

[0135] Optionally, the step of constructing a business template based on the view features includes:

[0136] Define the metric template name, associated engine operator, and description information for each business scenario;

[0137] The business metric name, metric storage type, storage configuration, and metric calculation logic are defined based on the metric template.

[0138] Each business function has its own set of metrics to focus on, and metric templates are used to configure near real-time metric calculation templates for these functions. Metric templates allow for the reuse of metric calculation logic within the same business scenario. First, the template name, associated engine operators, and description information for each metric need to be defined as its identifier. Second, the metric name, storage type, storage configuration, and calculation logic need to be defined. Since metric templates ultimately need to be implemented in specific business functions, dynamic parameter configuration is required; parameters can come from raw values, view features, environment variables, etc. Finally, the metric calculation job needs to be deployed, and its orchestration logic and specific calculation logic are configured through job settings.

[0139] From the template pool of model templates in the current business scenario, select target model templates whose model metrics match the aforementioned business metrics. These target model templates can sort the information to be recommended, obtain recommendation scores for each piece of information, and then sort them again. The top-ranked information in the sorting results will be used as the target recommended information.

[0140] This embodiment discloses a method for constructing a universal recommendation model applicable to different scenarios. By coordinating four parts—data management, template management, scenario management, and activity management—a highly efficient and flexible recommendation model construction system adaptable to various scenarios is built. This system boasts high reusability and flexible configuration, enabling rapid deployment and expansion, avoiding redundant development and resource waste, reducing development and maintenance costs, and improving recommendation model performance and efficiency, while also enhancing the accuracy and personalized service quality of the recommendation system. An interactive approach is adopted to build new recommendation links, allowing users to quickly and easily configure and adjust recommendation model templates and components to meet different business scenarios and recommendation needs. This provides users with personalized and efficient recommendation services and experiences, improving system scalability and maintainability, providing better support and assurance for recommendation model development and deployment, and reducing deployment and maintenance costs. By flexibly combining and calling different templates and components in different scenarios, diverse and personalized recommendation services and experiences are provided to users, improving the universality and flexibility of the recommendation model and making its construction more convenient.

[0141] Furthermore, this embodiment of the invention also proposes a recommendation model construction apparatus, which includes a memory, a processor, and a recommendation model construction program stored in the memory and executable on the processor. When the recommendation model construction program is executed by the processor, it implements the steps of the recommendation model construction method described in the above embodiments.

[0142] Furthermore, embodiments of the present invention also propose a computer-readable storage medium storing a recommendation model building program, which, when executed by a processor, implements the steps of the recommendation model building method described in the above embodiments.

[0143] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0144] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0145] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause the recommendation model building device to execute the methods described in the various embodiments of the present invention.

[0146] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A method for constructing a recommendation model, characterized in that, The method includes: Business data is acquired through multiple business channels and processed into data features. Data features corresponding to different business channels are processed into data features with the same calculation caliber and mapped to obtain view features; A business template is constructed based on the view features, and the business template includes a model template, an indicator template, and a rule template. When a user-triggered activity creation operation is received, the target business scenario corresponding to the activity creation operation is determined, and the template pool corresponding to the target business scenario is output. Here, the business scenario is a collection of templates and business data. In response to the user's selection action, determine the target business template selected by the user in the template pool; Construct a recommendation model adapted to the target business scenario based on the target business template; The step of constructing a business template based on the view features includes: By using the view features, different data features corresponding to different channels are indexed. For the different data features, algorithm models and optimization parameters, model templates are provided. The model templates are used to sort the items to be recommended. The model template is selected as the target model template, and model training and adaptation are performed based on the target model template.

2. The recommendation model construction method as described in claim 1, characterized in that, The step of constructing a recommendation model adapted to the target business scenario based on the target business template includes: Replace the corresponding data features with the view features used in the target business template; The target business templates are combined according to preset combination rules to obtain the recommendation model.

3. The recommendation model construction method as described in claim 2, characterized in that, The target business template includes a target back-end rule template and a target recall rule template. After the step of replacing the corresponding data features with the view features used in the target business template, the method further includes: The activity configuration interface receives the recall operator selected by the user for the target recall rule template, and the activity configuration interface also receives the back row operator selected by the user for the target back row rule template. Configure the target recall rule template according to the recall operator, and configure the target back row rule template according to the back row operator.

4. The recommendation model construction method as described in claim 2, characterized in that, Before the step of combining the target business templates according to preset combination rules to obtain the recommendation model, the method further includes: Determine the selection order of the rule templates after the user selects the target; The selected order will be used as the execution order of the target back row rule template; The execution order is used as one of the preset combination rules to execute the back row rules corresponding to the target back row rule template in a series on the sorting results.

5. The recommendation model construction method as described in claim 1, characterized in that, After the step of acquiring business data through multiple business channels and processing the business data into data features, the method further includes: Based on the data type corresponding to the data feature, the data feature is stored in the corresponding feature library; Based on the business requirements of different business scenarios, determine the business templates and feature libraries involved in the business scenarios; Establish a template pool corresponding to the business scenario based on the relevant business templates.

6. The recommendation model construction method as described in claim 5, characterized in that, The target business template includes a target model template. After the step of constructing a recommendation model adapted to the target business scenario based on the target business template, the method further includes: The activity configuration interface receives the field information selected by the user for the model template; Training parameters are constructed based on the field information and the model template; The recommendation model is trained based on the training parameters and the selected target feature library used for training.

7. The recommendation model construction method as described in claim 1, characterized in that, The step of constructing a business template based on the view features further includes: Define the metric template name, associated engine operator, and description information for each business scenario; The business metric name, metric storage type, storage configuration, and metric calculation logic are defined based on the metric template.

8. A recommendation model construction apparatus, characterized in that, The recommendation model building apparatus includes: a memory, a processor, and a recommendation model building program stored in the memory and executable on the processor. When the recommendation model building program is executed by the processor, it implements the steps of the recommendation model building method as described in any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a recommendation model building program, which, when executed by a processor, implements the steps of the recommendation model building method as described in any one of claims 1 to 7.

Images