User circle selection rule generation method, device, storage medium and program product

Abstract

The application discloses a user circle selection rule generation method and device, a storage medium and a program product. The application can analyze a received circle selection rule acquisition request to obtain a circle selection intention. The circle selection intention comprises user description information and business scenario information. A target reasoning strategy is determined based on the circle selection intention, and a target crowd portrait corresponding to the circle selection rule acquisition request is reasoned based on the target reasoning strategy. Label matching is performed according to the target crowd portrait to obtain a target label set. A user circle selection rule is generated according to the target label set. Therefore, the application can reduce manual intervention and improve the accuracy of a user circle selection result.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to user selection rule generation methods, devices, storage media, and program products. Background Technology

[0002] In user profiling and operational scenarios, to conduct marketing campaigns and membership operations targeting specific groups, the selection of target users is typically carried out through manual review of tag documents or keyword-based tag retrieval. Alternatively, it can be done by setting certain filtering rules and selecting users based on these rules and selection requirements. However, when users describe their selection needs in natural language, existing technologies struggle to accurately understand the semantics of complex business scenarios, and the retrieval process relies on manual judgment, resulting in low efficiency. Consequently, inaccurate or missing tag selections are prone to occur, affecting the accuracy of user selection results and business effectiveness. Summary of the Invention

[0003] In view of the above problems, this application provides a user selection rule generation method, device, storage medium and program product, which can obtain the user's selection intent by parsing the user's natural language and adaptively determine the target reasoning strategy to retrieve selection tags in the tag knowledge base, thereby improving the accuracy of understanding the selection needs described in natural language and the retrieval efficiency, reducing the degree of manual intervention, and thus improving the accuracy of user selection results and business effectiveness.

[0004] To address the aforementioned technical problems, this application provides a method for generating user selection rules, comprising:

[0005] The received request to obtain the selection rules is parsed to obtain the selection intent; the selection intent includes user description information and business scenario information.

[0006] Based on the selection intent, a target inference strategy is determined, and based on the target inference strategy, the selection rules are inferred to obtain the target population profile corresponding to the request.

[0007] Based on the target audience profile, tag matching is performed to obtain a target tag set;

[0008] User selection rules are generated based on the target tag set.

[0009] Optionally, parsing the received selection rule acquisition request to obtain the selection intent includes:

[0010] Extract the selection rules to obtain the business line information in the request;

[0011] If the business line information is empty, a prompt message will be output to instruct the user to select a business line;

[0012] If the business line information is not empty, the request is obtained by parsing the selection rules based on the session history information to extract user descriptions and business scenarios, and obtain the extraction results;

[0013] If the user description information and business scenario information in the extracted results are both empty, output a requirement supplement prompt message;

[0014] If the user description information in the extraction result is not empty and the business scenario information is empty, output supplementary business scenario prompts.

[0015] The information in the extracted results that is not empty, including at least one of the user description information and business scenario information, is identified as the selection intent.

[0016] Optionally, the step of determining the target inference strategy based on the selection intent, and inferring the target audience profile corresponding to the selection rule based on the target inference strategy, includes:

[0017] If the user description information in the selected intent is not empty and the business scenario information is empty, then the target reasoning strategy is determined to be the first reasoning strategy, and the target audience profile is obtained based on the user description information.

[0018] If the user description information in the selected intent is empty and the business scenario information is not empty, then the target reasoning strategy is determined to be the second reasoning strategy, and the target audience profile is obtained based on the business scenario information.

[0019] If both the user description information and the business scenario information in the selected intent are not empty, then the target reasoning strategy is determined to be the third reasoning strategy, and the target audience profile is obtained by jointly reasoning based on the user description information and the business scenario information.

[0020] Optionally, if the user description information in the selected intent is empty and the business scenario information is not empty, then the target inference strategy is determined to be the second inference strategy, and a target audience profile is obtained based on the business scenario information, including:

[0021] The business scenario information is input into a large language model, so that the large language model can infer supplementary user description information based on the business scenario information, and generate a target audience profile based on the supplementary user description information and the business scenario information.

[0022] Optionally, if both the user description information and the business scenario information in the selected intent are not empty, then the target inference strategy is determined to be the third inference strategy, and the target audience profile is obtained by jointly inferring based on the user description information and the business scenario information, including:

[0023] Based on the user description information in the selected intent, a first group profile is determined;

[0024] The business scenario information is input into the large language model so that the large language model can infer supplementary user description information based on the business scenario information.

[0025] Based on the supplemented user description information, the user description information in the selected intent, and the business scenario information, a second user profile is determined.

[0026] The first and second group profiles are integrated to obtain the target group profile.

[0027] Optionally, the step of inputting the business scenario information into a large language model so that the large language model can infer supplementary user description information based on the business scenario information includes:

[0028] Based on the business scenario information, retrieve historical selection cases related to the business scenario information from the historical selection knowledge base;

[0029] The business scenario information and the historical selected cases are input into the large language model, so that the large language model can infer the business scenario information based on the historical selected cases and output supplementary user description information.

[0030] Optionally, the step of inputting the business scenario information into a large language model so that the large language model can infer supplementary user description information based on the business scenario information includes:

[0031] The business scenario information is input into a large language model, which performs semantic analysis on the business scenario information, determines the target user group characteristics corresponding to the business scenario information, and generates supplementary user description information based on the target user group characteristics.

[0032] Optionally, the target tag set includes user feature tags; the tag knowledge base for tag matching includes a user feature tag knowledge base; the step of performing tag matching based on the target audience profile to obtain the target tag set includes:

[0033] Obtain the semantic text of the portrait used to characterize the target population profile;

[0034] Identify the target entity in the semantic text of the portrait, and perform supplementary retrieval based on the target entity to obtain supplementary semantic information related to the target entity;

[0035] The supplementary semantic information is fused with the image semantic text to obtain enhanced semantic text;

[0036] Named entity recognition is performed on the enhanced semantic text to obtain an entity list;

[0037] Based on the entity list, a search is performed in the user feature tag knowledge base corresponding to the selection intent to obtain user feature tags, and the user feature tags are added to the target tag set.

[0038] Optionally, the selected tags include a target tag set; the tag knowledge base for tag matching includes a behavioral feature tag knowledge base; the step of performing tag matching based on the target audience profile to obtain the target tag set includes:

[0039] Based on the target audience profile, a search is performed in the behavioral feature tag knowledge base corresponding to the selected intent to obtain candidate behavioral feature tags;

[0040] The large language model is invoked to determine the correlation score between the candidate behavioral feature labels and the target population profile;

[0041] Based on the relevance score, candidate behavioral feature labels with relevance scores lower than a preset relevance threshold are filtered to obtain behavioral feature labels, and the behavioral feature labels are added to the target label set.

[0042] Optionally, the selection tags include user feature tags and behavioral feature tags; the step of generating user selection rules based on the target tag set includes:

[0043] The target audience profile and the user feature tags and behavioral feature tags in the target tag set are input into the large language model, so that the large language model performs combined reasoning on the user feature tags and behavioral feature tags based on the target audience profile to obtain candidate user feature rules and candidate behavioral feature rules.

[0044] Obtain the tag metadata under the business line corresponding to the selection rule;

[0045] Based on the tag metadata, the candidate user feature rules and the candidate behavior feature rules are converted into front-end component configuration data;

[0046] The configuration data of the front-end components is validated, and the front-end component configuration data that passes the validation is determined as the user selection rule.

[0047] The user selection rules are output to the front-end configuration component.

[0048] This application also provides a user selection rule generation device, including:

[0049] The parsing module is used to parse the received request for obtaining selection rules to obtain the selection intent; the selection intent includes user description information and business scenario information.

[0050] The reasoning module is used to determine the target reasoning strategy based on the selection intent, and to infer the target audience profile corresponding to the selection rule acquisition request based on the target reasoning strategy.

[0051] The tag matching module is used to perform tag matching based on the target audience profile to obtain a target tag set;

[0052] The rule generation module is used to generate user selection rules based on the target tag set.

[0053] This application also provides an electronic device, including:

[0054] Memory, used to store computer programs;

[0055] A processor is used to implement the user selection rule generation method described above when executing the computer program.

[0056] This application also provides a computer program product, including a computer program or instructions, which, when executed by a processor, implement the user selection rule generation method described above.

[0057] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when loaded and executed by a processor, implement the user selection rule generation method described above.

[0058] By employing the aforementioned technical solution, this application obtains selection intent by parsing selection rules and acquiring requests, thus revealing selection intent containing user description information and business scenario information. Based on this selection intent, a target inference strategy is determined, and a target audience profile is formed. This transforms natural language requirements with varying degrees of completeness and expression into audience features suitable for tag matching. Furthermore, matching the target audience profile to obtain a set of target tags improves the completeness and accuracy of tag retrieval. Generating user selection rules based on the target tag set reduces manual tag searching, filtering, and combination operations, lowering reliance on human experience. Therefore, this application enhances the understanding of complex business semantics and improves the efficiency, accuracy, and intelligence of user selection rule generation. Attached Figure Description

[0059] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and the originals and elements are not necessarily drawn to scale.

[0060] Figure 1 This application provides a schematic flowchart of a user selection rule generation method.

[0061] Figure 2 A schematic diagram of another user selection rule generation method provided in this application embodiment;

[0062] Figure 3 This application provides a schematic diagram of a user selection rule generation interface.

[0063] Figure 4 A schematic diagram of a user selection rule generation device provided in an embodiment of this application;

[0064] Figure 5 This is a schematic diagram of an electronic device structure provided in an embodiment of this application. Detailed Implementation

[0065] The embodiments of this application are described below with reference to the accompanying drawings. The terminology used in the implementation section of this application is for explaining specific embodiments only and is not intended to limit the scope of this application.

[0066] The embodiments of this application will now be described with reference to the accompanying drawings. Those skilled in the art will recognize that, with technological advancements and the emergence of new scenarios, the technical solutions provided in the embodiments of this application are equally applicable to similar technical problems.

[0067] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such terms are interchangeable where appropriate; this is merely a way of distinguishing objects with the same attributes in the embodiments of this application. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, so that a process, method, system, product, or apparatus that comprises a series of elements is not necessarily limited to those elements, but may include other elements not explicitly listed or inherent to those processes, methods, products, or apparatuses.

[0068] In existing technologies, user selection rules are typically generated using the following methods. The first method is manual tag selection, where business personnel manually search for and judge applicable tags by browsing tag knowledge base documents and select tag combinations based on experience. For example, they might set conditions such as female gender, age at least 40, and middle-class user status. This process often relies on the experience of product managers or data analysts. The second method is rule engine-based user selection, which uses pre-defined, fixed user selection rule templates (e.g., users who have been active in the last 30 days and have accumulated spending of at least 100 yuan). Users select existing rules and set parameters through an interface. This method is only suitable for highly standardized user selection scenarios and lacks flexibility. The third method is traditional search and recommendation, which involves retrieving tags based on keywords (e.g., entering "active" returns all tags containing the word "active") or recommending popular tags based on tag usage frequency. However, the search results often have a low degree of relevance to the business scenario.

[0069] However, in existing technologies, business personnel need to manually sift through thousands of tags, which can take hours and is prone to missing key tags. Furthermore, traditional keyword retrieval cannot understand the deep semantics of business scenarios. For example, when a user description is suitable for a membership price increase experiment, the system cannot identify the inherent connection between the description and tags such as "high-value user" or "high historical spending." Additionally, existing solutions only match based on tag names, but different business lines within the same company (such as Group A, Interactive Entertainment Business Line A1, Platform and Content Business Line A2, and Cloud and Smart Industry Business Line A3) may have duplicate tag names or similar meanings but different tag identifiers, making it easy to miss or misuse tags during retrieval. All of these situations require users to repeatedly consult staff or review documents, affecting the efficiency and accuracy of the user's selection process.

[0070] To address the aforementioned issues, this application provides a method for generating user selection rules. The method described in this embodiment is as follows. This application can be applied to user profiling and audience selection scenarios, and is particularly suitable for tag management and intelligent selection scenarios across multiple business lines. The following section uses a user profiling platform system as an example to introduce several application scenarios implemented in products.

[0071] First, we will introduce the application scenarios of this application. This application can be applied, but is not limited to, to applications or cloud services provided by cloud-side servers that have user profile analysis, audience tag management, and intelligent selection functions, such as intelligent audience segmentation systems deployed on profile middleware platforms. This system can integrate multi-source data to build a tag knowledge base and realize audience segmentation intent parsing, target inference strategy determination, target audience profile inference, tag matching, and user segmentation rule generation through natural language interaction. It is suitable for various business scenarios such as membership operation, advertising, and strategy experimentation. These will be introduced in detail below.

[0072] First, the system architecture of this application embodiment is introduced. The system may include terminals and servers, and the server can provide the methods provided in this application embodiment to one or more terminals. In a specific implementation, the system adopts a layered architecture design and is deployed in a cloud server environment, including a front-end user interaction layer, a back-end application service layer, a knowledge base and vector retrieval layer, and an external service layer. Among them, the front-end user interaction layer can be deployed on a web server (such as Nginx) to host the front-end single-page application and provide interface proxy; the back-end application service layer can be built on a workflow platform (such as Dify) and deployed in a microservice cluster to realize artificial intelligence (AI) workflow orchestration, user permission management, session management, and large language model scheduling; the knowledge base and vector retrieval layer can be built on a vector retrieval engine (such as RagFlow) to complete document preprocessing, knowledge base management, and vector retrieval services; the external service layer may include a large language model inference cluster, Internet search interface, and database system, etc., to provide model inference capabilities, external information supplementation, and data storage capabilities. The layers can be connected at high speed through an intranet to achieve high availability and scalability.

[0073] The terminal may have user profiling analysis or user segmentation related applications or web applications installed. These applications or web pages can provide an interactive interface, allowing the terminal to receive parameters such as natural language segmentation requirements, business line selection information, and user identification information input by the user on this interface, and then send these parameters to the server. The server can execute the user segmentation rule generation method of this embodiment based on the received parameters to obtain the corresponding user segmentation rules, and then return the user segmentation rules to the terminal. In some optional implementations, the terminal may also directly integrate some or all of the processing logic to complete the segmentation rule generation process locally without server involvement; this application does not limit this approach.

[0074] The following describes the product form of the terminal. The terminal in this application embodiment can be a mobile phone, tablet computer, wearable device, in-vehicle device, augmented reality device, virtual reality device, laptop computer, super mobile personal computer, netbook, or personal digital assistant, etc. The terminal may include components such as a radio frequency unit, memory, input unit, display unit, processor, and external interfaces. The input unit can be used to receive natural language text or operation instructions input by the user, such as receiving the user's selection request via a touchscreen, keyboard, or other input device; the display unit can be used to display the interactive interface and the results of the selection rule generation, such as displaying a natural language dialogue interface, target audience profile inference results, target tag set, and user selection rule configuration interface; the memory is used to store program code related to the user selection rule generation method; the processor is used to execute the above program code to implement the method flow of this application; the radio frequency unit or communication interface is used for data interaction with the server.

[0075] At the system architecture level, the server side can further include an offline parsing module, an AI workflow module, and a user interaction support module. The offline parsing module collects and cleans user tags, behavioral tags, and business selection experience data from multi-source data, and performs standardized processing and vectorized modeling to build a high-quality tag knowledge base. The AI ​​workflow module adopts a multi-layered intelligent inference architecture to process natural language input requests from users, including business line verification, information preprocessing, selection intent recognition, target inference strategy determination, target audience profile inference, tag matching based on target audience profiles, and rule generation based on target tag sets, thereby transforming natural language requests into structured selection rules. The user interaction layer provides front-end interface interaction capabilities, enabling user input and system output through natural language input boxes, rule configuration components, and other methods.

[0076] In terms of front-end and back-end communication, the terminal and server can interact using streaming communication technology based on Server-Sent Events (SSE). Specifically, the terminal can initiate a request to the server through a preset communication interface, carrying parameters such as the user's input natural language selection requirements, business line selection information, and user identification information, while setting itself to streaming response mode. During processing, the server can continuously push intermediate and final results to the terminal. The terminal parses and displays the returned data by listening to different event types, such as progressively displaying the generated text content and the final selection rule results. This streaming communication method enables a simultaneous computation and display effect, reducing response latency and improving user experience. For example, the SSE communication process includes: the frontend initiates a POST request to the backend workflow interface through a preset communication interface (e.g., using the fetchEventSource library) to establish an SSE connection; while establishing the connection, the request carries parameters such as the user's input natural language query content, business line selection information, and user identification information, and sets the response mode to streaming mode to support streaming data return; after the connection is established, the backend continuously pushes data in the processing process to the frontend based on the SSE connection, and the frontend parses and displays the received data by listening to different types of events. Among them, the message event is used to receive text fragments generated by the backend in real time and append them step by step to form a typewriter-style display effect. The workflow_finished event is used to receive the final complete output when the workflow processing is completed. During communication, the connection can also be managed through control components (such as AbortController) to support manual interruption of the connection and automatic shutdown in case of abnormal situations. In addition, the data returned by the backend can contain preset formatted markup information. For example, the selection operation button data can be encapsulated by the HTML tag data-type="button", and the structured rule data and readable result information can be encapsulated by data-type="report-data" and data-type="result-read" respectively, so that the frontend can further parse and render it.

[0077] The memory can be used to store software code related to the user selection rule generation method; the processor can execute the steps of the user selection rule generation method and can also schedule other units to achieve corresponding functions; the radio frequency unit or communication interface is used to send data to the server and receive the processing results returned by the server; the power supply is used to power the various components; and the external interface can be used to connect other devices or perform data interaction. Those skilled in the art will understand that the above terminal structure is merely an example and does not constitute a limitation on this application.

[0078] The server includes a bus, a processor, a communication interface, and a memory. The processor, memory, and communication interface are connected via the bus. The memory stores program code related to the user selection rule generation method. The processor executes the program code to implement the method flow provided in this application and can schedule each module to complete functions such as parsing the selection rule acquisition request, determining the selection intent, determining the target reasoning strategy, reasoning the target audience profile, matching the target tag set, and generating user selection rules.

[0079] Reference Figure 1 , Figure 1 This application provides a flowchart illustrating a method for generating user selection rules, as shown in the embodiments below. Figure 1 As shown in the figure, the user selection rule generation method provided in this application embodiment may include steps S101 to S103, which are described in detail below.

[0080] S101. Parse the received request for obtaining the selection rules to obtain the selection intent;

[0081] In this embodiment, the selection rule acquisition request refers to the request information input by the user through an interactive interface to obtain the selection rules of the target user. The request information is usually expressed in natural language text form to express the user's selection needs. When parsing the selection rule acquisition request, the parsing process mainly involves semantic analysis of the natural language text to identify the descriptive content that represents the characteristics of the target user group and the descriptive content that represents the purpose, business goal, or application task of the selection. Among them, user description information (also known as "people") is used to describe the target user group's attributes, interests, identity categories, spending power, activity level, or other group characteristics, such as high-spending users, recently active young users, etc.; business scenario information (also known as "context") is used to describe the specific business purpose, application scenario, or task requirements served by the user's selection rules, such as the Double Eleven promotion or new product launch promotion.

[0082] S102. Determine the target inference strategy based on the selection intent, and infer the selection rules based on the target inference strategy to obtain the target population profile corresponding to the request;

[0083] In this embodiment, the target audience profile is used to characterize the target user group for which the selection rule request is directed, and may include information such as user attributes, identity characteristics, interests, consumption characteristics, and behavioral tendencies. The target inference strategy includes a first inference strategy, a second inference strategy, and a third inference strategy. The first inference strategy is used to infer the target audience profile based on the user description information when the user description information is not empty and the business scenario information is empty; the second inference strategy is used to infer supplementary user description information based on the business scenario information when the user description information is empty and the business scenario information is not empty, and then generate the target audience profile; the third inference strategy is used to perform inference based on both the user description information and the business scenario information, respectively, and integrate the inference results to obtain the target audience profile when both the user description information and the business scenario information are not empty.

[0084] S103. Perform tag matching based on the target audience profile to obtain the target tag set.

[0085] In this embodiment, semantic analysis can be performed on the target audience profile to identify features that are relevant to selection, and matching tags can be found based on the identified features. The matched tags can also be deduplicated, filtered, or merged to reduce the impact of duplicate or irrelevant tags on the selection results. After tag matching, tags that reflect the target audience profile and can be used to construct selection conditions are determined as target tags, and one or more target tags constitute a target tag set.

[0086] S104. Generate user selection rules based on the target tag set.

[0087] In this embodiment, the target tag set may include one or more target tags that match the target audience profile. The server processes the target tags in the target tag set to generate user selection rules. Specific processing operations may include deduplication, filtering, association, and rule-based processing to form rule results that directly represent the target user selection criteria. Deduplication refers to eliminating duplicate tags with the same semantics or identifiers; filtering refers to removing tags with low relevance to the target audience profile or unsuitable for the current selection needs; association refers to establishing combination, logical, or constraint relationships between different target tags; and rule-based processing refers to organizing the integrated target tags into executable or displayable user selection rules. User selection rules are rule results used to limit the scope of target users and can be formed by combining one or more target tags according to logical relationships. They are used for audience screening, marketing campaigns, membership operations, or other business applications. User selection rules are data objects that can be directly executed by downstream systems. Specifically, they can be a list of tag IDs, a logical combination of tags (e.g., tag A and tag B), or an interpretable rule description text.

[0088] By employing the aforementioned technical solution, this application obtains selection intent by parsing selection rules and acquiring requests, thus revealing selection intent containing user description information and business scenario information. Based on this selection intent, a target inference strategy is determined, and a target audience profile is formed. This transforms natural language requirements with varying degrees of completeness and expression into audience features suitable for tag matching. Furthermore, matching the target audience profile to obtain a set of target tags improves the completeness and accuracy of tag retrieval. Generating user selection rules based on the target tag set reduces manual tag searching, filtering, and combination operations, lowering reliance on human experience. Therefore, this application enhances the understanding of complex business semantics and improves the efficiency, accuracy, and intelligence of user selection rule generation.

[0089] To better understand the overall process and various feasible implementation methods, please refer to [link / reference]. Figure 2 , Figure 2 A flowchart of another user selection rule generation method provided in this application embodiment.

[0090] As a feasible implementation method, step S101 parses the received selection rule acquisition request to obtain the selection intent, including: extracting business line information from the selection rule acquisition request; if the business line is empty, outputting prompt information to instruct the user to select a business line; if the business line is not empty, parsing the selection rule acquisition request based on the session history information to extract the user description and business scenario to obtain the extraction result; if both the user description information and the business scenario information in the extraction result are empty, outputting supplementary requirement prompt information; if both the user description information and the business scenario information in the extraction result are not empty, outputting supplementary business scenario prompt information; and determining at least one of the non-empty information in the user description information and the business scenario information in the extraction result as the selection intent.

[0091] A selection rule retrieval request is a request submitted by the user through the front-end interface to obtain the user's selection rules. This request is typically expressed in natural language text and may also include business line information selected through the interface. Business line information is a data identifier representing the business domain to which the current selection rule retrieval request belongs, such as Group A, Business Line A1, Business Line A2, or Business Line A3. For example, assuming the product lines include an e-commerce user product line, a content recommendation product line, and a financial risk control product line, the business line information can be extracted by parsing the fields carried in the selection rule retrieval request, reading preset variables in the session context, or identifying the business line name specified in the user's input. If the user directly specifies in the request text that they want to select users within the e-commerce product line, the extracted business line information will be the e-commerce user product line; if the user pre-selects a product line through a product drop-down menu provided by the interface, the business line information will be read from the corresponding interface state variables.

[0092] If the extracted business line information is empty, meaning the user has not specified or selected any tag product line, a prompt message is output to instruct the user to select a business line. This prompt message can be a piece of natural language text, such as "Please select the tag product line in the lower left corner," and is displayed to the user through the interactive interface. Simultaneously, the subsequent steps of extracting user description information, extracting business scenario information, inferring target audience profiles, tag matching, and generating user selection rules are terminated. The reason for intercepting the request to obtain selection rules when the business line is empty in this step is that business line information typically affects the inference constraints of the target audience profile, the tag range corresponding to subsequent tag matching, and the tag semantic system. It is important to understand that in existing technologies, different business lines may have the same tag name but different meanings, or different tag identifiers but similar semantics. If the subsequent process continues without determining the business line, it may lead to inconsistencies between the inferred target audience profile and the actual business domain, or mismatches between the matched target tags and the real business environment, resulting in distorted selection results.

[0093] If the extracted business line information is not empty, the system further obtains requests based on the session history information parsing selection rules to extract user description information and business scenario information, thus obtaining the extraction results. Here, session history information refers to historical input content, historical system response content, and contextual information already determined in previous rounds associated with the current user session, such as supplementary explanations previously entered by the user or query results previously output by the system. The extraction results refer to the structured information results obtained after parsing, which include at least one or more of user description information and business scenario information. In practical applications, the system first updates the session memory variable, stores the current request content in the session history, and then calls a large language model (such as DeepSeek-V3) to extract keywords from the user input, thereby identifying keywords representing user characteristics as user description information and keywords representing business scenarios as business scenario information. For example, in a scenario where a user has already entered "link artist A's fans to push drama A's playlist" in the first round, "artist A's fans" can be extracted as user description information, and "push drama A's playlist" can be extracted as business scenario information. In a multi-round dialogue scenario, if the user has already entered "want to convert membership renewals" in the previous round, and the current round only enters "find college student users," the system can combine the conversation history information to extract college student users as user description information, and supplement the membership renewal conversions from previous rounds as business scenario information, thus obtaining a complete extraction result.

[0094] If the extracted user description and business scenario information are both empty, meaning the large language model failed to identify any demographic or business scenario descriptions from the user input, the current request is determined to be a non-targeting request (e.g., the user enters casual conversation such as "Hello" or "How's the weather today?", or unclear statements like "Help me check this out"). In this case, the system outputs a request supplement prompt, guiding the user to provide a specific description of the target audience and business scenario. This prompt can be expressed as "Please provide the demographic characteristics you wish to target and the applicable business scenario," and then ends the current process, waiting for the user to re-enter valid targeting rules to retrieve the request.

[0095] If the extracted user description information is not empty but the business scenario information is empty, it is determined that the user has provided a description of the target audience but has not specified the business scenario. Especially when the current dialogue is the first round (i.e., there is no previously entered scenario information in the conversation history), the system further determines that the information is insufficient and outputs a business scenario supplement prompt to guide the user to supplement the business scenario. For example, if the user only enters "help me find college student users" in the first round of dialogue, although the system can identify college student users as user description information, it cannot determine whether this audience is for membership renewal, advertising, or content push. At this time, a prompt message such as "Please supplement the applicable business scenario for this selection operation (such as membership renewal conversion, campus playlist push, advertising, etc.)" is output so that the user can further input a specific business scenario description. It should be noted that in this embodiment, outputting this prompt message does not terminate the process, but continues to execute the subsequent selection intent determination steps, that is, continues to determine part of the selection intent based on the existing non-empty user description information, while waiting for the user to supplement the business scenario information later, thus ensuring the user-friendliness of the interaction.

[0096] After completing the above judgments and prompts, the system will determine the selection intent as the information in the extracted results where at least one of the user description information and business scenario information is not empty. Specifically, if the user description information is not empty but the business scenario information is empty, the selection intent will only include the user description information; if the user description information is empty but the business scenario information is not empty, the selection intent will only include the business scenario information; if both are not empty, the selection intent will include both the user description information and the business scenario information.

[0097] As a feasible implementation method, step S102 determines the target inference strategy based on the selection intent, and infers the target audience profile corresponding to the selection rule acquisition request based on the target inference strategy, including: if the user description information in the selection intent is not empty and the business scenario information is empty, then the target inference strategy is determined to be the first inference strategy, and the target audience profile is obtained based on the user description information; if the user description information in the selection intent is empty and the business scenario information is not empty, then the target inference strategy is determined to be the second inference strategy, and the target audience profile is obtained based on the business scenario information; if both the user description information and the business scenario information in the selection intent are not empty, then the target inference strategy is determined to be the third inference strategy, and the target audience profile is obtained by jointly inferring based on the user description information and the business scenario information.

[0098] Based on the presence or absence of user description information and business scenario information in the selection intent, the target inference strategy used in the current selection request is determined, and a user profile is generated based on the corresponding target inference strategy. Specifically, the selection intent is obtained by parsing in step S101. If the user description information in the selection intent is not empty and the business scenario information is empty, the target inference strategy is determined to be the first inference strategy, and a target user profile is inferred based on the user description information. The first inference strategy is used to perform semantic understanding, feature extraction, and standardized expression of the target user group characteristics provided by the user, converting the user description information into profile information that can represent the target user group. Specifically, the current business line information and user description information can be used together as retrieval input, input to the underlying inference module, to identify user characteristics such as identity, attributes, interests, preferences, or behavioral tendencies contained in the user description information, and to generate a target user profile based on the identified user characteristics. For example, when a user enters "follow artist B," the system parses "artist B fan" as user description information, but not specific business scenario information. In this case, the first inference strategy can be directly applied to perform semantic analysis on "artist B fan" to infer a target audience profile representing the user group's preference for or interest in artist B-related content. The first inference strategy mainly utilizes the user-provided audience description to generate the audience profile, has a shorter processing path, and is suitable for situations where the user description information is relatively clear but no specific business scenario is provided.

[0099] When the user description information in the selected intent is empty but the business scenario information is not empty, the target inference strategy is determined to be the second inference strategy, and a target user profile is obtained based on the business scenario information. The second inference strategy is used to analyze the business objectives, application objects, and business constraints represented by the business scenario information to infer the characteristics of the target user group that is compatible with the business scenario. However, since the current selected intent does not directly provide user description information, the corresponding user description information needs to be generated based on the business scenario information before performing the inference of the target user profile.

[0100] As a feasible implementation method, if the user description information in the selected intent is empty and the business scenario information is not empty, then the target inference strategy is determined to be the second inference strategy, and the target audience profile is obtained based on the business scenario information. This includes: inputting the business scenario information into the large language model so that the large language model can infer the supplementary user description information based on the business scenario information, and generating the target audience profile based on the supplementary user description information and the business scenario information.

[0101] Specifically, business scenario information can be input into the semantic reasoning module, such as a large language model. This allows the large language model to perform semantic analysis based on the business scenario to determine user group characteristics suitable for that scenario, thus obtaining supplementary user description information. After obtaining the supplementary user description information, it is correlated with the business scenario information to generate a target audience profile representing the target user group. The target audience profile can include at least one of the following: user identity, user attributes, membership status, consumption characteristics, activity level, and behavioral tendencies. For example, when a user inputs "membership renewal conversion," the system only identifies the membership renewal conversion business scenario information but fails to identify the user description information. In this case, according to the second reasoning strategy, user description information related to the membership renewal conversion business scenario can be inferred first, such as user descriptions of expiring memberships, historical paying users, or highly active members. This information is then combined with the "membership renewal conversion" business scenario information to generate a target audience profile representing a user group whose membership is about to expire, has a history of paying, or has a high level of activity, such as "Green Diamond membership expires ≤ 30 days," "Paying users within the past year," or "High member activity level."

[0102] As another feasible implementation, inferring target audience profiles based on business scenario information can include: obtaining supplementary user description information from a predefined mapping rule library based on business scenario information; and generating target audience profiles based on the supplementary user description information and business scenario information. Specifically, a scenario-audience association mapping rule library can be pre-built. The mapping rule library can store multiple business scenarios and corresponding user description information in key-value pairs. For example, the business scenario information of "summer promotion" is mapped to user description information such as "users aged 18-35 who are interested in summer cooling products and have recent browsing history," and the business scenario information of "membership renewal conversion" is mapped to user description information such as "users whose membership expires 30 days ago, whose historical payment amount is greater than 100 yuan, and who have logged in in the past 7 days." When a selection intent containing only business scenario information (such as "new product launch promotion") is received, the system extracts keywords from the business scenario information and performs precise or fuzzy matching in the mapping rule library to obtain supplementary user description information. If multiple candidate supplementary user description information are matched, the most matching one can be selected based on scenario similarity. Subsequently, the selected supplementary user description information is fused with the original business scenario information to obtain the target audience profile corresponding to the current selection rule request. This eliminates the need for real-time inference using large language models and reliance on historical case libraries, instead enabling rapid generation of supplementary user description information through lightweight rule matching. This approach boasts advantages such as simplicity, low computational overhead, and fast response time.

[0103] As a feasible implementation method, if both the user description information and the business scenario information in the selected intent are not empty, then the target inference strategy is determined to be the third inference strategy, and a target audience profile is obtained by jointly inferring based on the user description information and the business scenario information. This includes: determining a first audience profile based on the user description information in the selected intent; inputting the business scenario information into a large language model so that the large language model can infer supplementary user description information based on the business scenario information; determining a second audience profile based on the supplementary user description information, the user description information in the selected intent, and the business scenario information; and integrating the first audience profile and the second audience profile to obtain the target audience profile.

[0104] The third inference strategy is configured to consider both user description information and business scenario information when inferring target audience profiles. On the one hand, it determines the first audience profile based on the user description information provided by the user. On the other hand, it infers supplementary user description information based on business scenario information. It then combines the supplementary user description information, the original user description information, and the business scenario information to determine the second audience profile. Finally, it integrates the first audience profile and the second audience profile to obtain the target audience profile.

[0105] In its implementation, the system executes two retrieval paths in parallel or sequentially: Path 1 takes user description information (people) as input, performs semantic parsing on the user description information, extracts the included demographic attributes, interests, identity features, or other user characteristics, and generates a first user profile; Path 2 takes business scenario information (context) as input, performs semantic analysis on the business scenario through a large language model, infers supplementary user description information that matches the current business scenario, and combines the supplementary user description information, the original user description information, and the business scenario information to generate a second user profile. Subsequently, the first and second user profiles are fused, removing duplicate features and retaining effective features that can jointly represent the target user group, resulting in the final target user profile.

[0106] For example, when a user inputs "convert fans of artist B into members," the system can describe the user based on "artist B's fans" and determine the first user profile, such as users who follow artist B's content or have a preference for artist B. Simultaneously, it identifies "membership conversion" as a business scenario, analyzes this scenario using a large language model, and infers supplementary user descriptions, such as highly active users, users with payment potential, or historical members. This, combined with information like "artist B's fans" and "membership conversion," determines the second user profile. Finally, by merging the first and second user profiles, a target user profile is obtained that includes both interest and preference characteristics and membership conversion-related features.

[0107] As another feasible implementation method, the target audience profile is obtained by reasoning based on user description information and business scenario information, including: weighting and combining user description information and business scenario information to generate a weighted semantic vector; performing semantic reasoning based on the weighted semantic vector to obtain a first profile feature; using user description information as an independent input for feature parsing to obtain a second profile feature; and fusing the first profile feature and the second profile feature to obtain the target audience profile.

[0108] Specifically, different weight values ​​can be pre-assigned to user description information and business scenario information. For example, in scenarios where the business scenario has strong tag constraints (such as holiday event placement, special content push, or specific marketing conversion scenarios), the weight of business scenario information is increased, and vice versa. The semantic vector generated by weighted combination can comprehensively reflect the difference in importance between the two. At the same time, feature parsing is performed based on user description information to obtain second profile features. If there is duplicate or conflicting information between the first and second profile features, deduplication and filtering are performed, and the processed profile features are combined to obtain the target audience profile. For example, if a user inputs "select high-spending users for the luxury shopping festival event," user features such as high spending power and high purchase frequency can be extracted based on the user description information "high-spending users." At the same time, combined with the business scenario information "luxury shopping festival event," user features such as interest in high-end brands and a tendency to consume luxury goods can be obtained through further inference. After fusing the above features, a target audience profile containing spending power and consumption interest features is generated. In this way, without relying on additional historical cases, the completeness and business adaptability of the target audience profile can be improved through comprehensive reasoning based on user description information and business scenario information, while reducing the computational complexity of the reasoning process.

[0109] When the user description information in the selection intent is empty but the business scenario information is not empty, there are two ways to supplement and generate the corresponding user description information based on the business scenario information. As a feasible implementation method, the business scenario information is input into a large language model, so that the large language model can infer the supplementary user description information based on the business scenario information. This includes: retrieving historical selection cases related to the business scenario information from the historical selection knowledge base based on the business scenario information; and inputting the business scenario information and historical selection cases into the large language model, so that the large language model can infer the business scenario information based on the historical selection cases and output the supplementary user description information.

[0110] In cases where users do not directly provide user description information, this application embodiment primarily generates supplementary user description information by combining historical selection experience data to semantically enhance and reverse-engineer the business scenario. Specifically, the historical selection knowledge base stores past successful selection rule generation cases. Each case includes at least a business scenario description and a corresponding audience characteristic description, such as the campaign scenario name and suggested audience characteristic descriptions (i.e., historical user description information). In actual processing, similarity retrieval can be performed from the historical selection knowledge base based on keywords, scenario categories, or semantic vectors in the current business scenario information to recall one or more historical selection cases related to the current business scenario information. Similarity can be used to characterize the semantic relevance between the current business scenario information and the business scenario descriptions in historical selection cases. For example, in a membership renewal conversion scenario, the corresponding historical cases may contain user group characteristic descriptions such as expiring membership users, historical paying users, or highly active users.

[0111] After obtaining historical selected cases, the current business scenario information and the retrieved historical selected cases are input into the large language model. This allows the large language model to infer the current business scenario information based on the historical selected cases and output supplementary user description information. Specifically, the large language model can compare and analyze the current business scenario information with the scenario descriptions and user characteristic descriptions in the historical selected cases to determine the fit between the user characteristics in the historical selected cases and the current business scenario, and extract user characteristics applicable to the current scenario. User characteristics that are irrelevant to the current business scenario, conflict with it, or are clearly inapplicable can be excluded from the supplementary user description information. For example, when the business scenario is a membership discount activity targeting students, the historical selected cases may contain user characteristics such as being between 18 and 24 years old, having the status of a university student, and having paid behavior within the past year. Based on these historical selected cases, the large language model can infer the supplementary user description information of "users aged between 18 and 24 years old, having the status of a university student, and having paid behavior within the past year."

[0112] After obtaining the supplementary user description information, the supplementary user description information, business scenario information, and current business line information can be associated or integrated to generate a target audience profile corresponding to the selection rule acquisition request.

[0113] As another feasible implementation method, business scenario information is input into a large language model so that the large language model can infer supplementary user description information based on the business scenario information. This includes: inputting business scenario information into a large language model so that the large language model can perform semantic analysis on the business scenario information, determine the target user group characteristics corresponding to the business scenario information, and generate supplementary user description information based on the target user group characteristics.

[0114] When users do not directly provide user description information, the parsed business scenario information is passed as input parameters to the large language model. Upon receiving the business scenario information, the large language model performs semantic analysis and logical deduction to understand the elements described in the business scenario information, such as the name, purpose, time characteristics, or target audience preferences of the business activity. Based on the identified elements, it determines the characteristics of the target user group corresponding to the business scenario information. Target user group characteristics may include user identity, basic attributes, interests, consumption characteristics, membership status, activity level, or behavioral tendencies. The large language model further summarizes and organizes the determined target user group characteristics to generate supplementary user description information. For example, if the business scenario information provided by the user is a promotion of a new summer refreshing beverage, the large language model, through autonomous reasoning, concludes that the core objective of this scenario is to attract consumers who want to try the new beverage. Therefore, suitable user characteristics may include interest in summer cooling, interest in new beverages, age between 18 and 35, and a preference for trying new things, thereby generating corresponding supplementary user description information. For example, when the business scenario is to push new song recommendations to increase user activity, the large language model can infer user descriptions such as frequent music listeners, recently active users, or users sensitive to new content based on its understanding of the goal of increasing activity. When the business scenario is to improve membership conversion rates, the large language model can infer user descriptions such as highly active non-member users with historical payment behavior or users with high usage frequency but no membership. Through the above reasoning process, user descriptions that meet business objectives can be generated even in the absence of input user description information.

[0115] After obtaining the supplementary user description information, it can be associated or integrated with business scenario information to generate a target audience profile corresponding to the selection rule request. Specifically, the supplementary user description information is used to characterize the audience characteristics inferred from the business scenario information, while the business scenario information is used to characterize the business objectives and application conditions to which the audience characteristics are applicable. Associating the two allows the generated target audience profile to simultaneously reflect the characteristics of the target user group and its corresponding business needs.

[0116] Understandably, in this implementation, the large language model can determine one or more target user group features based on information from the same business scenario. For target user group features with identical or similar semantics, they can be merged or deduplicated; for conflicting target user group features, the retention result can be determined based on business objectives, scenario constraints, or the inference confidence of the corresponding features; and for target user group features with low relevance to business scenario information, they can be filtered. After the above processing, the retained target user group features are organized into supplementary user description information to improve the semantic consistency and business adaptability between the supplementary user description information and the business scenario information.

[0117] Specifically, in the retrieval chain based on user feature tags, as a feasible implementation method, the target tag set includes user feature tags; the tag knowledge base used for tag matching includes a user feature tag knowledge base; tag matching is performed according to the target audience profile to obtain the target tag set, including: obtaining the profile semantic text used to represent the target audience profile; determining the target entity in the profile semantic text, and performing supplementary retrieval based on the target entity to obtain supplementary semantic information related to the target entity; fusing the supplementary semantic information with the profile semantic text to obtain enhanced semantic text; performing named entity recognition on the enhanced semantic text to obtain an entity list; and based on the entity list, searching in the user feature tag knowledge base corresponding to the selection intent to obtain user feature tags, and adding the user feature tags to the target tag set.

[0118] Specifically, when performing tag matching based on the target audience profile, and the target tag set includes user feature tags, semantic text representing the target audience profile can be obtained. The target audience profile can be in natural language text form or in a structured form including multiple profile fields. When the target audience profile is in a structured form, the profile fields and their corresponding field values ​​can be concatenated or serialized according to a preset field order to obtain the semantic text. The semantic text is used to centrally represent information such as the target user group's identity attributes, interests, behavioral tendencies, and business applicability characteristics. For example, if the user description information is a fan of singer AA, and the business scenario information is a promotional campaign for the new song "NEW SONGA," then the corresponding semantic text could be users who follow singer AA and related content and have potential interest in the promotional content of the new song "NEW SONGA."

[0119] After obtaining the semantic text of the portrait, the target entities within the semantic text are identified, and supplementary retrieval is performed based on these target entities to obtain supplementary semantic information related to them. Here, a target entity refers to an object with a specific semantic reference in the semantic text of the portrait, including but not limited to names of people, groups, works, brands, programs, objects of interest, or specific intellectual property (IP) names. Supplementary retrieval refers to calling external retrieval resources to obtain background and related information related to the target entity, thereby supplementing the content not explicitly recorded in the target audience portrait but helpful in understanding the semantics of the target entity. For example, for the target entity "singer AA" in the combined semantic text, the system can obtain supplementary semantic information through internet retrieval, such as singer AA being a Chinese pop male singer, his fan base being called "AA fans," his active age group being 20-30 years old, and his preference for English songs; for "NEW SONG A," information such as the album being released in 20XX, the title track being the same song, and its high popularity on a certain short video platform can be retrieved. The retrieved supplementary semantic information can be temporarily stored in text form for later integration with the original combined semantic text. In this way, when recalling tags for popular figures, trending events, or new content scenarios, supplementary searches can be used to compensate for the insufficient semantic coverage caused by the relatively lagging updates of the user feature tag knowledge base.

[0120] The retrieved supplementary semantic information is then fused with the profile semantic text to obtain enhanced semantic text. The fusion method can be to append the supplementary semantic information to the profile semantic text, or to insert the supplementary semantic information into the corresponding position in the profile semantic text according to the correspondence between the target entity and the supplementary semantic information. It can be understood that the fused enhanced semantic text retains both the user's original input of demographic characteristics and business scenario descriptions, while also incorporating real-time entity background knowledge and the latest developments obtained from external retrieval resources, thus significantly improving the accuracy of downstream semantic understanding tasks. Then, a large language model (such as DeepSeek-V3) is called to perform named entity recognition on the enhanced semantic text, extracting key entities and outputting the extracted entities to obtain an entity list. Named entity recognition using a large language model can identify synonyms, aliases, and context-related entity references, demonstrating stronger semantic understanding capabilities compared to traditional keyword matching. For example, by performing named entity recognition on the enhanced semantic text mentioned above, the following entities can be extracted: person entity AA (type: singer), group entity AAfan (type: fan group), work entity "NEW SONG A" (type: song), and the implicit demographic characteristic entity 20-30 years old (type: age range). These entities can be stored based on an array list or key-value pair JSON information, with each entity containing two fields: entity name and entity type.

[0121] Based on the entity list obtained in the previous step, the system retrieves user feature tags from the user feature tag knowledge base corresponding to the selected intent, and adds these tags to the target tag set. The user feature tag knowledge base is a tag storage system pre-built according to business lines, where each tag is associated with metadata such as tag name, tag description, and applicable entity type. During retrieval, the system first assigns the corresponding user feature tag knowledge base branch based on the business line information in the selected intent (e.g., Group A, Business Line A1, Business Line A2, etc.). Then, for each entity in the entity list, vector retrieval and keyword matching are performed in the corresponding knowledge base. For example, for the entity "Singer AA," retrieving the user feature tag knowledge base of business line A1 yields the tag "Singer AA Fan" (tag ID: 123, description: users who like AA's songs); for the entity "AA Fan," the synonym tag "AA Treasure" is obtained; and for the entity "20-30 years old," the age tag "20-30 years old" is obtained. The system performs deduplication on the tags retrieved from multiple entities, removing duplicate tags, and adds the processed user feature tags to the target tag set to provide a tag basis for generating user selection rules based on the target tag set in the future.

[0122] Meanwhile, in the retrieval chain based on behavioral feature tags, as a feasible implementation method, the selected tags include a target tag set; the tag knowledge base used for tag matching includes a behavioral feature tag knowledge base; tag matching is performed according to the target audience profile to obtain the target tag set, including: according to the target audience profile, searching in the behavioral feature tag knowledge base corresponding to the selection intent to obtain candidate behavioral feature tags; calling a large language model to determine the relevance score between the candidate behavioral feature tags and the target audience profile; based on the relevance score, filtering candidate behavioral feature tags with relevance scores lower than a preset relevance threshold to obtain behavioral feature tags, and adding the behavioral feature tags to the target tag set.

[0123] When the target tag set includes behavioral feature tags, and the tag knowledge base used for tag matching includes a behavioral feature tag knowledge base, the tag matching process based on the target audience profile first includes the initial retrieval of behavioral feature tags. Specifically, based on the target audience profile, candidate behavioral feature tags can be obtained by searching the behavioral feature tag knowledge base corresponding to the selected intent. Here, behavioral feature tags refer to tag information used to characterize user dynamic behavior, interaction history, temporal actions, or business event responses. It can be understood that, unlike user feature tags used to describe user static attributes and long-term preferences, behavioral feature tags focus more on describing the user's actual operational behavior and intensity within a certain time frame. The behavioral feature tag knowledge base refers to a pre-established data set used to store such behavioral tags. The behavioral feature tag knowledge base may include behavior names, behavior filtering items, behavior aggregation options, and metadata related to behavior rule generation. The behavior name is used to characterize the specific action performed by the user, such as playing, favorite, downloading, sharing, clicking, searching, purchasing, or activating a membership; the behavior filter is used to limit the behavior object, the environment in which the behavior occurs, or the conditions attached to the behavior, such as song name, artist name, content type, terminal platform, business entry point, or behavior occurrence channel; the behavior aggregation option is used to characterize the method of statistically analyzing, counting, or calculating the behavior over time, such as total number of behaviors, consecutive days, cumulative duration, the time of the most recent occurrence, or the frequency of the behavior within a certain time range.

[0124] During retrieval, the system can leverage user behavior tendencies, interests, business applicability characteristics, and other behavior-related demographic features represented by the target audience profile. It can then use vector similarity matching, keyword matching, or a combination of both to retrieve an initial set of behavioral feature tags as candidate behavioral feature tags from the behavioral feature tag knowledge base. For example, if the user description indicates they are a fan of singer AA, and the business scenario is a promotional campaign for the new song "NEW SONG A," the system can retrieve candidate behavioral feature tags from the behavioral feature tag knowledge base, such as playing "NEW SONG A," saving the song, sharing it on social media, recently searching for 'singer AA,' and downloading the song.

[0125] Understandably, candidate behavioral feature labels may contain multiple labels, some of which are highly relevant to the selection intent, while others may have a weaker correlation. Therefore, after obtaining the candidate behavioral feature labels, a large language model can be invoked to determine the relevance score between each candidate behavioral feature label and the target audience profile. The large language model receives metadata such as candidate behavioral feature labels and their corresponding behavioral names, behavioral filtering options, and behavioral aggregation options. It also receives the target audience profile as a basis for relevance judgment and analyzes whether the user dynamic behaviors represented by each candidate behavioral feature label conform to the typical behavioral patterns described in the target audience profile. Typically, the large language model outputs a numerical relevance score, such as a floating-point number ranging from 0 to 1. A higher score indicates a stronger relevance between the behavioral label and the selection intent. For example, regarding the candidate behavioral feature tag of playing "NEW SONG A", the large language model analysis suggests that singer AA's fans have a strong willingness to listen to the song in the context of promoting the new song. This behavior is highly consistent with the demographic characteristics and business scenario, thus giving it a relevance score of 0.95. The tag of downloading the song is also relevant, but perhaps not as direct as the playback behavior, scoring 0.75. The tag of sharing to social media is likely only performed by some active fans, scoring 0.60. While the tag of recently searching for 'singer AA' is relevant, it is a search behavior rather than song interaction, scoring 0.50. Therefore, the relevance scores can be used to effectively distinguish the importance of different behavioral tags to the current user selection rule generation task.

[0126] Based on the output relevance score, a preset relevance threshold, such as 0.6, is used to filter candidate behavioral feature tags. It should be noted that a higher relevance threshold results in more accurate retained tags but may reduce recall, while a lower relevance threshold retains more tags but may introduce noise. The relevance threshold can be adjusted according to actual business needs, and this embodiment does not limit this. Then, candidate behavioral feature tags with a relevance score greater than or equal to the preset relevance threshold are identified as behavioral feature tags and added to the target tag set for subsequent generation of user selection rules based on the target tag set.

[0127] The user selection rule generation method provided in this application relies on a pre-built tag knowledge base, which stores at least three types of data: user feature tags, behavioral feature tags, and historical selection cases. The data fields for user feature tags include tag identifier, tag name, tag description, operator, tag value type, tag value enumeration (if any), and secondary tag value enumeration (if any). For example, an instance of a user feature tag might be: tag identifier 87741478, tag name "student status," and tag description obtained from data filled in on the user's QQ profile card and the result predicted by the model. Currently based on a mirror image from 2018, future optimizations will be implemented. Operators include equal to, not equal to, contain, not contain, empty, not empty, fuzzy search, left fuzzy search, right fuzzy search, not equal to or empty, not contain or empty. The tag value type is an enumeration, with enumeration options including junior high school, university students, high school students, university freshmen, and primary school students.

[0128] The data fields for behavioral feature tags include the behavior name, behavior filter name, behavior filter operator, behavior filter value type, behavior aggregation option name, behavior aggregation option operator, and behavior aggregation option value type. For example, an instance of a behavioral feature tag might be: the behavior name is "[Conceptual Version] Remaining Days of VIP Access," the behavior aggregation option includes "Remaining Days of VIP Access," with operators including equal to, not equal to, less than, less than or equal to, greater than, greater than or equal to, and a range of values; the value type is numeric. The behavior filter includes "Statistical Date," with operators including equal to, not equal to, contain, do not contain, empty, not empty, fuzzy search, left fuzzy search, right fuzzy search, not equal to or empty, and do not contain or empty; the value type is string.

[0129] The historical targeting knowledge base records historical targeting cases. The data fields in the historical targeting knowledge base include the targeting scenario and suggested audience characteristics. For example, in a membership scenario, a targeting scenario is "Green Diamond membership discounts for students – recommended tab placements." The corresponding suggested audience characteristics are given in the form of logical expressions: {Green Diamond membership expiration days ≤ 30, membership status - continuous monthly subscription approximately ≠ 1, average monthly Green Diamond subscription price in the past year > 0, average monthly Green Diamond subscription price in the past year < 1500, and age between 8 and 24}, or {Whether Green Diamond Deluxe (real-time) does not contain "Yes" and Super Member monthly subscription (real-time) does not contain "Yes", QQ Music Membership ltv372_ (unit is points) ≤ 9600, QQ Music Membership ltv372_ (unit is points) > 0, and age between 8 and 24}. This historical targeting case can be used to infer supplementary user description information based on business scenario information when the user does not explicitly provide user description information, and further generate a target audience profile.

[0130] As a feasible implementation method, the selection tags include user feature tags and behavioral feature tags. Generating user selection rules based on the target tag set includes: inputting the target audience profile and the user feature tags and behavioral feature tags from the target tag set into a large language model, so that the large language model performs combined reasoning on the user feature tags and behavioral feature tags based on the target audience profile to obtain candidate user feature rules and candidate behavioral feature rules; obtaining tag metadata under the business line corresponding to the selection rule acquisition request; converting the candidate user feature rules and candidate behavioral feature rules into front-end component configuration data based on the tag metadata; validating the front-end component configuration data and determining the valid front-end component configuration data as the user selection rules; and outputting the user selection rules to the front-end configuration component.

[0131] The target audience profile and user feature tags and behavioral feature tags from the target tag set are input into the large language model. Based on a preset prompt word template, the large language model uses the target audience profile as the semantic constraint for rule generation. It performs combined reasoning on the user feature tags and behavioral feature tags, analyzing the correspondence between each tag and the audience attributes, interests, behavioral tendencies, and business needs represented by the target audience profile, and generating candidate user feature rules and candidate behavioral feature rules respectively. The candidate user feature rules describe selection conditions based on inherent features such as static attributes and interests, presented in the form of logical expressions in this embodiment, such as "student status equals college student AND age between 18 and 24". The candidate behavioral feature rules describe selection conditions based on time-sensitive features such as dynamic actions and interaction history, such as "played within a specified time range, and the artist name equals a certain value, and the number of plays is greater than a certain value". The output of the large language model can simultaneously include structured data (such as JSON format data) for front-end parsing and natural language text for user understanding. For example, when the user description information is "membership discount activities for students", the user feature tags include "student status" and "age", and the behavioral feature tags include "green diamond expiration days" and "playing behavior", the large language model can infer and output the candidate user feature rule as "student status equals college student AND age between 8 and 24", and the candidate behavioral feature rule as "green diamond expiration days less than or equal to 30 AND played behavior".

[0132] After obtaining candidate user feature rules and candidate behavior feature rules, the system retrieves the tag metadata under the business line corresponding to the selection rule retrieval request. The business line can be determined based on the business line information carried or parsed in the selection rule retrieval request. Tag metadata refers to the structured attribute information of each tag in the tag knowledge base corresponding to the current business line, including but not limited to: tag identifier, tag name, tag description, list of supported operators (such as equal to, not equal to, greater than, less than, etc.), tag value type (such as enumeration, numeric, string), tag value enumeration options (if any), and secondary tag information. Different business lines may have mutually isolated or partially different tag systems. Therefore, by retrieving the tag metadata under the business line corresponding to the selection rule retrieval request, the actual identifier, available operators, and legal value range of the tag referenced by the candidate rule in the current business line can be determined. In practical applications, the system can send a request to the backend server to call the corresponding data interface to obtain a complete tag list (e.g., vctFeatureList) under the current business line. This list can be returned in a structured data format (such as a JSON array). For example, for business line A, the obtained tag metadata may include the student status tag ID 87741478, which supports operators such as equal to, not equal to, and contain, and the value type is an enumeration with enumeration options such as junior high school, college student, high school student, college freshman, and primary school.

[0133] Based on the obtained tag metadata, the candidate user feature rules and candidate behavior feature rules are converted into front-end component configuration data that the front-end component can recognize and render. Specifically, the conversion process includes: (1) logical relationship mapping, that is, mapping the logical operators (such as and, or) in the candidate rules to the logical symbols (such as &, |) used inside the front-end component; (2) tag matching, that is, according to the tag names in the candidate user feature rules and candidate behavior feature rules, searching for the corresponding tag items in the tag metadata list and obtaining their complete information such as tag ID, operator list, and value type; (3) operator conversion, that is, converting the text form operators (such as equal to, greater than) in the candidate rules to the corresponding numeric operator ID or internal encoding; (4) value mapping processing, for enumeration type tags, mapping the text values ​​in the candidate rules (such as college students) to the corresponding enumeration value ID in the tag metadata; (5) secondary tag processing, if the candidate rules contain secondary tag conditions (such as artist names as filtering items for playback behavior), then further parsing the secondary tag names and configuring the corresponding sub-feature conditions. For behavioral feature rules, it is also necessary to handle the transformation of complex logic such as time range, aggregation conditions (such as total number of times, most recent time), and filtering items. After completing the transformation of the front-end component configuration data, the front-end component configuration data is verified, and the front-end component configuration data that passes the verification is determined as the user selection rule. The specific verification methods may include: (1) Tag existence verification, checking whether the tag name referenced in the candidate user feature rule and candidate behavioral feature rule exists in the tag metadata of the current business line. If it does not exist, an error is recorded and the condition is skipped; (2) Permission verification, checking whether the current user has the permission to use the tag. If the user does not have the permission, an error is recorded and the condition is skipped; (3) Operation compliance verification, checking whether the operator used in the candidate rule belongs to the list of operators supported by the tag; (4) Value validity verification, checking whether the value in the candidate rule is within the range of selectable values ​​of the tag; (5) For the time condition of the behavioral tag, checking whether the set date is within the valid time range of the tag data. If the verification fails, the system removes the corresponding condition and informs the user of the specific error information through a message prompt; if the key verification fails (such as the overall parsing of the tag rule fails), the process is terminated and a prompt is made that the generation of the user selection rule has failed. For front-end component configuration data that passes verification, it can be determined as an executable or displayable user selection rule.

[0134] Finally, the transformed and validated user feature rules (e.g., tagRule) and behavioral feature rules (e.g., actRule) are passed to the front-end rule rendering component in structured data format. The front-end configuration component, located in the user interface, is a rule selection component. Upon receiving rule data in the specified format, it automatically renders the user feature rules and behavioral feature rules into visual configuration items, including a tag selector (displaying tag names in a dropdown menu), an operator dropdown (displaying the operators supported by the tag), a value input box or enumeration selection box (displaying the corresponding input control based on the tag value type), and logical combination controls (such as adding condition groups, toggling AND / OR relationships). This allows the selection rules filled into various controls to be displayed on the interface, and can be manually adjusted and fine-tuned according to actual needs. Simultaneously, the system will determine whether to output a plain text reply or both a text reply and a selection button on the display interface based on front-end parameters (e.g., whether to display a selection button). If a selection button is displayed, after the user clicks it, the front-end will generate a submitable user selection scheme according to the above transformation and validation process.

[0135] Please see Figure 3 , Figure 3 This is an exemplary display of the user selection rules output by a user selection rule generation method provided in an embodiment of this application in an interactive interface. For example... Figure 3 As shown, the user selection rules generated based on the target tag set are divided into a user feature rule configuration area and a behavioral feature rule configuration area. The user feature rule configuration area is used to display the static attribute or interest preference selection conditions generated based on the user feature tags in the target tag set. In the illustration, the first user feature rule is "Group Artist Profile (Fan Segmentation) - Artist A - equals - High-Active Fans / B", which indicates that the user belongs to the high-active fan segment of Artist A; the second rule is "OST Preference Group - equals - 1", which indicates that the user belongs to the OST preference group. The above multiple user feature rules can be combined through logical relationships. For example, in the illustration, two rules are combined using an "AND" relationship, indicating that multiple user feature conditions must be met simultaneously, thereby achieving refined filtering of user static attributes and interest preferences.

[0136] Furthermore, the behavior feature rule configuration area is used to display dynamic behavior-based selection conditions generated based on behavior feature tags in the target tag set. In the illustration, the behavior rule is limited to a time range of "2026-01-01 to 2026-01-11", the behavior type is "playback behavior", and it is limited by the filter condition "artist name equals artist A" and the aggregation condition "playback count greater than 0", indicating that the user has played content related to artist A at least once within the specified time range. This type of behavior rule can use logical expressions such as "and satisfy" to achieve multi-dimensional constraints on user behavior features, thereby improving the timeliness and accuracy of the selection results. The right side of the behavior condition provides operation buttons such as filter, add, delete, and copy. Users can add rule conditions using the add button, remove invalid rules using the delete button, and quickly reuse existing rule structures using the copy button, thereby achieving flexible configuration and adjustment of rules. The user selection rules in this embodiment can be presented in a visual form through the front-end component, making complex tag combination relationships and behavior constraints intuitively displayed, thereby improving users' understanding of the selection rules and operational efficiency, while enhancing the usability and interactive experience of the system.

[0137] The above describes a user selection rule generation method provided by the embodiments of this application. The following will describe the apparatus for performing the above user selection rule generation method.

[0138] Please see Figure 4 , Figure 4 This is a schematic diagram of a user selection rule generation device provided in an embodiment of this application. Figure 4 As shown, the user selection rule generation device 400 includes:

[0139] The parsing module 401 is used to parse the received selection rule acquisition request to obtain the selection intent; the selection intent includes user description information and business scenario information.

[0140] The reasoning module 402 is used to determine the target reasoning strategy based on the selection intent, and to reason out the target population profile corresponding to the selection rule acquisition request based on the target reasoning strategy.

[0141] The tag matching module 403 is used to perform tag matching based on the target audience profile to obtain a target tag set.

[0142] The rule generation module 404 is used to generate user selection rules based on the target tag set.

[0143] Optionally, the parsing module 401 is used to extract business line information from the selection rule acquisition request; if the business line information is empty, a prompt message is output to instruct the user to select a business line; if the business line information is not empty, the selection rule acquisition request is parsed based on the session history information to extract the user description and business scenario, and obtain the extraction result; if both the user description information and the business scenario information in the extraction result are empty, a supplementary requirement prompt message is output; if the user description information in the extraction result is not empty and the business scenario information is empty, a supplementary business scenario prompt message is output; information in which at least one of the user description information and the business scenario information in the extraction result is not empty is determined as the selection intention.

[0144] Optionally, the inference module 402 is configured to: if the user description information in the selection intent is not empty and the business scenario information is empty, determine the target inference strategy as a first inference strategy and infer a target audience profile based on the user description information; if the user description information in the selection intent is empty and the business scenario information is not empty, determine the target inference strategy as a second inference strategy and infer a target audience profile based on the business scenario information; if both the user description information and the business scenario information in the selection intent are not empty, determine the target inference strategy as a third inference strategy and infer a target audience profile based on both the user description information and the business scenario information.

[0145] Optionally, the reasoning module 402 is used to input the business scenario information into the large language model, so that the large language model can infer supplementary user description information based on the business scenario information, and generate a target audience profile based on the supplementary user description information and the business scenario information.

[0146] Optionally, the inference module 402 is used to determine a first user profile based on the user description information in the selected intent; input the business scenario information into a large language model so that the large language model can infer supplementary user description information based on the business scenario information; determine a second user profile based on the supplementary user description information, the user description information in the selected intent, and the business scenario information; and integrate the first user profile and the second user profile to obtain a target user profile.

[0147] Optionally, the reasoning module 402 is used to retrieve historical selection cases related to the business scenario information from the historical selection knowledge base based on the business scenario information; input the business scenario information and the historical selection cases into the large language model, so that the large language model can reason about the business scenario information based on the historical selection cases and output supplementary user description information.

[0148] Optionally, the reasoning module 402 is used to input the business scenario information into a large language model, so that the large language model performs semantic analysis on the business scenario information, determines the target user group characteristics corresponding to the business scenario information, and generates supplementary user description information based on the target user group characteristics.

[0149] Optionally, the target tag set includes user feature tags; the tag knowledge base for tag matching includes a user feature tag knowledge base; the tag matching module 403 is used to acquire portrait semantic text representing the target population profile; determine the target entity in the portrait semantic text, and perform supplementary retrieval based on the target entity to obtain supplementary semantic information related to the target entity; fuse the supplementary semantic information with the portrait semantic text to obtain enhanced semantic text; perform named entity recognition on the enhanced semantic text to obtain an entity list; and based on the entity list, search in the user feature tag knowledge base corresponding to the selection intent to obtain user feature tags, and add the user feature tags to the target tag set.

[0150] Optionally, the selected tags include a set of target tags; the tag knowledge base for tag matching includes a behavioral feature tag knowledge base; the tag matching module 403 is used to retrieve candidate behavioral feature tags from the behavioral feature tag knowledge base corresponding to the selection intent based on the target audience profile; call a large language model to determine the relevance score between the candidate behavioral feature tags and the target audience profile; based on the relevance score, filter candidate behavioral feature tags whose relevance score is lower than a preset relevance threshold to obtain behavioral feature tags, and add the behavioral feature tags to the set of target tags.

[0151] Optionally, the selection tags include user feature tags and behavioral feature tags; the rule generation module 404 is used to input the target audience profile and the user feature tags and behavioral feature tags in the target tag set into a large language model, so that the large language model performs combined reasoning on the user feature tags and behavioral feature tags based on the target audience profile to obtain candidate user feature rules and candidate behavioral feature rules; obtain tag metadata under the business line corresponding to the selection rule acquisition request; convert the candidate user feature rules and candidate behavioral feature rules into front-end component configuration data according to the tag metadata; verify the front-end component configuration data, and determine the front-end component configuration data that passes the verification as the user selection rule; and output the user selection rule to the front-end configuration component.

[0152] This application also provides an electronic device in its embodiments. (See reference...) Figure 5The diagram illustrates a structural schematic of an electronic device suitable for implementing the user selection rule generation method in the embodiments of this application. The electronic device in the embodiments of this application may include, but is not limited to, fixed terminals such as mobile phones, laptops, PDAs (personal digital assistants), PADs (tablet computers), desktop computers, etc. Figure 5 The electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0153] like Figure 5 As shown, the electronic device may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 501, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 502 or a program loaded from a storage device 508 into a random access memory (RAM) 503. When the electronic device is powered on, the RAM 503 also stores various programs and data required for the operation of the electronic device. The processing unit 501, ROM 502, and RAM 503 are interconnected via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0154] Typically, the following devices can be connected to I / O interface 505: input devices 506 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 507 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 508 including, for example, memory cards, hard drives, etc.; and communication devices 509. Communication device 509 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 5 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown. More or fewer devices may be implemented or have alternatively.

[0155] This application also provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the user selection rule generation method as described in the above embodiments.

[0156] Since the embodiments of the computer program product part correspond to the embodiments of the user selection rule generation method part, please refer to the description of the embodiments of the user selection rule generation method part for the embodiments of the computer program product part, and will not be repeated here.

[0157] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the user selection rule generation method described in the above embodiment.

[0158] Since the embodiments of the computer-readable storage medium portion correspond to the embodiments of the user selection rule generation method portion, the embodiments of the storage medium portion are described in the description of the embodiments of the user selection rule generation method portion, and will not be repeated here.

[0159] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0160] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0161] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

Claims

1. A method for generating user selection rules, characterized in that, The method includes: The received request to obtain the selection rules is parsed to obtain the selection intent; the selection intent includes user description information and business scenario information. Based on the selection intent, a target inference strategy is determined, and based on the target inference strategy, the selection rules are inferred to obtain the target population profile corresponding to the request. Based on the target audience profile, tag matching is performed to obtain a target tag set; User selection rules are generated based on the target tag set.

2. The method according to claim 1, characterized in that, The step of parsing the received selection rule acquisition request to obtain the selection intent includes: Extract the selection rules to obtain the business line information in the request; If the business line information is empty, a prompt message will be output to instruct the user to select a business line; If the business line information is not empty, the request is obtained by parsing the selection rules based on the session history information to extract user descriptions and business scenarios, and obtain the extraction results; If the user description information and business scenario information in the extracted results are both empty, output a requirement supplement prompt message; If the user description information in the extraction result is not empty and the business scenario information is empty, output supplementary business scenario prompts. The information in the extracted results that is not empty, including at least one of the user description information and business scenario information, is identified as the selection intent.

3. The method according to claim 1, characterized in that, The step of determining the target inference strategy based on the selection intent, and inferring the selection rules to obtain the target audience profile corresponding to the request based on the target inference strategy, includes: If the user description information in the selected intent is not empty and the business scenario information is empty, then the target reasoning strategy is determined to be the first reasoning strategy, and the target audience profile is obtained based on the user description information. If the user description information in the selected intent is empty and the business scenario information is not empty, then the target reasoning strategy is determined to be the second reasoning strategy, and the target audience profile is obtained based on the business scenario information. If both the user description information and the business scenario information in the selected intent are not empty, then the target reasoning strategy is determined to be the third reasoning strategy, and the target audience profile is obtained by jointly reasoning based on the user description information and the business scenario information.

4. The method according to claim 3, characterized in that, If the user description information in the selected intent is empty and the business scenario information is not empty, then the target inference strategy is determined to be the second inference strategy, and a target audience profile is obtained based on the business scenario information, including: The business scenario information is input into a large language model, so that the large language model can infer supplementary user description information based on the business scenario information, and generate a target audience profile based on the supplementary user description information and the business scenario information.

5. The method according to claim 3, characterized in that, If both the user description information and the business scenario information in the selected intent are not empty, then the target inference strategy is determined to be the third inference strategy, and the target audience profile is obtained by jointly inferring based on the user description information and the business scenario information, including: Based on the user description information in the selected intent, a first group profile is determined; The business scenario information is input into the large language model so that the large language model can infer supplementary user description information based on the business scenario information. Based on the supplemented user description information, the user description information in the selected intent, and the business scenario information, a second user profile is determined. The first and second group profiles are integrated to obtain the target group profile.

6. The method according to claim 4 or 5, characterized in that, The step of inputting the business scenario information into a large language model, so that the large language model can infer supplementary user description information based on the business scenario information, includes: Based on the business scenario information, retrieve historical selection cases related to the business scenario information from the historical selection knowledge base; The business scenario information and the historical selected cases are input into the large language model, so that the large language model can infer the business scenario information based on the historical selected cases and output supplementary user description information.

7. The method according to claim 4 or 5, characterized in that, The step of inputting the business scenario information into a large language model, so that the large language model can infer supplementary user description information based on the business scenario information, includes: The business scenario information is input into a large language model, which performs semantic analysis on the business scenario information, determines the target user group characteristics corresponding to the business scenario information, and generates supplementary user description information based on the target user group characteristics.

8. The method according to claim 1, characterized in that, The target tag set includes user feature tags; The tag knowledge base used for tag matching includes a user feature tag knowledge base; the step of performing tag matching based on the target audience profile to obtain a target tag set includes: Obtain the semantic text of the portrait used to characterize the target population profile; Identify the target entity in the semantic text of the portrait, and perform supplementary retrieval based on the target entity to obtain supplementary semantic information related to the target entity; The supplementary semantic information is fused with the image semantic text to obtain enhanced semantic text; Named entity recognition is performed on the enhanced semantic text to obtain an entity list; Based on the entity list, a search is performed in the user feature tag knowledge base corresponding to the selection intent to obtain user feature tags, and the user feature tags are added to the target tag set.

9. The method according to claim 1, characterized in that, The selected tags include a set of target tags; the tag knowledge base used for tag matching includes a behavioral feature tag knowledge base; the step of performing tag matching based on the target audience profile to obtain the set of target tags includes: Based on the target audience profile, a search is performed in the behavioral feature tag knowledge base corresponding to the selected intent to obtain candidate behavioral feature tags; The large language model is invoked to determine the correlation score between the candidate behavioral feature labels and the target population profile; Based on the relevance score, candidate behavioral feature labels with relevance scores lower than a preset relevance threshold are filtered to obtain behavioral feature labels, and the behavioral feature labels are added to the target label set.

10. The method according to claim 1, characterized in that, The selection tags include user feature tags and behavioral feature tags; the step of generating user selection rules based on the target tag set includes: The target audience profile and the user feature tags and behavioral feature tags in the target tag set are input into the large language model, so that the large language model performs combined reasoning on the user feature tags and behavioral feature tags based on the target audience profile to obtain candidate user feature rules and candidate behavioral feature rules. Obtain the tag metadata under the business line corresponding to the selection rule; Based on the tag metadata, the candidate user feature rules and the candidate behavior feature rules are converted into front-end component configuration data; The configuration data of the front-end components is validated, and the front-end component configuration data that passes the validation is determined as the user selection rule. The user selection rules are output to the front-end configuration component.

11. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to implement the user selection rule generation method as described in any one of claims 1 to 10 when executing the computer program.

12. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed by a processor, they implement the user selection rule generation method as described in any one of claims 1 to 10.

13. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when loaded and executed by a processor, implement the user selection rule generation method as described in any one of claims 1 to 10.

Images