Method, device and equipment for generating crowd data

Abstract

The embodiment of the invention discloses a method, device and equipment for generating crowd data. In the embodiment of the invention, user demand information is acquired; the user demand information is input into a first large language model, the expanded user demand information is output, and the first large language model is used for expanding the input user demand information; the expanded user demand information of the user is input into a second large language model, at least one target tag is output, and the target tag is an unstructured tag extracted according to the expanded user demand information; inputting the at least one target label into a representation model, and outputting at least one target entity; and determining crowd data according to the at least one target entity, the crowd data being data conforming to the user demand information. Through the method, crowd data meeting refined operation requirements can be accurately and efficiently generated.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and more specifically, to a method, apparatus, and device for generating population data. Background Technology

[0002] With the continuous development of big data technology, big data-based applications and services are expanding in various scenarios. For example, user tags are used to segment people to support needs such as precision marketing.

[0003] In existing technologies, users manually combine existing user behavior tags or user profile tags to construct audience segmentation logic and obtain corresponding audience data. These user behavior tags or user profile tags are defined by experts and generated after processing and accumulating data based on specific rules, behaviors, or profile fields. Users need to combine user tags using Boolean logic in the tag selection interface. For ordinary users, it is difficult to accurately select the desired tags during the tag selection process. To improve user convenience, an AI-based audience segmentation method based on Large Language Model (LLM) has been proposed. This method utilizes the natural language understanding capabilities of LLM to understand user input questions, associates them with existing user tags, and reorganizes the existing user tags to complete the construction of audience data. However, all of the above methods rely on existing user tags. With increasingly sophisticated operations, most user needs can no longer be met using existing user tags.

[0004] In conclusion, how to generate audience data that meets the needs of refined operations is a problem that needs to be solved. Summary of the Invention

[0005] In view of this, embodiments of the present invention provide a method, apparatus, and device for generating population data, which can accurately and efficiently generate population data that meets the needs of refined operations.

[0006] In a first aspect, embodiments of the present invention provide a method for generating population data, the method comprising: acquiring user demand information; inputting the user demand information into a first large language model and outputting expanded user demand information, wherein the first large language model is used to expand the input user demand information; inputting the expanded user demand information into a second large language model and outputting at least one target label, wherein the target label is an unstructured label extracted based on the expanded user demand information; inputting the at least one target label into a representation model and outputting at least one target entity; and determining population data based on the at least one target entity, wherein the population data is data that conforms to the user demand information.

[0007] Optionally, the method further includes: inputting the expanded user demand information into a second language model and outputting at least one basic tag, wherein the basic tag is a pre-set structured tag.

[0008] Optionally, the step of inputting the at least one target label into the representation model and outputting at least one target entity specifically includes: inputting the at least one target label and the at least one base label into the representation model and outputting at least one target entity.

[0009] Optionally, the step of inputting the at least one target label into the representation model and outputting at least one target entity further includes: obtaining information on multiple candidate entities; inputting the at least one target label and the information on the multiple candidate entities into the representation model and outputting at least one target entity.

[0010] Optionally, the candidate entity information includes short text information and / or long text information, wherein the short text information is generated by the candidate entity title information through the first large language model through understanding enhancement, and the long text information is generated by the short text information through the first large language model through understanding enhancement.

[0011] Optionally, determining the crowd data based on the at least one target entity specifically includes: determining initial crowd data based on the at least one target entity; determining the interaction score between each person in the initial crowd data and the at least one target entity; sorting the people in the initial crowd data in descending order of the interaction score; and determining the multiple people ranked at the top of the set ranking as the crowd data.

[0012] Optionally, the step of inputting the expanded user demand information into the second language model and outputting at least one target label specifically includes:

[0013] The expanded user demand information is input into the second language model, and at least one target label and at least one exemplary entity are output.

[0014] Optionally, the target label includes a label name and a label description.

[0015] Optionally, the second large language model includes multiple agents, and the training process of the second large language model includes:

[0016] The second major language model, which includes multiple agents, is trained based on a pre-set labeling system and sample data.

[0017] Secondly, embodiments of the present invention provide an apparatus for generating population data, the apparatus comprising: an acquisition unit for acquiring user demand information; an expansion unit for inputting the user demand information into a first large language model and outputting expanded user demand information, wherein the first large language model is used to expand the input user demand information; a generation unit for inputting the expanded user demand information into a second large language model and outputting at least one target label, wherein the target label is an unstructured label extracted based on the expanded user demand information; a matching unit for inputting the at least one target label into a representation model and outputting at least one target entity; and a determination unit for determining population data based on the at least one target entity, wherein the population data is data that conforms to the user demand information.

[0018] Optionally, the generation unit is further configured to: input the expanded user demand information into the second language model and output at least one basic tag, wherein the basic tag is a pre-set structured tag.

[0019] Optionally, the matching unit is specifically used to: input the at least one target label and the at least one basic label into the representation model, and output at least one target entity.

[0020] Optionally, the matching unit is further configured to: acquire information on multiple candidate entities; input the at least one target label and the information on the multiple candidate entities into a representation model, and output at least one target entity.

[0021] Optionally, the candidate entity information includes short text information and / or long text information, wherein the short text information is generated by the candidate entity title information through the first large language model through understanding enhancement, and the long text information is generated by the short text information through the first large language model through understanding enhancement.

[0022] Optionally, the determining unit is specifically used for: determining initial crowd data based on the at least one target entity; determining the interaction score between each person in the initial crowd data and the at least one target entity; sorting the people in the initial crowd data in descending order of the interaction score; and determining the multiple people ranked at the top of the set ranking as the crowd data.

[0023] Optionally, the generation unit is specifically used to: input the expanded user demand information into the second language model, and output at least one target label and at least one exemplary entity.

[0024] Optionally, the target label includes a label name and a label description.

[0025] Optionally, the second large language model includes multiple agents, and the device further includes: a training model for training the second large language model including multiple agents based on a pre-set label system and sample data.

[0026] Thirdly, embodiments of the present invention provide an electronic device, including a memory and a processor, wherein the memory is used to store one or more computer program instructions, wherein the one or more computer program instructions are executed by the processor to implement the method as described in the first aspect or any one of the possible methods of the first aspect.

[0027] Fourthly, embodiments of the present invention provide a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the method as described in the first aspect or any one of the possibilities of the first aspect.

[0028] In this embodiment of the invention, user demand information is acquired; this user demand information is input into a first large language model, which outputs expanded user demand information, wherein the first large language model is used to expand the input user demand information; the expanded user demand information is input into a second large language model, which outputs at least one target label, wherein the target label is an unstructured label extracted based on the expanded user demand information; the at least one target label is input into a representation model, which outputs at least one target entity; and user data is determined based on the at least one target entity, wherein the user data is data that conforms to the user demand information. Through the above method, user data that meets the needs of refined operations can be generated accurately and efficiently. Attached Figure Description

[0029] The above and other objects, features and advantages of the present invention will become clearer from the following description of embodiments of the invention with reference to the accompanying drawings, in which:

[0030] Figure 1 This is a flowchart of a method for generating population data according to an embodiment of the present invention;

[0031] Figure 2 This is a schematic diagram of a labeling system in an embodiment of the present invention;

[0032] Figure 3 This is a schematic diagram illustrating the result of a multi-agent system in an embodiment of the present invention;

[0033] Figure 4 This is a flowchart of another method for generating crowd data in an embodiment of the present invention;

[0034] Figure 5This is a flowchart of another method for generating crowd data in an embodiment of the present invention;

[0035] Figure 6 This is a flowchart of another method for generating population data in an embodiment of the present invention;

[0036] Figure 7 This is a schematic diagram of a device for generating population data in an embodiment of the present invention;

[0037] Figure 8 This is a schematic diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0038] The present application is described below based on embodiments, but it is not limited to these embodiments. In the detailed description of the present application below, certain specific details are described in detail. Those skilled in the art can fully understand the present application without these details. To avoid obscuring the substance of the present application, well-known methods, processes, flows, elements, and circuits are not described in detail.

[0039] Furthermore, those skilled in the art should understand that the accompanying drawings provided herein are for illustrative purposes only and are not necessarily drawn to scale.

[0040] Unless the context explicitly requires it, words such as "including" or "contains" throughout the application should be interpreted as including rather than exclusive or exhaustive; that is, meaning "including but not limited to".

[0041] In the description of this application, it should be understood that the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance. Furthermore, in the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0042] In existing technologies, users manually combine existing user tags, including user behavior tags or user profile tags, to construct audience segmentation logic and obtain corresponding audience data. For example, if a user selects user tags such as "within 30 days," "browsed milk powder," and "female user," then audience segmentation is performed based on "female users who browsed milk powder within 30 days." Audience segmentation refers to operators filtering user groups that meet specific conditions based on user tags, user behavior, or user characteristics for targeted promotion or strategy execution. User tags, also known as tag profiles, refer to a structured feature set constructed using user behavior and attributes on the platform, such as "post-90s female" or "highly active in beauty products." Specifically, users need to combine user tags using Boolean logic in the tag selection interface. For ordinary users, it is difficult to accurately select the desired tags during the tag selection process. To improve user convenience, a method based on a Large Language Model is proposed. The Large Language Model (LLM) approach uses artificial intelligence (AI) to segment users, leveraging the LLM's natural language understanding capabilities to interpret user input questions and associate them with existing user tags. This reorganization of existing user tags completes the construction of user data. The user input questions are natural language expressions of the target audience, meaning users describe their target demographic through free text, such as "people who recently bought lightweight down jackets and browsed winter skiing-related content." The large language model understands these questions and associates them with existing user tags. These expressions can also be termed unstructured audience intent, meaning users express their needs without adhering to a tag system, resulting in vague, broad, or cross-tag descriptions. However, all these methods rely heavily on existing user tags, leading to a heavy dependence on tag assets and a lack of support for new scenarios and directions. With increasingly sophisticated operations, most user needs can no longer be met using existing user tags. This sophisticated operation refers to achieving targeted and highly efficient marketing and service strategies through more precise data analysis and audience insights. Therefore, generating audience data that meets the needs of sophisticated operations is a problem that needs to be solved.

[0043] In this embodiment of the invention, the large language model can also be called a large model or a large-scale language model, etc. The large language model is a deep learning model based on a transformer architecture, which can process and generate natural language text. It is usually trained on a large amount of text data, has the ability to understand and generate language, and is widely used in dialogue systems, text generation and other natural language processing tasks.

[0044] In this embodiment of the invention, to solve the above problems, a method for generating population data is proposed, specifically as follows: Figure 1 As shown, the method includes:

[0045] Step S101: Obtain user demand information.

[0046] For example, the user demand information is the screen interface of the platform that generates the audience data, which is entered by the user. For example, the user demand information is "I want to operate a group of people who are interested in 'Cat Cat Hero' on the animation channel". The above user demand information is only an example and can be determined according to the actual situation. It can involve the generation of audience data in multiple fields and multiple platforms, and can also be called audience data selection or audience generation.

[0047] Step S102: Input the user demand information into the first language model and output the expanded user demand information.

[0048] Specifically, the first large language model is used to expand the input user demand information.

[0049] In one possible implementation, the intent recognition and function calling module of the platform generating the crowd data determines the user's purpose as crowd generation based on the user's demand information, and calls the first large language model. This first large language model performs discovery-based reasoning under the guidance of In-Context Learning (ICL) and K-Nearest Neighbors for Few-Shot Learning (KNN-fewshot) to expand the user's demand information. For example, inputting "wanting to operate a group of people interested in 'Neko Atsume' on an anime channel" into the first large language model outputs "People who show a strong interest in the classic Japanese anime 'Neko Atsume' may frequently visit anime-related websites, forums, or social media groups, participating in discussions about plot development, character analysis, and other topics. Furthermore, this group may also be keen on collecting 'Neko Atsume' related merchandise, such as figurines, comic books, and peripheral clothing, and may participate in anime conventions or other related activities." This is merely an illustrative example; the actual generation will depend on the specific circumstances.

[0050] In one possible implementation, the intent recognition module can also determine the user's purpose, such as crowd exploration or asset exploration, based on the user's demand information.

[0051] Step S103: Input the expanded user demand information into the second language model and output at least one target label.

[0052] Specifically, the target tags are unstructured tags extracted based on the expanded user demand information, including tag names and tag descriptions.

[0053] In one possible implementation, the phrase "People who show a strong interest in the classic Japanese anime 'Neko Atsume' may frequently visit anime-related websites, forums, or social media groups to discuss plot developments, character analyses, and other topics. Furthermore, this group may be enthusiastic about collecting 'Neko Atsume' merchandise, such as figurines, comic books, and related apparel, and may attend anime conventions or other related events." is input into the second language model, which outputs target tags such as "Neko Atsume-themed figurines" and "Neko Atsume hero." These target tags can also be referred to as scene orientations.

[0054] In one possible implementation, the second large language model includes multiple agents, and the second large language model is trained and generated based on a pre-set label system and sample data.

[0055] For example, suppose a pre-set labeling system is as follows: Figure 2 As shown, the demand tags include subjective and objective demand perspectives. The subjective demand perspective includes time, space (location), audience (people), scene (event), function, senses, identification, and consumer psychology. Time includes Qixi Festival, Mid-Autumn Festival, weekends, etc.; space includes office, park, car, etc.; audience includes infants, fathers, college students, etc.; scene includes dating, skiing, postpartum care, etc.; function includes warmth, whitening, rust removal, etc.; senses include vision (color, shape, measurement), hearing (sound quality), smell (odor), taste (flavor), touch (tactile sensation), and bodily sensation (head, neck, hands, feet). The labels include: style (Lolita, punk, retro, design, etc.), celebrities (A, B, C, etc.), and IPs (AA, BB, CC, etc.); consumer psychology includes price comparison, promotions, curiosity, and secondhand goods; the objective supply perspective includes brand, category, ingredients, and place of origin, etc., where brands include AA, BB, CC, etc.; categories include long dresses, baby seats, notebooks, etc.; ingredients include 100% cotton, 100% wheat, hyaluronic acid, etc.; and place of origin includes regions A, B, C, etc. The above labeling system is only an illustrative example and can be set according to actual conditions. This embodiment of the invention does not limit it.

[0056] In one possible implementation, the sample data includes a label name and a label description. For example, the label name is "Protection for Women During Special Periods," and the label description is "Personal hygiene and health management measures and products for women during menstruation." The sample data is described from multiple information dimensions within the label system, including target audience: women; time description: menstruation; function: protection; and associated exemplary entities such as sanitary napkins and handmade brown sugar. Another example is the label name "Daily Breath Management," with the label description "For image-conscious individuals, regularly removing bad breath using long-lasting breath freshening patches or high-efficiency mouthwash to maintain confidence." This sample data is also described from multiple information dimensions within the label system, including scenario / event: breath management; time description: daily; and the label name "Home Improvement Needs." The label "Marketplace" is described as "a physical or online marketing venue that meets the needs of users seeking home building materials, home furnishings, and soft furnishings products." The sample data described above is based on multiple information dimensions within the label system, including: scenario event: home decoration needs; category: home decoration; location: home furnishing marketplace; and associated exemplary entities, including home furnishing store 1 and home furnishing store 2. Under the guidance of the label system, the sample data trains a multi-agent collaborative second language model, internalizing the ability to associate entities with the target label into the second language model. Exemplary entities are generated concurrently during the label name and label description process to assist users in understanding the label name and label description. In this embodiment of the invention, the target label can also be simply referred to as a label.

[0057] In this embodiment of the invention, the second major language model of multi-agent collaboration includes Agent 1, Agent 2, and Agent 3. Agent 1 is responsible for core dimension generation for product cluster understanding, primarily used for defining and generating core tag elements and recognizing tag information. Agent 2 is a comprehensive reasoning model, mainly used for supervised fine-tuning (SFT), multi-dimensional tag decomposition, and tag-associated product generation. Agent 3 is responsible for information merging, self-guidance, and correction, used for merging decomposition results, disambiguation correction, and guidance. Agent 1, Agent 2, and Agent 3 are processed sequentially, and the output of Agent 3 is fed back to Agent 2. The outputs of Agent 1 and Agent 2 jointly train Agent 3. A schematic diagram of the multi-agent model is shown below. Figure 3As shown; here, auxiliary information is used to assist in the training of the second language model, wherein the auxiliary information includes global entities, existing assets (e.g., existing structured labels) and external knowledge; after multiple rounds of optimization, the second language model can directly generate target labels through Agent2. Under the guidance of the label system and similar semantic aggregation, the labels are clustered at multiple levels to complete the normalization and mounting of similar semantic labels, and merge to generate label descriptions; the target labels generated under the guidance of the label system can include multiple information dimensions, for example, 2-4 information dimensions.

[0058] For example, the target label generated by the second language model is "Fun Breakfast for Children: Fun breakfasts designed specifically for children, featuring brightly colored and cute cartoon shapes, aiming to stimulate children's appetite and make breakfast time fun and nutritionally balanced. Expected label: Directional, not overly focused on a specific category, and rich in descriptive information." Before generating the above target label, the untrained second language model might have generated target labels such as "Fun Breakfast for Children: Breakfast options that attract children, making even picky eaters love breakfast time," "Fun Breakfast for Children: Breakfast options designed specifically for children, rich in various grains, cute shapes, stimulating children's appetite," and "Fun Breakfast for Children: To attract children's attention and ensure their nutritional balance, choose brightly colored and cute cartoon steamed buns and milk-flavored small steamed buns when preparing breakfast, allowing children to enjoy delicious food while increasing the fun of eating." These target labels have various problems such as being too simple or too detailed. After multiple rounds of optimization, the final second language model and the better target labels generated by the optimized large language model were generated.

[0059] In one possible implementation, the expanded user demand information is input into a second language model, and at least one target label and at least one exemplary entity are output. The exemplary entity can be an entity that provides supplementary explanations for the target label, such as a product, item, address, or store name.

[0060] Step S104: Input the at least one target label into the representation model and output at least one target entity.

[0061] Specifically, multiple candidate entity information is obtained; the at least one target label and the multiple candidate entity information are input into a representation model, and at least one target entity is output; wherein, the candidate entity information includes short text information and / or long text information, wherein the short text information is generated by the candidate entity title information through the first large language model through understanding enhancement, and the long text information is generated by the short text information through the first large language model through understanding enhancement.

[0062] For example, the candidate entities can be entities involving multiple fields and platforms, such as goods, items, tickets, food, TV series names, and locations. Specific examples are as follows:

[0063] Example 1: The title information of the candidate entity is "Amusement Park 2025 New Year Mascot Pendant [Choose Face]". The title "Amusement Park 2025 New Year Mascot 1 Pendant [Choose Face]" is input into the first large language model for understanding enhancement to generate short text information. The short text information includes discovery tags, specifically: "Usage Scenarios: Holiday Decorations, Daily Accessories; Purpose: Pendant, Collection; Needs: New Year Gifts, Personalized Decorations, Disney Merchandise; Interest Tags: Amusement Park Merchandise, Holiday Merchandise, Anime Models". The short text information is input into the first... The large language model enhances understanding and generates long text information, specifically: "This product is a limited-edition Lunar New Year pendant designed with the theme of the amusement park character 'Mascot 1,' allowing users to choose their preferred face shape. The product is suitable for decorating during the Lunar New Year or as an accessory for everyday items, such as keychains and backpack charms. The target audience is primarily amusement park IP enthusiasts, anime merchandise collectors, and young female consumers who value the emotional connection and uniqueness of the product. They may purchase it for personal collection, as a holiday gift, or to complement their cute style outfits. The core need is to convey a joyful atmosphere through IP merchandise and satisfy their collecting preferences."

[0064] Example 2: The title information of the candidate entity is "Location H International Circus Ticket + Wildlife World 1-Day Combo Ticket for Adults". Inputting this into the first large language model for enhanced understanding generates the short text information as "Destination: Location H; Key Elements: International Circus, Wildlife World, 1-Day Combo Ticket; Interest Tags: Family Trip, Animal Viewing, Circus Performance, Location H Tourism". Inputting this short text information into the first large language model for enhanced understanding generates the long text information as "Location H International Circus and Wildlife World 1-Day Combo Ticket for Adults, providing a one-stop themed entertainment experience. Visitors can enjoy circus performances, including high-altitude acrobatics, animal interactions, and other exciting programs, and explore the wildlife world for close-up animal observation. This package is suitable for adult tourists and families seeking efficient travel, meeting the needs for entertainment performances, animal science education, and convenient itineraries. The target audience is mainly urban tourists who love theme parks, cultural performances, and natural ecology, especially suitable for travelers with limited time but who wish to combine multiple experiences."

[0065] Example 3: The title information of the candidate entity is "Solid wood tea table and chair set, boat wood Chinese Kung Fu tea table, office leisure tea art table, creative tea brewing table integrated". Inputting this into the first large language model for understanding enhancement generates the short text information as "Usage scenarios: home tea art, office, tea room; Purpose: tea tasting, entertaining guests, office; Needs: Chinese style, multifunctionality; Interest tags: Chinese furniture, tea culture, solid wood enthusiasts, home decoration". Inputting the above short text into the first large language model for understanding enhancement generates the long text information as "This product is a solid wood tea table and chair set made of boat wood". This wooden tea table and chair set combines the functions of a traditional Chinese tea table and a tea ceremony table, serving both office and leisure purposes. The design emphasizes the natural wood grain texture, making it suitable for family tea rooms, Chinese-style clubs, or office lounges. The modular structure caters to various scenarios such as tea brewing, tea ceremony demonstrations, and small meetings. The target audience is tea enthusiasts or business professionals who value traditional cultural heritage and prefer natural materials. The ship timber's characteristics give the product a unique sense of historical patina, making it suitable for consumers who appreciate Zen-like spatial aesthetics and value practicality in furniture. It can serve as a home cultural decoration or meet the cultural style requirements of business receptions.

[0066] Example 4: The title information of the candidate entity is "MOCHA Cat Cafe". Inputting this into the first large language model for enhanced understanding generates the short text information as follows: "Location characteristics: pet interaction, coffee leisure; Needs: cat-petting for stress relief, afternoon tea leisure; Interest tags: cat lovers, coffee and light food enthusiasts, healing experience". Inputting the above short text into the first large language model for enhanced understanding generates the long text information as follows: "MOCHA Cat Cafe is a composite space integrating pet services and coffee culture, focusing on a dual healing experience of 'coffee + cute cats'. The store has a warm and comfortable environment, offering freshly ground coffee, handmade desserts, and light meals. It also houses several gentle purebred cats, allowing customers to freely interact with them. It is both a social hub for cat lovers and a haven for urban dwellers to relax, suitable for afternoon tea with friends, couples' leisure, or simply unwinding alone. Regularly held cat knowledge salons and adoption events further strengthen its pet-friendly attributes and sense of community."

[0067] Example 5: The title information of the candidate entity is "Original Flavor Luosifen + Double the Sour Bamboo Shoots". Inputting this into the first large language model for understanding enhancement generates the following short text: "Type: Rice Noodles; Main Ingredient: Luosifen; Characteristics: Stinky, Spicy and Sour; Flavor: Spicy and Sour, Stinky; Interest Tags: Rice Noodle Lovers, Liuzhou Snack Lovers, Strong Flavor Lovers, Sour Bamboo Shoot Lovers". Inputting this short text into the first large language model for understanding enhancement generates the following long text: "Luosifen, a local delicacy, originates from Liuzhou. Double the sour bamboo shoots enhance the tangy and refreshing taste, creating a rich and layered flavor experience. Suitable for diners who seek authentic Liuzhou flavor, enjoy strong flavors, and have a special preference for sour bamboo shoots."

[0068] Examples 1 to 5 above are merely illustrative examples, and the specific details should be determined according to the actual situation. The embodiments of the present invention do not limit them.

[0069] In this embodiment of the invention, the short text information and / or long text information of the at least one target label and the multiple candidate entity information are input into the representation model for matching, and at least one target entity is output.

[0070] In one possible implementation, the target label "Cat Hero Theme Figurine Model, Cat Hero, etc." is input into the representation model for matching with multiple candidate entity information in multiple fields or multiple platforms to determine the target entity "Domain 1 Cat Hero T-shirt, Domain 2 Cat Hero Band-Aid, Domain 3 Cat Hero Transformation Forms Review, etc." This is only an illustrative example.

[0071] Step S105: Determine the population data based on the at least one target entity.

[0072] The population data refers to data that matches the user's needs.

[0073] In one possible implementation, the process of determining the population data based on the at least one target entity is as follows: Figure 4 As shown, it includes the following steps:

[0074] Step S401: Determine initial population data based on the at least one target entity.

[0075] Specifically, initial population data associated with each target entity is determined, which can be the population that has searched or browsed the target entity.

[0076] Step S402: Determine the interaction score between each person in the initial crowd data and the at least one target entity, and sort the people in the initial crowd data in descending order of the interaction score.

[0077] Suppose that the initial population data obtained is 100,000 people. Based on the interaction score of each person with the at least one target entity, the 100,000 people are sorted in descending order of the interaction score.

[0078] Step S403: Determine the multiple people ranked first in the set ranking as the population data.

[0079] For example, the top 20,000 people are identified as the population data.

[0080] In one possible implementation, the method further includes the following steps, specifically as follows: Figure 5 As shown, it includes:

[0081] Step S106: Display the analysis data of the population data.

[0082] Specifically, the analysis data includes the name, scale, validity period, data source, and significant features of the population data, enabling users to intuitively understand the population data.

[0083] In an embodiment of the present invention, the restrictive requirements of the user are identified, and the extended user requirement information of the user is input into the second large language model to output at least one basic label, where the basic label is a pre-set structured label; the structured label can also be referred to as a basic label asset. For example, gender, age, marital and childbearing status, point of interest (POI), education level, consumption level, whether there is a car, gender of children, etc. Through the generation of the above basic labels, the conditional cross-logic is completed, and the association between the population data and the basic label assets is achieved.

[0084] In a possible implementation manner, after determining the basic label, inputting the at least one target label into a characterization model to output at least one target entity specifically includes: inputting the at least one target label and the at least one basic label into the characterization model to output at least one target entity.

[0085] The method for generating population data will be described in detail below through a complete embodiment, specifically as Figure 6 So, it includes the following:

[0086] User demand information is input into a platform that generates crowd data. This platform can also be called a task-oriented intelligent agent. The task-oriented intelligent agent has tool invocation capabilities and user demand understanding. Through an intent recognition module, it determines the user's intent as crowd generation, crowd exploration, or asset exploration. After determining the user's intent based on intent recognition, a function call is made. The user demand understanding function uses a first major language model to expand and extract constraints from the user demand information. Constraint extraction involves generating basic tags from the user demand information; refining the expanded user demand information; and then defining and generating tags by calling a second major language model, specifically including scene direction generation, user-defined direction, and scene keyword generation. Scene keyword generation involves associating celebrity names or specific words with internet terms to obtain multiple polysemous words. The scene direction generation and scene keyword generation correspond to different domains. The custom direction is further refined or understood. The scene direction generation involves single-stage and multi-stage generation to generate scene directions and exemplary entities. The single-stage and multi-stage generation can also generate scene directions and exemplary entities. The scene directions and exemplary entities are matched using a representation model and by word segmentation matching of the text. The representation matching and word segmentation matching are entity matching performed through a production engine. This process is used to calculate and generate audience data, and finally, audience display and exploration are performed. The constraint extraction, which generates basic tags through tag asset association, can also affect audience data calculation and generation.

[0087] Through the above embodiments, the natural language input by the user can be semantically understood through the first large language model, automatically parsing out the required features, behaviors, interests, etc., and expanding the natural language input by the user. The expanded natural language is then input into the second large language model, which can not only match existing basic tags, but also intelligently infer unlabeled feature combinations, i.e., target tags, breaking through the original tag granularity limitations, realizing intelligent understanding and utilization of unlabeled data, reducing reliance on expert-preset tag systems, and supporting more personalized and real-time audience building. In addition, the above method also supports cold start. Even for new businesses, new products, and new scenarios without clear tag coverage, the large oracle model can complete the construction of zero-sample audience data through semantic generalization, similar user behavior inference, etc., improving the audience reach capability in the early stage of new businesses. Furthermore, it reduces the tag selection and combination steps, directly generating audience data through natural language, which greatly lowers the usage threshold of the audience data generation platform. With a single step of inputting user intent, recommended audience data or automatic construction of audience selection logic can be obtained, improving the user experience.

[0088] In this embodiment of the invention, an apparatus for generating population data is provided, such as... Figure 7As shown, it specifically includes: an acquisition unit 701, an expansion unit 702, a generation unit 703, a matching unit 704, and a determination unit 705;

[0089] The acquisition unit 701 is used to acquire user demand information; the expansion unit 702 is used to input the user demand information into a first large language model and output expanded user demand information, wherein the first large language model is used to expand the input user demand information; the generation unit 703 is used to input the expanded user demand information into a second large language model and output at least one target label, wherein the target label is an unstructured label extracted based on the expanded user demand information; the matching unit 704 is used to input the at least one target label into a representation model and output at least one target entity; the determination unit 705 is used to determine crowd data based on the at least one target entity, wherein the crowd data is data that conforms to the user demand information.

[0090] Furthermore, the generation unit is also used to: input the expanded user demand information into the second language model and output at least one basic tag, wherein the basic tag is a pre-set structured tag.

[0091] Furthermore, the matching unit is specifically used to: input the at least one target label and the at least one basic label into the representation model, and output at least one target entity.

[0092] Furthermore, the matching unit is specifically used to: acquire information on multiple candidate entities; input the at least one target label and the information on the multiple candidate entities into the representation model, and output at least one target entity.

[0093] Furthermore, the candidate entity information includes short text information and / or long text information, wherein the short text information is generated by the candidate entity title information through the first large language model through understanding enhancement, and the long text information is generated by the short text information through the first large language model through understanding enhancement.

[0094] Further, the determining unit is specifically used for: determining initial crowd data based on the at least one target entity; determining the interaction score between each person in the initial crowd data and the at least one target entity; sorting the people in the initial crowd data in descending order of the interaction score; and determining the multiple people ranked at the top of the set ranking as the crowd data.

[0095] Furthermore, the generation unit is specifically used to: input the expanded user demand information into the second language model, and output at least one target label and at least one exemplary entity.

[0096] Furthermore, the target label includes a label name and a label description.

[0097] Furthermore, the second large language model includes multiple agents, and the device further includes: a training model for training the second large language model including multiple agents based on a pre-set label system and sample data.

[0098] Figure 8 This is a schematic diagram of the structure of the electronic device described in an embodiment of the present invention. Figure 8 As shown, it includes a general computer hardware architecture, which includes at least a processor 801 and a memory 802. The processor 801 and the memory 802 are connected via a bus 803. The memory 802 is adapted to store instructions or programs executable by the processor 801. The processor 801 can be a standalone microprocessor or a collection of one or more microprocessors. Thus, the processor 801 executes the instructions stored in the memory 802 to perform the method flow of the embodiments of the present invention as described above, thereby realizing data processing and control of other devices. The bus 803 connects the above-mentioned components together, and also connects the above-mentioned components to a display controller 804, a display device, and an input / output (I / O) device 805. The input / output (I / O) device 805 can be a mouse, keyboard, modem, network interface, touch input device, motion-sensing input device, printer, and other devices known in the art. Typically, the input / output device 805 is connected to the system via an input / output (I / O) controller 806.

[0099] The instructions stored in memory 802 are executed by at least one processor 801 to: acquire user demand information; input the user demand information into a first large language model and output expanded user demand information, wherein the first large language model is used to expand the input user demand information; input the expanded user demand information into a second large language model and output at least one target label, wherein the target label is an unstructured label extracted based on the expanded user demand information; input the at least one target label into a representation model and output at least one target entity; determine population data based on the at least one target entity, wherein the population data is data that conforms to the user demand information.

[0100] Specifically, the electronic device includes: one or more processors 801 and a memory 802. Figure 8Take a processor 801 as an example. The processor 801 and the memory 802 can be connected via a bus or other means. Figure 8 Taking a bus connection as an example, memory 802, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules. Processor 801 executes various functional applications and data processing of the device by running the non-volatile software programs, instructions, and modules stored in memory 802, thereby implementing the aforementioned method for determining and generating population data.

[0101] Memory 802 may include a program storage area and a data storage area, wherein the program storage area may store the operating system and applications required for at least one function; the data storage area may store an option list, etc. Furthermore, memory 802 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, memory 802 may optionally include memory remotely located relative to processor 801, and these remote memories can be connected to external devices via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0102] One or more modules are stored in memory 802, and when executed by one or more processors 801, they perform the method for generating crowd data in any of the above method embodiments.

[0103] As those skilled in the art will recognize, various aspects of the embodiments of the present invention can be implemented as a system, method, or computer program product. Therefore, various aspects of the embodiments of the present invention can take the form of a completely hardware implementation, a completely software implementation (including firmware, resident software, microcode, etc.), or an implementation combining software and hardware aspects, which may generally be referred to herein as a "circuit," "module," or "system." Furthermore, various aspects of the embodiments of the present invention can take the form of a computer program product implemented in one or more computer-readable media having computer-readable program code implemented thereon.

[0104] Any combination of one or more computer-readable media can be used. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium can be, for example, (but not limited to) an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or apparatus, or any suitable combination thereof. More specific examples (not an exhaustive list) of computer-readable storage media will include: an electrical connection having one or more wires, a portable computer floppy disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable optical disc read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In the context of embodiments of the present invention, a computer-readable storage medium can be any tangible medium capable of containing or storing a program used by or in conjunction with an instruction execution system, device, or apparatus.

[0105] Computer-readable signal media may include propagated digital signals having computer-readable program code implemented therein, such as in baseband or as part of a carrier wave. Such propagated signals may take any of a variety of forms, including, but not limited to, electromagnetic, optical, or any suitable combination thereof. A computer-readable signal medium may be any computer-readable medium that is not a computer-readable storage medium and can communicate, propagate, or transmit a program used by or in conjunction with an instruction execution system, device, or apparatus.

[0106] Program code implemented on a computer-readable medium may be transmitted using any suitable medium, including but not limited to wireless, wired, fiber optic cable, RF, or any suitable combination thereof.

[0107] Computer program code for performing operations relating to various aspects of embodiments of the present invention can be written in any combination of one or more programming languages, including: object-oriented programming languages ​​such as Java, Smalltalk, C++, etc.; and conventional procedural programming languages ​​such as the "C" programming language or similar programming languages. The program code can be executed as a standalone software package entirely on the user's computer, partially on the user's computer, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the latter case, the remote computer can be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (e.g., via the Internet provided by an Internet service provider).

[0108] The flowchart illustrations and / or block diagrams of the methods, apparatus (systems), and computer program products according to embodiments of the present invention describe various aspects of the embodiments of the present invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus to produce a machine such that the instructions (executed via the processor of the computer or other programmable data processing apparatus) create means for implementing the functions / actions specified in the flowchart and / or block diagram blocks or blocks.

[0109] These computer program instructions may also be stored in a computer-readable medium that can direct a computer, other programmable data processing apparatus or other means to operate in a particular manner, such that the instructions stored in the computer-readable medium produce an article of writing that includes instructions that implement the functions / actions specified in flowchart and / or block diagram blocks or blocks.

[0110] Computer program instructions may also be loaded onto a computer, other programmable data processing apparatus or other device to cause a series of operable steps to be performed on the computer, other programmable apparatus or other device to produce a computer-implemented process, such that the instructions, which execute on the computer or other programmable apparatus, provide for implementing the functions / actions specified in flowchart and / or block diagram blocks or blocks.

[0111] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0112] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use, and processing of such data must comply with the relevant laws, regulations, and standards of the relevant countries and regions, and corresponding access points are provided for users to choose to authorize or refuse processing. A user's refusal to process personal information beyond what is necessary for basic functions will not affect the user's use of basic functions.

Claims

1. A method for generating population data, characterized in that, The method includes: Obtain user demand information; The user demand information is input into the first large language model, and the expanded user demand information is output. The first large language model is used to expand the input user demand information. The expanded user demand information is input into the second language model, and at least one target label is output, wherein the target label is an unstructured label extracted based on the expanded user demand information; Input the at least one target label into the representation model and output at least one target entity; Crowd data is determined based on the at least one target entity, wherein the crowd data is data that conforms to the user's needs information.

2. The method according to claim 1, characterized in that, The method further includes: The expanded user demand information is input into the second language model, and at least one basic tag is output, wherein the basic tag is a pre-set structured tag.

3. The method according to claim 2, characterized in that, The step of inputting the at least one target label into the representation model and outputting at least one target entity specifically includes: Input the at least one target label and the at least one base label into the representation model, and output at least one target entity.

4. The method according to claim 1, characterized in that, The step of inputting the at least one target label into the representation model and outputting at least one target entity further includes: Obtain information on multiple candidate entities; The at least one target label and the information of the plurality of candidate entities are input into the representation model, and at least one target entity is output.

5. The method according to claim 4, characterized in that, The candidate entity information includes short text information and / or long text information, wherein the short text information is generated by the candidate entity title information through the first large language model through understanding enhancement, and the long text information is generated by the short text information through the first large language model through understanding enhancement.

6. The method according to claim 1, characterized in that, The step of determining the population data based on the at least one target entity specifically includes: Initial population data is determined based on at least one target entity; Determine the interaction score between each person in the initial crowd data and the at least one target entity, and sort the people in the initial crowd data in descending order of the interaction scores; The individuals ranked at the top of the set ranking are identified as the population data.

7. The method according to claim 1, characterized in that, The step of inputting the expanded user demand information into the second language model and outputting at least one target label specifically includes: The expanded user demand information is input into the second language model, and at least one target label and at least one exemplary entity are output.

8. The method according to claim 1, characterized in that, The target label includes a label name and a label description.

9. The method according to claim 1, characterized in that, The second major language model comprises multiple agents, and its training process includes: The second major language model, which includes multiple agents, is trained based on a pre-set labeling system and sample data.

10. An apparatus for generating crowd data, characterized in that, The device includes: The acquisition unit is used to acquire user demand information; An extension unit is used to input the user demand information into a first large language model and output extended user demand information, wherein the first large language model is used to extend the input user demand information; The generation unit is used to input the expanded user demand information into the second language model and output at least one target label, wherein the target label is an unstructured label extracted based on the expanded user demand information; A matching unit is used to input the at least one target label into the representation model and output at least one target entity; A determining unit is configured to determine crowd data based on the at least one target entity, wherein the crowd data is data that conforms to the user demand information.

11. An electronic device comprising a memory and a processor, characterized in that, The memory is used to store one or more computer program instructions, wherein the one or more computer program instructions are executed by the processor to implement the method as described in any one of claims 1-9.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method as described in any one of claims 1-9.

Images