Audience selection method and apparatus, device, and storage medium

Abstract

Embodiments of the present application relate to the technical field of computers, and provide an audience selection method and apparatus, an electronic device, and a computer-readable storage medium. The method comprises: inputting first information into prompt engineering to construct first data, wherein the first information is related to an operational issue, and the first data comprises issue features, user features and data features; inputting the first data into a pre-trained large language model (LLM) to obtain a first index and a corresponding dimension; and on the basis of the first index, the dimension, and data corresponding to the dimension, performing a greedy algorithm to obtain an audience. The embodiments of the present application achieve the objective of improving the accuracy and efficiency of audience selection.

Description

Method, device, equipment and storage medium for crowd selection Technical Field

[0001] The present application relates to the field of computer technology, and more specifically, to a method, device, electronic device, and computer-readable storage medium for selecting a group of people. Background Art

[0002] Existing crowd selection usually requires a lot of manpower and time for manual screening, which is not only time-consuming and labor-intensive, but also prone to errors and omissions, resulting in inaccurate and inefficient crowd selection.

[0003] Summary of the Invention

[0004] The embodiments of the present application provide a method, device, electronic device, and computer-readable storage medium for crowd selection, which are used to solve the technical problems of low accuracy and efficiency in crowd selection.

[0005] According to one aspect of an embodiment of the present application, a method for selecting a group of people is provided, the method comprising:

[0006] Constructing first data by inputting first information into a prompt word project, wherein the first information is related to an operational problem, and the first data includes problem characteristics, user characteristics, and data characteristics;

[0007] Inputting the first data into a pre-trained large language model LLM to obtain a first indicator and a corresponding dimension;

[0008] A greedy algorithm is performed based on the first indicator and dimension, as well as the data corresponding to the dimension, to obtain a population package.

[0009] In one possible implementation, inputting the first data into a pre-trained LLM to obtain a first indicator and a corresponding dimension includes:

[0010] Input the first data into the LLM to obtain and display the second indicator;

[0011] determining the first indicator according to a first instruction, wherein the first instruction is used to instruct adjustment of the second indicator to obtain the first indicator;

[0012] According to the first indicator, second data is constructed in the prompt word project and input into the LLM to obtain corresponding dimensions.

[0013] In another possible implementation, the method further includes:

[0014] Second information is displayed, where the second information includes the physical meaning represented by the first indicator and dimension, and / or the reason for selecting the first indicator.

[0015] In yet another possible implementation, the method further includes:

[0016] A static materialized view is established based on the third information associated with the user ID in the crowd package, wherein the third information includes user portraits and user behavior data.

[0017] In yet another possible implementation, the method further includes:

[0018] Based on the static materialized view, a chart for reflecting the significant features of the crowd package is obtained and displayed.

[0019] In another possible implementation, a greedy algorithm is performed based on the first indicator and the dimension, as well as data corresponding to the dimension, to obtain a crowd package, including:

[0020] If the first indicator is one, all combinations of data corresponding to the dimension of the first indicator are determined as sub-populations based on a greedy algorithm;

[0021] Use the pre-calculated cube to perform statistical analysis on the data corresponding to the first indicator of each sub-population to obtain the indicator results of the corresponding sub-population;

[0022] According to the ranking of the indicator results of all sub-populations from high to low, a preset number of sub-populations ranked first are determined as the population package.

[0023] In another possible implementation, a greedy algorithm is performed based on the first indicator and the dimension, as well as data corresponding to the dimension, to obtain a crowd package, including:

[0024] If there are multiple first indicators, perform the following operations for each first indicator:

[0025] Based on a greedy algorithm, all combinations of data corresponding to the dimension of the first indicator are determined as sub-populations;

[0026] Use the pre-calculated cube to perform statistical analysis on the data corresponding to the first indicator of each sub-population to obtain the indicator results of the corresponding sub-population;

[0027] Performing weighted calculation on the indicator results of the sub-populations corresponding to all the first indicators according to the weights of the multiple first indicators to obtain the indicator results of all the sub-populations;

[0028] According to the ranking of the indicator results of all sub-populations from high to low, a preset number of sub-populations ranked first are determined as the population package.

[0029] In yet another possible implementation, the method further includes:

[0030] Adjusting parameters of the crowd package according to a second instruction, where the second instruction is used to instruct adjustment of parameters of the crowd package, the parameters including at least one of crowd size, crowd distribution, and crowd characteristics;

[0031] A greedy algorithm is performed based on the first indicator and dimension and the adjusted crowd data to redefine the crowd package.

[0032] In yet another possible implementation, the method further includes:

[0033] Fine-tune the original LLM based on the acquired training samples to obtain a trained LLM and a first result, wherein the training samples include: the second information and the corresponding third indicator and dimension, the first result includes the fourth indicator and dimension, and the second information is related to the operational issue;

[0034] Inputting the first information into the trained LLM to obtain a second result, wherein the second result includes a fifth indicator and a dimension;

[0035] Determining a bad example based on the first result, the second result, and the training sample;

[0036] Bad examples are labeled and used as new training samples for subsequent training.

[0037] In yet another possible implementation, determining a bad example according to the first result, the second result, and the training sample includes:

[0038] Comparing the same indicators in the first result and the second result with the indicators of the training samples respectively, and determining the indicators whose comparison results are lower than a first preset threshold as bad examples;

[0039] Comparing the dimensions in the first result and the second result with the dimensions of the training sample respectively, and determining the dimension whose comparison result is lower than a second preset threshold as a bad example;

[0040] Comparing the indicators in the first result and the second result with the indicators of the training sample respectively, and determining that the indicator is a bad example if the semantic similarity of the indicator whose comparison result is higher than the first preset threshold is lower than the first preset value;

[0041] Comparing the dimensions in the first result and the second result with the dimensions of the training sample respectively, and determining a dimension as a bad example if the semantic similarity of the dimension whose comparison result is higher than the second preset threshold is lower than the second preset value;

[0042] Comparing the indicators in the first result and the second result with the dimensions of the training sample respectively, and determining that the indicator is a bad example if the comparison result is lower than the first preset threshold and does not meet the first condition;

[0043] The dimensions in the first result and the second result are respectively compared with the dimensions of the training sample. If the dimension whose comparison result is lower than the second preset threshold does not meet the second condition, it is determined to be a bad example.

[0044] According to another aspect of an embodiment of the present application, a device for selecting a group of people is provided, the device comprising:

[0045] A data analysis module, configured to construct first data by inputting first information into a prompt word project, wherein the first information is related to an operational problem, and the first data includes problem characteristics, user characteristics, and data characteristics;

[0046] A model calculation module, configured to input the first data into a pre-trained large language model (LLM) to obtain a first indicator and a corresponding dimension;

[0047] The data analysis module is further configured to perform a greedy algorithm based on the first indicator and dimension, as well as data corresponding to the dimension, to obtain a population package.

[0048] According to another aspect of an embodiment of the present application, an electronic device is provided, comprising: a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the method for selecting a group of people described in the first aspect above.

[0049] According to another aspect of the embodiments of the present application, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method for selecting a group of people described in the first aspect are implemented.

[0050] The beneficial effects of the technical solution provided by the embodiments of the present application are:

[0051] By inputting first information related to operational issues into the prompt word project, first data is constructed. This first data includes problem characteristics, user characteristics, and data characteristics. This first data is then input into a pre-trained large language model (LLM) to obtain first indicators and corresponding dimensions. A greedy algorithm is then applied to these first indicators and dimensions, along with the data corresponding to these dimensions, to generate a population package. This method not only lowers the barrier to entry for operational personnel but also uses the large language model to break down operational issues described in natural language into reasonable target indicators and population dimensions, and combines these with the corresponding data to generate a population package, thereby improving the accuracy and efficiency of population package selection. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for describing the embodiments of the present application.

[0053] FIG1 is a system structure diagram of a method for selecting a group of people according to an embodiment of the present application;

[0054] FIG2 is a flow chart of a method for selecting a group of people provided in an embodiment of the present application;

[0055] FIG3 is a flow chart of a method for selecting a group of people provided in an embodiment of the present application;

[0056] FIG4 is a flow chart of a method for selecting a group of people provided in an embodiment of the present application;

[0057] FIG5 is a flow chart of a method for selecting a group of people provided in an embodiment of the present application;

[0058] FIG6 is a schematic diagram of the LLM training process in a method for selecting a group of people provided in an embodiment of the present application;

[0059] FIG7 is a schematic structural diagram of a device for selecting a group of people provided in an embodiment of the present application;

[0060] FIG8 is a schematic structural diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0061] The following describes the embodiments of the present application in conjunction with the accompanying drawings. It should be understood that the embodiments described below in conjunction with the accompanying drawings are exemplary descriptions for explaining the technical solutions of the embodiments of the present application and do not constitute a limitation on the technical solutions of the embodiments of the present application.

[0062] Those skilled in the art will understand that, unless otherwise stated, the singular forms "a", "an", "said" and "the" used herein may also include plural forms. It should be further understood that the terms "including" and "comprising" used in the embodiments of the present application mean that the corresponding features can be implemented as the presented features, information, data, steps, operations, elements and / or components, but do not exclude implementation as other features, information, data, steps, operations, elements, components and / or combinations thereof supported by the present technical field. It should be understood that when an element is said to be "connected" or "coupled" to another element, the element can be directly connected or coupled to the other element, or it can refer to the element and the other element establishing a connection relationship through an intermediate element. In addition, the "connection" or "coupling" used here can include wireless connection or wireless coupling. The term "and / or" used here indicates at least one of the items defined by the term, for example, "A and / or B" can be implemented as "A", or as "B", or as "A and B".

[0063] In order to make the objectives, technical solutions and advantages of this application clearer, the implementation methods of this application will be further described in detail below with reference to the accompanying drawings.

[0064] With existing technologies, crowd selection typically requires a significant amount of manpower and time for manual screening. This is not only time-consuming and labor-intensive, but also prone to errors and omissions. It also faces the following three major issues:

[0065] 1. The threshold for operators is high:

[0066] Traditional crowd selection requires heavy manual involvement. Operators must possess professional skills and a deep understanding of the business to accurately select qualified audiences. Furthermore, when selecting audiences, operators must break down operational issues into multiple target indicators and select the audience dimensions associated with each indicator. This requires in-depth business knowledge; otherwise, the selected audiences may not align with business objectives, thus affecting marketing effectiveness.

[0067] 2. Data real-time performance is poor:

[0068] Traditional crowd selection requires operators to manually screen target groups. However, these operators often struggle to accurately understand the connections between different indicators and cannot meet operational needs in real time, resulting in inaccurate and inefficient crowd selection. Furthermore, data from various data systems is not yet fully integrated, making it difficult to organically integrate, analyze, and process information from these disparate sources.

[0069] 3. Crowd selection takes a long time and has poor performance:

[0070] Traditional crowdsourcing methods require significant manpower and time for manual screening, resulting in long selection times and low efficiency. Furthermore, due to the sheer volume of data, some traditional algorithms may face performance bottlenecks and be unable to meet the demands of large-scale data processing.

[0071] In addition to the three major issues mentioned above, there are other issues, such as insufficiently precise crowd targeting and the lack of transparency in the black box of selected crowds. These issues affect the accuracy and efficiency of crowd selection and require new technical solutions.

[0072] Traditional crowd selection methods mainly include the following three selection logics:

[0073] a. Demographic-based crowd selection: This involves dividing different groups based on demographic characteristics, such as age, gender, region, and occupation. This approach typically relies on market research and statistical data analysis, combining different demographic characteristics to form different crowd classification criteria, which are then used to select groups.

[0074] b. Crowd selection based on behavioral data: Crowd selection is based on user behavioral data, such as search history, browsing behavior, and purchase history. By analyzing and mining this behavioral data, we can understand users' interests, hobbies, and needs, thereby achieving more accurate crowd selection.

[0075] c. Crowd selection based on social network data. This method utilizes user behavior data on social network platforms, such as user friendships, comment history, and sharing behavior, to achieve crowd selection. By analyzing and mining this social network data, we can understand users' interests and social relationships, thereby achieving more accurate crowd selection.

[0076] The selection logic for an operational problem depends entirely on the operator's understanding of the problem. Based on past experience, they operate the database to cross-reference people of different dimensions to obtain the final result.

[0077] The following describes several exemplary embodiments to illustrate the technical solutions of the embodiments of the present application and the technical effects produced by the technical solutions of the present application. It should be noted that the following embodiments can refer to, draw on, or combine with each other, and the same terms, similar features, and similar implementation steps in different embodiments will not be repeated.

[0078] The core issues of crowd selection mainly include: what are the indicators related to operational issues, what are the dimensions most relevant to these indicators; and how to quickly calculate crowd packages based on these indicators and dimensions.

[0079] As shown in FIG1 , the system for implementing a method for selecting a group of people provided in an embodiment of the present application mainly includes: a user interaction module, a data analysis module, a model calculation module, a data engine module and a business analysis module.

[0080] Before using the system, you can load data, which can include: loading data into the data engine module, performing relevant optimization, and establishing materialized views. In addition, during the data loading process, it is necessary to solve the problem of confusion in the user's actual data fields and build and maintain a logical indicator library.

[0081] Using the above system to obtain the appropriate crowd package for a specific operational problem requires the following steps:

[0082] Step 1: User 1 (who may be an operations staff) raises an operations question.

[0083] Step 2: Based on the user's question, the data analysis module uses prompt engineering to construct a detailed data question from User 1's vague question (for example, adding User 2's role (which can be the group to be circled), User 2's goal, data table items, etc.). This question is then transmitted to the Large Language Model (LLM). To enable the LLM to better understand the user's data, the actual indicators and dimension names must be mapped to logical indicators and dimensions.

[0084] Step 3: LLM outputs recommended indicators based on detailed data issues.

[0085] Step 4: User 1 evaluates whether the indicators recommended by LLM are reasonable based on the questions he or she raises. If not, he or she can specify the target indicators in the indicator library.

[0086] Step 5: The data analysis module uses the pre-aggregated data of the data engine based on the target indicators and related dimensions, and uses a greedy algorithm to drill down and calculate the indicator data of the sub-populations. Based on the calculated indicator ranking of the sub-populations, the top n are selected, where n is an integer greater than 2.

[0087] Step 6: The data analysis module associates the portrait, behavior and other data in the database based on the final calculated sub-populations and establishes a static materialized view.

[0088] Step 7: LLM performs natural language translation based on the target indicators and dimensions, so that operators can better understand the reasons for selecting these indicators and dimensions.

[0089] Step 8: In the Business Intelligence (BI) tool, establish an association between the dataset and the static materialized view, and create a new chart to display the data distribution, thereby achieving white-box processing of the crowd package and allowing operations personnel to better understand the details of the crowd package.

[0090] Step 9. After the operator receives the crowd package, he can make adjustments based on the actual situation. The adjustments may include adjustments to the crowd size, crowd distribution, and crowd characteristics.

[0091] Step 10. Repeat steps 5-9.

[0092] Therefore, the crowd selection solution based on a large language model proposed in the embodiment of the present application can lower the usage threshold for operators. It can break down operational problems described in natural language into reasonable target indicators and crowd dimensions, and provide more reasonable and effective crowd packages to operators for decision-making based on actual data conditions, thereby improving the accuracy and efficiency of crowd package selection.

[0093] In the technical solution of the embodiment of the present application, the performance of data analysis and calculation can be improved by constructing a pre-calculated cube in the data engine. Usually, when performing data analysis and calculation, it is necessary to associate (join) multiple tables. The performance of join determines the time consumption of ad hoc query calculation. By pre-making materialized views, commonly used wide tables are associated in advance, and a reasonable data update cycle is set. In the actual calculation process, the most time-consuming join part can be directly saved.

[0094] In addition, commonly used indicators and roll-up and drill-down calculations can also be pre-calculated in advance, and materialized views can be created for periodic updates, further reducing the time consumption of ad hoc queries.

[0095] Precomputed cubes are constructed in three ways. The first is synchronous construction during data loading. The second is construction based on preset rules. For example, this involves analyzing Structured Query Language (SQL) to count wide tables and metrics with the highest number of calculations. When a certain threshold is reached, a precomputed cube is created. The third is model construction. For example, this involves modeling the SQL within the system, evaluating the most valuable compute nodes, creating precomputed cubes for these high-value nodes, and promptly deleting existing precomputed cubes with lower value.

[0096] This embodiment of the present application provides a method for selecting a group of people, as shown in FIG2 , the method comprising:

[0097] S11. Constructing first data by inputting first information into a prompt word project, wherein the first information is related to an operational problem, and the first data includes problem characteristics, user characteristics, and data characteristics;

[0098] S12. Input the first data into a pre-trained large language model LLM to obtain a first indicator and a corresponding dimension;

[0099] S13. Perform a greedy algorithm based on the first indicator and dimension, as well as the data corresponding to the dimension, to obtain a crowd package.

[0100] The solution of the embodiment of the present application is to input the first information related to the operational problem into the prompt word project to construct the first data, which includes the problem characteristics, user characteristics and data characteristics. The first data is then input into the pre-trained large language model (LLM) to obtain the first indicator and the corresponding dimension. A greedy algorithm is then performed based on the first indicator and dimension, as well as the data corresponding to the dimension, to obtain a crowd package. This process not only lowers the threshold for operators, but also decomposes the operational problems described in natural language into reasonable target indicators and crowd dimensions through the large language model, and combines the corresponding data to obtain the crowd package, which can improve the accuracy and efficiency of crowd package selection.

[0101] In this embodiment, the prompt word engineering approach decomposes operational problems into three parts: problem features, user features, and data features. Problem features include the user's original question, the goal of the question (e.g., increasing the user's deposit rate), and relevant information about the question (e.g., operational strategy). User features are supplemented with relevant user characteristics based on the original question, such as user roles. Data features are relevant data used to solve the problem, such as feature importance analysis data and feature distribution data. Generally speaking, the detailed nature of the data features directly determines whether the LLM can infer the correct results.

[0102] It should be noted that, in this embodiment, the dimension corresponding to the first indicator may be one or more, and the dimension may also be called a population dimension.

[0103] In the embodiments of the present application, "a plurality of" refers to two or more. In some embodiments, terms such as "at least one (at least one item, at least one)", "one or more" and the like can be used interchangeably.

[0104] In the embodiment of the present application, the acquisition of data, information, etc. can comply with the laws and regulations of the country where the user is located. In the embodiment of the present application, the acquisition of data, information, etc. can be performed after obtaining the user's consent.

[0105] In addition, in this embodiment, the data features may also include some preset indicators and dimensions to provide reference for LLM. These preset indicators and dimensions may be obtained based on empirical solutions to actual operational problems.

[0106] In some embodiments, the above step S12 may include:

[0107] Input the first data into the LLM to obtain and display the second indicator;

[0108] determining the first indicator according to a first instruction, wherein the first instruction is used to instruct adjustment of the second indicator to obtain the first indicator;

[0109] According to the first indicator, second data is constructed in the prompt word project and input into the LLM to obtain corresponding dimensions.

[0110] In an embodiment of the present application, better results can be achieved through a step-by-step approach. For example, when selecting a group of people, you can first construct a prompt request (which can correspond to the first data above), obtain reasonable data analysis indicators, and then construct a prompt request (which can correspond to the second data above), add the data distribution under the indicators, to obtain the optimal dimension set, and the final dimensions will be more accurate.

[0111] In this embodiment, the first indicator may be completely identical to the second indicator, may be partially identical to the second indicator, or may be partially identical to a portion of the second indicator.

[0112] Exemplarily, the second indicator includes indicator 1 and indicator 2, and the first indicator includes indicator 1 and indicator 2; the second indicator includes indicator 1 and indicator 2, and the first indicator includes indicator 1, indicator 2, and indicator 3; the second indicator includes indicator 1 and indicator 2, and the first indicator includes indicator 1, or the first indicator includes indicator 1 and indicator 3.

[0113] It should be noted that in this embodiment, if the first indicator and the second indicator are exactly the same, the second data is consistent with the first data. If the first indicator and the second indicator are not exactly the same (for example, part of the first indicator is the same as the second indicator), the second data is data obtained based on the information of operational issues related to the first indicator entered in the prompt word project.

[0114] In this embodiment, the first instruction may also be used to instruct deletion of some of the second indicators, and / or to instruct acquisition of at least one indicator from the indicator library. That is, indicator 3 in the above example may be an indicator selected from the indicator library.

[0115] This embodiment of the present application provides a possible implementation method, as shown in FIG3 , the above method may further include:

[0116] S14. Display second information, where the second information includes the physical meaning represented by the first indicator and dimension, and / or the reason for selecting the first indicator.

[0117] In this embodiment, after the final indicators and corresponding dimensions are determined, they can be translated into natural language, and the physical meaning represented by the indicators and dimensions, and / or the reasons for selecting the indicators can be displayed to user 1, thereby helping user 1 to choose an operating strategy and improve user experience.

[0118] For example, indicator 1 can measure the user's purchasing activity on the platform. By analyzing this indicator, we can understand the user's purchasing power and consumption habits during the shopping process.

[0119] This embodiment of the present application provides a possible implementation method, as shown in FIG4 , the above method may further include:

[0120] S15. Establish a static materialized view based on the third information associated with the user ID in the crowd package, wherein the third information includes user portraits and user behavior data.

[0121] In this embodiment, a static materialized view can be established based on the user portrait and behavior data associated with the ID of user 2 in the crowd package, thereby preparing for subsequent white box processing.

[0122] This embodiment of the present application provides a possible implementation method, as shown in FIG4 , the above method may further include:

[0123] S16. Based on the static materialized view, a chart for reflecting the significant features of the crowd package is obtained and displayed.

[0124] In this embodiment, the data set can be associated with the static materialized view through the BI tool, and then a significant feature analysis can be performed to obtain and display the corresponding chart, so that the user 1 can intuitively see the distribution characteristics and specific distribution status of the population data, thereby improving the user experience.

[0125] The present application embodiment provides a possible implementation method, as shown in FIG5 , the above embodiment may further include:

[0126] S10: Determine whether to adjust the parameters of the crowd package. If yes, return to execute S13; otherwise, execute S14 and / or S15 and S16.

[0127] The parameters include at least one of crowd size, crowd distribution, and crowd characteristics.

[0128] Specifically, in this embodiment, if a second instruction input by user 1 is received, the corresponding parameters of the crowd package are adjusted. Exemplarily, the second instruction is used to instruct to expand the crowd size, or to reduce the crowd size, etc. Otherwise, the parameters of the crowd package may not be adjusted.

[0129] S13 may specifically be: performing a greedy algorithm according to the first indicator and dimension and the adjusted crowd data to re-determine the crowd package.

[0130] In the solutions of the above embodiments, step S13 may include the following two implementation solutions:

[0131] The first solution is the case where the number of the first indicator is one, specifically:

[0132] Based on a greedy algorithm, all combinations of data corresponding to the dimension of the first indicator are determined as sub-populations;

[0133] Use the pre-calculated cube to perform statistical analysis on the data corresponding to the first indicator of each sub-population to obtain the indicator results of the corresponding sub-population;

[0134] According to the order of the indicator results of all sub-populations from high to low, a preset number of sub-populations ranked first are determined as the population package.

[0135] For example, the first indicator includes indicator 1, and the dimensions corresponding to indicator 1 include dimensions 1, 2, and 3. Dimension 1 corresponds to 5 user data points, dimension 2 corresponds to 3 user data points, and dimension 3 corresponds to 4 user data points. Therefore, all possible combinations of data corresponding to the three dimensions, obtained using a greedy algorithm, are: 5*3*4 = 60 sub-populations. If a pre-computed cube is used to perform statistical analysis on the data corresponding to indicator 1 for each of these 60 sub-populations, the indicator results for all 60 sub-populations are obtained. These 60 sub-populations are then sorted from high to low, and the top five sub-populations are identified as the population group.

[0136] The second solution is when there are multiple first indicators, specifically:

[0137] For each first indicator, do the following:

[0138] Based on a greedy algorithm, all combinations of data corresponding to each dimension of the first indicator are determined as sub-populations;

[0139] Use the pre-calculated cube to perform statistical analysis on the data corresponding to the first indicator of each sub-population to obtain the indicator results of the corresponding sub-population;

[0140] Then, according to the weights of the multiple first indicators, the indicator results of the sub-populations corresponding to all the first indicators are weighted and calculated to obtain the indicator results of all the sub-populations;

[0141] According to the ranking of the indicator results of all sub-populations from high to low, a preset number of sub-populations ranked first are determined as the population package.

[0142] For example, the first indicator includes indicator 1 and indicator 2. The dimensions corresponding to indicator 1 include dimensions 1, 2, and 3, and the dimensions corresponding to indicator 2 include dimensions 4 and 5. The weight of indicator 1 is w1, and the weight of indicator 2 is w2. Dimension 1 corresponds to 5 user data points, dimension 2 corresponds to 3 user data points, and dimension 3 corresponds to 4 user data points. Therefore, all possible combinations of the data corresponding to the three dimensions, based on the greedy algorithm, are: 5*3*4 = 60 sub-populations. Dimension 4 corresponds to 10 user data points, and dimension 5 corresponds to 8 user data points. Therefore, all possible combinations of the data corresponding to the two dimensions, based on the greedy algorithm, are: 10*8 = 80 sub-populations.

[0143] For example, if we use the pre-calculated cube to perform statistical analysis on the data corresponding to indicator 1 for each of the 60 sub-populations, we obtain the indicator results for the 60 sub-populations. These 60 sub-population indicator results are then multiplied by w1 to obtain the weighted 60 indicator results. If we use the pre-calculated cube to perform statistical analysis on the data corresponding to indicator 2 for each of the 80 sub-populations, we obtain the indicator results for the 80 sub-populations. These 80 sub-population indicator results are then multiplied by w2 to obtain the weighted 80 indicator results. Then, we sort the weighted 60 indicator results and the weighted 80 indicator results from high to low, and identify the top 15 sub-populations as the population package.

[0144] In the solution of this application, by breaking down the operational problems described in natural language into reasonable target indicators and population dimensions, and combining the actual data situation with pre-calculation to improve the computing performance of population selection, a more reasonable and effective population package is provided to the operation personnel for decision-making, thereby improving the accuracy and efficiency of the population package selection.

[0145] It should be noted that when the number of dimensions corresponding to the first indicator is one, all user data under the dimension are traversed for statistical analysis to obtain the indicator results of the sub-populations, and the indicator results of these sub-populations are sorted from high to low, and the top n sub-populations are determined as the population package.

[0146] In some other embodiments, a training process of the LLM is further included, comprising the following steps:

[0147] Fine-tune the original LLM based on the acquired training samples to obtain a trained LLM and a first result, wherein the training samples include the second information and the corresponding third indicator and dimension, the first result includes the fourth indicator and dimension, and the second information is related to the operational issue;

[0148] Inputting the second information into the trained LLM to obtain a second result, wherein the second result includes the fifth indicator and the dimension;

[0149] Determining a bad example based on the first result, the second result, and the training sample;

[0150] Bad examples are labeled and used as new training samples for subsequent training.

[0151] In this embodiment, for the determined bad cases, the bad cases can be marked as expected results and used as new training samples. For example, if a bad case in the first result is determined to be the sixth indicator, it can be marked as the fourth indicator and used as a new training sample.

[0152] In some embodiments, bad cases may be determined by at least one of the following methods:

[0153] Method 1: Compare the indicators in the first result and the second result with the indicators of the training sample respectively, and determine the indicator whose comparison result is lower than a first preset threshold as a bad example;

[0154] Method 2: Compare the dimensions in the first result and the second result with the dimensions of the training sample respectively, and determine the dimension whose comparison result is lower than a second preset threshold as a bad example;

[0155] In this embodiment, the first preset threshold and the second preset threshold may be the same or different.

[0156] That is, samples whose hit rates relative to training samples in the first result and the second result are lower than a preset threshold may be determined as bad examples.

[0157] Method 3: Compare the indicators in the first result and the second result with the indicators of the training sample respectively. If the semantic similarity of the indicator whose comparison result is higher than the first preset threshold is lower than the first preset value, it is determined to be a bad example;

[0158] Method 4: Compare the dimensions in the first result and the second result with the dimensions of the training sample respectively. If the semantic similarity of the dimension whose comparison result exceeds the second preset threshold is less than the second preset value, it is determined to be a bad example.

[0159] In this embodiment, the first preset value and the second preset value may be the same or different.

[0160] That is, samples in the first result and the second result whose hit rates relative to the training samples are higher than a preset threshold are further semantically compared, and samples whose semantic similarity is lower than a preset value are determined as bad examples.

[0161] Method 5: Compare the indicators in the first result and the second result with the dimensions of the training sample respectively. If the indicator whose comparison result is lower than the first preset threshold does not meet the first condition, it is determined to be a bad example;

[0162] Method 6: Compare the dimensions in the first result and the second result with the dimensions of the training sample respectively. If the dimension whose comparison result is lower than the second preset threshold does not meet the second condition, it is determined to be a bad example.

[0163] In this embodiment, the first condition and the second condition may be manual screening conditions.

[0164] That is, samples in the first result and the second result whose hit rates relative to the training samples are lower than a preset threshold are further manually screened, and samples that do not meet the screening conditions are determined as bad examples.

[0165] In some embodiments, as shown in FIG6 , the LLM training process may specifically include:

[0166] 1. Original samples come from natural language processing tools and manual labor. Different operational problems are constructed using natural language processing tools (corresponding to the second information mentioned above), and the recommended indicators and dimensions of these operational problems are calculated using natural language processing tools. The results of the natural language processing tools and the operational problems are then combined into training samples.

[0167] Since this method can be automatically constructed continuously and utilizes the large corpus capabilities of natural language processing tools, it can supplement and optimize the basic capabilities of LLM. In addition, it can also supplement artificial samples. These samples come from actual problems of system users and also from manually annotated and constructed samples.

[0168] 2. Based on the original samples, LLM performs fine-tune training.

[0169] 3. After LLM training, perform inference calculations to obtain the model results for the same problem (which can correspond to the second result above).

[0170] 4. Perform semantic evaluation on the model results.

[0171] Specifically, for all samples, the original samples are used as the reference set, and the LLM results (which can correspond to the first and second results mentioned above) are used as the candidate set. Semantic similarity scores are calculated for the indicators and dimensions in the reference and candidate sets. If the scores of the same indicators and / or dimensions are below a preset threshold, it is considered a bad case; otherwise, it is a good case.

[0172] In this embodiment, the BLEU algorithm in the translation field may be used for semantic evaluation.

[0173] 5. For bad cases, the natural language processing tool can provide a set of similar questions and put them into the next iteration for training. The bad cases also need to be manually labeled and modified as new samples and put into the next round of iteration.

[0174] In the embodiment of the present application, the LLM is continuously trained and iterated through a semi-automated link to improve the computing effect of the model.

[0175] The following describes the implementation process of the technical solution provided in the embodiment of the present application with reference to an example.

[0176] Step 1: User 1 asks: How to improve user activity?

[0177] Step 2. The data analysis module uses prompt engineering to construct a more detailed data problem for this problem: I am an operations expert in the e-commerce field. I need to improve user activity and select a reasonable group of people for this problem. The data includes the following indicators {'sum_quantity', 'sum_booking', 'sum_revenue',}, dimensions {'customer_gender', 'order_status', 'part_brand', 'lineitem_returnflag', 'lineitem_linestatus', 'order_status', 'order_priority'}, and there are 1,000 users in total, including 400 male users and 600 female users. The sum_quantity indicator of male users is 2,100, and the sum_quantity indicator of female users is 3,400. Please give the relevant indicator set and dimension set.

[0178] That is to say, after the prompt word engineering, a simple operation problem is enriched into a professional and detailed operation problem, including as many user and data features as possible, among which,

[0179] User characteristics include: industry field: e-commerce field, role: operation expert, business goal: to improve activity, question target: to circle the crowd, time: question time.

[0180] Data features include: product user scale: 1,000 people, gender distribution: 400 male users, 600 female users, indicator distribution: sum_quantity indicator of male users is 2,100, and sum_quantity indicator of female users is 3,400. Data features are all actual data obtained through data analysis.

[0181] Among them, the indicators {'sum_quantity', 'sum_booking', 'sum_revenue'} and the dimensions {'customer_mktsegment', 'order_status', 'part_brand'} are all logical indicators and logical dimensions, which are converted using the pre-built logical indicator and dimension mapping table.

[0182] Step 3. Based on the above data problem, LLM returns the recommended indicator sum_quantity (which can correspond to the second indicator above) with dimensions {'customer_gender', 'order_status', 'part_brand'}.

[0183] Step 4: After viewing the recommended indicators, user 1 believes that the indicators are reasonable based on his or her historical experience, but the part_brand dimension is not reasonable, so he or she adjusts it to the lineitem_return flag.

[0184] Step 5: The data analysis module begins greedy drilling down based on this indicator and the dimension set. In the constructed materialized view, it explores all possible combinations of the three dimensions. Each combination represents a sub-population. Statistical analysis is performed on the sum_quantity indicator of each sub-population. As many sub-populations as possible are traversed, sorted by indicator results, and the top n sub-population packages are selected as the final population.

[0185] Because materialized views are prepared in advance, data is updated in a timely manner, and indicator calculation results can be obtained with performance far higher than traditional relational databases and offline data warehouse batch computing.

[0186] Step 6: The data analysis module associates the user profile and behavior data based on the user ID of the final calculated population package, and stores it as a static materialized view in the data engine for subsequent BI analysis.

[0187] Step 7: LLM performs natural language translation based on the finally selected indicators and dimensions, and provides an explanation (which may correspond to the second information above) to help user 1 understand the reasons for selecting the indicators and dimensions.

[0188] For example: {'Based on the given indicators and dimensions, the following reasonable explanations can be given:

[0189] "sum_quantity" (total purchase quantity): This metric measures user purchasing activity on the platform. By analyzing this metric, we can understand users' purchasing power and spending habits during the shopping process.

[0190] Dimension "customer_gender": Gender is an important dimension that helps understand the differences in purchasing behavior between male and female users. By comparing the purchase volume of users of different genders, we can determine which gender is more likely to purchase products on e-commerce platforms.

[0191] Dimension "order_status": Order status refers to the current status of a user's order, such as pending, shipped, or completed. By analyzing the purchase quantity under different order statuses, we can understand users' purchasing intentions and behaviors at different stages.

[0192] The dimension "lineitem_returnflag" can be used to assess user satisfaction and repurchase intention, identify product quality issues, and implement personalized measures to increase user engagement. Analyzing user groups with high return rates can improve product quality, provide better after-sales support, and rebuild user trust. Personalized marketing strategies, such as coupons and compensation, can be implemented for returning users to encourage repeat purchases. This is a crucial dimension for enhancing user satisfaction, sustainable business growth, and user loyalty.

[0193] In summary, by analyzing the metric "sum_quantity" and the dimensions "customer_gender", "order_status", and "lineitem_returnflag", we can gain in-depth insights into user activity. This data can help identify the purchasing preferences, purchase stages, and post-purchase feedback of specific demographics, thereby optimizing operational strategies and improving overall user activity.

[0194] Step 8. After creating a static physical view for business analysis in Step 6, you need to display the selected demographic group's performance. Based on the significance analysis and calculation of all dimensions of the demographic group, it is determined that the demographic group has significant characteristics in terms of regional and age distribution. Two charts are displayed: one showing the regional distribution, indicating the provinces and cities where the selected demographic group is primarily concentrated, and the other showing the age distribution of the demographic group, which can be compared with the overall population.

[0195] Step 9: After user 1 views the crowd package, he can make further adjustments.

[0196] For example, to expand the population size to 10,000 people, the data analysis module uses the seed population (which corresponds to the population package obtained in step 5) as the positive sample and the non-seed population as the negative sample to train the model to obtain the LR model. The non-seed population is then calculated and sorted by score, and the top 10,000 are taken as the new population package.

[0197] In this embodiment, user 1 obtains an ideal crowd package from a vague operational problem through simple interaction. LLM can quickly obtain a reasonable crowd package for operational activity planning based on large corpus data, the user's problem, the real-time status of the data, and the historical experience of other users (cases recorded in the system are fed back to model training) with a low threshold.

[0198] The embodiment of the present application provides a device for selecting a crowd. As shown in FIG7 , the device for selecting a crowd may include: a data analysis module 101 and a model calculation module 102, wherein:

[0199] The data analysis module 101 is configured to construct first data by inputting first information into a prompt word project, wherein the first information is related to an operational problem, and the first data includes problem characteristics, user characteristics, and data characteristics;

[0200] The model calculation module 102 is used to input the first data into a pre-trained large language model LLM to obtain a first indicator and a corresponding dimension;

[0201] The data analysis module 101 is further configured to perform a greedy algorithm based on the first indicators and dimensions, as well as data corresponding to each dimension, to obtain a population package.

[0202] Furthermore, the model calculation module 102 is specifically configured to:

[0203] Input the first data into the LLM to obtain and display the second indicator;

[0204] determining the first indicator according to a first instruction, wherein the first instruction is used to instruct adjustment of the second indicator to obtain the first indicator;

[0205] According to the first indicator, second data is constructed in the prompt word project and input into the LLM to obtain corresponding dimensions.

[0206] Furthermore, the above-mentioned device also includes: a user interaction module, which is used to display second information, wherein the second information includes the physical meaning represented by the first indicator and dimension, and / or the reason for selecting the first indicator.

[0207] It should be noted that, in this embodiment, the user interaction module is also used to receive the first information input by the user.

[0208] Furthermore, the above-mentioned device also includes: a data engine module, which is used to establish a static materialized view based on the third information associated with the user ID in the crowd package, wherein the third information includes user portraits and user behavior data.

[0209] Furthermore, the above-mentioned device also includes: a business analysis module, which is used to obtain and display a chart that reflects the significant characteristics of the population group based on the static materialized view.

[0210] Furthermore, the data analysis module 101 is specifically configured to:

[0211] If the number of the first indicator is one, all combinations of data corresponding to each dimension of the first indicator are determined as sub-populations based on a greedy algorithm;

[0212] Use the pre-calculated cube to perform statistical analysis on the data corresponding to the first indicator of each sub-population to obtain the indicator results of the corresponding sub-population;

[0213] According to the ranking of the indicator results of all sub-populations from high to low, a preset number of sub-populations ranked first are determined as the population package.

[0214] Furthermore, the data analysis module 101 is specifically configured to: if there are multiple first indicators, perform the following operations for each first indicator:

[0215] Based on a greedy algorithm, all combinations of data corresponding to each dimension of the first indicator are determined as sub-populations;

[0216] Use the pre-calculated cube to perform statistical analysis on the data corresponding to the first indicator of each sub-population to obtain the indicator results of the corresponding sub-population;

[0217] Performing weighted calculation on the indicator results of the sub-populations corresponding to all the first indicators according to the weights of the multiple first indicators to obtain the indicator results of all the sub-populations;

[0218] According to the ranking of the indicator results of all sub-populations from high to low, a preset number of sub-populations ranked first are determined as the population package.

[0219] Furthermore, the data analysis module 101 is also used to:

[0220] Adjusting parameters of the crowd package according to a second instruction, where the second instruction is used to instruct adjustment of parameters of the crowd package, the parameters including at least one of crowd size, crowd distribution, and crowd characteristics;

[0221] A greedy algorithm is performed based on the first indicator and dimension and the adjusted crowd data to redefine the crowd package.

[0222] Furthermore, the model calculation module 102 is further configured to: fine-tune the original LLM based on the acquired training samples to obtain a trained LLM and a first result, wherein the training samples include: the second information and the corresponding third indicator and dimension, the first result includes the fourth indicator and dimension, and the second information is related to the operational problem;

[0223] Inputting the first information into the trained LLM to obtain a second result, wherein the second result includes a fifth indicator and a dimension;

[0224] Determining a bad example based on the first result, the second result, and the training sample;

[0225] Bad examples are labeled and used as new training samples for subsequent training.

[0226] Furthermore, the model calculation module 102 is specifically configured to determine bad cases by at least one of the following methods:

[0227] Comparing the indicators in the first result and the second result with the indicators of the training sample respectively, and determining the indicators whose comparison results are lower than a first preset threshold as bad examples;

[0228] Comparing the dimensions in the first result and the second result with the dimensions of the training sample respectively, and determining the dimension whose comparison result is lower than a second preset threshold as a bad example;

[0229] Comparing the indicators in the first result and the second result with the indicators of the training sample respectively, and determining that the indicator is a bad example if the semantic similarity of the indicator whose comparison result is higher than the first preset threshold is lower than the first preset value;

[0230] Comparing the dimensions in the first result and the second result with the dimensions of the training sample respectively, and determining a dimension as a bad example if the semantic similarity of the dimension whose comparison result is higher than the second preset threshold is lower than the second preset value;

[0231] Comparing the indicators in the first result and the second result with the dimensions of the training sample respectively, and determining that the indicator is a bad example if the comparison result is lower than the first preset threshold and does not meet the first condition;

[0232] The dimensions in the first result and the second result are respectively compared with the dimensions of the training sample. If the dimension whose comparison result is lower than the second preset threshold does not meet the second condition, it is determined to be a bad example.

[0233] The device of the embodiment of the present application can execute the method provided by the embodiment of the present application, and its implementation principle is similar. The actions performed by each module in the device of each embodiment of the present application correspond to the steps in the method of each embodiment of the present application. For the detailed functional description of each module of the device, please refer to the description in the corresponding method shown in the previous text, and will not be repeated here.

[0234] It should be noted that the modules involved in the embodiments of the present application can be implemented by software or by hardware. Among them, the name of the module does not constitute a limitation on the module itself under certain circumstances. For example, the data analysis module can also be described as "a module that constructs the first data by inputting the first information into the prompt word project", or "a module that obtains the crowd package by performing a greedy algorithm based on the first indicator and dimension, and the data corresponding to each dimension". In other words, the modules used to implement different functions can be integrated into one module, or they can be different modules that are independent of each other. The names of the modules in the embodiments of the present application are for illustrative purposes only, and are not used to limit the functions of these modules.

[0235] The functions described above herein may be performed, at least in part, by one or more hardware logic components. For example, and without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chip (SOCs), complex programmable logic devices (CPLDs), and the like.

[0236] An embodiment of the present application provides an electronic device (computer device / equipment / system), including a memory, a processor, and a computer program stored on the memory, wherein the processor executes the above-mentioned computer program to implement the steps of a method for crowd selection. Compared with the related art, it can be achieved as follows: by inputting first information related to the operational problem into a prompt word project, constructing first data, the first data including problem features, user features, and data features, and then inputting the first data into a pre-trained large language model LLM to obtain a first indicator and a corresponding dimension, and performing a greedy algorithm based on the first indicator and dimension, as well as the data corresponding to each dimension, to obtain a crowd package. This process can not only lower the threshold for operators, but also decompose the operational problems described in natural language into reasonable target indicators and crowd dimensions through a large language model, and obtain crowd packages in combination with the corresponding data, which can improve the accuracy and efficiency of crowd package selection.

[0237] In an optional embodiment, an electronic device is provided, as shown in FIG8 , and the electronic device 4000 shown in FIG8 includes: a processor 4001 and a memory 4003. The processor 4001 and the memory 4003 are connected, such as through a bus 4002. Optionally, the electronic device 4000 may further include a transceiver 4004, which may be used for data interaction between the electronic device and other electronic devices, such as data transmission and / or data reception. It should be noted that in actual applications, the transceiver 4004 is not limited to one, and the structure of the electronic device 4000 does not constitute a limitation on the embodiments of the present application.

[0238] Processor 4001 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It may implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 4001 may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, and the like.

[0239] Bus 4002 may include a path for transmitting information between the aforementioned components. Bus 4002 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, for example. Bus 4002 may be divided into an address bus, a data bus, a control bus, and the like. For ease of illustration, FIG8 shows a single thick line, but this does not imply that there is only one bus or only one type of bus.

[0240] The memory 4003 may be a ROM (Read Only Memory) or other types of static storage devices that can store static information and instructions, a RAM (Random Access Memory) or other types of dynamic storage devices that can store information and instructions, or an EEPROM (Electrically Erasable Programmable Read Only Memory), a CD-ROM (Compact Disc Read Only Memory) or other optical disk storage, optical disc storage (including compact discs, laser discs, optical discs, digital versatile discs, Blu-ray discs, etc.), magnetic disk storage media, other magnetic storage devices, or any other medium that can be used to carry or store computer programs and can be read by a computer, without limitation herein.

[0241] The memory 4003 is used to store the computer program for executing the embodiment of the present application, and the execution is controlled by the processor 4001. The processor 4001 is used to execute the computer program stored in the memory 4003 to implement the steps shown in the above method embodiment.

[0242] An embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps and corresponding contents of the aforementioned method embodiment can be implemented.

[0243] An embodiment of the present application also provides a computer program product, including a computer program, which can implement the steps and corresponding contents of the aforementioned method embodiment when executed by a processor.

[0244] The terms "first," "second," "third," "fourth," "1," "2," and the like (if any) in the specification and claims of this application and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a particular order or sequential sequence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments of the application described herein can be implemented in an order other than that shown or described in the drawings.

[0245] It should be understood that, although each operation step is indicated by arrows in the flowchart of the embodiment of the present application, the order of implementation of these steps is not limited to the order indicated by the arrows. Unless otherwise clearly stated herein, in some implementation scenarios of the embodiment of the present application, the implementation steps in each flowchart can be performed in other orders according to demand. In addition, some or all of the steps in each flowchart can include multiple sub-steps or multiple stages based on actual implementation scenarios. Some or all of these sub-steps or stages can be executed at the same time, and each sub-step or stage in these sub-steps or stages can also be executed at different times respectively. Under different scenarios at the execution time, the execution order of these sub-steps or stages can be flexibly configured according to demand, and the embodiment of the present application does not limit this.

[0246] The above description is only an optional implementation method for some implementation scenarios of this application. It should be pointed out that for ordinary technicians in this technical field, without departing from the technical concept of the solution of this application, the use of other similar implementation methods based on the technical ideas of this application also falls within the protection scope of the embodiments of this application.

Claims

1. A method for selecting a group of people, characterized in that: include: Constructing first data by inputting first information into a prompt word project, wherein the first information is related to an operational problem, and the first data includes problem characteristics, user characteristics, and data characteristics; Inputting the first data into a pre-trained large language model LLM to obtain a first indicator and a corresponding dimension; A greedy algorithm is performed according to the first indicator and dimension, as well as the data corresponding to the dimension, to obtain a population package.

2. The method according to claim 1, characterized in that The first data is input into the pre-trained LLM to obtain a first indicator and a corresponding dimension, including: Input the first data into the LLM to obtain and display the second indicator; Determine the first indicator according to a first instruction, wherein the first instruction is used to instruct to adjust the second indicator to obtain the first indicator; According to the first indicator, second data is constructed in the prompt word project and input into the LLM to obtain corresponding dimensions.

3. The method according to claim 1, characterized in that Also includes: The second information is displayed, wherein the second information includes the physical meaning represented by the first indicator and dimension, and / or the reason for selecting the first indicator.

4. The method according to claim 1, characterized in that Also includes: A static materialized view is established based on the third information associated with the user ID in the crowd package, wherein the third information includes user portraits and user behavior data.

5. The method according to claim 4, characterized in that Also includes: Based on the static materialized view, a chart for reflecting the significant features of the crowd package is obtained and displayed.

6. The method according to any one of claims 1 to 5, characterized in that A greedy algorithm is performed based on the first indicator and dimension, as well as the data corresponding to the dimension, to obtain a crowd package, including: If the number of the first indicator is one, all combinations of data corresponding to the dimension of the first indicator are determined as sub-populations based on a greedy algorithm; Use the pre-calculated cube to perform statistical analysis on the data corresponding to the first indicator of each sub-population to obtain the indicator results of the corresponding sub-population; According to the ranking of the indicator results of all sub-populations from high to low, a preset number of sub-populations ranked first are determined as the population package.

7. The method according to any one of claims 1 to 5, characterized in that A greedy algorithm is performed based on the first indicator and dimension, as well as the data corresponding to the dimension, to obtain a crowd package, including: If there are multiple first indicators, for each first indicator, perform the following operations: Based on a greedy algorithm, all combinations of data corresponding to the dimension of the first indicator are determined as sub-populations; Use the pre-calculated cube to perform statistical analysis on the data corresponding to the first indicator of each sub-population to obtain the indicator results of the corresponding sub-population; According to the weights of the multiple first indicators, weighted calculation is performed on the indicator results of the sub-populations corresponding to all the first indicators to obtain the indicator results of all the sub-populations; According to the ranking of the indicator results of all sub-populations from high to low, a preset number of sub-populations ranked first are determined as the population package.

8. The method according to any one of claims 1 to 5, characterized in that Also includes: According to a second instruction, adjusting the parameters of the crowd package, the second instruction is used to instruct to adjust the parameters of the crowd package, the parameters including at least one of the size of the crowd, the distribution of the crowd, and the characteristics of the crowd; A greedy algorithm is performed according to the first indicator and dimension and the adjusted crowd data to redefine the crowd package.

9. The method according to claim 1, characterized in that: Also includes: Fine-tune the original LLM according to the acquired training samples to obtain the trained LLM and the first result, wherein the training samples include: the second information and the corresponding third indicator and dimension, the first result includes the fourth indicator and dimension, and the second information is related to the operation problem; Inputting the second information into the trained LLM to obtain a second result, wherein the second result includes a fifth indicator and a dimension; Determine a bad example according to the first result, the second result, and the training sample; Bad examples are labeled and used as new training samples for subsequent training.

10. The method according to claim 9, characterized in that Determining a bad example according to the first result, the second result, and the training sample includes at least one of the following methods: Compare the indicators in the first result and the second result with the indicators of the training sample respectively, and determine the indicators whose comparison results are lower than the first preset threshold as bad examples; Compare the dimensions in the first result and the second result with the dimensions of the training sample respectively, and determine the dimensions whose comparison results are lower than a second preset threshold as bad examples; Compare the indicators in the first result and the second result with the indicators of the training sample respectively. The semantic similarity of the indicator whose result is higher than the first preset threshold is less than the first preset value and is determined to be a bad example; Compare the dimensions in the first result and the second result with the dimensions of the training sample respectively, and if the semantic similarity of the dimension whose comparison result is higher than the second preset threshold is lower than the second preset value, determine it as a bad example; Compare the indicators in the first result and the second result with the dimensions of the training sample respectively, and if the indicator whose comparison result is lower than the first preset threshold does not meet the first condition, determine it as a bad example; The dimensions in the first result and the second result are compared with the dimensions of the training sample respectively. If the dimension whose comparison result is lower than the second preset threshold does not meet the second condition, it is determined to be a bad example.

11. A device for selecting a group of people, characterized in that: include: A data analysis module, configured to construct first data by inputting first information into a prompt word project, wherein the first information is related to an operation problem, and the first data includes problem characteristics, user characteristics, and data characteristics; A model calculation module, used for inputting the first data into a pre-trained large language model LLM to obtain a first indicator and a corresponding dimension; The data analysis module is further used to perform a greedy algorithm based on the first indicator and dimension, and data corresponding to the dimension, to obtain a population package.

12. An electronic device comprising: A memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the method for selecting a group of people as described in any one of claims 1-10.

13. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for selecting a group of people according to any one of claims 1 to 10 are implemented.

Images