Service pushing method and device and related equipment

Abstract

The invention provides a service pushing method and device and related equipment, and relates to the technical field of artificial intelligence. According to the technical scheme of the invention, historical service data of a target cell before a target time period is acquired; therefore, the target service type which is most suitable for being pushed to the target cell is determined according to the plurality of data dimensions of the historical service data, so that the pre-trained large language model is used for generating a pushing scheme corresponding to the target service type, and the target service type is pushed. The most suitable service pushing scheme is generated for the pushing cell during service pushing, and the service pushing effect is improved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, specifically to a method, apparatus, and related equipment for pushing services. Background Technology

[0002] Telecommunications service grid operation and marketing involves telecom operators dividing their service areas into several finely managed "grids." Based on data such as geography, population, and consumer behavior, differentiated service push strategies are developed for each grid. This refined and data-driven marketing management approach is a crucial means of increasing market share. In traditional marketing campaign generation methods, operators rely solely on basic metrics such as historical user numbers and revenue, lacking real-time scenario data to develop targeted strategies, resulting in poor performance in pushing services. Summary of the Invention

[0003] This application provides a method, apparatus, and related equipment for pushing services, which solves the problem of poor service push performance in related technologies.

[0004] To solve the above problems, this application is implemented as follows:

[0005] In a first aspect, embodiments of this application provide a method for pushing services, the method comprising:

[0006] Obtain historical service data of the target cell before the target time period, wherein the historical service data includes service processing data corresponding to multiple service types in the target cell;

[0007] Based on the historical business data, a target business type is determined among the multiple business types. Among the multiple business types, different business types correspond to different business data. The target business type is the business type with the largest number of users processed within the target time period among the multiple business types.

[0008] The target business type is input into a pre-trained large language model to generate a solution, thereby obtaining a business push solution corresponding to the target business type.

[0009] During the target time period, services are pushed to the target cell based on the service push scheme.

[0010] Optionally, obtaining the historical service data of the target cell before the target time period includes:

[0011] The business opportunity data of the target community before the target time period is evaluated to obtain first business opportunity data and second business opportunity data. The first business opportunity data is the business opportunity data corresponding to a first business type of the target community, and the second business opportunity data is the business opportunity data corresponding to a second business type of the target community. The multiple business types include the first business type and the second business type. The user churn rate of the first business type in the target community is lower than the user churn rate of the second business type in the target community. The business opportunity data is used to indicate the number of potential users corresponding to the corresponding business type in the target community.

[0012] Predict the pedestrian flow distribution and spatiotemporal scene of the target cell before the target time period to obtain pedestrian flow prediction data and spatiotemporal scene detail data. The spatiotemporal scene includes the geographical location and scene events corresponding to the target cell. The scene events are base station-related events that occur in the target cell.

[0013] The business opportunity value of the target community before the target time period is evaluated to obtain the total business opportunity efficiency data;

[0014] The marketing effectiveness of the target community before the target time period is evaluated to obtain historical marketing data;

[0015] The historical business data includes the first business opportunity data, the second business opportunity data, the traffic flow prediction data, the spatiotemporal scenario detail data, the total business opportunity efficiency data, and the historical marketing data.

[0016] Optionally, determining the target business type from the plurality of business types based on the historical business data includes:

[0017] The historical business data is filled into a preset data table, which includes a first indicator, a second indicator, a third indicator, and a fourth indicator. The influence of the data corresponding to the first indicator in the preset data table is lower than that of the data corresponding to the second indicator in the preset data table. The influence of the data corresponding to the second indicator in the preset data table is lower than that of the data corresponding to the third indicator in the preset data table. The influence of the data corresponding to the third indicator in the preset data table is lower than that of the data corresponding to the fourth indicator in the preset data table.

[0018] The data matrices corresponding to the first indicator, the second indicator, and the third indicator are respectively subjected to positive transformation processing to convert the first indicator, the second indicator, and the third indicator into the fourth indicator in the preset data table;

[0019] According to the fourth indicator, the multiple service types are scored to obtain multiple score values. The multiple score values ​​correspond one-to-one with the multiple service types. The score value is used to indicate the push conversion rate of the corresponding service type in the target cell. The push conversion rate is the ratio of the subscription volume to the push volume of the corresponding service type.

[0020] The business type corresponding to the highest score among the multiple score values ​​is determined as the target business type.

[0021] Optionally, the scoring of the multiple business types based on the fourth indicator yields multiple score values, including:

[0022] Obtain the number of business marketing events and the duration of business marketing events for the target cell before the target time period;

[0023] In the preset data table, based on the indicator type and indicator format of the second indicator, the number of business marketing events and the business marketing time are converted to obtain the fifth indicator;

[0024] The data matrix corresponding to the fifth indicator is positiveized to obtain the target matrix;

[0025] Based on the data matrix corresponding to the fourth indicator and the target matrix, the push conversion rate corresponding to the multiple business types is scored to obtain multiple score values.

[0026] Optionally, before inputting the target business type into a pre-trained large language model to generate a solution and obtain the business push solution corresponding to the target business type, the method further includes:

[0027] Obtain feature data of the target cell prior to the target time period, the feature data including traffic usage information, consumption information and bandwidth information corresponding to the target cell;

[0028] Based on the feature data, the historical business data, and the target time period, the initial large language model is trained to obtain the large language model.

[0029] Optionally, the step of training the initial large language model based on the feature data, the historical business data, and the target time period to obtain the large language model includes:

[0030] The feature data, the historical business data, and the target time period are filled into a preset prompt word template to obtain the prompt word;

[0031] The prompt words are input into the initial large language model, and the initial large language model is trained based on the low-rank adaptive algorithm LoRA to obtain the large language model.

[0032] Secondly, embodiments of this application also provide a service push device, the device comprising:

[0033] The acquisition module is used to acquire historical service data of the target cell before the target time period. The historical service data includes service processing data corresponding to multiple service types in the target cell.

[0034] The determination module is used to determine a target business type among the multiple business types based on the historical business data. Among the multiple business types, the business data corresponding to different business types are different. The target business type is the business type with the largest number of users processed within the target time period among the multiple business types.

[0035] The generation module is used to input the target business type into a pre-trained large language model to generate a solution, thereby obtaining a business push solution corresponding to the target business type.

[0036] The push module is used to push services to the target cell based on the service push scheme within the target time period.

[0037] Thirdly, this application also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the method described in the first aspect above.

[0038] Fourthly, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method described in the first aspect above.

[0039] Fifthly, this application also provides a computer program product, including computer instructions that, when executed by a processor, implement the steps of the method described in the first aspect above.

[0040] This application provides a service push method, apparatus, and related equipment, relating to the field of artificial intelligence technology. The method includes: acquiring historical service data of a target cell before a target time period, the historical service data including service processing data corresponding to multiple service types within the target cell; determining a target service type among the multiple service types based on the historical service data, wherein different service types correspond to different service data, and the target service type is the service type with the most users processed within the target time period; inputting the target service type into a pre-trained large language model to generate a scheme, obtaining a service push scheme corresponding to the target service type; and pushing the service to the target cell based on the service push scheme within the target time period. The technical solution of this application, by acquiring historical service data of a target cell before a target time period, determines the most suitable target service type to push to the target cell based on multiple data dimensions of the historical service data, and then uses a pre-trained large language model to generate a push scheme corresponding to the target service type to push the target service type. This achieves the generation of the most suitable service push scheme for the push cell during service push, improving the effectiveness of service push. Attached Figure Description

[0041] To more clearly illustrate the technical solution of this application, the drawings used in the description of this application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 A flowchart illustrating a service push method provided in an embodiment of this application;

[0043] Figure 2 This application provides a system architecture diagram for generating a service push application.

[0044] Figure 3 A flowchart for generating business recommendations provided in the embodiments of this application;

[0045] Figure 4 A schematic diagram of the structure of a service push device provided in an embodiment of this application;

[0046] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0047] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0048] The terms "first," "second," etc., used in the embodiments of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products, or devices. Additionally, the use of "and / or" in this application indicates at least one of the connected objects, such as A and / or B and / or C, representing seven possibilities: including A alone, B alone, C alone, and the presence of both A and B, both B and C, both A and C, and the presence of A, B, and C.

[0049] See Figure 1 , Figure 1 This is a flowchart illustrating the push method for services provided in an embodiment of this application. For example... Figure 1 As shown, the method for pushing services may include the following steps:

[0050] Step 101: Obtain historical business data of the target cell before the target time period. The historical business data includes business processing data corresponding to multiple business types in the target cell.

[0051] In this embodiment, the target cell is the cell that needs to be pushed with services. By analyzing the target cell, historical service data of the target cell in multiple dimensions before the target time period is obtained. The target time period is the time period during which services need to be pushed to the target cell.

[0052] Historical business data may include key business performance assessments, corresponding business opportunity tags, user hourly locations, historical grid business opportunities, etc. This embodiment does not impose specific limitations. By associating and processing marketing data from different dimensions, the comprehensiveness and accuracy of marketing campaigns can be improved.

[0053] Step 102: Based on the historical business data, determine the target business type among the multiple business types. Among the multiple business types, the business data corresponding to different business types are different. The target business type is the business type with the largest number of users processed within the target time period among the multiple business types.

[0054] In this embodiment, based on the acquired historical business data, the target business type most suitable for being pushed to the target cell within a target time period is determined from multiple different business types. For example, in this embodiment, the Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) can be used to rank multiple business types. Specifically, TOPSIS is a comprehensive evaluation and analysis method for multiple indicators and solutions, which can fully utilize raw data information and accurately reflect the differences between various evaluation solutions. This module summarizes the outputs of each specialized model and, through TOPSIS multi-indicator comprehensive evaluation, quantifies and ranks the optimal marketing strategy to inform grid operators of the best marketing time, location, and content.

[0055] It should be noted that the target service type in this embodiment is the service type that has the best push effect on the target cell within the target time period.

[0056] Step 103: Input the target business type into the pre-trained large language model to generate a scheme, and obtain the business push scheme corresponding to the target business type.

[0057] In this embodiment, the generated target business type is input into a trained Large Language Model (LLM). The LLM automatically generates a push notification scheme, resulting in a business push notification scheme corresponding to the target business type. The Large Language Model refers to a deep learning model trained on a large amount of text data, enabling it to generate natural language text or understand the meaning of language text. The LLM can provide in-depth knowledge and language production on various topics through training on massive datasets. Its core idea is to learn the patterns and structures of natural language through large-scale unsupervised training, simulating the human language cognition and generation process to a certain extent.

[0058] This embodiment builds a marketing campaign generation model based on a large language model. The large language model possesses powerful semantic understanding, logical reasoning, and language generation capabilities, effectively improving the completeness and readability of the intelligent recommendation marketing campaign text results. Simultaneously, the system employs an efficient parameter fine-tuning algorithm to train the large language model, reducing the training and iteration costs.

[0059] Step 104: During the target time period, push services to the target cell based on the service push scheme.

[0060] In this embodiment, within a given time period, a service push plan is used to push services to the target community. For example, the generated service push plan is sent to grid operation personnel, who then push the plan to the target community. This application leverages multi-dimensional data from grid-based operation scenarios, including comprehensive indicators such as pedestrian flow characteristics, business opportunities, and scenario efficiency, combined with multi-criteria decision analysis and intelligent semantic generation technology, to accurately identify the business shortcomings and potential growth points of each grid. On one hand, by quantifying the impact of multi-dimensional factors on marketing effectiveness and combining the TOPSIS algorithm to quickly identify the optimal marketing plan, it achieves scientific allocation of marketing resources and dynamic optimization of marketing strategies, thereby significantly improving marketing conversion rates and business growth rates. On the other hand, relying on the semantic understanding and reasoning capabilities of a large language model, it automatically generates detailed and clear marketing scripts, providing intelligent recommendation services throughout the entire marketing cycle (before, during, and after). This not only improves the execution efficiency of operation personnel but also enhances the accuracy of customer reach and perceived satisfaction, ultimately promoting the intelligence, precision, and efficiency of the overall marketing system.

[0061] The technical solution of this application obtains historical business data of the target cell before the target time period, and then determines the most suitable target business type to be pushed to the target cell based on multiple data dimensions of the historical business data. Then, a pre-trained large language model is used to generate a push plan corresponding to the target business type to push the target business type. This achieves the generation of the most suitable business push plan for the push cell when pushing business, thus improving the effect of business push.

[0062] In some feasible implementations, obtaining the historical service data of the target cell before the target time period includes:

[0063] The business opportunity data of the target community before the target time period is evaluated to obtain first business opportunity data and second business opportunity data. The first business opportunity data is the business opportunity data corresponding to a first business type of the target community, and the second business opportunity data is the business opportunity data corresponding to a second business type of the target community. The multiple business types include the first business type and the second business type. The user churn rate of the first business type in the target community is lower than the user churn rate of the second business type in the target community. The business opportunity data is used to indicate the number of potential users corresponding to the corresponding business type in the target community.

[0064] Predict the pedestrian flow distribution and spatiotemporal scene of the target cell before the target time period to obtain pedestrian flow prediction data and spatiotemporal scene detail data. The spatiotemporal scene includes the geographical location and scene events corresponding to the target cell. The scene events are base station-related events that occur in the target cell.

[0065] The business opportunity value of the target community before the target time period is evaluated to obtain the total business opportunity efficiency data;

[0066] The marketing effectiveness of the target community before the target time period is evaluated to obtain historical marketing data;

[0067] The historical business data includes the first business opportunity data, the second business opportunity data, the traffic flow prediction data, the spatiotemporal scenario detail data, the total business opportunity efficiency data, and the historical marketing data.

[0068] In this embodiment, specific analysis and prediction models are constructed for key business assessments, corresponding business opportunity tags, user hourly locations, and historical grid business opportunities. These specific analysis and prediction models are software algorithms that accurately generate corresponding model data.

[0069] In some embodiments, such as Figure 2 As shown, Figure 2 This embodiment presents a system architecture diagram for business push generation. Starting with business opportunities arising from weak performance indicators within the current grid, the system uses the TOPSIS multi-criteria decision recommendation framework to comprehensively process the outputs of multi-dimensional specialized small models, obtaining the optimal marketing time-space scenario. It then leverages the semantic understanding and intelligent reasoning capabilities of a large language model to generate marketing scripts, providing marketing recommendation services to operations personnel before, during, and after marketing campaigns, thus improving the comprehensiveness and accuracy of automated marketing plan generation. This application primarily includes a grid marketing multi-dimensional factor reasoning module, a TOPSIS multi-criteria decision recommendation module, a marketing script generation module, and a daily updated intelligent recommendation and display module for marketing plans.

[0070] Specifically, the business opportunity data of the target cell before the target time period is first evaluated to obtain first business opportunity data and second business opportunity data. The first business opportunity data is the lagging business indicator data, and the second business opportunity data is the non-lagging business indicator data. Lagging business refers to data typically used to measure the poor performance or challenges faced by a telecommunications service provider in certain business areas. This data helps the company identify problem areas and develop improvement measures. The first and second business opportunity data are generated through a business opportunity intelligent promotion model. Specifically, the business opportunity intelligent promotion model associates lagging business indicators with multi-dimensional business opportunity tags to accurately locate marketing opportunities. The model obtains the latest data from the business assessment table, obtains the lagging business indicators for each grid, associates them with the business indicator-business opportunity tag mapping table, and outputs the lagging business opportunity tags for that grid.

[0071] Secondly, the pedestrian flow distribution and spatiotemporal scenarios of the target cell before the target time period are predicted, thereby generating pedestrian flow prediction data and detailed spatiotemporal scenario data. In this embodiment, the spatiotemporal pedestrian flow prediction model is used. The spatiotemporal pedestrian flow prediction model is an analysis framework based on mobile positioning data. Its core is to achieve refined pedestrian flow distribution prediction by integrating user spatiotemporal behavior patterns and scene semantic information. The model marks the user's hourly base station location data with time period and date type data, counts the total time the user stays at the base station in different time periods (weekdays / non-weekdays × morning / afternoon / evening), obtains the TOP1 base station, retrieves its associated scene ID in the base station scene correspondence data, obtains the predicted scene for each user in different time periods, and finally outputs the predicted pedestrian flow data according to the scene, date type, and time period type format.

[0072] Then, an opportunity table is obtained through the business opportunity prediction model. Specifically, based on the pedestrian flow prediction, the business opportunities of pedestrian flow are predicted. The model associates user-level spatiotemporal scene location data output by the spatiotemporal pedestrian flow prediction model with user business opportunity data in the smart community user tag data through the user ID field to obtain scene business opportunity prediction data. The data is then summarized according to the format of date type, time period type, scene ID, and business opportunity to obtain the business opportunity table.

[0073] The business opportunity value of the target community before the target time period is evaluated to obtain the total business opportunity effectiveness data. In this embodiment, the total business opportunity effectiveness data is generated through a business opportunity effectiveness evaluation model. The business opportunity effectiveness evaluation model evaluates the value of different business opportunities based on the user volume, contact volume, contact rate, conversion rate, and revenue conversion per unit of business opportunity in historical grids. The model first takes the data from the end of the previous 3 months and the latest day of the current month from the effectiveness data of the business opportunity tag dimension to obtain information fields such as grid ID. Secondly, it calculates the reach rate, conversion rate, and unit revenue based on the data of the previous 3 months, and deletes abnormal records with reach rate and conversion rate greater than 1. Then, it logarithmically normalizes the revenue, number of people reached, and target number of people, and then weights them with the reach rate and conversion rate by 0.2. After merging, it is normalized again to obtain the effectiveness data (0-1) of each tag under the grid. Finally, it calculates the mean of the business opportunity tag under all grids according to the business opportunity tag. When there is no effectiveness data for the business opportunity, the grid ID is empty, and the average effectiveness weight data of the business opportunity is output.

[0074] Finally, the marketing effectiveness of the target community before the target time period is evaluated to obtain historical marketing data. In this embodiment, historical marketing data is generated through a marketing effectiveness evaluation model, which mainly reflects the degree of user cooperation and marketing effect level in this scenario. The model first retrieves task effectiveness data from the task-level effectiveness data using the task ID as the key, including the end of the last three months and the latest day of the current month. It calculates the total number of marketing events for different task IDs in a single scenario, as well as the average customer contact rate and conversion rate of multiple tasks in the community, to obtain the unit revenue. When the scenario ID is empty, the number of marketing events in the last three months is set to 0, and the last marketing time is set to the first day of the month from which the data is retrieved. Next, the number of marketing events in the grid scenario over the last three months is retained, the most recent marketing time is converted to the date of the last marketing event, and the unit revenue, number of users reached, and number of customers are logarithmized and normalized. Then, a weighted average is calculated with the reach rate and conversion rate at a weight of 0.2 to obtain the comprehensive evaluation value of the historical marketing effectiveness of the scenario. If there have been no marketing activities in the past 3 months, the Scene ID and Grid ID will be empty. The comprehensive evaluation value is the average of the historical marketing effectiveness scores of all scenes. The number of marketing activities in the past 3 months is 0, and the number of days since the last marketing activity is fixed at 90 days.

[0075] This application collects multi-factor data on grid marketing, develops multiple specific analysis models for multi-dimensional source data, quantifies the impact weight of multi-dimensional data on marketing effectiveness, and realizes intelligent analysis of multi-objective factors, which can effectively improve the scientificity and comprehensiveness of marketing plan design.

[0076] Optionally, determining the target business type from the plurality of business types based on the historical business data includes:

[0077] The historical business data is filled into a preset data table, which includes a first indicator, a second indicator, a third indicator, and a fourth indicator. The influence of the data corresponding to the first indicator in the preset data table is lower than that of the data corresponding to the second indicator in the preset data table. The influence of the data corresponding to the second indicator in the preset data table is lower than that of the data corresponding to the third indicator in the preset data table. The influence of the data corresponding to the third indicator in the preset data table is lower than that of the data corresponding to the fourth indicator in the preset data table.

[0078] The data matrices corresponding to the first indicator, the second indicator, and the third indicator are respectively subjected to positive transformation processing to convert the first indicator, the second indicator, and the third indicator into the fourth indicator in the preset data table;

[0079] According to the fourth indicator, the multiple service types are scored to obtain multiple score values. The multiple score values ​​correspond one-to-one with the multiple service types. The score value is used to indicate the push conversion rate of the corresponding service type in the target cell. The push conversion rate is the ratio of the subscription volume to the push volume of the corresponding service type.

[0080] The business type corresponding to the highest score among the multiple score values ​​is determined as the target business type.

[0081] In this embodiment, the TOPSIS algorithm is used to comprehensively evaluate multiple business types using multiple indicators, and finally determine the target task type. Specifically, traffic prediction data, detailed spatiotemporal scene data, business opportunity data corresponding to lagging and non-lagging businesses in the scene, total business opportunity efficiency data, and historical marketing data generated by the special model are collected and summarized into a preset input data table, namely a wide table.

[0082] In this embodiment, the first indicator is a very small indicator, the second indicator is a medium indicator, the third indicator is a range indicator, and the fourth indicator is a very large indicator. The data matrix corresponding to the very small indicator, the medium indicator, and the range indicator in the wide table is positiveized into the very large indicator.

[0083] Therefore, based on the determined maximum indicators, multiple business types are scored, resulting in multiple score values. A higher score indicates a better solution. Specifically, the distance between each solution and the optimal and worst solutions is calculated, and a score is calculated based on the optimal and worst solutions to obtain the score for each marketing solution. A higher score indicates a better solution. The calculation formula is as follows:

[0084] .

[0085] .

[0086] .

[0087] in, The optimal connection distance, The optimal connection distance, Rating value and This is the matrix after normalization.

[0088] The solution with the highest score in each scenario among all solutions under a single grid manager is retained, and the rest are removed. The optimal solutions for each scenario under a single grid manager are sorted in descending order, and the output is the TOP recommended solution. The business type corresponding to the highest score is then determined as the target business type.

[0089] Optionally, the step of scoring the multiple business types according to the fourth indicator to obtain multiple score values ​​includes: obtaining the number of business marketing events and the business marketing time of the target cell before the target time period;

[0090] In the preset data table, based on the indicator type and indicator format of the second indicator, the number of business marketing events and the business marketing time are converted to obtain the fifth indicator;

[0091] The data matrix corresponding to the fifth indicator is positiveized to obtain the target matrix;

[0092] Based on the data matrix corresponding to the fourth indicator and the target matrix, the push conversion rate corresponding to the multiple business types is scored to obtain multiple score values.

[0093] In this embodiment, the optimal marketing effectiveness is analyzed based on historical stall performance data in a preset data table. The Sharpe ratio is used to comprehensively view the mean and standard deviation, where the benchmark mean uses the mean effectiveness of 0.34176 to obtain the number of marketing campaigns and the time of the last marketing campaign.

[0094] Specifically, by taking the number of marketing campaigns and the time of the last marketing campaign as intermediate indicators, we obtain the fifth indicator, and denote the optimal value as x. best The formula for positive transformation of intermediate indicators is as follows:

[0095] .

[0096] .

[0097] in, For the optimal value, For marketing frequency, This is the i-th marketing campaign.

[0098] The data matrix corresponding to the fifth indicator is positively oriented to obtain the target matrix. Then, based on the data matrix corresponding to the fourth indicator and the target matrix, the push conversion rate for multiple business types is scored, resulting in multiple score values. According to the determined maximal indicator, multiple business types are scored, resulting in multiple score values; the higher the score, the better the solution. The solution with the highest score for each scenario under a single grid manager is retained, and the remaining solutions are discarded. The optimal solutions for each scenario under a single grid manager are sorted in descending order, and the output is the TOP recommended solution. Thus, the business type corresponding to the highest score value is determined as the target business type.

[0099] Optionally, before inputting the target business type into a pre-trained large language model to generate a solution and obtain the business push solution corresponding to the target business type, the method further includes:

[0100] Obtain feature data of the target cell prior to the target time period, the feature data including traffic usage information, consumption information and bandwidth information corresponding to the target cell;

[0101] Based on the feature data, the historical business data, and the target time period, the initial large language model is trained to obtain the large language model.

[0102] In this embodiment, when using a large language model, it is necessary to train the initial large language model based on the acquired feature data, historical business data, and target time period to obtain a trained large language model.

[0103] The feature data is generated through a scenario wide table. The construction of the scenario wide table includes: extracting basic information data of the community, calculating the average ARPU, new network installation users, FTTR penetration rate, and labeling low-income areas, consumption, broadband, and elderly users.

[0104] Therefore, the initial large language model is trained based on the feature data, the acquired historical business data, and the target time period to obtain the trained large language model.

[0105] It should be noted that each large language model is bound to a community. For example, a smart marketing campaign generation program based on TOPSIS multi-criteria decision recommendation is deployed and launched. Data from the two days before today is used to calculate batches of offline data for marketing campaign recommendations and display for the day after today. The APP displays the intelligent recommendation results, with the homepage featuring the top 10 recommended campaigns. Each recommended campaign includes two sub-contents: an explanation of the recommendation reason and information to be filled in by the operations personnel participating in the campaign.

[0106] Optionally, the step of training the initial large language model based on the feature data, the historical business data, and the target time period to obtain the large language model includes:

[0107] The feature data, the historical business data, and the target time period are filled into a preset prompt word template to obtain the prompt word;

[0108] The prompt words are input into the initial large language model, and the initial large language model is trained based on the low-rank adaptive algorithm LoRA to obtain the large language model.

[0109] In this embodiment, the recommendation results and wide table data corresponding to the scenario are input into the initial large language model, and prompt words are designed and constructed to guide the large language model to generate marketing copy text.

[0110] Historically accumulated marketing script data, corresponding neighborhood information, and business opportunity information are used for fine-tuning the large model. A low-rank adaptive algorithm (LoRA) is employed for efficient fine-tuning. LoRA freezes the pre-trained model weights and injects the trainable low-rank factorization matrix into the Transformer layer for parallel computation, significantly reducing the number of trainable parameters for downstream tasks. Furthermore, LoRA eliminates the need to reserve a portion of the sequence length for constructing cue words (extra cue words would reduce the sequence length available for downstream tasks). The output includes a marketing script and a scenario description text.

[0111] This embodiment builds a marketing campaign generation model based on a large language model. The large language model possesses powerful semantic understanding, logical reasoning, and language generation capabilities, effectively improving the completeness and readability of the intelligent recommendation marketing campaign text results. Simultaneously, the system employs an efficient parameter fine-tuning algorithm to train the large language model, reducing the training and iteration costs.

[0112] In some implementations, such as Figure 3 As shown, Figure 3 The business recommendation generation flowchart in this embodiment includes the following steps: Step 1: Obtain multi-dimensional marketing-related data such as key business indicator data and grid-grid manager / grid leader correspondence data. Step 2: Input the raw data into the corresponding business opportunity intelligent promotion model, spatiotemporal flow prediction model, business opportunity prediction model, business opportunity efficiency evaluation model, and marketing efficiency evaluation model, and save the multi-dimensional prediction and reasoning results. Step 3: Use the TOPSIS multi-criteria decision recommendation algorithm to integrate multi-dimensional data such as flow, business opportunities, efficiency, number of marketing campaigns in the past 3 months, and time since the last marketing campaign output by the specialized models to recommend the optimal spatiotemporal setting. Step 4: Through the large model prompt project, convert the recommendation time, scenario, and description information into input instructions for marketing case generation. Step 5: Configure the relevant parameters for fine-tuning the training of the large language model, and use the LoRA efficient fine-tuning training algorithm to train the intelligent marketing case generation model. Step 6: Confirm the marketing case generation model through fine-tuning training. Step 7: Use the semantic understanding and reasoning capabilities of the large model to integrate the recommendation results and scenario information to generate a marketing script that conforms to the specifications. Step 8: The results of the large model inference are displayed on the APP to recommend the best marketing plan of the day to the grid operation staff.

[0113] The technical solution of this application obtains historical business data of the target cell before the target time period, and then determines the most suitable target business type to be pushed to the target cell based on multiple data dimensions of the historical business data. Then, a pre-trained large language model is used to generate a push plan corresponding to the target business type to push the target business type. This achieves the generation of the most suitable business push plan for the push cell when pushing business, thus improving the effect of business push.

[0114] See Figure 4 , Figure 4 This is a structural diagram of the service push device provided in the embodiments of this application. For example... Figure 4 As shown, the service push device 400 includes:

[0115] The acquisition module 410 is used to acquire historical service data of the target cell before the target time period. The historical service data includes service processing data corresponding to multiple service types in the target cell.

[0116] The determining module 420 is used to determine a target business type among the multiple business types based on the historical business data. Among the multiple business types, the business data corresponding to different business types are different. The target business type is the business type with the largest number of users processed within the target time period among the multiple business types.

[0117] The generation module 430 is used to input the target business type into a pre-trained large language model to generate a scheme, thereby obtaining a business push scheme corresponding to the target business type.

[0118] The push module 440 is used to push services to the target cell based on the service push scheme during the target time period.

[0119] Optionally, the acquisition module 410 includes:

[0120] The first evaluation submodule is used to evaluate the business opportunity data of the target cell before the target time period to obtain first business opportunity data and second business opportunity data. The first business opportunity data is the business opportunity data corresponding to the first business type of the target cell, and the second business opportunity data is the business opportunity data corresponding to the second business type of the target cell. The multiple business types include the first business type and the second business type. The user churn rate of the first business type in the target cell is lower than the user churn rate of the second business type in the target cell. The business opportunity data is used to indicate the number of potential users corresponding to the corresponding business type in the target cell.

[0121] The prediction submodule is used to predict the pedestrian flow distribution and spatiotemporal scene of the target cell before the target time period, and obtain pedestrian flow prediction data and spatiotemporal scene detailed data. The spatiotemporal scene includes the geographical location and scene events corresponding to the target cell. The scene events are base station-related events that occur in the target cell.

[0122] The second evaluation submodule is used to evaluate the business opportunity value of the target cell before the target time period and obtain the total business opportunity efficiency data.

[0123] The third evaluation submodule is used to evaluate the marketing effectiveness of the target community before the target time period and obtain historical marketing data.

[0124] The historical business data includes the first business opportunity data, the second business opportunity data, the traffic flow prediction data, the spatiotemporal scenario detail data, the total business opportunity efficiency data, and the historical marketing data.

[0125] Optionally, the determining module 420 includes:

[0126] The input submodule is used to input the historical business data into a preset data table. The preset data table includes a first indicator, a second indicator, a third indicator, and a fourth indicator. The influence of the data corresponding to the first indicator in the preset data table is lower than that of the data corresponding to the second indicator in the preset data table. The influence of the data corresponding to the second indicator in the preset data table is lower than that of the data corresponding to the third indicator in the preset data table. The influence of the data corresponding to the third indicator in the preset data table is lower than that of the data corresponding to the fourth indicator in the preset data table.

[0127] The processing submodule is used to perform forward processing on the data matrices corresponding to the first indicator, the second indicator, and the third indicator, respectively, so as to convert the first indicator, the second indicator, and the third indicator into the fourth indicator in the preset data table;

[0128] The scoring submodule is used to score the multiple service types according to the fourth indicator to obtain multiple score values. The multiple score values ​​correspond one-to-one with the multiple service types. The score value is used to indicate the push conversion rate of the corresponding service type in the target cell. The push conversion rate is the ratio of the subscription volume to the push volume of the corresponding service type.

[0129] The determination submodule is used to determine the business type corresponding to the highest score among the multiple score values ​​as the target business type.

[0130] Optionally, the scoring submodule includes:

[0131] The acquisition unit is used to acquire the number of business marketing activities and the business marketing time of the target cell before the target time period;

[0132] A conversion unit is used to convert the number of business marketing events and the business marketing time in the preset data table based on the indicator type and indicator format of the second indicator to obtain a fifth indicator;

[0133] The processing unit is used to perform positive transformation on the data matrix corresponding to the fifth indicator to obtain the target matrix;

[0134] The scoring unit is used to score the push conversion rate corresponding to the multiple business types based on the data matrix corresponding to the fourth indicator and the target matrix, and obtain multiple score values.

[0135] Optionally, the generation module 430 includes:

[0136] The acquisition submodule is used to acquire the feature data of the target cell before the target time period, and the feature data includes traffic usage information, consumption information and bandwidth information corresponding to the target cell;

[0137] The training submodule is used to train the initial large language model based on the feature data, the historical business data, and the target time period to obtain the large language model.

[0138] Optionally, the training submodule includes:

[0139] The input unit is used to input the feature data, the historical business data and the target time period into a preset prompt word template to obtain prompt words;

[0140] The training unit is used to input the prompt words into the initial large language model and train the initial large language model based on the low-rank adaptive algorithm LoRA to obtain the large language model.

[0141] The technical solution of this application obtains historical business data of the target cell before the target time period, and then determines the most suitable target business type to be pushed to the target cell based on multiple data dimensions of the historical business data. Then, a pre-trained large language model is used to generate a push plan corresponding to the target business type to push the target business type. This achieves the generation of the most suitable business push plan for the push cell when pushing business, thus improving the effect of business push.

[0142] This application also provides an electronic device. Please refer to [link to relevant documentation]. Figure 5The electronic device may include a processor 501, a memory 502, and a program 5021 stored in the memory 502 and capable of running on the processor 501.

[0143] When program 5021 is executed by processor 501, it can achieve the following: Figure 1 Any step in the corresponding method embodiment:

[0144] Obtain historical service data of the target cell before the target time period, wherein the historical service data includes service processing data corresponding to multiple service types in the target cell;

[0145] Based on the historical business data, a target business type is determined among the multiple business types. Among the multiple business types, different business types correspond to different business data. The target business type is the business type with the largest number of users processed within the target time period among the multiple business types.

[0146] The target business type is input into a pre-trained large language model to generate a solution, thereby obtaining a business push solution corresponding to the target business type.

[0147] During the target time period, services are pushed to the target cell based on the service push scheme.

[0148] Optionally, obtaining the historical service data of the target cell before the target time period includes:

[0149] The business opportunity data of the target community before the target time period is evaluated to obtain first business opportunity data and second business opportunity data. The first business opportunity data is the business opportunity data corresponding to a first business type of the target community, and the second business opportunity data is the business opportunity data corresponding to a second business type of the target community. The multiple business types include the first business type and the second business type. The user churn rate of the first business type in the target community is lower than the user churn rate of the second business type in the target community. The business opportunity data is used to indicate the number of potential users corresponding to the corresponding business type in the target community.

[0150] Predict the pedestrian flow distribution and spatiotemporal scene of the target cell before the target time period to obtain pedestrian flow prediction data and spatiotemporal scene detail data. The spatiotemporal scene includes the geographical location and scene events corresponding to the target cell. The scene events are base station-related events that occur in the target cell.

[0151] The business opportunity value of the target community before the target time period is evaluated to obtain the total business opportunity efficiency data;

[0152] The marketing effectiveness of the target community before the target time period is evaluated to obtain historical marketing data;

[0153] The historical business data includes the first business opportunity data, the second business opportunity data, the traffic flow prediction data, the spatiotemporal scenario detail data, the total business opportunity efficiency data, and the historical marketing data.

[0154] Optionally, determining the target business type from the plurality of business types based on the historical business data includes:

[0155] The historical business data is filled into a preset data table, which includes a first indicator, a second indicator, a third indicator, and a fourth indicator. The influence of the data corresponding to the first indicator in the preset data table is lower than that of the data corresponding to the second indicator in the preset data table. The influence of the data corresponding to the second indicator in the preset data table is lower than that of the data corresponding to the third indicator in the preset data table. The influence of the data corresponding to the third indicator in the preset data table is lower than that of the data corresponding to the fourth indicator in the preset data table.

[0156] The data matrices corresponding to the first indicator, the second indicator, and the third indicator are respectively subjected to positive transformation processing to convert the first indicator, the second indicator, and the third indicator into the fourth indicator in the preset data table;

[0157] According to the fourth indicator, the multiple service types are scored to obtain multiple score values. The multiple score values ​​correspond one-to-one with the multiple service types. The score value is used to indicate the push conversion rate of the corresponding service type in the target cell. The push conversion rate is the ratio of the subscription volume to the push volume of the corresponding service type.

[0158] The business type corresponding to the highest score among the multiple score values ​​is determined as the target business type.

[0159] Optionally, the scoring of the multiple business types based on the fourth indicator yields multiple score values, including:

[0160] Obtain the number of business marketing events and the duration of business marketing events for the target cell before the target time period;

[0161] In the preset data table, based on the indicator type and indicator format of the second indicator, the number of business marketing events and the business marketing time are converted to obtain the fifth indicator;

[0162] The data matrix corresponding to the fifth indicator is positiveized to obtain the target matrix;

[0163] Based on the data matrix corresponding to the fourth indicator and the target matrix, the push conversion rate corresponding to the multiple business types is scored to obtain multiple score values.

[0164] Optionally, before inputting the target business type into a pre-trained large language model to generate a solution and obtain the business push solution corresponding to the target business type, the method further includes:

[0165] Obtain feature data of the target cell prior to the target time period, the feature data including traffic usage information, consumption information and bandwidth information corresponding to the target cell;

[0166] Based on the feature data, the historical business data, and the target time period, the initial large language model is trained to obtain the large language model.

[0167] Optionally, the step of training the initial large language model based on the feature data, the historical business data, and the target time period to obtain the large language model includes:

[0168] The feature data, the historical business data, and the target time period are filled into a preset prompt word template to obtain the prompt word;

[0169] The prompt words are input into the initial large language model, and the initial large language model is trained based on the low-rank adaptive algorithm LoRA to obtain the large language model.

[0170] The technical solution of this application obtains historical business data of the target cell before the target time period, and then determines the most suitable target business type to be pushed to the target cell based on multiple data dimensions of the historical business data. Then, a pre-trained large language model is used to generate a push plan corresponding to the target business type to push the target business type. This achieves the generation of the most suitable business push plan for the push cell when pushing business, thus improving the effect of business push.

[0171] This application also provides a computer-readable storage medium storing a computer program. When executed by a processor, this computer program implements the various processes of the above-described push service embodiment and achieves the same technical effect. To avoid repetition, it will not be described again here. The computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc.

[0172] This application also provides a computer program product, which is stored in a storage medium and executed by at least one processor to implement the various processes of the above-described push method embodiment for the business, and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0173] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0174] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a communication device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0175] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A method for pushing services, characterized in that, The method includes: Obtain historical service data of the target cell before the target time period, wherein the historical service data includes service processing data corresponding to multiple service types in the target cell; Based on the historical business data, a target business type is determined among the multiple business types. Among the multiple business types, different business types correspond to different business data. The target business type is the business type with the largest number of users processed within the target time period among the multiple business types. The target business type is input into a pre-trained large language model to generate a solution, thereby obtaining a business push solution corresponding to the target business type. During the target time period, services are pushed to the target cell based on the service push scheme.

2. The method according to claim 1, characterized in that, The acquisition of historical service data of the target cell before the target time period includes: The business opportunity data of the target community before the target time period is evaluated to obtain first business opportunity data and second business opportunity data. The first business opportunity data is the business opportunity data corresponding to a first business type of the target community, and the second business opportunity data is the business opportunity data corresponding to a second business type of the target community. The multiple business types include the first business type and the second business type. The user churn rate of the first business type in the target community is lower than the user churn rate of the second business type in the target community. The business opportunity data is used to indicate the number of potential users corresponding to the corresponding business type in the target community. Predict the pedestrian flow distribution and spatiotemporal scene of the target cell before the target time period to obtain pedestrian flow prediction data and spatiotemporal scene detail data. The spatiotemporal scene includes the geographical location and scene events corresponding to the target cell. The scene events are base station-related events that occur in the target cell. The business opportunity value of the target community before the target time period is evaluated to obtain the total business opportunity efficiency data; The marketing effectiveness of the target community before the target time period is evaluated to obtain historical marketing data; The historical business data includes the first business opportunity data, the second business opportunity data, the traffic flow prediction data, the spatiotemporal scenario detail data, the total business opportunity efficiency data, and the historical marketing data.

3. The method according to claim 1, characterized in that, The step of determining the target business type from the plurality of business types based on the historical business data includes: The historical business data is filled into a preset data table, which includes a first indicator, a second indicator, a third indicator, and a fourth indicator. The influence of the data corresponding to the first indicator in the preset data table is lower than that of the data corresponding to the second indicator in the preset data table. The influence of the data corresponding to the second indicator in the preset data table is lower than that of the data corresponding to the third indicator in the preset data table. The influence of the data corresponding to the third indicator in the preset data table is lower than that of the data corresponding to the fourth indicator in the preset data table. The data matrices corresponding to the first indicator, the second indicator, and the third indicator are respectively subjected to positive transformation processing to convert the first indicator, the second indicator, and the third indicator into the fourth indicator in the preset data table; According to the fourth indicator, the multiple service types are scored to obtain multiple score values. The multiple score values ​​correspond one-to-one with the multiple service types. The score value is used to indicate the push conversion rate of the corresponding service type in the target cell. The push conversion rate is the ratio of the subscription volume to the push volume of the corresponding service type. The business type corresponding to the highest score among the multiple score values ​​is determined as the target business type.

4. The method according to claim 3, characterized in that, The process involves scoring the multiple business types based on the fourth indicator to obtain multiple score values, including: Obtain the number of business marketing events and the duration of business marketing events for the target cell before the target time period; In the preset data table, based on the indicator type and indicator format of the second indicator, the number of business marketing events and the business marketing time are converted to obtain the fifth indicator; The data matrix corresponding to the fifth indicator is positiveized to obtain the target matrix; Based on the data matrix corresponding to the fourth indicator and the target matrix, the push conversion rate corresponding to the multiple business types is scored to obtain multiple score values.

5. The method according to claim 1, characterized in that, Before inputting the target business type into a pre-trained large language model to generate a scheme and obtain the business push scheme corresponding to the target business type, the method further includes: Obtain feature data of the target cell prior to the target time period, the feature data including traffic usage information, consumption information and bandwidth information corresponding to the target cell; Based on the feature data, the historical business data, and the target time period, the initial large language model is trained to obtain the large language model.

6. The method according to claim 5, characterized in that, The process of training an initial large language model based on the feature data, the historical business data, and the target time period to obtain the large language model includes: The feature data, the historical business data, and the target time period are filled into a preset prompt word template to obtain the prompt word; The prompt words are input into the initial large language model, and the initial large language model is trained based on the low-rank adaptive algorithm LoRA to obtain the large language model.

7. A service delivery device, characterized in that, The device includes: The acquisition module is used to acquire historical service data of the target cell before the target time period. The historical service data includes service processing data corresponding to multiple service types in the target cell. The determination module is used to determine a target business type among the multiple business types based on the historical business data. Among the multiple business types, the business data corresponding to different business types are different. The target business type is the business type with the largest number of users processed within the target time period among the multiple business types. The generation module is used to input the target business type into a pre-trained large language model to generate a solution, thereby obtaining a business push solution corresponding to the target business type. The push module is used to push services to the target cell based on the service push scheme within the target time period.

8. An electronic device, characterized in that, include: A processor, a memory, and a program stored in the memory and executable on the processor, wherein the program, when executed by the processor, implements the steps of the method as described in any one of claims 1 to 6.

9. 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 steps of the method as described in any one of claims 1 to 6.

10. A computer program product, characterized in that, Includes computer instructions that, when executed by a processor, implement the steps of the method as described in any one of claims 1 to 6.

Images