Enterprise business intelligence auxiliary decision-making method and device based on multi-source large language model, equipment and storage medium

Abstract

The application discloses an enterprise business intelligent auxiliary decision-making method and device based on a multi-source large language model, equipment and a storage medium, relates to the field of artificial intelligence, is applied to a terminal device loaded with a target intelligent auxiliary decision-making system, and comprises the following steps: obtaining a to-be-decided task related to a target enterprise business to determine target label data; acquiring corresponding target portrait data from a target business portrait database in a corresponding dimension based on each target label data by using a target business knowledge platform to generate a corresponding structured parameter package; determining a target large language model corresponding to a task type based on the task type and a target mapping table acquired in advance; and generating an intelligent auxiliary decision related to the target enterprise business based on the structured parameter package and a target prompt word template by using the target large language model, so as to complete the target enterprise business based on the intelligent auxiliary decision. According to the application, different large models can be called according to business requirements.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence, and in particular to a method, apparatus, device, and storage medium for enterprise business intelligent auxiliary decision-making based on a multi-source large language model. Background Technology

[0002] When using large-scale language models in omnichannel business scenarios with heavy offline services, such as beauty and retail, there are common technical problems such as severe data silos and limited model capabilities. Specifically, "product recommendation" content on social media cannot effectively drive customers to in-store appointments, and offline data such as user tags in offline operating systems, consumption records in POS (Point of Sales) systems, and in-store consultants' understanding of users are fragmented with online user data, failing to form a complete user profile. At the same time, different business tasks (such as generating creative product recommendation copy, analyzing sales data, and reviewing the compliance of communication content) have different requirements for AI capabilities. Relying on a single, general-purpose large model makes it difficult to achieve the best cost-effectiveness ratio for all tasks.

[0003] Therefore, how to achieve the integration of data from multiple channels and intelligently call different large models according to business needs is an urgent problem to be solved. Summary of the Invention

[0004] In view of this, the purpose of this invention is to provide an enterprise business intelligent auxiliary decision-making method, device, equipment, and storage medium based on a multi-source large language model, which can realize the fusion of multi-channel data and intelligently call different large models according to business needs. The specific solution is as follows:

[0005] In a first aspect, this application discloses an enterprise business intelligent auxiliary decision-making method based on a multi-source large language model, applied to a terminal device equipped with a target intelligent auxiliary decision-making system, comprising:

[0006] Acquire decision-making tasks related to the target enterprise's business, and parse the decision-making tasks to determine target tag data; the target tag data includes product category, channel category, and customer category;

[0007] The target business knowledge platform of the target intelligent auxiliary decision-making system is used to obtain target profile data corresponding to each target tag data from the target business profile database of the corresponding dimension based on each target tag data, and to generate a structured parameter package corresponding to the task to be decided based on all the target profile data; the target business profile database includes a product information database, a channel content database, and a customer information database.

[0008] Using the target model scheduler of the target intelligent auxiliary decision-making system, based on the task type corresponding to the task to be decided and a pre-acquired target mapping table, a target large language model corresponding to the task type is determined from a plurality of pre-created large language models; wherein, the target mapping table is used to store the mapping relationship between each task type and each of the large language models in the target intelligent auxiliary decision-making system; different large language models are models built based on different types of data sources;

[0009] Obtain a pre-configured target prompt word template related to the target enterprise's business, and use the target large language model to generate intelligent auxiliary decision-making related to the target enterprise's business based on the structured parameter package and the target prompt word template, so as to complete the target enterprise's business based on the intelligent auxiliary decision-making.

[0010] Optionally, the step of generating intelligent auxiliary decision-making related to the target enterprise's business based on the target large language model using the structured parameter package and target prompt word template includes:

[0011] The initial auxiliary decision corresponding to the target large language model is obtained based on the structured parameter package and the target prompt word template using the target large language model;

[0012] The target decision review model is determined using the target mapping table, and the target decision review model is scheduled using the target model scheduler to determine whether the initial auxiliary decision meets the decision output conditions based on preset decision review rules.

[0013] If the initial auxiliary decision meets the decision output conditions, then the initial auxiliary decision is determined as the intelligent auxiliary decision corresponding to the task to be decided.

[0014] Optionally, the target model scheduler of the target intelligent auxiliary decision-making system determines the target large language model corresponding to the task type from a pre-created set of large language models based on the task type corresponding to the task to be decided and a pre-acquired target mapping table, including:

[0015] The target model scheduler of the target intelligent auxiliary decision-making system determines all target large language models based on the task type corresponding to the task to be decided and the pre-acquired target mapping table, and determines the target call chain of all the target large language models;

[0016] Accordingly, the step of generating intelligent auxiliary decision-making related to the target enterprise's business based on the target large language model using the structured parameter package and the target prompt word template includes:

[0017] Based on the target call chain, each of the target large language models is invoked to obtain the intelligent auxiliary decision-making corresponding to the target large language model based on the structured parameter package and the target prompt word template.

[0018] Optionally, the enterprise business intelligent auxiliary decision-making method based on a multi-source large language model further includes:

[0019] The target receiving end corresponding to the intelligent assisted decision is determined based on the task output label in the target label data of the task to be decided;

[0020] Determine the target decision output rule corresponding to the target receiving end, and send the intelligent auxiliary decision to the target receiving end based on the target decision output rule.

[0021] Optionally, the enterprise business intelligent auxiliary decision-making method based on a multi-source large language model further includes:

[0022] The feedback results corresponding to the intelligent assisted decision-making are obtained using the target data integration interface;

[0023] The intelligent assisted decision-making is quantitatively evaluated based on the feedback results using a target quantification model to determine the corresponding quantitative evaluation results, and the quantitative evaluation results are visualized and output using target visualization rules.

[0024] The target large language model is updated using a preset target model update rule based on the task to be decided and the corresponding quantitative evaluation result, so as to execute the task to be decided using the updated target large language model.

[0025] Optionally, after using the target quantification model to quantify the intelligent assisted decision based on the feedback results to determine the corresponding quantification evaluation results, the method further includes:

[0026] Based on all the tasks to be decided and the corresponding quantitative evaluation results, the customers to be activated and the corresponding customer activation tasks are determined from the customer information database.

[0027] The customer activation task is identified as the new task to be decided, and the process proceeds to the step of parsing the task to be decided to determine the target tag data.

[0028] Optionally, the enterprise business intelligent auxiliary decision-making method based on a multi-source large language model further includes:

[0029] Obtain a mapping table update instruction, and update the model registry in the target model scheduler based on the mapping table maintenance instruction, so as to update the target mapping table based on the updated model registry.

[0030] Secondly, this application discloses an enterprise business intelligent auxiliary decision-making device based on a multi-source large language model, applied to a terminal device equipped with a target intelligent auxiliary decision-making system, comprising:

[0031] The task parsing module is used to acquire decision-making tasks related to the target enterprise's business, and to parse the decision-making tasks to determine target tag data; the target tag data includes product category, channel category, and customer category;

[0032] The parameter package generation module is used to utilize the target business knowledge platform of the target intelligent auxiliary decision-making system to obtain target profile data corresponding to each target tag data from the target business profile database of the corresponding dimension based on each target tag data, and generate a structured parameter package corresponding to the task to be decided based on all the target profile data; the target business profile database includes a product information database, a channel content database, and a customer information database;

[0033] The model determination module is used to determine the target large language model corresponding to the task type from a plurality of pre-created large language models based on the task type corresponding to the task to be decided and a pre-acquired target mapping table, using the target model scheduler of the target intelligent auxiliary decision-making system; wherein, the target mapping table is used to store the mapping relationship between each task type and each of the large language models in the target intelligent auxiliary decision-making system; different large language models are models built based on different types of data sources;

[0034] The auxiliary decision generation module is used to obtain a pre-configured target prompt word template related to the target enterprise's business, and use the target large language model to generate intelligent auxiliary decisions related to the target enterprise's business based on the structured parameter package and the target prompt word template, so as to complete the target enterprise's business based on the intelligent auxiliary decisions.

[0035] Thirdly, this application discloses an electronic device, including:

[0036] Memory, used to store computer programs;

[0037] A processor is used to execute the computer program to implement the aforementioned enterprise business intelligent auxiliary decision-making method based on a multi-source large language model.

[0038] Fourthly, this application discloses a computer-readable storage medium for storing a computer program, wherein the computer program, when executed by a processor, implements the aforementioned enterprise business intelligent auxiliary decision-making method based on a multi-source large language model.

[0039] In this application, when using a terminal device equipped with a target intelligent auxiliary decision-making system to perform intelligent auxiliary decision-making for enterprise business, a decision-making task related to the target enterprise business is acquired, and the decision-making task is parsed to determine target tag data; the target tag data includes product category, channel category, and customer category; using the target business knowledge platform of the target intelligent auxiliary decision-making system, target profile data corresponding to each target tag data is obtained from the target business profile database of the corresponding dimension based on each target tag data, and a structured parameter package corresponding to the decision-making task is generated based on all the target profile data; the target business profile database includes a product information database, a channel content database, and a customer information database; using the target intelligent auxiliary decision-making system... A target model scheduler for an auxiliary decision-making system determines a target large language model corresponding to the task type from multiple pre-created large language models based on the task type corresponding to the task to be decided and a pre-acquired target mapping table. The target mapping table stores the mapping relationship between each task type and each large language model within the target intelligent auxiliary decision-making system. Different large language models are models built based on different types of data sources. A pre-configured target prompt word template related to the target enterprise's business is obtained. The target large language model, based on the structured parameter package and the target prompt word template, generates intelligent auxiliary decisions related to the target enterprise's business, thereby completing the target enterprise's business based on the intelligent auxiliary decisions. As can be seen, when generating intelligent auxiliary decision-making for a task to be decided, this application dynamically retrieves and combines target profile data from the target business profile database across one or more dimensions based on the target tag data corresponding to the task, forming a comprehensive structured parameter package. This structured parameter package is then used as the core input. A target model scheduler, based on the target mapping table and the task type corresponding to the task to be decided, automatically selects the most suitable large language model as the target large language model, driving the target large language model to perform intelligent analysis or content generation, thereby optimizing the overall system performance, effectiveness, and cost. This application utilizes a configurable target business knowledge platform to achieve the integration of multi-source heterogeneous data and knowledge, and organically combines it with a target model scheduler that can autonomously select the most suitable one, forming a target intelligent auxiliary decision-making system. This provides enterprises with a standardized methodology and technical framework for achieving intelligent operation, systematically improving their ability to utilize data and artificial intelligence for business management. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0041] Figure 1 This application discloses a flowchart of an enterprise business intelligent auxiliary decision-making method based on a multi-source large language model.

[0042] Figure 2 This is a specific signaling diagram of an intelligent assisted decision-making process disclosed in this application;

[0043] Figure 3 This is a schematic diagram of the business profile database architecture of a specific target intelligent auxiliary decision-making system disclosed in this application;

[0044] Figure 4 This is a schematic diagram of the structure of an enterprise business intelligent auxiliary decision-making device based on a multi-source large language model disclosed in this application;

[0045] Figure 5 This is a structural diagram of an electronic device disclosed in this application. Detailed Implementation

[0046] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0047] When using large language models in omnichannel business scenarios with heavy offline services, such as beauty and retail, there are common technical problems such as severe data silos and limited model capabilities. Specifically, "product recommendation" content on social media cannot effectively drive customers to in-store appointments, and offline data such as user tags in the offline operating system, consumption records in the POS system, and in-store consultants' understanding of users are fragmented with online user data, failing to form a complete user profile. At the same time, different business tasks (such as generating creative product recommendation copy, analyzing sales data, and reviewing the compliance of communication content) have different requirements for AI capabilities. Relying on a single general-purpose large model makes it difficult to achieve the best cost-effectiveness ratio for all tasks. To solve the above technical problems, this application discloses an enterprise business intelligent auxiliary decision-making method based on a multi-source large language model, which can realize the integration of multi-channel data and intelligently call different large models according to business needs.

[0048] SeeFigure 1 As shown, this invention discloses an enterprise business intelligent auxiliary decision-making method based on a multi-source large language model, applied to a terminal device equipped with a target intelligent auxiliary decision-making system, including:

[0049] Step S11: Obtain decision-making tasks related to the target enterprise's business, and parse the decision-making tasks to determine target tag data; the target tag data includes product category, channel category and customer category.

[0050] In this embodiment, the decision-making tasks related to the target enterprise's business acquired by the terminal device equipped with the target intelligent auxiliary decision-making system can be various customer management tasks, such as product recommendation, new customer development, and churn customer activation, or they can be tasks awaiting decision-making, such as generating historical business analysis reports and generating and reviewing marketing copy. Figure 2 The diagram illustrates the entire process of handling a customer management task. Taking a specific business scenario as an example, the task instruction received by the target intelligent decision support system is: "For customer ID: 123 (a high-value customer whose historical consumption records show a preference for high-end anti-aging projects), she just browsed our 'Summer Sunscreen Festival' event (Event ID: 456) on social platform A. Please recommend the new product 'Waterproof and Sweatproof Isolation Cream' (Product ID: 789) to her via WeChat and guide her to her frequently visited 'XX Store' (Channel ID: 101) for an experience." The task instruction can be parsed by the task parsing module of the target intelligent decision support system to obtain target tag data including product category, channel category, and customer category. For example, in the target tag data corresponding to this task instruction, the product category is Waterproof and Sweatproof Isolation Cream (Product ID: 789), the channel category is XX Store (Channel ID: 101), the customer category is Customer ID: 123 (a high-value customer whose historical consumption records show a preference for high-end anti-aging projects), and the event category is Summer Sunscreen Festival (Event ID: 456).

[0051] Step S12: Utilize the target business knowledge platform of the target intelligent auxiliary decision-making system to obtain target profile data corresponding to each target tag data from the target business profile database of the corresponding dimension based on each target tag data, and generate a structured parameter package corresponding to the task to be decided based on all the target profile data; the target business profile database includes a product information database, a channel content database, and a customer information database.

[0052] In this embodiment, as Figure 3As shown, the target business knowledge center of the target intelligent assisted decision-making system is a configurable multi-dimensional business knowledge center. In a specific implementation, the target business knowledge center is configured for the omni-channel business scenarios of industries such as beauty and retail that focus on offline services, and is responsible for structurally aggregating the multi-source knowledge of the enterprise's online and offline. The target business knowledge center systematically aggregates and structures the data from various channels of the enterprise (such as skin test data from mini-programs, social media interaction data, store POS transaction data), internal business system data (such as CRM, Customer Relationship Management), and even the experiential knowledge of front-line employees (such as the communication content of gold medal salespersons), transforming it into "fuel" that the target model scheduler can understand and utilize, achieving in-depth modeling of business knowledge. By summarizing these aggregated knowledge according to the core business objects, an extensible multi-dimensional business portrait model is formed as the target business portrait database. When the target intelligent assisted decision-making system obtains the target label data corresponding to the task instruction, the target label data can be used as the input of the multi-dimensional business portrait model to obtain the target portrait data corresponding to the dimension of the target label data, so as to generate a structured parameter package corresponding to the decision-making task based on all the target portrait data. In another specific implementation, the target business portrait database can be constructed based on RAG or graph database modeling of unstructured documents, directly storing documents such as brand manuals and product introductions into the vector database, simplifying the knowledge entry process, but regular maintenance is required.

[0053] In this embodiment, the target business portrait database includes, but is not limited to, a product information database, a channel content database, and a customer information database. In a specific implementation, the dimensions of the target business portrait database and the corresponding database representation examples are as follows:

[0054] (1) Brand dimension (Brand): Stores personified information such as the positioning, values, story, and professional tone of the brand.

[0055] Plain Text

[0056] CREATE TABLE brand_dimension (

[0057] id BIGINT PRIMARY KEY AUTO_INCREMENT,

[0058] brand_name VARCHAR(255) NOT NULL COMMENT 'Brand name',

[0059] brand_positioning TEXT COMMENT 'Brand positioning',

[0060] brand_values TEXT COMMENT 'Brand values',

[0061] brand_tone VARCHAR(255) COMMENT 'Brand tone',

[0062] brand_story TEXT COMMENT 'Brand story',

[0063] is_deleted INT DEFAULT 0,

[0064] create_time DATETIME DEFAULT CURRENT_TIMESTAMP,

[0065] update_time DATETIME DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP)。

[0066] (2)Product / Service dimension: Store the core selling points, features, suitable skin types, usage scenarios, standard operating procedures (SOP, Standard Operating Procedure), etc. of the product or service.

[0067] Plain Text

[0068] CREATE TABLE product_dimension (

[0069] id BIGINT PRIMARY KEY AUTO_INCREMENT,

[0070] product_name VARCHAR(255) NOT NULL COMMENT 'Product / Service name',

[0071] product_type VARCHAR(100) COMMENT 'Type (product / service)',

[0072] selling_points TEXT COMMENT 'Core selling points',

[0073] features TEXT COMMENT 'Product features',

[0074] usage_scenarios TEXT COMMENT 'Usage scenarios',

[0075] target_users TEXT COMMENT 'Target Users',

[0076] is_deleted INT DEFAULT 0,

[0077] create_time DATETIME DEFAULT CURRENT_TIMESTAMP,

[0078] update_time DATETIME DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP).

[0079] (3) Customer dimension (User): This dimension deeply integrates data from multiple systems such as CRM, POS, and mini-programs, and stores a complete profile of customers, including skin condition reports, consumption records and preferences, treatment progress, loyalty level, content preferences, repurchase cycle prediction, etc.

[0080] Plain Text

[0081] CREATE TABLE user_dimension (

[0082] id BIGINT PRIMARY KEY AUTO_INCREMENT,

[0083] user_id VARCHAR(255) NOT NULL UNIQUE COMMENT 'Customer ID associated with CRM',

[0084] user_persona TEXT NOT NULL COMMENT 'Customer Profile Description',

[0085] skin_analysis_report JSON COMMENT 'Skin texture analysis report',

[0086] purchase_history JSON COMMENT 'purchase history',

[0087] content_preference text comment 'Content preferences',

[0088] life_cycle_stage VARCHAR(100) COMMENT 'Lifecycle stage',

[0089] is_deleted INT DEFAULT 0,

[0090] create_time DATETIME DEFAULT CURRENT_TIMESTAMP,

[0091] update_time DATETIME DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP).

[0092] (4) Activity dimension: Store the core scenarios, objectives, content strategies, social media tags, and discount rules of marketing activities.

[0093] Plain Text

[0094] CREATE TABLE activity_dimension (

[0095] id BIGINT PRIMARY KEY AUTO_INCREMENT,

[0096] activity_name VARCHAR(255) NOT NULL COMMENT 'Activity Name',

[0097] core_scenario VARCHAR(255) NOT NULL COMMENT 'Core Scenario',

[0098] content_strategy TEXT COMMENT 'Content Strategy',

[0099] required_tags TEXT COMMENT 'Required tags',

[0100] activity_goals TEXT COMMENT 'Activity Goals',

[0101] rules TEXT COMMENT 'Event Rules',

[0102] is_deleted INT DEFAULT 0,

[0103] create_time DATETIME DEFAULT CURRENT_TIMESTAMP,

[0104] update_time DATETIME DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP).

[0105] (5) Channel / Touchpoint Dimension: This is the key hub connecting online and offline. It stores the attributes, service capabilities, equipment, current inventory, special activities, and real-time transaction data generated by the POS system for each channel (offline stores, mini-program stores, etc.).

[0106] Plain Text

[0107] CREATE TABLE channel_dimension (

[0108] id BIGINT PRIMARY KEY AUTO_INCREMENT,

[0109] `channel_name VARCHAR(255) NOT NULL COMMENT 'Channel Name'`

[0110] `channel_type VARCHAR(100) COMMENT 'Channel type (e.g., offline stores, mini-programs)'`

[0111] channel_address TEXT COMMENT 'address or URL',

[0112] service_capabilities TEXT COMMENT 'Service Capabilities Description',

[0113] current_promotions JSON COMMENT 'Current activity (related activity dimension)',

[0114] is_deleted INT DEFAULT 0,

[0115] create_time DATETIME DEFAULT CURRENT_TIMESTAMP,

[0116] update_time DATETIME DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP);

[0117] (6) Employee / Expert Dimension: This is the core of making tacit knowledge explicit. It stores the personal profiles of top consultants / experts, their areas of expertise, as well as proven and effective sales communication methods, service processes, and customer follow-up cases. It also includes the expert's diagnostic knowledge base.

[0118] Plain Text

[0119] CREATE TABLE employee_dimension (

[0120] id BIGINT PRIMARY KEY AUTO_INCREMENT,

[0121] employee_name VARCHAR(255) NOT NULL COMMENT 'Employee Name',

[0122] employee_role VARCHAR(100) COMMENT 'Employee role (e.g., consultant, store manager)'

[0123] expertise_area TEXT COMMENT 'Areas of expertise (e.g., anti-aging, acne treatment)',

[0124] sales_scripts JSON COMMENT 'Gold Medal Communication Library (Scenario -> Communication)',

[0125] successful_cases TEXT COMMENT 'Successful Cases Sharing'

[0126] is_deleted INT DEFAULT 0,

[0127] create_time DATETIME DEFAULT CURRENT_TIMESTAMP,

[0128] update_time DATETIME DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP).

[0129] It is understandable that the profile data corresponding to all the above dimensions collectively constitute the complete and dynamic business context required by the target intelligent auxiliary decision-making system. By calling the data management service, the target intelligent auxiliary decision-making system can accurately retrieve the corresponding and complete target profile data from each target business profile database according to the ID in the task instruction, for example:

[0130] Brand: "Our brand image is professional and technological."

[0131] Product: "Product No. 789 is waterproof and sweatproof, suitable for outdoor sports."

[0132] User: "User No. 123 is an outdoor sports enthusiast with strong historical spending power and pays attention to product ingredients."

[0133] Event: "Event No. 456 is the Summer Sun Protection Festival, focusing on outdoor scenarios."

[0134] Channel: "'XX Store' has advanced equipment and offers exclusive discounts for new product trials this week."

[0135] Employee: "The key communication points for top consultant A at XX store when communicating with high-value clients are: emphasizing exclusive technological ingredients and, in conjunction with the customer's skin test report, pointing out that the product can address their potential photoaging risks."

[0136] Subsequently, the data management service can combine all the retrieved target profile data into a structured, in-depth context parameter package (such as a JSON object), which is also the structured parameter package corresponding to the decision-making task. It should be noted that this embodiment, through a structured approach, stores tacit knowledge such as the best practices of frontline employees (top-notch communication) and the decision-making logic of experts (diagnostic knowledge) in the corresponding dimension of the database. Therefore, this valuable knowledge is no longer lost due to staff turnover. A new employee can quickly master the communication skills of top consultants with the assistance of AI, achieving an overall improvement in enterprise capabilities. This transforms individual strengths into standardized, replicable enterprise competitiveness, solving the core pain point of uneven capabilities among chain stores.

[0137] Understandably, in this embodiment, enterprises can adjust or add knowledge dimensions to the target business knowledge platform (such as adding a "competitor dimension") to enable the same system architecture to quickly adapt to various business scenarios, from marketing (generating Xiaohongshu notes), sales service (generating sales communication methods), to internal management (executing SOPs), greatly reducing the marginal cost of applying AI in different scenarios and maximizing the reuse of technology assets.

[0138] Step S13: Using the target model scheduler of the target intelligent auxiliary decision-making system, based on the task type corresponding to the task to be decided and the pre-acquired target mapping table, determine the target large language model corresponding to the task type from a number of pre-created large language models; wherein, the target mapping table is used to store the mapping relationship between each task type and each of the large language models in the target intelligent auxiliary decision-making system; different large language models are models built based on different types of data sources.

[0139] In this embodiment, the target model scheduler of the target intelligent auxiliary decision-making system is a multi-source large language model scheduler. It is not simply a call to an AI interface, but an intelligent routing system with perception, analysis, and decision-making capabilities. The target model scheduler mainly includes the following core components:

[0140] (1) Model Capability Registry:

[0141] This is an internal registry (also known as the model registry) where system administrators can maintain all integrated and available LLM (Large Language Model) resources. Each model is not just an API (Application Programming Interface) address, but is assigned a series of structured "capability tags" and "cost parameters".

[0142] Model names: such as DeepSeek-R1, DeepSeek-V3.1, Qwen2-72B, Claude-3-Opus, etc.

[0143] Capability tags: Qualitative descriptions of model characteristics, such as strong creativity, strong logical reasoning, coding ability, low cost, high speed, following instructions, and low illusion rate.

[0144] Cost parameters: quantified cost metrics, such as input cost / million tokens, output cost / million tokens.

[0145] Performance metrics, such as average response time (ms).

[0146] (2) Task Parsing & Classification Module;

[0147] When a task instruction (such as "generate a promotional note for user 123") enters the system, the module will first parse and "tag" it, transforming it from a natural language instruction into a structured task that the machine can understand.

[0148] Input: "Generate a promotional note about product 789 for user 123"

[0149] Output: (structured tags).

[0150] Task type: Creative content generation.

[0151] Output channel: social media.

[0152] Core requirements: attractiveness and empathy.

[0153] Cost sensitivity: Medium.

[0154] (3) Intelligent Routing & Decision Engine:

[0155] This is the "brain" of the scheduler. It receives the structured labels output by the task parsing and classification module, and then selects the most suitable LLM from the model capability library to execute the task according to preset routing rules. Its decision-making logic mainly relies on a dynamic Task-Model Mapping Table, also known as the target mapping table. The target mapping table is used to store the mapping relationship between each task type and the major language models within the target intelligent auxiliary decision-making system; different major language models are models built based on different types of data sources. When the target intelligent auxiliary decision-making system receives a mapping table update instruction, it updates the model registry in the target model scheduler based on the mapping table maintenance instruction, and then updates the target mapping table based on the updated model registry.

[0156] In this embodiment, the target model scheduler analyzes the task to be decided to determine the task label data of the task, thereby determining the task type corresponding to the task. Then, the target model scheduler can determine the target large language model corresponding to the task type from all large language models of the target intelligent auxiliary decision-making system according to the target mapping table.

[0157] In one specific implementation, the task to be decided is to generate a personalized social media product promotion copy for high-value customers. The target intelligent auxiliary decision-making system analyzes the task and finds that the corresponding task tag data is "personalized marketing communication" and "creative copywriting." Therefore, the task type is determined to be creative content generation. The core requirements for the large language model are creativity and empathy. The intelligent routing decision engine, based on the target mapping table, selects a large language model whose capability tags include strong creativity and empathy as the main model (i.e., the target large language model) to ensure the attractiveness of the generated content. Through rule matching, DeepSeek-R1 and Claude-3-Opus are found in the model capability library to meet the requirements. The engine compares the cost parameters and performance indicators of the two and may choose DeepSeek-R1, which has a lower current load or better overall cost, as the main model.

[0158] In one specific implementation, the decision-making task is to analyze the sales data of a store in the previous quarter and generate an operational analysis report. The task parsing module tags the task type as data analysis and report generation; the core requirements are strong logical reasoning and low illusion rate. The intelligent routing decision engine queries the target mapping table and matches according to rules. For the decision-making task of data analysis, it must select a model whose capability tags include strong logical reasoning and low illusion rate. Qwen2-72B is found in the model capability library that meets the requirements. Understandably, models with strong creativity (such as DeepSeek-R1) have a lower priority in this type of task.

[0159] In one specific implementation, the task to be decided is to review whether a promotional text message written by an employee contains any illegal promises. The task parsing module tags the task as compliance review; the core requirements are following instructions, low cost, and high speed. The intelligent routing decision engine queries the target mapping table and matches rules. For the task to be decided, which is compliance review, models with low cost and high speed capabilities are prioritized. DeepSeek-V3.1 is found to be the most suitable in the model capability library. Understandably, using an expensive, large model for such a simple rule matching task would be wasteful.

[0160] It should be noted that for more complex decision-making tasks, the target model scheduler can determine all target large language models based on the task type corresponding to the decision-making task and the pre-acquired target mapping table, and determine the target call chain of all the target large language models. For example, for the decision-making task "Based on the latest market trends (Document A) and the company's product characteristics (Document B), plan three themes for next month's marketing campaign and generate corresponding social media promotion copy," the scheduling process of the target model scheduler may specifically include:

[0161] Step 1 (Information Extraction): The scheduler calls a model with strong logical reasoning (such as Qwen2-72B) to read documents A and B respectively and extract the core points.

[0162] Step 2 (Brainstorming): Hand over the extracted key points to a highly creative model (such as DeepSeek-R1) for brainstorming to generate three activity themes.

[0163] Step 3 (Copywriting Generation): Submit each topic to a highly creative model (such as DeepSeek-R1) to generate specific Xiaohongshu copy.

[0164] Step 4 (Compliance Review): Submit all generated text to a low-cost review model (such as DeepSeek-V3.1) to check for any prohibited words.

[0165] Through the aforementioned multi-source large language model scheduling mechanism, this embodiment ensures that specialized tasks are handled by the most capable "expert model," achieving optimal results in terms of creativity, logic, and compliance, thus maximizing task effectiveness. Furthermore, this embodiment does not rely on any single model vendor. When a model is upgraded, its price decreases, or it malfunctions, simply updating its tags and parameters in the model capability library allows the entire system to seamlessly and intelligently switch, ensuring business continuity and future scalability. Moreover, based on this multi-source large language model scheduling mechanism, this embodiment always uses the most cost-effective model to handle matching tasks, avoiding wasted model computing resources and significantly reducing the total cost of AI operations, thereby ensuring optimal cost-effectiveness. It should be noted that, based on this multi-source large language model scheduling mechanism and multi-dimensional business knowledge platform, this embodiment can quickly train any general-purpose large language model into a "domain expert" with a deep understanding of a specific enterprise's business. This maintainable and scalable multi-dimensional business knowledge platform is a key infrastructure for enterprises to build core competitiveness in the era of artificial intelligence. Its application prospects cover all aspects of user operation and internal management, such as intelligent customer service, automated marketing, private domain traffic operation, data analysis and insight, employee training, and SOP execution monitoring.

[0166] Step S14: Obtain a pre-configured target prompt word template related to the target enterprise's business, and use the target large language model to generate intelligent auxiliary decision-making related to the target enterprise's business based on the structured parameter package and the target prompt word template, so as to complete the target enterprise's business based on the intelligent auxiliary decision-making.

[0167] In this embodiment, the target prompt word template is a pre-configured Prompt template related to the target enterprise's business, which can also be updated periodically by the system administrator through the template configuration interface. The intelligent auxiliary decision-making related to the target enterprise's business is generated using the target large language model based on the structured parameter package and the target prompt word template. This includes: calling each target large language model based on the target call chain, so as to obtain the intelligent auxiliary decision corresponding to the target large language model based on the structured parameter package and the target prompt word template.

[0168] In this embodiment, intelligent auxiliary decision-making related to the target enterprise's business is generated using a target large language model based on a structured parameter package and target prompt word templates. This includes: obtaining the initial auxiliary decision corresponding to the target large language model using the target large language model based on the structured parameter package and target prompt word templates; determining the target decision review model using a target mapping table, and scheduling the target decision review model using a target model scheduler to determine whether the initial auxiliary decision meets the decision output conditions based on preset decision review rules; if the initial auxiliary decision meets the decision output conditions, then the initial auxiliary decision is determined as the intelligent auxiliary decision corresponding to the task to be decided. Subsequently, the target receiving end corresponding to the intelligent auxiliary decision can be determined based on the task output tags in the target tag data of the task to be decided; the target decision output rules corresponding to the target receiving end are determined, and the intelligent auxiliary decision is sent to the target receiving end based on the target decision output rules.

[0169] For example, when the task to be decided is to recommend products to customer A online, the target model scheduler injects the structured parameter package corresponding to the task into the Prompt template designed for the task type. It then calls the selected creative LLM to generate a highly personalized initial draft for social media communication. This draft can then proceed to the review stage. At this point, the target model scheduler again plays a role. Recognizing the task type as "content compliance review," it selects a lower-cost, more rule-compliant, and logically rigorous large language model (such as DeepSeek-V3.1) based on the target mapping table to perform the review task, checking for exaggerated claims and achieving an optimal balance between cost and effectiveness. Once the content passes review, it is output to the downstream social media SCRM system, where customer consultants can send it with a single click.

[0170] In this embodiment, when a frontline consultant is receiving a customer, the system can access the customer's "user dimension" profile, "product / service dimension" knowledge, and "employee / expert dimension" gold standard communication and standard operating procedures (SOPs) in real time. This provides the consultant with real-time communication methods and product pairing suggestions tailored to the current customer. More importantly, the system can automatically integrate preliminary skin analysis data from online mini-programs and in-depth testing data from offline professional instruments, and call upon diagnostic knowledge from the "employee / expert dimension" to generate a standardized "diagnostic basis + recommended treatment plan." This not only enhances professionalism and customer trust but also makes the target intelligent decision-making system an indispensable intelligent assistant for employees, thus solving the problem of system implementation and application. It helps new employees quickly master standardized communication and treatment recommendations, saving time and increasing conversion rates. Store managers can also use this data to monitor employee communication quality and conversion rates in real time for review and training, while the corporate headquarters can use this data to assess employee potential and create cross-store efficiency comparisons and experience models, thereby improving store efficiency and sales stability.

[0171] In this embodiment, the target data integration interface can be used to obtain the feedback results corresponding to the intelligent assisted decision-making; the target quantification model can be used to perform quantitative evaluation of the intelligent assisted decision-making based on the feedback results to determine the corresponding quantitative evaluation results, and the target visualization rules can be used to visualize the quantitative evaluation results; the target large language model can be updated based on the task to be decided and the corresponding quantitative evaluation results using the preset target large language model update rules, so as to use the updated target large language model to execute the task to be decided.

[0172] In one specific implementation, this embodiment uses a built-in target data integration interface to connect public domain data, private domain data, and offline store POS system data from the channel dimension, constructing an end-to-end behavior tracking link. The target intelligent auxiliary decision-making system calls upon a powerful LLM (Logical Learning Model) with strong logical reasoning capabilities to analyze the integrated end-to-end data and generate a visualized ROI analysis dashboard. To achieve accurate quantitative evaluation, the AI ​​Agent incorporates the following core computational models:

[0173] a. Full-Funnel ROI:

[0174] ;

[0175] in, For POS terminal revenue of marketing campaign i, For advertising costs, The cost of using the system for generating the target content. The activity operating cost of the marketing campaign i.

[0176] b. Multi-level Conversion Funnel:

[0177] ;

[0178] in, The conversion rate from public domain traffic to private domain leads was calculated. correspond The denominator is the count of public domain traffic (such as the number of people reached in the public domain, the number of public domain impressions / clicks, and other initial traffic on the public domain side). correspond The molecule is the result of the conversion of "public domain traffic" (private domain leads) (such as the number of leads that are transferred from the public domain to the private domain, such as the number of users who add the contact information of social software).

[0179] It calculated the conversion rate from private domain leads to offline store visits, accurately identified bottlenecks in the marketing funnel, and improved organic traffic and marketing ROI. correspond The denominator is the count of private domain leads (e.g., the total number of private domain leads entering the conversion process, i.e., the number of private domain leads in the pool awaiting conversion). correspond The molecule is the result of the conversion of "private domain leads" (offline in-store visits), such as the number of users who actually visit the store from the private domain leads.

[0180] Through the aforementioned quantitative assessment process, this embodiment can generate an operational health report, providing enterprises with clear, data-driven decision support, explicitly linking every input with output, and turning data into a growth asset. Furthermore, system administrators can define task templates containing objectives, materials, and execution standards within the system and store them in the "Employee / Expert Dimension." Through the organizational structure of the "Channel Dimension," the system can issue tasks to designated stores or employees with a single click. Each execution unit reports task completion status back through the system (e.g., photos, data entry), thus forming a management closed loop of "headquarters strategy issuance - store process tracking - effect data feedback." The enterprise headquarters can view real-time task progress heatmaps and completion quality for each region and store, ensuring that the enterprise's marketing strategies are implemented efficiently and transparently at all levels, significantly shortening the management radius and feedback cycle, and forming a replicable management closed loop.

[0181] It is understood that this embodiment can also determine the customers to be activated and the corresponding customer activation tasks from the customer information database based on all pending decision tasks and corresponding quantitative evaluation results; determine the customer activation tasks as new pending decision tasks, and jump to the step of parsing the pending decision tasks to determine the target tag data. In other words, the system can continuously monitor data changes in the "user dimension" (such as treatment end time, estimated product expiration period, points about to expire, etc.). When the preset customer activation rule is triggered, the system will determine the customer as a pending activation customer, and automatically combine the user's profile, historical preferences and the marketing strategy in the matching "activity dimension", and then drive the creative LLM in the multi-source LLM scheduler to generate highly personalized communication content such as repurchase reminders, birthday care, and new product recommendations. Stores can also automatically generate recall lists and messages, and automatically reach customers through the corresponding channels at the best time, thereby establishing an "time-right and person-right" automatic recall mechanism, realizing refined and automated management of the entire life cycle from customer activation, conversion to retention and recall, reducing customer churn rate and increasing product repurchase rate.

[0182] As can be seen, when generating intelligent auxiliary decision-making for a task to be decided, this application dynamically retrieves and combines target profile data from the target business profile database across one or more dimensions based on the target tag data corresponding to the task, forming a comprehensive structured parameter package. This structured parameter package is then used as the core input. A target model scheduler, based on the target mapping table and the task type corresponding to the task to be decided, automatically selects the most suitable large language model as the target large language model, driving the target large language model to perform intelligent analysis or content generation, thereby optimizing the overall system performance, effectiveness, and cost. This application utilizes a configurable target business knowledge platform to achieve the integration of multi-source heterogeneous data and knowledge, and organically combines it with a target model scheduler that can autonomously select the most suitable one, forming a target intelligent auxiliary decision-making system. This provides enterprises with a standardized methodology and technical framework for achieving intelligent operation, systematically improving their ability to utilize data and artificial intelligence for business management.

[0183] See Figure 4 As shown, this application discloses an enterprise business intelligent auxiliary decision-making device based on a multi-source large language model, applied to a terminal device equipped with a target intelligent auxiliary decision-making system, comprising:

[0184] The task parsing module 11 is used to acquire decision-making tasks related to the target enterprise's business, and to parse the decision-making tasks to determine target tag data; the target tag data includes product category, channel category and customer category;

[0185] The parameter package generation module 12 is used to utilize the target business knowledge platform of the target intelligent auxiliary decision-making system to obtain target profile data corresponding to each target tag data from the target business profile database of the corresponding dimension based on each target tag data, and generate a structured parameter package corresponding to the task to be decided based on all the target profile data; the target business profile database includes a product information database, a channel content database, and a customer information database.

[0186] The model determination module 13 is used to determine the target large language model corresponding to the task type from a plurality of pre-created large language models based on the task type corresponding to the task to be decided and a pre-acquired target mapping table using the target model scheduler of the target intelligent auxiliary decision-making system; wherein, the target mapping table is used to store the mapping relationship between each task type and each of the large language models in the target intelligent auxiliary decision-making system; different large language models are models built based on different types of data sources;

[0187] The auxiliary decision generation module 14 is used to obtain a pre-configured target prompt word template related to the target enterprise's business, and use the target large language model to generate intelligent auxiliary decisions related to the target enterprise's business based on the structured parameter package and the target prompt word template, so as to complete the target enterprise's business based on the intelligent auxiliary decisions.

[0188] As can be seen, when generating intelligent auxiliary decision-making for a task to be decided, this application dynamically retrieves and combines target profile data from the target business profile database across one or more dimensions based on the target tag data corresponding to the task, forming a comprehensive structured parameter package. This structured parameter package is then used as the core input. A target model scheduler, based on the target mapping table and the task type corresponding to the task to be decided, automatically selects the most suitable large language model as the target large language model, driving the target large language model to perform intelligent analysis or content generation, thereby optimizing the overall system performance, effectiveness, and cost. This application utilizes a configurable target business knowledge platform to achieve the integration of multi-source heterogeneous data and knowledge, and organically combines it with a target model scheduler that can autonomously select the most suitable one, forming a target intelligent auxiliary decision-making system. This provides enterprises with a standardized methodology and technical framework for achieving intelligent operation, systematically improving their ability to utilize data and artificial intelligence for business management.

[0189] In one specific embodiment, the auxiliary decision generation module 14 may include:

[0190] The first decision generation unit is used to obtain the initial auxiliary decision corresponding to the target large language model based on the structured parameter package and the target prompt word template using the target large language model;

[0191] The condition judgment unit is used to determine the target decision review model using the target mapping table, and to schedule the target decision review model using the target model scheduler, so as to determine whether the initial auxiliary decision meets the decision output conditions based on the preset decision review rules;

[0192] The auxiliary decision determination unit is used to determine the initial auxiliary decision as the intelligent auxiliary decision corresponding to the task to be decided if the initial auxiliary decision meets the decision output conditions.

[0193] In one specific implementation, the model determination module 13 may include:

[0194] The call chain determination unit is used to determine all target large language models based on the task type corresponding to the task to be decided and the pre-acquired target mapping table using the target model scheduler of the target intelligent auxiliary decision-making system, and to determine the target call chain of all the target large language models;

[0195] Accordingly, the auxiliary decision generation module 14 may specifically include:

[0196] The second decision acquisition unit is used to call each of the target large language models based on the target call chain, so as to use the target large language model to obtain the intelligent auxiliary decision corresponding to the target large language model based on the structured parameter package and the target prompt word template.

[0197] In one specific embodiment, the device may further include:

[0198] The receiver determination module is used to determine the target receiver corresponding to the intelligent assisted decision based on the task output label in the target label data of the task to be decided;

[0199] The decision sending module is used to determine the target decision output rule corresponding to the target receiving end, and send the intelligent auxiliary decision to the target receiving end based on the target decision output rule.

[0200] In one specific embodiment, the device may further include:

[0201] The feedback result acquisition module is used to acquire the feedback result corresponding to the intelligent auxiliary decision using the target data integration interface;

[0202] The decision quantification module is used to perform quantitative evaluation of the intelligent assisted decision based on the feedback results using the target quantification model to determine the corresponding quantitative evaluation results, and to visualize the quantitative evaluation results using target visualization rules.

[0203] The model update module is used to update the target large language model based on the task to be decided and the corresponding quantitative evaluation result using a preset target model update rule, so as to use the updated target large language model to execute the task to be decided.

[0204] In one specific embodiment, the device may further include:

[0205] The activation task generation module is used to determine the customers to be activated and the corresponding customer activation tasks from the customer information database based on all the tasks to be decided and the corresponding quantitative evaluation results.

[0206] The task execution module is activated to identify the customer activation task as a new task to be decided and to proceed to the step of parsing the task to be decided to determine the target tag data.

[0207] In one specific embodiment, the device may further include:

[0208] The mapping table update module is used to obtain the mapping table update instruction and update the model registry in the target model scheduler based on the mapping table maintenance instruction, so as to update the target mapping table based on the updated model registry.

[0209] Furthermore, embodiments of this application also disclose an electronic device, Figure 5 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content of the diagram should not be construed as limiting the scope of this application.

[0210] Figure 5 This is a schematic diagram of the structure of an electronic device 20 provided in an embodiment of this application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the enterprise business intelligent auxiliary decision-making method based on a multi-source large language model disclosed in any of the foregoing embodiments. Alternatively, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0211] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0212] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk, or optical disk, etc. The resources stored thereon can include an operating system 221, computer programs 222, etc., and the storage method can be temporary storage or permanent storage.

[0213] The operating system 221 is used to manage and control the various hardware devices on the electronic device 20 and the computer program 222, which may be Windows Server, Netware, Unix, Linux, etc. In addition to including a computer program capable of performing the enterprise business intelligent auxiliary decision-making method based on a multi-source large language model executed by the electronic device 20 as disclosed in any of the foregoing embodiments, the computer program 222 may further include computer programs capable of performing other specific tasks.

[0214] Furthermore, this application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned disclosed enterprise business intelligent auxiliary decision-making method based on a multi-source large language model. Specific steps of this method can be found in the corresponding content disclosed in the foregoing embodiments, and will not be repeated here.

[0215] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0216] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0217] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0218] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, 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. Without further limitations, 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 said element.

[0219] The technical solutions provided in this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A business intelligence-assisted decision-making method based on a multi-source large language model, characterized in that, Applications to terminal devices equipped with a target-based intelligent auxiliary decision-making system include: Acquire decision-making tasks related to the target enterprise's business, and parse the decision-making tasks to determine target tag data; the target tag data includes product category, channel category, and customer category; The target business knowledge platform of the target intelligent auxiliary decision-making system is used to obtain target profile data corresponding to each target tag data from the target business profile database of the corresponding dimension based on each target tag data, and to generate a structured parameter package corresponding to the task to be decided based on all the target profile data; the target business profile database includes a product information database, a channel content database, and a customer information database. Using the target model scheduler of the target intelligent auxiliary decision-making system, based on the task type corresponding to the task to be decided and a pre-acquired target mapping table, a target large language model corresponding to the task type is determined from a plurality of pre-created large language models; wherein, the target mapping table is used to store the mapping relationship between each task type and each of the large language models in the target intelligent auxiliary decision-making system; different large language models are models built based on different types of data sources; Obtain a pre-configured target prompt word template related to the target enterprise's business, and use the target large language model to generate intelligent auxiliary decision-making related to the target enterprise's business based on the structured parameter package and the target prompt word template, so as to complete the target enterprise's business based on the intelligent auxiliary decision-making.

2. The enterprise business intelligent auxiliary decision-making method based on a multi-source large language model according to claim 1, characterized in that, The process of generating intelligent auxiliary decision-making related to the target enterprise's business using the target large language model based on the structured parameter package and target prompt word template includes: The initial auxiliary decision corresponding to the target large language model is obtained based on the structured parameter package and the target prompt word template using the target large language model; The target decision review model is determined using the target mapping table, and the target decision review model is scheduled using the target model scheduler to determine whether the initial auxiliary decision meets the decision output conditions based on preset decision review rules. If the initial auxiliary decision meets the decision output conditions, then the initial auxiliary decision is determined as the intelligent auxiliary decision corresponding to the task to be decided.

3. The enterprise business intelligent auxiliary decision-making method based on a multi-source large language model according to claim 1, characterized in that, The target model scheduler of the target intelligent auxiliary decision-making system determines the target large language model corresponding to the task type from a pre-created pool of large language models based on the task type corresponding to the task to be decided and a pre-acquired target mapping table, including: The target model scheduler of the target intelligent auxiliary decision-making system determines all target large language models based on the task type corresponding to the task to be decided and the pre-acquired target mapping table, and determines the target call chain of all the target large language models; Accordingly, the step of generating intelligent auxiliary decision-making related to the target enterprise's business based on the target large language model using the structured parameter package and the target prompt word template includes: Based on the target call chain, each of the target large language models is invoked to obtain the intelligent auxiliary decision-making corresponding to the target large language model based on the structured parameter package and the target prompt word template.

4. The enterprise business intelligent auxiliary decision-making method based on a multi-source large language model according to claim 1, characterized in that, Also includes: The target receiving end corresponding to the intelligent assisted decision is determined based on the task output label in the target label data of the task to be decided; Determine the target decision output rule corresponding to the target receiving end, and send the intelligent auxiliary decision to the target receiving end based on the target decision output rule.

5. The enterprise business intelligent auxiliary decision-making method based on a multi-source large language model according to claim 1, characterized in that, Also includes: The feedback results corresponding to the intelligent assisted decision-making are obtained using the target data integration interface; The intelligent assisted decision-making is quantitatively evaluated based on the feedback results using a target quantification model to determine the corresponding quantitative evaluation results, and the quantitative evaluation results are visualized and output using target visualization rules. The target large language model is updated using a preset target model update rule based on the task to be decided and the corresponding quantitative evaluation result, so as to execute the task to be decided using the updated target large language model.

6. The enterprise business intelligent auxiliary decision-making method based on a multi-source large language model according to claim 5, characterized in that, After using the target quantification model to quantify the intelligent assisted decision-making based on the feedback results to determine the corresponding quantification evaluation results, the method further includes: Based on all the tasks to be decided and the corresponding quantitative evaluation results, the customers to be activated and the corresponding customer activation tasks are determined from the customer information database. The customer activation task is identified as the new task to be decided, and the process proceeds to the step of parsing the task to be decided to determine the target tag data.

7. The enterprise business intelligent auxiliary decision-making method based on a multi-source large language model according to claim 1, characterized in that, Also includes: Obtain a mapping table update instruction, and update the model registry in the target model scheduler based on the mapping table maintenance instruction, so as to update the target mapping table based on the updated model registry.

8. An enterprise business intelligent auxiliary decision-making device based on a multi-source large language model, characterized in that, Applications to terminal devices equipped with a target-based intelligent auxiliary decision-making system include: The task parsing module is used to acquire decision-making tasks related to the target enterprise's business, and to parse the decision-making tasks to determine target tag data; the target tag data includes product category, channel category, and customer category; The parameter package generation module is used to utilize the target business knowledge platform of the target intelligent auxiliary decision-making system to obtain target profile data corresponding to each target tag data from the target business profile database of the corresponding dimension based on each target tag data, and generate a structured parameter package corresponding to the task to be decided based on all the target profile data; the target business profile database includes a product information database, a channel content database, and a customer information database; The model determination module is used to determine the target large language model corresponding to the task type from a plurality of pre-created large language models based on the task type corresponding to the task to be decided and a pre-acquired target mapping table, using the target model scheduler of the target intelligent auxiliary decision-making system; wherein, the target mapping table is used to store the mapping relationship between each task type and each of the large language models in the target intelligent auxiliary decision-making system; different large language models are models built based on different types of data sources; The auxiliary decision generation module is used to obtain a pre-configured target prompt word template related to the target enterprise's business, and use the target large language model to generate intelligent auxiliary decisions related to the target enterprise's business based on the structured parameter package and the target prompt word template, so as to complete the target enterprise's business based on the intelligent auxiliary decisions.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the enterprise business intelligent auxiliary decision-making method based on a multi-source large language model as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, Used to store computer programs, wherein the computer programs, when executed by a processor, implement the enterprise business intelligent auxiliary decision-making method based on a multi-source large language model as described in any one of claims 1 to 7.

Images