Marketing decision-making method and device, equipment and medium

Abstract

The invention discloses a marketing decision-making method and device, equipment and a medium. The method comprises the steps of outputting marketing decision-making data for marketing data through a large model, and performing supervised fine tuning training on a basic inference model based on the marketing data, the marketing decision-making data and an experience library to obtain an initial marketing decision-making model; performing reinforcement learning on the initial marketing decision model by using the real marketing data to obtain a marketing decision model; and carrying out marketing decision making by using the marketing decision making model and the experience library. In this way, on one hand, the basic reasoning model learns prior knowledge provided by a large model, richer marketing experience data is obtained, and the accuracy of marketing decision making is improved. And on the other hand, the marketing decision model obtained by training the basic reasoning model can perform reasoning analysis on the marketing data, features do not need to be selected through feature engineering, and the problem of low decision accuracy caused by incomplete features is avoided.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a marketing decision-making method, device, equipment and medium. Background Art

[0002] Intelligent decision-making refers to the process of automatically filtering and analyzing data, and making decisions based on these analyses, using technologies like machine learning and deep learning. It has widespread applications in fields such as finance, healthcare, and logistics, effectively improving efficiency, reducing costs, and helping people make more informed decisions.

[0003] Currently, training marketing decision-making systems primarily relies on feature engineering, which aims to select more representative features to improve model efficiency and accuracy. Such systems can only train machine learning or deep learning models based on a few manually selected representative features and limited training data. However, these manually selected features are imperfect, resulting in insufficient generalization capabilities after training. Furthermore, the decision-making process based on neural networks is a "black box," lacking transparency and explainability, making its reasoning process untrustworthy.

[0004] Therefore existing technology still needs to be improved and improved. Summary of the Invention

[0005] The technical problem to be solved by this application is to provide a marketing decision-making method, device, equipment and medium to address the shortcomings of the existing technology.

[0006] In order to solve the above technical problems, the first aspect of the present application provides a marketing decision-making method, wherein the marketing decision-making method specifically includes: Outputting marketing decision data from the marketing data using a large model, wherein the marketing data includes a target to be marketed and a number of potential customers, and the marketing decision data includes a marketing decision conclusion and marketing decision reasons for each potential customer as to whether to purchase the target to be marketed; Performing supervised fine-tuning training on the basic reasoning model based on the marketing data, the marketing decision data, and the experience library to obtain an initial marketing decision model; Performing reinforcement learning on the initial marketing decision model using real marketing data to obtain a marketing decision model; The marketing decision model and the experience database are used to make marketing decisions.

[0007] The marketing decision method, wherein, in the process of outputting marketing decision data from the marketing data through the large model, first experience data is obtained; The first experience data is clustered into the experience database to update the experience database.

[0008] The marketing decision-making method, before performing supervised fine-tuning training on the basic inference model based on the marketing data, the marketing decision data, and the experience library to obtain the initial marketing decision model, further comprises: Performing consistency check on the marketing decision conclusion and the marketing decision reasons; If the consistency check passes, the marketing decision data is used as the label data of the corresponding marketing data; If the consistency check fails, the marketing decision data and its corresponding marketing data are filtered out.

[0009] In the marketing decision-making method, the step of performing reinforcement learning on the initial marketing decision model to obtain the marketing decision model specifically includes: Obtain a real marketing data set, each of which includes several potential customers, marketing targets, and the actual sales conclusions corresponding to each potential customer; Based on the actual marketing data and the experience database, generating predicted marketing decision data through the initial marketing decision model, wherein the predicted marketing decision data includes a predicted marketing decision conclusion and a reason for the predicted marketing decision of whether each potential customer in the actual marketing data will purchase the target marketing object; Determining a reinforcement learning reward based on the predicted marketing decision data and the actual sales conclusion; The initial marketing decision model is optimized based on the reward to obtain a marketing decision model.

[0010] The marketing decision-making method, wherein, in the process of performing reinforcement learning on the initial marketing decision model using real marketing data, second experience data is also obtained; The second experience data is clustered into the experience database to update the experience database.

[0011] In the marketing decision-making method, determining the reinforcement learning reward based on the predicted marketing decision data and the actual sales conclusion specifically includes: Obtaining the number of correct customers in the predicted marketing decision conclusion in the predicted marketing decision data, wherein the predicted marketing decision conclusion for the correct customers is consistent with its corresponding real sales conclusion; The reward is calculated based on the number of correct customers and the number of potential customers in the real marketing data.

[0012] The marketing decision-making method, wherein the multiple real marketing data include the real marketing data of the target to be marketed and the real marketing data of other targets to be marketed, wherein the real sales conclusions in the real marketing data include sales success and / or sales failure.

[0013] A second aspect of the present application provides a marketing decision-making device, wherein the marketing decision-making device specifically includes: A distillation module is configured to output marketing decision data based on the marketing data using a large model, wherein the marketing data includes a target to be marketed and a number of potential customers, and the marketing decision data includes a marketing decision conclusion and marketing decision reasons for each potential customer regarding whether to purchase the target to be marketed; A training module, configured to perform supervised fine-tuning training on a basic inference model based on the marketing data, the marketing decision data, and an experience database to obtain an initial marketing decision model; A reinforcement learning module, configured to perform reinforcement learning on the initial marketing decision model using real marketing data to obtain a marketing decision model; A decision-making module is used to make marketing decisions using the marketing decision model and the experience library.

[0014] A third aspect of the present application provides a computer-readable storage medium, which stores one or more programs. The one or more programs can be executed by one or more processors to implement the steps in any of the marketing decision-making methods described above.

[0015] A fourth aspect of the present application provides a terminal device, comprising: a processor and a memory; The memory stores a computer-readable program executable by the processor; When the processor executes the computer-readable program, the steps in any of the above-mentioned marketing decision-making methods are implemented.

[0016] Beneficial effects: Compared with the existing technology, the present application provides a marketing decision method, device, equipment and medium, the method comprising: outputting marketing decision data from marketing data through a large model; performing supervised fine-tuning training on a basic inference model based on the marketing data, the marketing decision data and the experience library to obtain an initial marketing decision model; performing reinforcement learning on the initial marketing decision model using real marketing data to obtain a marketing decision model; and making marketing decisions using the marketing decision model and the experience library. In this way, on the one hand, the basic inference model learns the prior knowledge provided by the large model, obtains richer marketing experience data, and improves the accuracy of marketing decisions. On the other hand, the marketing decision model obtained after training the basic inference model can perform inference analysis on marketing data without the need to select features through feature engineering, thereby avoiding the problem of low decision accuracy due to incomplete features. In addition, the marketing decision model outputs the reasons for the marketing decision while outputting the marketing decision conclusion, thereby improving the comprehensibility of the marketing decision conclusion. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0018] Figure 1 A flowchart of a marketing decision-making method provided in an embodiment of the present application.

[0019] Figure 2 A schematic diagram of the principle flow of the marketing decision-making method provided in an embodiment of the present application.

[0020] Figure 3 This is a functional block diagram of the marketing decision-making device provided in an embodiment of the present application.

[0021] Figure 4 This is a block diagram of the principles of the terminal device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0022] The embodiments of this application provide a marketing decision-making method, apparatus, device, and medium. To make the objectives, technical solutions, and effects of this application more clear and explicit, the application is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are intended only to explain this application and are not intended to limit this application.

[0023] It will be understood by those skilled in the art that, unless otherwise stated, the singular forms "a," "an," "said," and "the" used herein may also include the plural forms. It should be further understood that the term "comprising" used in the specification of this application refers to the presence of the stated features, integers, steps, operations, elements, and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. The term "and / or" used herein includes all or any unit and all combinations of one or more associated listed items.

[0024] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by those skilled in the art to which this application belongs. It should also be understood that terms such as those defined in common dictionaries should be understood to have meanings consistent with their meanings in the context of the prior art and will not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0025] It should be understood that the sequence numbers and sizes of the steps in this embodiment do not imply the order of execution. The order of execution of each process is determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of this application.

[0026] Research has shown that intelligent decision-making refers to the process of automatically filtering and analyzing data, and making decisions based on these analyses, using technologies such as machine learning and deep learning. It has widespread application in fields such as finance, healthcare, and logistics, effectively improving efficiency, reducing costs, and helping people make more informed decisions.

[0027] Currently, marketing decision-making systems rely primarily on feature engineering to train them. The goal of feature engineering is to select more representative features to improve the efficiency and accuracy of the model. However, existing decision-making systems commonly suffer from the following problems: 1. Incomplete features: Machine learning models or deep learning models can only be trained based on some manually selected representative features and limited training data. However, the manually selected features are not perfect, and there may be problems such as insufficient features or even omissions.

[0028] 2. Poor explainability: The decision-making process of neural network-based intelligent agents is a black box, lacking transparency and explainability, making it impossible for humans to understand their reasoning process and trust their decision conclusions.

[0029] 3. Insufficient generalization: The training process is designed for a single specific task, lacks interactive capabilities and has no prior knowledge, making it difficult to apply to real-world decision-making tasks that require natural language understanding, resulting in insufficient generalization capabilities.

[0030] In order to solve the above problems, in an embodiment of the present application, a large model is used to output marketing decision data for marketing data, and a basic reasoning model is fine-tuned based on the marketing data, the marketing decision data, and the experience library to obtain an initial marketing decision model; the initial marketing decision model is reinforced with real marketing data to obtain a marketing decision model; and marketing decisions are made using the marketing decision model and the experience library. In this way, on the one hand, the basic reasoning model learns the prior knowledge provided by the large model, obtains richer marketing experience data, and can improve the accuracy of marketing decisions. On the other hand, the marketing decision model obtained after training the basic reasoning model can perform reasoning analysis on marketing data without the need to select features through feature engineering, thereby avoiding the problem of low decision accuracy due to incomplete features. In addition, the marketing decision model outputs the reasons for the marketing decision while outputting the marketing decision conclusion, thereby improving the comprehensibility of the marketing decision conclusion.

[0031] The application content will be further explained below through description of embodiments in conjunction with the accompanying drawings.

[0032] This embodiment provides a marketing decision-making method, such as Figure 1 and Figure 2 As shown, the method includes: S10. Output marketing decision data based on marketing data through a large model.

[0033] Specifically, marketing data includes several potential customers and potential marketing targets. The potential marketing targets include information such as the target's name and functional description. Potential customers include basic information such as their name, business scope, and founding date. They may also include commercial information such as job postings, financial reports, and news. Potential customers can be businesses or individuals, while potential marketing targets can be products or services. For example, potential marketing targets include software product A and video editing services. Several potential customers include potential customer A and potential customer B. Potential customer A is XXX Co., Ltd., a company specializing in the production of packaging materials; potential customer B is XXX agency, whose primary businesses include live streaming and short videos.

[0034] The marketing decision data includes the marketing decision conclusion and marketing decision reasons for each potential customer regarding whether or not to purchase the marketing target. Marketing decision conclusions include conclusions regarding whether the potential customer will purchase the marketing target or not, and marketing decision reasons provide the rationale for the marketing decision conclusion. For example, the marketing decision conclusion for potential customer A is: unlikely to purchase; the marketing decision reason is: potential customer A specializes in the production of packaging materials, with a focus on manufacturing and supply chain management. Software Product A's features (such as video editing) are not directly related to potential customer A's business information (packaging material production). While they may occasionally need to create product promotional videos, their business information does not mention video editing or content creation needs, and Software Product A's features cannot address their production optimization issues.

[0035] The large model in this application is used to distill marketing data. Marketing data is input into the large model to generate marketing decision data. In the process of outputting marketing decision data from the marketing data through the large model, the language, world, and common sense knowledge learned by the large model itself are used to provide prior knowledge / experience for marketing decisions, so as to obtain a data set including marketing data and marketing decision data for training. Then, high-quality data sets are screened out from the data set to perform supervised learning on the basic reasoning model. Among them, the large model selects a large language model with a large number of model parameters (such as Qwen3-235B-A32B, etc.), and uses the large language model with large-scale parameters to screen the data set to obtain higher-quality training data. Using high-quality training data to train the basic reasoning model can not only reduce the training cost of supervised learning, but also improve the decision-making efficiency of the trained marketing decision model.

[0036] In one embodiment, the big model distills the marketing data to obtain several high-quality data groups including marketing data and marketing decision data, wherein the distillation process of the big model is: constructing input data for the big model based on the marketing data, and the big model generates output data based on the input data, the output data includes an explainable marketing decision process and marketing decision data, the explainable marketing decision process is the reasoning process for generating the marketing decision data, and the marketing decision data includes the marketing decision conclusion of whether each potential customer purchases the marketing object and the marketing decision reasons for making the marketing decision conclusion.

[0037] The input data can include marketing data, experience libraries, decision prompts, and output format requirements. Decision prompts can include marketing decision tasks and experience library supplementation tasks. These prompts inform the macro model of the tasks to be performed. Output format requirements guide the macro model in generating marketing decision data according to a specific output format. For example, the output format requirement may include outputting each potential customer, the marketing decision conclusion, and the rationale for the marketing decision.

[0038] The experience database can be pre-built and used as prior knowledge for marketing decisions, providing the big model with reasons why potential customers might be interested in purchasing the target. The experience database can include business rules and expert experience. Business rules indicate the overlap between a potential customer's basic data and / or business information and the target. For example, a potential customer whose business scope overlaps with the target's functionality is likely to purchase the target. Expert experience indicates the relevance between a potential customer's business information and the target's functionality. For example, if a potential customer's job posting requires talent that uses the target, the customer is likely to purchase the target.

[0039] It should be noted that the pre-built experience database may include a number of business rules and / or expert experience, or it may be empty. In the process of the large model outputting marketing decision data based on marketing data, the large model may obtain experience data (recorded as first experience data) from the reasoning process and cluster the first experience data into the experience database to expand the experience database. In other words, the first experience data is used as supplementary information for the experience database. Clustering the first experience data into the experience database allows the experience database to include more appropriate reasons for purchase, thereby improving the completeness of the reasons for purchase included in the experience database. In this embodiment, the experience database is updated using a clustering method, which can automatically extract the first experience data when business rules and / or expert experience are difficult to express in words, providing a more comprehensive perspective for subsequent decision-making.

[0040] To further illustrate the distillation process of large models, a specific example is given here.

[0041] Example 1 The target market includes software product A and video editing services. Potential customers include potential customer A and potential customer B. Potential customer A includes a limited company specializing in the production of packaging materials with a staff of 10-20 people. Potential customer B includes a brokerage firm with a staff of 500-900 people, specializing in live streaming and short video. The experience base includes potential customers whose business scope overlaps with the target market's functionalities and who may purchase the target market. Potential customers whose job postings include a talent need for the target market may also purchase the target market. Therefore, the distillation process of the large model can be: a. Construct input data. The input data is: Decision-making prompts (including marketing decision tasks and experience library supplementation tasks): You need to market a software product A. Select potential customers who are likely to purchase software product A from a given pool of potential customers. Analyze and explain in detail the reasons why these potential customers would purchase software product A. For each given potential customer, output a marketing decision conclusion and reasons for the decision. Also, supplement and improve the given experience library and make analytical decisions based on this library.

[0042] Experience Library: Empirical data 1 (business rule): Potential customers whose business scope overlaps with the functions of the target market are likely to purchase the target market; Experience data 2 (expert experience): Potential customers whose recruitment information contains talent needs for the target market are likely to purchase the target market.

[0043] Target audience: Name of the marketing target: Software Product A; Functional description of the target audience: Provide video editing services.

[0044] Potential Customers: Potential Customer A: A limited company specializing in the production of packaging materials, with a staff of 10-20 people. Potential client B: A brokerage company with main businesses such as live streaming and short videos, and a staff size of 500-900 people.

[0045] Output format requirements: For each potential customer, output the potential customer name, marketing decision conclusion, and marketing decision reasons; Lead Name: Conclusion; Reason for marketing decision: xxx.

[0046] b. The large model makes marketing decision inferences based on the input data, and outputs explainable marketing decision processes and marketing decision data according to the output format requirements.

[0047] The output example is ( <think> Marketing decision-making process< / think>, marketing decision data), for example, explainable marketing decision-making process: <think>Potential Customer A is a manufacturing company whose business focuses on producing physical products, not content creation. Video editing software may not be directly relevant to them, unless they have marketing needs, such as creating product promotional videos. However, the user-provided information does not indicate any video production needs. Potential Customer B is an agency whose business includes live streaming and short videos. Live streaming and short videos require video editing tools to produce high-quality content.

[0048] Therefore, potential customer B is more likely to purchase software product A.

[0049] Now, let’s analyze the reasons why potential customer B purchased in detail: 1. Business relevance: The company's business involves live streaming and short videos, which frequently require video editing. Software Product A's features can directly improve content quality and attract more viewers.

[0050] 2. Cost-effectiveness: As a popular software, it may be more economical than professional software and suitable for the small and medium-sized teams that may be in the brokerage firm.

[0051] 3. Influence: The high download volume of software product A indicates that it is reliable and popular, and enterprises may be more likely to adopt it.

[0052] For potential customer a, the reason for purchase is not valid.

[0053] < / think> Marketing decision data: In this scenario, the size of the enterprise will affect its willingness to purchase. Generally, larger enterprises are more likely to purchase genuine software due to concerns about legal risks.

[0054] Potential Customer A's Marketing Decision Conclusion: Unlikely to Purchase; Marketing Decision Reasoning: Potential Customer A specializes in the production of packaging materials, with a focus on manufacturing and supply chain management. Software Product A's functionality (video editing) is not directly related to their core business (packaging material production). While they may occasionally need to produce product promotional videos, the information does not mention video editing or content creation needs, and Software Product A's functionality cannot solve their production optimization problems.

[0055] Potential Customer B's Marketing Decision Conclusion: Highly Likely to Purchase; Marketing Decision Reasoning: Potential Customer B's business model involves live streaming and short videos, which are highly dependent on the creation, editing, and distribution of video content. Furthermore, Potential Customer B is large in scale, and the features of Software Product A will directly improve their company's operational efficiency, content quality, and user engagement.

[0056] The first empirical data output by the large model is: The size of the potential customer will affect the willingness to buy from the sales target.

[0057] Clustering the first experience data into the experience database to expand and update the experience database can provide richer marketing experience for subsequent decision-making.

[0058] S20. Perform supervised fine-tuning training on the basic reasoning model based on the marketing data, the marketing decision data, and the experience library to obtain an initial marketing decision model.

[0059] Specifically, the basic reasoning model does not have the logical reasoning ability of the large model before training, but after supervised fine-tuning training, it can be used as the initial marketing decision model. The model parameters of the basic reasoning model are smaller than those of the large model. For example, the model parameters of the large model are more than 70B (such as Qwen3-235B-A32B), and the model parameters of the basic reasoning model are 7B (such as Qwen2.5-7B-Instruct). The embodiment of the present application uses a basic reasoning model with a small number of model parameters for training to obtain reasoning capabilities comparable to those of a large model with a large number of model parameters, thereby improving the decision-making efficiency of marketing decisions while ensuring the accuracy of marketing decisions.

[0060] Furthermore, the marketing data and experience database are used to construct the input data for the basic inference model, while the marketing decision data is used as label data for supervised learning of the basic inference model, enabling it to match the marketing decision-making capabilities of the large model. Since the experience database is expanded and updated using the first experience data when the large model generates marketing decision data, the updated experience database can be used when training the basic inference model, allowing the basic inference model to leverage the decision-making knowledge learned by the large model.

[0061] It should be noted that the data format of the input data and output data of the basic inference model is the same as that of the large model, both of which are text input and text output. That is, the input data and output data are all text descriptions, and there is no need for complex feature engineering, text feature extraction, feature conversion and other processes.

[0062] Furthermore, in practical applications, since marketing decision data is generated through decision reasoning using large models, the reasoning process of the large models may result in inference errors, which may lead to erroneous data in the marketing decision data. Therefore, to ensure the training effect of the basic inference model, before using the marketing decision data as labeled data for supervised learning of the basic inference model, the marketing decision data can be screened to improve the data quality of the marketing decision data used to train the basic inference model and ensure the training effect of the basic inference model.

[0063] Exemplarily, before performing supervised fine-tuning training on the basic inference model based on the marketing data, the marketing decision data, and the experience library to obtain the initial marketing decision model, the method further includes: Performing consistency check on the marketing decision conclusion and the marketing decision reasons; If the consistency check passes, the marketing decision data is used as the label data of the corresponding marketing data; If the consistency check fails, the marketing decision data and its corresponding marketing data are filtered out.

[0064] Specifically, consistency verification involves verifying whether the marketing decision conclusion is consistent with the marketing decision reasoning. Consistency verification is used to verify the data quality of marketing decision data to determine whether it can be used to train the basic reasoning model. This consistency serves as the basis for the marketing decision data to be used to train the basic reasoning model. Marketing decision data that passes the consistency verification can be used to train the basic reasoning model, meaning that the marketing decision data can be used as label data for its corresponding marketing data. Conversely, marketing decision data that fails the consistency verification cannot be used to train the basic reasoning model, meaning that the marketing decision data cannot be used as label data for its corresponding marketing data. Instead, the marketing decision data and its corresponding marketing data can be directly filtered out.

[0065] In one embodiment, the consistency check between the marketing decision conclusion and the marketing decision reason also includes an integrity check. The integrity check includes: (1) Potential customer integrity verification: used to verify whether the marketing decision data includes every potential customer in the marketing data (i.e., whether marketing decision reasoning is performed for each potential customer); (2) Decision integrity verification: This is used to verify whether each potential customer has a corresponding marketing decision conclusion and marketing decision reason. In other words, by performing integrity verification on the marketing decision conclusion and marketing decision reason, it can be verified whether the marketing decision data includes the marketing decision conclusion and marketing decision reason for each potential customer in the marketing data.

[0066] Specifically, the verification process of performing integrity verification on the marketing decision conclusion and the marketing decision reason may be: Detect whether the marketing decision data includes each potential customer to verify the integrity of the marketing decision data; If the potential customer integrity check passes, then check whether each potential customer has a corresponding marketing decision conclusion and marketing decision reason, so as to perform decision integrity check on the marketing decision data; If the decision integrity check passes, then it is determined that the marketing decision conclusion and the marketing decision reason integrity check pass; If the potential customer integrity check fails, or the decision integrity check fails, it is determined that the marketing decision conclusion and the marketing decision reason integrity check fail.

[0067] Specifically, a successful potential customer integrity check indicates that the marketing decision data includes every potential customer. Conversely, a failed potential customer integrity check indicates that at least one potential customer is not included in the marketing decision data. A successful decision integrity check indicates that every potential customer has a corresponding marketing decision conclusion and marketing decision reason. Conversely, a failed decision integrity check indicates that at least one potential customer does not have a corresponding marketing decision conclusion and / or marketing decision reason.

[0068] For example: Assume that potential customers include potential customer a and potential customer b; If the marketing decision data includes potential customer a and potential customer b (i.e., the potential customer integrity check passes), and both potential customer a and potential customer b have corresponding marketing decision conclusions and marketing decision reasons (decision integrity check passes), then the consistency check of the marketing decision conclusions and marketing decision reasons passes; If the marketing decision data does not include potential customer a and / or potential customer b (i.e., the decision integrity check passes), or the marketing decision data of potential customer a and / or the marketing decision data of potential customer b do not include the marketing decision conclusion and / or the marketing decision reason (i.e., the decision integrity check fails), then it is determined that the integrity check of the marketing decision conclusion and the marketing decision reason fails.

[0069] This application performs potential customer integrity verification, decision integrity verification and consistency verification on marketing decision data. Potential customer integrity verification ensures that the marketing decision data includes every potential customer in the marketing data. Decision integrity verification ensures that every potential customer has a corresponding marketing decision conclusion and marketing decision reason. Consistency verification ensures whether the marketing decision conclusion is consistent with the marketing decision reason, thereby screening out high-quality marketing data and marketing decision data pairs, and then using high-quality data to train the initial marketing decision model, which can improve the training quality and model performance of the model.

[0070] In another embodiment, the consistency check between the marketing decision conclusion and the marketing decision reason can be performed using a large model to verify the consistency between the decision conclusion in the decision reasoning process and the marketing decision conclusion in the marketing decision data. Specifically, the consistency check between the marketing decision conclusion and the marketing decision reason includes: Construct consistency check prompt words; The consistency check prompt words and the marketing decision data are input into the large model, and a consistency check result is output, wherein the consistency check result is a check pass or a check fail.

[0071] Specifically, the consistency verification prompt is used to inform the large model of the tasks that need to be performed, which may include consistency verification tasks. The consistency verification tasks are used to verify the consistency between the decision conclusions in the decision reasoning process and the marketing decision conclusions in the marketing decision data.

[0072] For example: Consistency check prompt words: You need to market a software product A. You have analyzed potential customers A and B to see whether they will purchase the software product A. Please determine whether the decision-making process and marketing decision conclusion are consistent.

[0073] Decision-making process: xxx Marketing decision conclusion: xxx Output format: [“potential customer a”: “agree / inconsistent”, “potential customer b”: “agree / inconsistent”].

[0074] The embodiment of the present application utilizes the knowledge information learned in the process of generating marketing decision data using a large model to verify the generated decision marketing data, which can quickly filter out marketing decision data whose decision conclusions in the decision process are inconsistent with the marketing decision conclusions. This not only improves the data quality of the marketing decision data, but also utilizes the knowledge information learned in the large model reasoning process to quickly screen the marketing decision data, thereby improving the screening speed of the marketing decision data.

[0075] In another embodiment, the potential customer integrity and decision integrity verification can be combined with the consistency verification, that is, the marketing decision data can be verified first to see whether it includes the marketing decision conclusions and marketing decision reasons for each potential customer in the marketing data, and then the consistency of the decision conclusions in the decision reasoning process and the marketing decision conclusions in the marketing decision data can be verified when the integrity verification passes; or, the consistency of the decision conclusions in the decision reasoning process and the marketing decision conclusions in the marketing decision data can be verified first, and then the marketing decision data can be verified after the consistency verification passes to see whether it includes the marketing decision conclusions and marketing decision reasons for each potential customer in the marketing data.

[0076] Exemplarily, performing consistency check on the marketing decision conclusion and the marketing decision reason may specifically include: Verify the integrity of potential customers and decisions based on marketing decision conclusions and reasons; Based on the marketing decision conclusion, the consistency between the decision conclusion in the decision reasoning process and the marketing decision conclusion in the marketing decision data is verified through the big model; If the potential customer integrity check, decision integrity check, and consistency of the decision conclusion in the decision reasoning process with the marketing decision conclusion in the marketing decision data are all verified, the consistency check is considered to have passed; If at least one of the potential customer integrity check, decision integrity check, and consistency between the decision conclusion in the decision reasoning process and the marketing decision conclusion in the marketing decision data has not been verified, the consistency check is determined to have failed.

[0077] It should be noted that the verification process for the integrity of potential customers, the integrity of decisions, and the consistency of the decision conclusions in the decision reasoning process with the marketing decision conclusions in the marketing decision data is the same as the implementation process of the above embodiment and will not be described here. The consistency verification of the integrity of potential customers, the integrity of decisions, and the consistency of the decision conclusions in the decision reasoning process with the marketing decision conclusions in the marketing decision data can also be performed simultaneously, and there is no specific restriction here. The high-quality marketing quantity, marketing decision data, and experience library obtained after the consistency verification is passed will be used for supervised fine-tuning training of the basic reasoning model to obtain the initial marketing decision model.

[0078] S30. Perform reinforcement learning on the initial marketing decision model using real marketing data to obtain a marketing decision model.

[0079] Specifically, the initial marketing decision model is a basic reasoning model (such as Qwen2.5-7B-Instruct) that has been trained through supervised learning and has acquired certain marketing decision-making capabilities. The initial marketing decision model is then trained through reinforcement learning to improve the model performance of the learned marketing decision model. At the same time, this application uses a basic reasoning model (such as a large language model, etc.) as the marketing decision model. Various forms of text descriptions can be used as input data for the marketing decision model. There is no need for complex feature engineering, text feature extraction, feature conversion, etc., and there is no need to select features for the marketing decision model through feature engineering. This avoids the problem of insufficient or even missing features due to manually selected features, and solves the problem of inaccurate decision conclusions due to incomplete features.

[0080] In one embodiment, the using of real marketing data to perform reinforcement learning on the initial marketing decision model to obtain the marketing decision model specifically includes: Obtain real marketing datasets; Based on the actual marketing data and the experience database, generating predicted marketing decision data through the initial marketing decision model, wherein the predicted marketing decision data includes a predicted marketing decision conclusion and a reason for the predicted marketing decision of whether each potential customer in the actual marketing data will purchase the target marketing object; Determining a reinforcement learning reward based on the predicted marketing decision data and the actual sales conclusion; The initial marketing decision model is optimized based on the reward to obtain a marketing decision model.

[0081] Specifically, real marketing data is relatively small. Each piece of real marketing data includes a number of potential customers, target customers, and real sales conclusions corresponding to each potential customer. Real sales conclusions can include successful sales and / or unsuccessful sales. The embodiments of this application construct a real marketing dataset by mixing data on successful and unsuccessful sales. This increases the richness and diversity of the real marketing dataset, thereby improving the generalization capability of the marketing decision model.

[0082] It should be noted that, in actual applications, the real marketing data set may include only one real marketing data set of the object to be marketed, or, when the marketing data of the object to be marketed is very small, it may also include some real marketing data sets of other marketing objects similar to the object to be marketed, etc., to enrich the real marketing data.

[0083] Furthermore, the experience database can be an experience database updated with the first experience data, or a pre-built experience database. In the embodiment of the present application, the experience database is an experience database updated with the first experience data, so that more experience data can be utilized in the reinforcement learning structure. Since the initial marketing decision model is a trained basic inference model, the input data and output data of the initial marketing decision model are the same as the input data and output data of the large model, and will not be further described here.

[0084] Furthermore, the reward is the reward earned through reinforcement learning, which is used to optimize the initial marketing decision model. The reward is determined based on the accuracy of the initial marketing decision model's predicted marketing decision data and is positively correlated with the accuracy of the predicted marketing decision data. That is, the higher the accuracy of the predicted marketing decision data, the higher the reward, and conversely, the lower the accuracy of the predicted marketing decision data, the lower the reward.

[0085] In one embodiment, determining the reinforcement learning reward based on the predicted marketing decision data and the actual sales conclusion specifically includes: Obtaining the correct number of customers in the predicted marketing decision conclusion in the predicted marketing decision data; The reward is calculated based on the number of correct customers and the number of potential customers in the real marketing data.

[0086] Specifically, the correct customer is a potential customer whose predicted marketing decision conclusion is consistent with its corresponding real sales conclusion. That is to say, after obtaining the predicted marketing decision data, the predicted marketing decision conclusion of each potential customer can be compared with its corresponding real sales conclusion. If the predicted marketing decision conclusion of several potential customers is consistent with the real sales conclusion, then the potential customer is determined to be a correct customer. Then, the ratio of the number of correct customers whose predicted marketing decision conclusion is consistent with the real sales conclusion to the number of all potential customers is used as a reward, that is, reward = number of correct customers / number of potential customers. The embodiment of the present application constructs a reward based on the potential customers whose predicted marketing decision conclusion is consistent with their corresponding real sales conclusion, and optimizes the initial marketing decision model through the reward, so that the initial marketing decision model learns the real sales conclusion (that is, a manifestation of expert experience), thereby improving the model performance of the initial marketing decision model.

[0087] In one embodiment, a basic inference model is first trained to obtain an initial marketing decision model, during which first empirical data is obtained. Further, when reinforcement learning is performed on the initial marketing decision model, second empirical data can also be obtained while the initial marketing decision model outputs predicted marketing decision data. Therefore, second empirical data is also obtained during the reinforcement learning of the initial marketing decision model using real marketing data. The second experience data is clustered into the experience database to update the experience database.

[0088] Specifically, the second experience data is obtained by the initial marketing decision model learning the real sales results. It is the implicit expert experience or business rules learned by the initial marketing model when learning the real sales results. In this way, the expert experience or business rules implicit in the real sales results are made explicit through the reinforcement learning process and added to the experience library, further improving the completeness of the experience library and providing richer prior knowledge / experience for the marketing decision model obtained after reinforcement learning.

[0089] S40: Make marketing decisions using the marketing decision model and the experience database.

[0090] Specifically, the experience database can adopt any of the experience databases mentioned in the above process. In a specific implementation, the experience database is an experience database that clusters the first experience data and the second experience data, which improves the adequacy of the experience data in the experience database and provides the marketing decision model with richer prior knowledge / experience. In other words, the experience database can provide accurate and refined experience data for the marketing decision model, so that the marketing decision model can more comprehensively consider the reasons why potential customer information can purchase the marketing object, thereby making more accurate marketing decisions. At the same time, it will also reduce repeated calculations in the reasoning process of the marketing decision model and improve the reasoning speed. On the other hand, since a small amount of real data can be used to train small and medium-sized reasoning models, a marketing decision model comparable to the capabilities of a large model can be quickly obtained, which solves the problem of insufficient generalization ability of traditional marketing decision models.

[0091] Furthermore, when utilizing the marketing decision model and experience database to make marketing decisions, potential customer information and target audiences are first obtained. Input data is then constructed based on this information, target audiences, and experience database. This input data is then fed into the marketing decision model, which then generates marketing decision data. The input data construction process and the data information included in the marketing decision data are identical to those described above and will not be further elaborated here. The marketing decision conclusions output by the marketing decision model are explainable and highly consistent with the rationale for the marketing decision, ensuring complete trust in potential customers.

[0092] In summary, this embodiment provides a marketing decision-making method, apparatus, device, and medium. The method includes outputting marketing decision data from marketing data using a large model; performing supervised fine-tuning training on a basic inference model based on the marketing data, the marketing decision data, and an experience library to obtain an initial marketing decision model; performing reinforcement learning on the initial marketing decision model using real marketing data to obtain a marketing decision model; and making marketing decisions using the marketing decision model and the experience library. This allows the basic inference model to learn the prior knowledge provided by the large model, obtaining richer marketing experience data and improving the accuracy of marketing decisions. Furthermore, the marketing decision model obtained after training the basic inference model can perform inference analysis on marketing data without the need for feature engineering to select features, thus avoiding the problem of low decision accuracy caused by incomplete features. Furthermore, the marketing decision model outputs the marketing decision reasoning along with the marketing decision conclusion, improving the comprehensibility of the marketing decision conclusion.

[0093] Based on the above marketing decision-making method, this embodiment provides a marketing decision-making device, such as Figure 3 As shown, the marketing decision-making device specifically includes: A distillation module 100 is configured to output marketing decision data based on the marketing data using a large model. The marketing data includes a target to be marketed and a number of potential customers. The marketing decision data includes a marketing decision conclusion and marketing decision reasons for each potential customer regarding whether to purchase the target to be marketed. A training module 200 is configured to perform supervised fine-tuning training on a basic inference model based on the marketing data, the marketing decision data, and the experience database to obtain an initial marketing decision model; A reinforcement learning module 300 is configured to perform reinforcement learning on the initial marketing decision model using real marketing data to obtain a marketing decision model; The decision module 400 is used to make marketing decisions using the marketing decision model and the experience database.

[0094] Based on the above-mentioned marketing decision-making method, this embodiment provides a computer-readable storage medium, which stores one or more programs. The one or more programs can be executed by one or more processors to implement the steps in the marketing decision-making method as described in the above-mentioned embodiment.

[0095] Based on the above marketing decision-making method, this application also provides a terminal device, such as Figure 4As shown, it includes at least one processor 20; a display screen 21; and a memory 22. It may also include a communications interface 23 and a bus 24. The processor 20, display screen 21, memory 22, and communications interface 23 can communicate with each other via bus 24. The display screen 21 is configured to display a preset user guidance interface in the initial setup mode. The communications interface 23 can transmit information. The processor 20 can invoke logic instructions in the memory 22 to execute the method described in the above embodiment.

[0096] In addition, the logic instructions in the memory 22 can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product.

[0097] The memory 22, as a computer-readable storage medium, can be configured to store software programs or computer-executable programs, such as program instructions or modules corresponding to the methods in the embodiments of the present disclosure. The processor 20 executes the software programs, instructions, or modules stored in the memory 22 to perform functional applications and data processing, thereby implementing the methods in the above embodiments.

[0098] The memory 22 may include a program storage area and a data storage area. The program storage area may store an operating system and at least one application required for a function; the data storage area may store data created based on the use of the terminal device. In addition, the memory 22 may include high-speed random access memory and non-volatile memory. For example, various media that can store program code, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, may also be transient storage media.

[0099] In addition, the specific process of loading and executing the multiple instructions in the storage medium and the processor in the terminal device has been described in detail in the above method and will not be described here one by one.

[0100] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A marketing decision-making method, characterized in that: The marketing decision-making method specifically includes: Outputting marketing decision data from the marketing data using a large model, wherein the marketing data includes a target to be marketed and a number of potential customers, and the marketing decision data includes a marketing decision conclusion and marketing decision reasons for each potential customer as to whether to purchase the target to be marketed; Performing supervised fine-tuning training on the basic reasoning model based on the marketing data, the marketing decision data, and the experience library to obtain an initial marketing decision model; Performing reinforcement learning on the initial marketing decision model using real marketing data to obtain a marketing decision model; The marketing decision model and the experience database are used to make marketing decisions.

2. The marketing decision-making method according to claim 1, characterized in that: In the process of outputting marketing decision data from the marketing data through the large model, first experience data is obtained; The first experience data is clustered into the experience database to update the experience database.

3. The marketing decision-making method according to claim 1, characterized in that: Before performing supervised fine-tuning training on the basic inference model based on the marketing data, the marketing decision data, and the experience database to obtain an initial marketing decision model, the method further includes: Performing consistency check on the marketing decision conclusion and the marketing decision reasons; If the consistency check passes, the marketing decision data is used as the label data of the corresponding marketing data; If the consistency check fails, the marketing decision data and its corresponding marketing data are filtered out.

4. The marketing decision-making method according to claim 1, characterized in that: The performing reinforcement learning on the initial marketing decision model to obtain the marketing decision model specifically includes: Obtain a real marketing data set, each of which includes several potential customers, marketing targets, and the actual sales conclusions corresponding to each potential customer; Based on the actual marketing data and the experience database, generating predicted marketing decision data through the initial marketing decision model, wherein the predicted marketing decision data includes a predicted marketing decision conclusion and a reason for the predicted marketing decision of whether each potential customer in the actual marketing data will purchase the target marketing object; Determining a reinforcement learning reward based on the predicted marketing decision data and the actual sales conclusion; The initial marketing decision model is optimized based on the reward to obtain a marketing decision model.

5. The marketing decision-making method according to claim 4, characterized in that: In the process of using real marketing data to perform reinforcement learning on the initial marketing decision model, second experience data is also obtained; The second experience data is clustered into the experience database to update the experience database.

6. The marketing decision-making method according to claim 4, characterized in that: The reward for reinforcement learning determined based on the predicted marketing decision data and the actual sales conclusion specifically includes: Obtaining the number of correct customers in the predicted marketing decision conclusion in the predicted marketing decision data, wherein the predicted marketing decision conclusion for the correct customers is consistent with its corresponding real sales conclusion; The reward is calculated based on the number of correct customers and the number of potential customers in the real marketing data.

7. The marketing decision-making method according to claim 4, characterized in that: The multiple real marketing data include real marketing data of the target to be marketed and real marketing data of other targets to be marketed, wherein the real sales conclusions in the real marketing data include successful sales and / or unsuccessful sales.

8. A marketing decision-making device, characterized in that: The marketing decision-making device specifically includes: A distillation module is configured to output marketing decision data based on the marketing data using a large model, wherein the marketing data includes a target to be marketed and a number of potential customers, and the marketing decision data includes a marketing decision conclusion and marketing decision reasons for each potential customer regarding whether to purchase the target to be marketed; A training module, configured to perform supervised fine-tuning training on a basic inference model based on the marketing data, the marketing decision data, and an experience database to obtain an initial marketing decision model; A reinforcement learning module, configured to perform reinforcement learning on the initial marketing decision model using real marketing data to obtain a marketing decision model; A decision-making module is used to make marketing decisions using the marketing decision model and the experience library.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps in the marketing decision-making method according to any one of claims 1 to 7.

10. A terminal device, characterized in that: include: processor and memory; The memory stores a computer-readable program executable by the processor; When the processor executes the computer-readable program, the steps of the marketing decision-making method according to any one of claims 1 to 7 are implemented.

Images