An activity evaluation processing method, apparatus, storage medium, and electronic device.

Abstract

This specification discloses an activity evaluation processing method, apparatus, storage medium, and electronic device. The method includes: obtaining a reference platform service promotion activity for an online platform service processing system; configuring a sandbox platform service simulation environment for the reference platform service promotion activity; controlling the sandbox platform service processing system to run the reference platform service promotion activity for each reference virtual user agent in the sandbox platform service simulation environment; driving the reference virtual user agents to make platform service operation decisions based on the reference platform service promotion activity based on the platform service processing large model to obtain the reference platform service decision results and the reference agent decision behavior interpretation; adjusting the activity configuration features of the reference platform service promotion activity based on the reference platform service decision results and the reference agent decision behavior interpretation to obtain the target platform service promotion activity for the online platform service processing system.

Description

Technical Field

[0001] This specification relates to the field of computer online platform service promotion technology, and in particular to an activity evaluation processing method, apparatus, storage medium and electronic device. Background Technology

[0002] In related technologies, service platforms provide services such as shopping, finance, express delivery, and insurance. To offer a better service experience to platform users, promotional activities are typically developed for these services. For example, in e-commerce, consumer finance, payment, and membership service scenarios, these promotions usually revolve around elements such as discounts, coupons, credit limits, points, or product bundles, aiming to guide users in their decision-making process regarding whether to participate in these promotions. To ensure consistent user experience and the security and reliability of the promotional activity process, the industry commonly adopts the practice of evaluating the effectiveness of promotional activities before their official launch.

[0003] With the popularization of big data and intelligent tools, service platforms often use offline playback, event tracking analysis, and data visualization to evaluate the expected effects of platform service promotion activities in combination with multi-dimensional indicators (such as conversion rate, average order value, user retention, etc.). Based on the evaluation results, they iteratively optimize the activity strategy so that it can more accurately reach target users in the real online platform service environment, enabling target users to better enjoy the benefits brought by the platform service promotion activities, while improving the platform service processing efficiency. Summary of the Invention

[0004] This specification provides an activity evaluation processing method, apparatus, storage medium, and electronic device, the technical solution of which is as follows:

[0005] Firstly, this specification provides an activity evaluation processing method, the method comprising:

[0006] Obtain reference platform service promotion activities for online platform service processing systems;

[0007] Configure a sandbox platform service simulation environment for the promotion activities of the reference platform service. The sandbox platform service simulation environment includes a sandbox platform service processing system and multiple reference virtual user agents.

[0008] In the sandbox platform service simulation environment, the sandbox platform service processing system controls the sandbox platform service to run the reference platform service promotion activities for each reference virtual user agent, and uses persona prompts based on the platform service processing big model to drive the reference virtual user agents to make platform service operation decisions based on the reference platform service promotion activities to obtain reference platform service decision results and reference agent decision behavior explanations. The reference agent decision behavior explanations include natural language descriptions of the subjective motivations, psychological activities, and cognitive logic of the decision behavior behind the platform service operation decision results using the platform service processing big model.

[0009] Based on the decision results of the reference platform service and the interpretation of the decision behavior of the reference agent, the activity configuration features of the promotion activities of the reference platform service are adjusted to obtain the target platform service promotion activities for the online platform service processing system.

[0010] In one feasible implementation, the sandbox platform service simulation environment is configured for the reference platform service promotion activity. The sandbox platform service simulation environment includes a sandbox platform service processing system and multiple reference virtual user agents, comprising:

[0011] Construct a sandbox platform service processing system for the online platform service processing system, and determine multiple reference virtual user agents from the virtual user agent pool for the promotion activities of the reference platform service;

[0012] Configure a sandbox platform service simulation environment based on the multiple reference virtual user agents and the sandbox platform service processing system.

[0013] In one feasible implementation, the method further includes:

[0014] User clustering is performed on historical user profile feature information to obtain a user group feature matrix;

[0015] Based on the user group feature matrix, feature combination is performed to obtain multiple individual user feature vectors, and prompt word engineering is performed on the individual user feature vectors to obtain persona prompt word information;

[0016] The aforementioned character prompts are associated with the platform service processing model to generate multiple virtual user agents;

[0017] A virtual user agent pool is formed based on all the aforementioned virtual user agents.

[0018] In one feasible implementation, the step of combining features based on the user group feature matrix to obtain multiple individual user feature vectors includes:

[0019] Based on the user group feature matrix, user group transformation is performed to obtain multiple user subgroup features. A benchmark user feature vector is selected within each user subgroup feature. A random perturbation of a preset amplitude is applied to the benchmark user feature vector to obtain multiple derived user feature vectors. The derived user feature vectors are used to express the individual behavioral differences within a homogeneous user group.

[0020] Multiple individual user feature vectors are obtained by random feature fusion based on the baseline user feature vector and the derived user feature vector.

[0021] In one feasible implementation, after forming a virtual user agent pool based on all the virtual user agents, the method further includes:

[0022] In the sandbox platform service simulation environment, the virtual decision-making results of each virtual user agent based on the historical platform service promotion activities are continuously collected, and the virtual decision-making results are compared with the historical behavior data of real users to obtain behavior difference information.

[0023] In the virtual user agent pool, the prompt words of the virtual user agent are adjusted based on the behavioral difference information to obtain the adjusted personality prompt words.

[0024] In one feasible implementation, the platform service processing big model utilizes persona prompt information to drive the reference virtual user agent to make platform service operation decisions based on the reference platform service promotion activities, obtaining reference platform service decision results and reference agent decision behavior interpretations, including:

[0025] Determine the persona prompt information corresponding to the reference virtual user agent, and generate agent platform service behavior decision prompts based on the persona prompt information and the reference platform service promotion activities;

[0026] The agent platform service behavior decision prompts are input into the platform service processing big model to make platform service operation decisions and obtain reference platform service decision results. The reference platform service decision results are then processed to obtain reference agent decision behavior explanations. Finally, the reference platform service decision results and the reference agent decision behavior explanations are output.

[0027] In one feasible implementation, the step of adjusting the activity plan for the reference platform service promotion activity based on the reference platform service decision results and the interpretation of reference agent decision behavior to obtain a target platform service promotion activity for the online platform service processing system includes:

[0028] Based on the decision-making results of the reference platform service and the interpretation of the decision-making behavior of the reference agent, the characteristic factors of the reference platform service promotion activities are adjusted to obtain the target platform service promotion activities for the online platform service processing system.

[0029] In one feasible implementation, obtaining reference platform service promotion activities for the online platform service processing system includes:

[0030] Obtain a multi-dimensional activity feature set, and generate a reference platform service promotion activity for the online platform service processing system based on the feature combination of the multi-dimensional activity feature set;

[0031] The characteristic factors of the reference platform service promotion activity are adjusted based on the reference platform service decision results and the interpretation of reference agent decision behavior to obtain the target platform service promotion activity for the online platform service processing system, including:

[0032] The reference platform service decision results and the reference agent decision behavior interpretation are subjected to feature mapping processing to obtain the contribution score of each feature factor in the multi-dimensional activity feature set;

[0033] The sensitive feature factors whose contribution scores exceed a preset threshold are identified, and local search optimization processing is performed in the feature factor value space of the sensitive feature factors to obtain a reference feature factor combination that meets a predetermined local search index.

[0034] Based on the combination of reference feature factors, the structured description language script of the reference platform service promotion activity is rewritten to form an optimized target platform service promotion activity for the online platform service processing system.

[0035] Secondly, this specification provides an activity evaluation processing device, the device comprising:

[0036] The activity acquisition module is used to acquire reference platform service promotion activities for the online platform service processing system;

[0037] The environment simulation module is used to configure a sandbox platform service simulation environment for the promotion activities of the reference platform service. The sandbox platform service simulation environment includes a sandbox platform service processing system and multiple reference virtual user agents.

[0038] The operation decision module is used to control the sandbox platform service processing system to run the reference platform service promotion activities for each reference virtual user agent in the sandbox platform service simulation environment, and to drive the reference virtual user agents to make platform service operation decisions based on the reference platform service promotion activities using persona prompt information based on the platform service processing big model, so as to obtain reference platform service decision results and reference agent decision behavior explanations. The reference agent decision behavior explanations include natural language descriptions of the subjective motivations, psychological activities of decision-making, and cognitive logic of decision-making behavior behind the platform service operation decision results using the platform service processing big model.

[0039] The activity adjustment module is used to adjust the activity configuration features of the reference platform service promotion activity based on the reference platform service decision results and the reference agent decision behavior interpretation, so as to obtain the target platform service promotion activity for the online platform service processing system.

[0040] In one feasible implementation, the sandbox platform service simulation environment is configured for the reference platform service promotion activity. The sandbox platform service simulation environment includes a sandbox platform service processing system and multiple reference virtual user agents, comprising:

[0041] Construct a sandbox platform service processing system for the online platform service processing system, and determine multiple reference virtual user agents from the virtual user agent pool for the promotion activities of the reference platform service;

[0042] Configure a sandbox platform service simulation environment based on the multiple reference virtual user agents and the sandbox platform service processing system.

[0043] In one feasible implementation, the method further includes:

[0044] User clustering is performed on historical user profile feature information to obtain a user group feature matrix;

[0045] Based on the user group feature matrix, feature combination is performed to obtain multiple individual user feature vectors, and prompt word engineering is performed on the individual user feature vectors to obtain persona prompt word information;

[0046] The aforementioned character prompts are associated with the platform service processing model to generate multiple virtual user agents;

[0047] A virtual user agent pool is formed based on all the aforementioned virtual user agents.

[0048] In one feasible implementation, the step of combining features based on the user group feature matrix to obtain multiple individual user feature vectors includes:

[0049] Based on the user group feature matrix, user group transformation is performed to obtain multiple user subgroup features. A benchmark user feature vector is selected within each user subgroup feature. A random perturbation of a preset amplitude is applied to the benchmark user feature vector to obtain multiple derived user feature vectors. The derived user feature vectors are used to express the individual behavioral differences within a homogeneous user group.

[0050] Multiple individual user feature vectors are obtained by random feature fusion based on the baseline user feature vector and the derived user feature vector.

[0051] In one feasible implementation, after forming a virtual user agent pool based on all the virtual user agents, the method further includes:

[0052] In the sandbox platform service simulation environment, the virtual decision-making results of each virtual user agent based on the historical platform service promotion activities are continuously collected, and the virtual decision-making results are compared with the historical behavior data of real users to obtain behavior difference information.

[0053] In the virtual user agent pool, the prompt words of the virtual user agent are adjusted based on the behavioral difference information to obtain the adjusted personality prompt words.

[0054] In one feasible implementation, the platform service processing big model utilizes persona prompt information to drive the reference virtual user agent to make platform service operation decisions based on the reference platform service promotion activities, thereby obtaining reference platform service decision results and reference agent decision behavior explanations. The reference agent decision behavior explanations include a natural language description, using the platform service processing big model, of the subjective motivations, psychological activities, and cognitive logic behind the platform service operation decision results, including:

[0055] Determine the persona prompt information corresponding to the reference virtual user agent, and generate agent platform service behavior decision prompts based on the persona prompt information and the reference platform service promotion activities;

[0056] The agent platform service behavior decision prompts are input into the platform service processing big model to make platform service operation decisions and obtain reference platform service decision results. The reference platform service decision results are then processed to obtain reference agent decision behavior explanations. Finally, the reference platform service decision results and the reference agent decision behavior explanations are output.

[0057] In one feasible implementation, the step of adjusting the activity plan for the reference platform service promotion activity based on the reference platform service decision results and the interpretation of reference agent decision behavior to obtain a target platform service promotion activity for the online platform service processing system includes:

[0058] Based on the decision-making results of the reference platform service and the interpretation of the decision-making behavior of the reference agent, the characteristic factors of the reference platform service promotion activities are adjusted to obtain the target platform service promotion activities for the online platform service processing system.

[0059] In one feasible implementation, obtaining reference platform service promotion activities for the online platform service processing system includes:

[0060] Obtain a multi-dimensional activity feature set, and generate a reference platform service promotion activity for the online platform service processing system based on the feature combination of the multi-dimensional activity feature set;

[0061] The characteristic factors of the reference platform service promotion activity are adjusted based on the reference platform service decision results and the interpretation of reference agent decision behavior to obtain the target platform service promotion activity for the online platform service processing system, including:

[0062] The reference platform service decision results and the reference agent decision behavior interpretation are subjected to feature mapping processing to obtain the contribution score of each feature factor in the multi-dimensional activity feature set;

[0063] The sensitive feature factors whose contribution scores exceed a preset threshold are identified, and local search optimization processing is performed in the feature factor value space of the sensitive feature factors to obtain a reference feature factor combination that meets a predetermined local search index.

[0064] Based on the combination of reference feature factors, the structured description language script of the reference platform service promotion activity is rewritten to form an optimized target platform service promotion activity for the online platform service processing system.

[0065] Thirdly, this specification provides a computer storage medium storing at least one instruction adapted for loading by a processor and executing method steps of one or more embodiments of this specification.

[0066] Fourthly, this specification provides a computer program product storing at least one instruction adapted to be loaded by a processor and to execute the method steps of one or more embodiments of this specification.

[0067] Fifthly, this specification provides an electronic device that may include: a processor and a memory; wherein the memory stores a computer program adapted to be loaded by the processor and to execute the method steps of one or more embodiments of this specification.

[0068] The beneficial effects of the technical solutions provided in some embodiments of this specification include at least the following:

[0069] In one or more embodiments of this specification, a sandbox platform service simulation environment is configured for the reference platform service promotion activities of the online platform service processing system. The sandbox platform service simulation environment includes the sandbox platform service processing system and multiple reference virtual user agents. In the sandbox platform service simulation environment, the sandbox platform service processing system is controlled to run reference platform service promotion activities for each reference virtual user agent. Based on the platform service processing big model, the reference virtual user agents are driven to make platform service operation decisions based on the reference platform service promotion activities to obtain reference platform service decision results and reference agent decision behavior interpretations. Based on the reference platform service decision results and reference agent decision behavior interpretations, the activity plan for the reference platform service promotion activities is adjusted to obtain the target platform service promotion activities for the online platform service processing system. On the one hand, without affecting the real platform service processing system, it can drive virtual user agents to make behavioral decisions and provide semantic feedback on the motivations behind these decisions in a simulated sandbox environment to promote services on a reference platform. This represents a leap from simply "knowing what" based on the virtual user agent's decision results to "knowing why" by introducing explanations of the agent's decision-making behavior, providing a technical path for exploring the subjective motivations behind virtual user decisions. On the other hand, based on the platform service decision results and the explanations of the agent's decision-making behavior, the introduction of additional motivational explanations can directly pinpoint weaknesses in the activity plan, enabling precise optimization that addresses the root cause and avoiding blind optimization. This adjustment enables a closed-loop evolution of platform service promotion activities from configuration and execution to optimization. It not only improves the verifiability and predictability of the activity plan before launch, but also obtains highly simulated evaluation results with causal explanations before the activity goes live. This effectively avoids user experience damage and platform risks caused by improper plan design in the real environment, reduces the trial and error cost of platform service operation, and enhances the accuracy of strategy optimization through user behavior simulation driven by large models. It provides a replicable, quantifiable and controllable technical path for the intelligent design and stable launch of platform service promotion activities. Attached Figure Description

[0070] To more clearly illustrate the technical solutions in this specification 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 some embodiments of this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0071] Figure 1 This is a schematic diagram of a scenario for an activity evaluation and processing system provided in this manual;

[0072] Figure 2This is a flowchart illustrating an activity evaluation and processing method provided in this manual;

[0073] Figure 3 This is a schematic diagram illustrating a scenario for simulating the environment of a sandbox platform service, as provided in this manual.

[0074] Figure 4 This is a flowchart illustrating the configuration process of a sandbox platform service simulation environment provided in this manual;

[0075] Figure 5 This is a flowchart illustrating the maintenance process of a virtual user agent pool, as provided in this manual.

[0076] Figure 6 This is a schematic diagram illustrating a scenario for virtual user agent pool processing provided in this manual;

[0077] Figure 7 This is a flowchart illustrating the operational decision-making process of a platform service provided in this manual;

[0078] Figure 8 This is a flowchart illustrating another activity evaluation process provided in this manual;

[0079] Figure 9 This is a schematic diagram illustrating a scenario for optimizing a platform service promotion activity, as provided in this manual.

[0080] Figure 10 This is a schematic diagram of the structure of an activity evaluation and processing device provided in this manual;

[0081] Figure 11 This is a schematic diagram of the structure of an electronic device provided in this specification. Detailed Implementation

[0082] The technical solutions in this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this specification.

[0083] In the description of this specification, it should be understood that the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance. In the description of this specification, it should be noted that, unless otherwise expressly specified and limited, "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices. Those skilled in the art can understand the specific meaning of the above terms in this specification based on the specific circumstances. Furthermore, in the description of this specification, unless otherwise stated, "multiple" means two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship.

[0084] In related technologies, platform service promotion activities (such as platform service sales promotions, platform service add-to-cart activities, and membership benefit pushes) typically rely on online A / B experiments or small-scale, gradual rollouts to evaluate their effectiveness before launch. These methods can only observe the results of user behavior (e.g., conversion rates, bounce rates), but cannot reveal the subjective motivations behind these behaviors (e.g., whether users leave because the copy is uninteresting or because the discount threshold is too high). This lack of understanding of the 'why' leads to activity optimization heavily relying on the experience and guesswork of promotion personnel, resulting in high trial-and-error costs and difficulty in conducting accurate, data-driven attribution analysis and iteration. Therefore, there is an urgent need in this field for a new evaluation technology that can simulate and explain the motivations behind user decisions. To reduce the risk of negative feedback, some platforms perform static verification in test environments using scripts or rule engines, but this pure rule verification cannot truly reproduce the user decision-making process, nor can it quantify the subtle impact of activity factors on metrics.

[0085] Meanwhile, as platform service chains become increasingly complex (cross-system inventory, payment, risk control, membership points, etc.), simple data playback or offline statistics are no longer sufficient to reveal the fine-grained differences in the responses of different user groups to the same activity. Service platform teams lack a fast, low-cost, and repeatable offline evaluation method. Therefore, under the premise of ensuring service and user experience, how to obtain highly reliable activity effect predictions in a quantifiable and traceable manner before the launch of platform promotion activities, and use this to guide the rapid iteration of activity parameters, has become an urgent technical problem to be solved in the platform service promotion scenario.

[0086] The present specification will now be described in detail with reference to specific embodiments.

[0087] Please see Figure 1 This is a schematic diagram of a scenario for an activity evaluation and processing system provided in this specification. Figure 1 As shown, the activity evaluation and processing system may include at least a client cluster and a service platform 100.

[0088] The client cluster may include at least one client, such as Figure 1 As shown, it specifically includes client 1 corresponding to user 1, client 2 corresponding to user 2, ..., client n corresponding to user n, where n is an integer greater than 0.

[0089] Each client in a client cluster can be an electronic device with communication capabilities, including but not limited to: wearable devices, handheld devices, personal computers, tablets, in-vehicle devices, smartphones, computing devices, or other processing devices connected to a wireless modem. Electronic devices may have different names in different networks, such as: user equipment, access terminal, user unit, user station, mobile station, mobile station, remote station, remote terminal, mobile device, user terminal, electronic device, wireless communication device, user agent or user device, cellular phone, cordless phone, personal digital assistant (PDA), and electronic devices in 5G networks or future evolved networks.

[0090] The service platform 100 can be a standalone server device, such as a rack-mount, blade, tower, or cabinet-type server device, or a workstation, mainframe, or other hardware device with strong computing power; or it can be a server cluster composed of multiple servers. The servers in the service cluster can be composed in a symmetrical manner, wherein each server is functionally and functionally equivalent in the platform service link, and each server can provide services independently. The independent provision of services can be understood as not requiring the assistance of other servers.

[0091] In one or more embodiments of this specification, the service platform 100 can establish a communication connection with at least one client in the client cluster, and complete the data interaction during the activity evaluation process based on the communication connection, such as online platform service data interaction. For example, the service platform 100 can realize the platform service promotion activity delivery to the client based on the target neural network model obtained by the activity evaluation processing method of this specification.

[0092] It should be noted that the service platform 100 establishes a communication connection with at least one client in the client cluster via a network for interactive communication. This network can be a wireless network or a wired network. Wireless networks include, but are not limited to, cellular networks, wireless LANs, infrared networks, or Bluetooth networks. Wired networks include, but are not limited to, Ethernet, universal serial bus (USB), or controller area networks. In one or more embodiments of the specification, technologies and / or formats including Hyper Text Markup Language (HTML), Extensible Markup Language (XML), etc., are used to represent data exchanged over the network (such as target compressed packets). Furthermore, conventional encryption technologies such as Secure Socket Layer (SSL), Transport Layer Security (TLS), Virtual Private Network (VPN), and Internet Protocol Security (IPsec) can be used to encrypt all or some links. In other embodiments, customized and / or dedicated data communication technologies can be used to replace or supplement the aforementioned data communication technologies.

[0093] The activity evaluation processing system embodiments provided in this specification and the activity evaluation processing methods described in one or more embodiments belong to the same concept. The execution entity corresponding to the activity evaluation processing method involved in one or more embodiments of the specification can be the aforementioned service platform 100. The implementation process embodied in the activity evaluation processing system embodiments can be found in the following method embodiments, which will not be repeated here.

[0094] based on Figure 1 The following is a detailed description of the activity evaluation processing method provided by one or more embodiments of this specification, as illustrated in the scene diagram.

[0095] Please see Figure 2 This document provides a flowchart illustrating an activity evaluation processing method according to one or more embodiments. This method can be implemented using a computer program and can run on an activity evaluation processing device based on the von Neumann architecture. The computer program can be integrated into an application or run as a standalone utility application. The activity evaluation processing device can be a service platform.

[0096] Specifically, the evaluation and processing methods for this activity include:

[0097] S102: Obtain reference platform service promotion activities for the online platform service processing system;

[0098] An online platform service processing system refers to a platform service processing system deployed in a production environment that directly handles platform service requests from end users or external systems. For example, in some platform service scenarios, it completes platform service processing and returns results with legal or financial validity by calling irreversible platform service processing logic such as orders, payments, inventory, risk control, and billing. Typical examples include e-commerce order systems, payment clearing systems, credit lending systems, and membership billing systems. In contrast, this specification constructs a sandbox platform service processing system for online platform service processing systems. The sandbox platform service processing system is only equivalent in interface and logic and is isolated from production data. It is used to conduct offline simulation verification of the promotion activities of reference platform services that are about to be launched.

[0099] In this specification, "reference platform service promotion activity" refers to a promotion plan that has not yet been deployed to the production environment and whose effectiveness is yet to be verified. This plan typically consists of several configurable activity feature factors. For example, in a reference platform service promotion activity within a specific platform service scenario, activity feature factors include, but are not limited to, price / rate, discount amount, copywriting style, exposure entry point, and time period restrictions.

[0100] In one feasible implementation, the promotional activities for the reference platform service are manually generated by the promotion end through a graphical configuration interface, for example:

[0101] The first step is to input the activity elements: the promoter logs in to the activity management front end and enters the activity name, target indicators (such as increasing the add-to-cart rate by 3%), activity period and target audience in sequence.

[0102] Then, feature factors are set. For example, in the factor template provided by the system, promoters select or fill in the values ​​of each factor: Price factor: Discount rate 90%, Incentive factor: 20 yuan off for purchases over 200 yuan, Visual factor: Main color #3A7DFF, Exposure factor: Homepage "Limited Time Discount" section.

[0103] Next, the system automatically generates a DSL (Domain-Specific Language). It maps the aforementioned feature factors to the domain-specific language script, timestamps the script, generates a version number, and then stores it in the activity configuration repository. Upon completion of this step, a reference platform service promotion activity for the online platform service processing system is obtained.

[0104] In one feasible implementation, the promotional activities for the reference platform services are automatically generated by the system based on a historical activity database and an algorithm, for example:

[0105] First, historical data is extracted. The service platform's system scans past online activities within a preset time period, extracts activity factors and outcome indicators, and constructs an activity characteristic factor-effect comparison table.

[0106] Then, factor contribution modeling is performed, which uses gradient boosting trees to train multiple activity feature factors to obtain the ranking of each factor's contribution to the conversion rate.

[0107] Next, a candidate factor combination search is performed. For a specified number of recommended feature factors with the highest contribution, the Bayesian optimization algorithm is used to perform a preset number of iterations to search and output the better candidate factor combinations under preset constraints (such as budget ≤ 30 yuan / order; discount rate ≥ 85%).

[0108] For example: Candidate factor combination discount rate 88%, coupon amount 15 yuan; visual color #FF6B00; exposure entry "product details page overlay".

[0109] Finally, a DSL script is generated for the candidate factor combinations. The system maps these candidate factor combinations to domain-specific language scripts, timestamps the scripts, generates a version number, and then stores them in the activity configuration repository. Upon completion of this step, a reference platform service promotion activity for the online platform service processing system is obtained.

[0110] S104: Configure a sandbox platform service simulation environment for the reference platform service promotion activity. The sandbox platform service simulation environment includes a sandbox platform service processing system and multiple reference virtual user agents.

[0111] The sandbox platform service simulation environment can be understood in this specification as a test field that is physically or logically isolated from the production environment, but equivalent to the online platform service processing system in terms of interface specifications and platform service processing logic. The sandbox platform service simulation environment includes the sandbox platform service processing system and multiple reference virtual user agents.

[0112] The sandbox platform service processing system in this manual can be understood as a copy of the service business chain deployed in the sandbox platform service simulation environment. Taking common platform services such as shopping services and loan services as examples, it can completely execute platform service processes such as order placement, payment, billing, risk control, and points crediting.

[0113] In this specification, the virtual user agent can be understood as an intelligent agent interaction object service driven by a large model of platform service processing based on persona prompts, etc. It can be called an agent, which is used to simulate the decision-making behavior (i.e., the platform service decision result) of the target user group being represented when the platform service promotion activity is triggered, and the interpretation of the decision-making behavior that makes the decision.

[0114] In one feasible implementation, such as Figure 3 As shown, Figure 3 This is a schematic diagram illustrating a sandbox platform service simulation environment. It can determine the platform service processing system image and platform service processing chain logic of a real business system, create an initial sandbox platform service simulation environment, and configure the platform service processing flow image and platform service processing chain logic within the sandbox platform service simulation environment as follows: Figure 3 The sandbox platform service processing system shown is as follows: Figure 3 As shown, a batch of virtual reference virtual user agents are selected for onboarding in the sandbox platform service simulation environment using a preset virtual user agent onboarding strategy. The preset virtual user agent onboarding strategy can be a random sampling strategy (without user feature screening, simulating the distribution of real users), or a user screening strategy (ensuring homogeneity and performing hypothesis testing), etc.

[0115] For example: Create an initial sandbox platform service simulation environment. Copy the microservice images (order service, inventory service, payment gateway, risk control verification, etc.) related to the platform service processing logic in the online platform service processing system to the isolated namespace of the sandbox platform service simulation environment, ensuring that API signatures, request-response formats, and platform service consistency logic are consistent with the production version. Perform structure masking or empty table initialization on the production database, loading only the necessary enumeration and dictionary data related to functional verification. Then, extract several reference virtual user agents from the virtual user agent pool according to a preset sample size or population distribution, and assign a unique sandbox identity to each agent. Use network policies, access control lists (ACLs), or namespace routing to ensure that any write operations within the sandbox platform service simulation environment will not affect production data. Enable real-time monitoring of the system processing status of the sandbox platform service processing system. For example, for a specific platform service, this could include real-time monitoring of order status, inventory deduction, payment transaction history, and log errors.

[0116] Example: In an e-commerce flash sale platform service scenario, it is necessary to verify the "¥30 off for purchases over ¥300 + homepage banner exposure" promotion scheme of a reference platform service. The system first mirrors the production order-inventory-payment chain to the xxx namespace of the sandbox platform service simulation environment and creates an empty inventory table. Then, it reads the DSL script to write the discount rules into the sandbox configuration center of the sandbox platform service simulation environment to configure the sandbox platform service processing system. Simultaneously, it automatically mounts page resources on the sandbox homepage of the sandbox platform service processing system. Finally, it extracts a preset number of reference virtual user agents from the virtual user agent pool and loads their historical purchase frequency and membership level. At this point, the sandbox platform service simulation environment is configured, and subsequent steps can control the sandbox platform service processing system and reference virtual user agents to jointly run the reference platform service promotion activity.

[0117] S106: In the sandbox platform service simulation environment, the sandbox platform service processing system is controlled to run the reference platform service promotion activities for each reference virtual user agent, and based on the platform service processing big model, the reference virtual user agents are driven to make platform service operation decisions based on the reference platform service promotion activities using persona prompt information to obtain reference platform service decision results and reference agent decision behavior explanations. The reference agent decision behavior explanations include natural language descriptions of the subjective motivations, behavioral decision psychological activities, and cognitive logic of decision behavior behind the platform service operation decision results using the platform service processing big model.

[0118] Platform service processing big model: can be understood as a big language model with task understanding, user behavior reasoning and decision-making capabilities. Its input is at least prompt words, and its output is the reference platform service decision results and reference agent decision behavior explanations of anthropomorphic users.

[0119] Optionally, the platform service can use general multimodal large language models (MLLM) to process large models, such as: GPT series large models, Wenxin Yiyan series large models, DeepSeek series large models, etc.

[0120] Platform service operation decisions: These refer to the platform service behavior choices made by virtual user agents in the context of referencing platform service promotion activities, inferred by the platform service processing model based on their persona prompts. Platform service operation decisions include whether to participate, how to respond, and at what point to exit.

[0121] Reference platform service decision results: refers to the actual platform service behavior results that occur during the sandbox simulation of platform service processing driven by the platform service processing big model and the reference virtual user agent, such as whether the order is completed, whether the coupon is redeemed, and whether the subsequent process is entered.

[0122] Reference agent decision-making behavior interpretation: This refers to the semantic and interpretable natural language description of the subjective motivations, psychological activities, and cognitive logic behind the platform service operation decisions made by reference virtual user agents, using a platform service processing model. It can be understood as exploring 'why' the decision was made, rather than simply recording 'what was decided'. For example, the semantic interpretation results of virtual user behavior motivations, decision-making hesitation points, and decision-making response paths are usually in natural language form.

[0123] Indicatively, the reference platform service promotion activities configured in step S104 are injected into the sandbox platform service processing system. These platform service promotion activities are structured data sets used to define multiple platform service factors, including but not limited to the activity's reach methods, discount mechanisms, interaction paths, resource usage rules, and visualization styles. The sandbox platform service processing system is a simulation runtime environment with a structure consistent with the online platform service processing system. It possesses the technical attributes of interface consistency and processing rule equivalence, but is isolated from production data, enabling full-link simulation of any set of platform service activities.

[0124] To support user behavior simulation, the system controller invokes multiple pre-generated reference virtual user agents. These reference virtual user agents are personas generated through historical real user profile clustering and prompt word engineering, and their behavioral decisions are driven by the platform service processing big model. After the activity simulation starts, each virtual user agent is processed by the platform service processing big model based on its bound persona prompt words (such as age, income, interests, consumption preferences, etc.) and the current promotional activity content information of the reference platform service promotion activity. The platform service processing big model is a large language model with user preference modeling and context understanding capabilities. Its input is a combination of "virtual user agent persona prompt words + activity context description of the platform service promotion activity" as a task prompt. Its output includes a platform service decision result composed of at least one behavioral decision information and an agent decision behavior explanation composed of each behavioral explanation information, respectively representing whether the user participates in the platform service operation and the motivation behind the behavior.

[0125] Subsequently, based on the behavioral decision control system output by the model, the system executes corresponding platform services, including but not limited to simulated order placement, adding to cart, coupon selection, and payment redirection in e-commerce platform scenarios. Each platform service behavior calls the internal interface of the sandbox system according to the real business logic. The sandbox system records platform service data such as operation path, interface response status, resource consumption, and execution results, and writes the reference platform service decision result and its corresponding behavioral explanation into the simulation result set.

[0126] Ultimately, the reference platform service decision results reflect the user agent's behavioral choices under the current platform service promotion campaign settings. The reference agent's decision behavior explanation is attributed to the underlying logic of the choice in natural language, such as "the high threshold of the promotion discount caused users to abandon the order" or "the visual style lacked appeal and did not trigger clicks." Through this process, a set of data with quantifiable structure and traceable interpretation can be obtained for subsequent evaluation of the promotion campaign's effectiveness and optimization of the plan.

[0127] This step introduces a large model of platform service processing into the sandbox platform service environment and drives multiple virtual user agents to complete the full-link simulation, realizing the closed-loop construction of platform service promotion activities from "deduction" to "feedback", which significantly improves the accuracy and low risk of pre-verification of platform service decisions.

[0128] S108: Based on the reference platform service decision results and the interpretation of the reference agent decision behavior, adjust the activity configuration features of the reference platform service promotion activity to obtain the target platform service promotion activity for the online platform service processing system.

[0129] In a schematic way, by analyzing the behavioral data (reference platform service decision results) and semantic feedback (reference agent decision behavior interpretation) generated by multiple reference virtual user agents participating in the promotion activities of the reference platform service in the sandbox platform service simulation environment, the key configuration factors of the current promotion activities are evaluated and optimized in a targeted manner to form a version of the platform service activities that can be officially launched and deployed in the online platform service processing system.

[0130] In this step, the system first integrates the reference platform service decision results with the behavioral interpretation to construct an activity response distribution model corresponding to each type of user group. This model identifies behavioral conversion paths, task completion rates, and semantic preference expression characteristics under different user profiles, thereby quantifying and attributing the performance of the current platform service activity to obtain model analysis results.

[0131] Subsequently, the system determines whether the preset launch conditions for the activity are met based on the analysis results of the aforementioned model. If the preset launch conditions are not met, the system identifies the activity factors that have a significant impact on the conversion rate based on the analysis results of the aforementioned model, such as: insufficient discount, low entry point click-through rate, copywriting that fails to arouse interest, and slow response of the target audience to the activity strategy. After identifying the key influencing factors, the system adjusts the relevant parameters of the service promotion activities on the reference platform, including but not limited to optimizing coupon thresholds, adjusting exposure positions, modifying visual style, redesigning wording content, and changing the order of entry point layout, in order to form a revised activity plan with better performance potential.

[0132] After being redefined using DSL structured scripting, the revised activity plan is executed as the new reference platform service promotion activity (S106-S108) until the model analysis results confirm that the preset activity launch conditions are met. This yields the target platform service promotion activity, whose structure, semantics, and behavioral objectives have been verified and optimized through simulation feedback mechanisms, demonstrating a high degree of controllability in expected results. This target platform service promotion activity can serve as the official version for subsequent canary releases or full deployments in the online platform service processing system, achieving the dual goals of business improvement and user experience enhancement.

[0133] Through the implementation of this step, an activity optimization mechanism based on simulation feedback results is established, thereby realizing a data-driven evolution path for platform service promotion activities, effectively avoiding the trial-and-error risks brought about by traditional experience-based configuration, and ensuring that the activities have strong effect prediction capabilities and behavioral adaptability before going online.

[0134] In one or more embodiments of this specification, a sandbox platform service simulation environment is configured for the reference platform service promotion activities of the online platform service processing system. The sandbox platform service simulation environment includes the sandbox platform service processing system and multiple reference virtual user agents. In the sandbox platform service simulation environment, the sandbox platform service processing system is controlled to run reference platform service promotion activities for each reference virtual user agent. Based on the platform service processing big model, the reference virtual user agents are driven to make platform service operation decisions based on the reference platform service promotion activities to obtain reference platform service decision results and reference agent decision behavior interpretations. Based on the reference platform service decision results and reference agent decision behavior interpretations, the activity plan for the reference platform service promotion activities is adjusted to obtain the target platform service promotion activities for the online platform service processing system. Without affecting the real platform service processing system, it can drive virtual user agents to make behavioral decisions and provide semantic feedback on the promotion activities of reference platform services based on a simulation sandbox environment. This enables a closed-loop evolution of platform service promotion activities from configuration, execution to optimization. It not only improves the verifiability and predictability of the activity plan before launch and reduces the trial and error cost of platform service operation, but also enhances the accuracy of strategy optimization through user behavior simulation driven by a large model. It provides a replicable, quantifiable and controllable technical path for the intelligent design and robust launch of platform service promotion activities.

[0135] Optional, please see Figure 4 , Figure 4 This is a flowchart illustrating the configuration of a sandbox platform service simulation environment according to one or more embodiments of this specification. Specifically, configuring the sandbox platform service simulation environment for the reference platform service promotion activity, wherein the sandbox platform service simulation environment includes a sandbox platform service processing system and multiple reference virtual user agents, can be performed in the following ways:

[0136] S202: Construct a sandbox platform service processing system for the online platform service processing system, and determine multiple reference virtual user agents from the virtual user agent pool for the promotion activities of the reference platform service;

[0137] In step S202, the system constructs a sandbox platform service processing system with the same logical structure as the online platform service processing system. This sandbox platform service processing system is a closed environment dedicated to simulation operation. It replicates the interface specifications, processing rules, and state machine processes of the real system to form a simulated execution platform that is equivalent to but data-isolated from the real platform service environment. This system supports simulated responses to various platform service modules in the business process, including but not limited to core operations such as user login, discount distribution, order settlement, payment processing, points changes, coupon redemption, and path redirection, ensuring that each process can be completely closed-loop during simulation execution.

[0138] Simultaneously, the system selects multiple reference virtual user agents from the previously established pool of virtual user agents, tailored to the current promotional activities of the reference platform service. These reference virtual user agents are simulated user individuals with persona-driven prompts, and their feature vectors can cover specific user profiles (such as new users, users making high-frequency behavioral decisions, price-sensitive users, etc.). Guided by the subsequent large model, they possess the ability to generate behavior and interpret feedback. This user selection can be completed according to preset strategies, such as proportional sampling, specific profile filtering, or strategies that maximize behavioral feature coverage, ensuring a representative user structure in the simulation.

[0139] S204: Configure a sandbox platform service simulation environment based on the plurality of reference virtual user agents and the sandbox platform service processing system.

[0140] In a schematic manner, the system binds the multiple reference virtual user agents to the constructed sandbox platform service processing system. Through logical configuration and environmental parameter settings, it generates a sandbox platform service simulation environment. This sandbox platform service simulation environment is a closed platform service evolution space, including independent resource snapshots, simulation session channels, and data isolation mechanisms. It can support the concurrent operation of multiple user agents in the same scenario and ensure that operation trajectories and platform service output data do not affect the production system. This environment is also configured with monitoring components and result archiving interfaces to support subsequent simulation behavior analysis, semantic interpretation collection, and activity plan optimization. Through the configuration and operation of this sandbox platform service simulation environment, a reliable experimental foundation can be provided for subsequent deployment of platform service promotion activities, observation of user behavior, and collection of feedback data in virtual simulation, effectively achieving pre-evaluation and risk control of activity plans.

[0141] In this specification, through steps S202 to S204, a platform service simulation platform with complete business links and representative user profiles can be efficiently constructed in a sandbox environment that is structurally consistent with but logically isolated from the online platform service processing system. By introducing multiple reference virtual user agents and co-configuring with the sandbox platform service processing system, the service platform achieves realistic behavioral extrapolation and full-process closed-loop simulation of promotional activities in a non-real user environment. This provides a low-cost, high-fidelity, controllable, and observable technical foundation for subsequent effectiveness evaluation and optimization of platform service solutions, significantly improving the scientific nature of decision-making and the security of strategies before the platform service promotion activities go live.

[0142] Optional, please see Figure 5 , Figure 5 This is a flowchart illustrating the maintenance process of a virtual user agent pool as proposed in one or more embodiments of this specification. The following method can be used as a reference:

[0143] S302: Perform user clustering on historical user profile feature information to obtain a user group feature matrix;

[0144] By clustering users based on historical user profile features, one or more user group feature matrices are obtained. The aim is to achieve structural summarization and group modeling of multi-dimensional profile information of historical real users, and to provide a feature foundation for the subsequent construction of virtual user agents.

[0145] User profile feature information refers to the multidimensional static and dynamic attribute data of users accumulated by the system during long-term platform service processing and interaction. Optionally, the attribute data includes, but is not limited to, basic information (such as age and gender), geographical attributes (such as city level and residential area), social attributes (such as occupation type and education level), financial attributes (such as payment ability, consumption preferences, and credit behavior), and behavioral preferences (such as page clicks, platform service completion rate, and incentive response). After being cleaned, organized, and standardized, this information is used as input feature matrix for cluster analysis.

[0146] User clustering refers to using unsupervised learning algorithms to measure the similarity and divide user feature sample sets into groups. Specifically, mainstream methods such as K-means clustering, Gaussian Mixture Model (GMM), DBSCAN, or hierarchical clustering can be used. Taking K-means clustering as an example, the system randomly initializes cluster centers based on a set number of clusters K, iteratively assigns feature samples to different clusters and updates the centers until convergence, ultimately forming K user groups. Each group exhibits high similarity in user features, while the groups show significant differences, demonstrating discriminability and representativeness.

[0147] After clustering is completed, the system constructs a corresponding user group feature matrix for each clustering result. This feature matrix is ​​a two-dimensional structure with feature dimensions as columns and statistical values ​​as elements, including the mean, standard deviation, maximum / minimum values, principal component vectors, and feature correlation indices for each dimension. This feature matrix serves as an expression model for user profiles and can be used for subsequent operations such as constructing typical user vectors, extracting representative user attributes, and generating persona prompts, demonstrating high abstraction and generalization capabilities.

[0148] By executing step S302, the system can rationally divide the user structure and model the feature structure while maintaining the diversity of real user distribution, forming a user group profile with interpretable behavior and representativeness, which provides a solid feature foundation for the subsequent generation of diverse and semantically rich virtual user agents.

[0149] S304: Based on the user group feature matrix, feature combination is performed to obtain multiple individual user feature vectors, and prompt word engineering is performed on the individual user feature vectors to obtain persona prompt word information;

[0150] Based on the user group feature matrix, feature combination is performed to obtain multiple individual user feature vectors. The individual user feature vectors are then transformed into character-based prompt word information through prompt word engineering. The aim is to transform the abstract user group modeling results into virtual user control instructions with character-based context through typical feature instantiation and semantic prompt word construction, which can be used to drive the language model or agent model to generate virtual user agents with specific behavioral characteristics.

[0151] An individual user feature vector refers to a representative and differentiated multidimensional user feature instance constructed from a user group feature matrix obtained through clustering, using feature selection and combination strategies. Specifically, the system can select statistically representative data points (e.g., mean, median, maximum, upper and lower quartiles, etc.) for each dimension of the group feature distribution, and perform Cartesian product combinations or restricted combinations on the multidimensional features to generate individual user feature vectors with different combinations of dimensions. Each vector constitutes a concrete feature representation of a user identity, reflecting the potential decision-making behavior of a specific user type in business. For example, a vector might include: gender as female, age range as 25-34, region as a first-tier city, education level as undergraduate, consumption level as moderate, and financial preference as high security preference, etc.

[0152] During the character-based prompt information construction phase, the system performs prompt engineering on the aforementioned individual user feature vectors. This process includes, but is not limited to, field mapping, semantic expansion, natural language conversion, and context constraint concatenation, so that the final output can be directly understood and invoked by the language model or virtual agent module. Conversion methods can employ structured template mapping (e.g., "age is {age}, occupation is {job}..."), vector semantic projection (e.g., embedding-to-text method), or expert rule generation.

[0153] For example, input an individual user feature vector: {Gender: Female, Age: 28, City Tier: First-tier, Education: Bachelor's Degree, Financial Preference: Risk-Averse, Promotional Response Propensity: High},

[0154] The corresponding prompt could be: "A 28-year-old female user who graduated with a bachelor's degree, lives in a first-tier city, is sensitive to price promotions, and prefers stable financial management solutions."

[0155] The persona prompt information serves as semantic input for controlling the behavior of the virtual user agent. It can be used to drive the platform service processing large language model (LLM) or task agent module to exhibit language output and platform service response behavior consistent with the characteristics of the target user during the simulation process, ensuring that the user simulation results are controllable, differentiated, and interpretable.

[0156] Through the above methods, step S304 achieves a smooth transition from user group structure to individualized virtual user identity, constructs a virtual user semantic description layer with behavioral intent control capabilities, and lays a key foundation for subsequent agent generation and sandbox platform service behavior simulation.

[0157] In one feasible implementation, the process of combining features based on the user group feature matrix to obtain multiple individual user feature vectors can be performed in the following manner:

[0158] A2: Based on the user group feature matrix, user group transformation is performed to obtain multiple user subgroup features. A benchmark user feature vector is selected within each user subgroup feature. A random perturbation of a preset amplitude is applied to the benchmark user feature vector to obtain multiple derived user feature vectors. The derived user feature vectors are used to express the individual behavioral differences within a homogeneous user group.

[0159] Based on the user group feature matrix, user group transformation is performed to obtain multiple user subgroup features. Within each user subgroup feature, a benchmark user feature vector is selected. The benchmark user feature vector is subjected to a random perturbation of a preset amplitude to obtain multiple derived user feature vectors. The purpose is to further generate individualized user vectors with internal differences based on the user clustering results, so as to enhance the diversity and fine-grained behavior simulation capabilities of subsequent virtual user agents.

[0160] The user group feature matrix is ​​a statistical matrix obtained during the user clustering stage, used to characterize the distribution of key features of each user group. User group transformation refers to further subdividing the internal structure of the already segmented user groups into subgroups, generating multiple user subgroup feature sets with finer granularity and more concentrated structure. This transformation can be based on the degree of local aggregation in the original group feature space, differences in feature density, or predefined rules (such as user consumption frequency, promotion response type, etc.) to split into smaller subgroups with more homogeneous structures.

[0161] After obtaining the user subgroup features, the system selects at least one benchmark user feature vector within the feature vector space corresponding to each user subgroup feature. The benchmark user feature vector is a representative point of the subgroup in the feature space, which can be calculated from the center point of the subgroup features (such as the mean vector), selected from the mode feature point, or determined by sampling along the median of the principal component direction. The benchmark user feature vector constitutes the core structure expressing the group characteristics of the subgroup, reflecting the typical attribute configuration of the subgroup in dimensions such as age, gender, region, spending power, and risk preference.

[0162] To further generate diverse individual vectors, the system implements random perturbations of a preset amplitude based on the aforementioned baseline user feature vectors, resulting in multiple derived user feature vectors. The perturbation process involves introducing controlled random offsets into each dimension of the original vectors. These offsets can be implemented using Gaussian, uniform, or custom perturbation strategies, and their amplitude must be limited within the subgroup feature boundaries to ensure that the perturbed features remain within a reasonably acceptable semantic range. For example, the perturbation amplitude for the age dimension can be set to ±3 years, the perturbation for the geographic dimension can be limited to replacements between cities of the same level, and the consumption level dimension can introduce a fine-tuning mapping between "medium" and "high". Through this perturbation mechanism, multiple differentiated virtual characters reflecting individual biases, behavioral randomness, and user volatility can be generated within a unified feature structure of a homogeneous user group.

[0163] The derived user feature vector serves as the input basis for subsequently constructing persona prompts and virtual user agents. It possesses high expressive flexibility and behavioral diversity, which can effectively improve the sandbox simulation system's ability to fit the differences in user behavior distribution and policy response, and enhance the simulation results' ability to simulate the complexity of real-world populations.

[0164] A4: Based on the baseline user feature vector and the derived user feature vector, random feature fusion is performed to obtain multiple individual user feature vectors.

[0165] Based on the baseline user feature vector and the derived user feature vector, random feature fusion is performed to obtain multiple individual user feature vectors. This is used to further introduce cross-individual dimension information mixing while maintaining the consistency of the user group structure, and to construct a more expressive and behaviorally diverse individual user model, so as to enhance the personality diversity and situational adaptability of the subsequent virtual user agent.

[0166] Random feature fusion refers to the partial recombination of features from multiple source vectors (including a baseline user feature vector and its derived vectors) in a dimensional combination manner to form new individual feature combinations. This process aims to break the direct inheritance relationship between features from the same source and generate virtual user features that more closely resemble the complex behaviors of real-world users through random reconstruction across individual feature dimensions. In practice, the system can set fusion rules on each feature dimension, for example:

[0167] Perform operations such as weighted averaging, random selection, and normal distribution sampling on numerical features (such as age, income level, etc.);

[0168] Samples are taken from categorical features (such as occupation, city tier, educational background, etc.) based on probability distribution, or randomly selected from multiple source vectors with set weights;

[0169] For behavioral preference features (such as loan willingness and promotion sensitivity), multi-directional cross generation can be performed by combining the label probability of the cluster center.

[0170] For example, if there is a set of vectors A, B, and C from the same subgroup, where A is the baseline user feature vector and B and C are its derived vectors, then "age" and "educational background" can be extracted from A, "city tier" can be extracted from B, and "financial preferences" and "promotional response tendency" can be extracted from C. Finally, these vectors are concatenated to generate a new individual feature vector D, representing "a young user who has received an undergraduate education, lives in a second-tier city, prefers low-risk financial products, and is sensitive to discounts and promotions".

[0171] The aforementioned fusion process not only maintains the structural consistency of the data sources but also introduces combinatorial diversity and individual differences, effectively improving the coverage of feature distribution and the richness of simulated performance during the virtual user construction process. This provides a more representative input foundation for subsequently constructing contextualized prompts and driving behavior models to generate user responses. The individual user feature vectors generated in this way possess high adjustability in feature dimension combinations and controllable behavioral distribution, making them widely applicable to virtual user generation and strategy evaluation tasks in various platform service promotion scenarios.

[0172] S306: Associate the aforementioned character prompt information with the platform service processing big model to generate multiple virtual user agents;

[0173] The aforementioned character prompts are associated with the platform service processing model to generate multiple virtual user agents. These agents are used to drive the language model or multimodal platform service processing model to generate virtual user agent individuals with behavioral autonomy and contextual consistency based on the user semantic description information constructed above, thereby achieving a concrete user behavior simulation capability.

[0174] The persona prompt information refers to the natural language description information generated in step S304 above, used to express the identity attributes and behavioral preferences of virtual users. This prompt information includes, but is not limited to, dimensions such as age, gender, region, education level, occupation type, consumption willingness, promotional sensitivity, and risk preference, and is presented in a semantically coherent natural language format for easy parsing and use by downstream large-scale models. For example, a persona prompt information could be: "This is a 35-year-old married woman living in a second-tier city, working in the education industry. She is usually cautious about high-risk investments but is quite sensitive to new user promotions on the platform."

[0175] The platform service processing big data model refers to a language generation model or multimodal task agent model with contextual understanding, user behavior simulation, and multi-turn task decision-making capabilities. This model can automatically generate platform service behaviors, response content, and feedback judgments that match the user's persona after receiving user prompts and platform service context information. This model can be a general-purpose big data language model (such as LLM) trained on platform service interaction data or a multi-task platform service agent model specifically tuned and optimized for platform service promotion scenarios.

[0176] During this step, the system takes each of the above-mentioned persona prompt words as a prefix or condition input, and inputs it into the platform service processing model through a specific prompt word template or context arrangement mechanism. It then combines the platform service content to be simulated (such as marketing copy, page style, activity mechanism, etc.) to construct a complete input sequence so that the model can generate user agent response logic with consistent behavioral preferences.

[0177] For example, when the input prompt and platform service task is: "A user with a rational consumption tendency receives an installment payment offer invitation", the model needs to automatically infer the user's possible behavioral choices (such as rejecting, continuing to browse, or clicking to view) and the reasons expressed (such as "I am not inclined to consume first and then pay later").

[0178] "Model-agent association" refers to the system generating a virtual user agent with autonomous behavior for each user characteristic through prompt word binding and model input configuration. Each agent can independently run on the large model based on the platform service in different sandbox platform service environments, and continuously maintain the consistency of its persona behavior under different input contexts. The system can maintain agent identifiers, agent policy states, and persona prompt context history during this process for subsequent behavior tracking, feedback analysis, and behavior verification.

[0179] By performing this step, the system realizes the logical mapping from user profile information to executable agent individuals. The generated virtual user agent has the ability to simulate real user behavior and can complete task decision-making, status response and behavior feedback in the sandbox platform service scenario, providing a controllable, repeatable and measurable simulation subject for subsequent strategy evaluation and promotion activity optimization.

[0180] S308: Form a virtual user agent pool based on all the aforementioned virtual user agents.

[0181] Based on all the aforementioned virtual user agents, a virtual user agent pool is formed. After completing the construction of multiple large model agents with personality prompts, the system will structurally summarize and uniformly manage all generated virtual user agent instances to build a standardized set of virtual user resources for scheduling and invocation in the sandbox platform service processing environment, referred to as the "virtual user agent pool".

[0182] The virtual user agent refers to an intelligent agent entity with behavioral generation and platform service response capabilities, generated by inputting individual user persona prompts into the platform's service processing model and combining them with specific control strategies. This virtual user agent can maintain consistency in its persona across different platform service scenarios and perform platform service operations consistent with the user profile, including but not limited to decision-making, intent expression, questionnaire completion, and feedback generation.

[0183] A virtual user agent pool refers to a large-scale collection of agent entities organized in a data structure manner. It possesses unified identity identification, context management, behavior policy binding, and scheduling interface capabilities, facilitating the system to select, configure, load, and schedule virtual user agents in batches according to simulation requirements. Each agent in this pool can record the following core attribute information: user persona prompt text, reference to the large model of the platform service it is bound to, user feature vector (for quick filtering and grouping), optional behavior tags (such as high-frequency clickers, risk-averse users, etc.), runtime history cache, and context state (for supporting multi-turn interactions).

[0184] During the formation of the virtual user agent pool, the system can perform quality verification and rationality screening on all generated virtual user agents, including but not limited to duplicate detection, behavior consistency assessment, and abnormal prompt word filtering, to ensure that the users in the agent pool are representative, have distinct personalities, and exhibit stable behavior.

[0185] Once the proxy pool is built, it will support the following features:

[0186] Sandbox simulation deployment scheduling: Select agent combinations that meet the characteristics of specific groups of people for deployment based on scenario or experimental needs;

[0187] Sampling in a population-controlled experiment: population pairing and control grouping based on feature labels;

[0188] Promotion strategy coverage assessment: Analyze whether the agent pool covers all key profile types required by the business;

[0189] Behavioral simulation result traceability: The simulation path is traced back through the prompts and behavioral processes recorded by the agent pool, which improves the interpretability of the simulation.

[0190] Through the implementation of this step, the system has built a structured, highly controllable, and easily schedulable virtual user resource management system, which can support subsequent large-scale, multi-dimensional, and multi-scenario platform service promotion strategy simulation and evaluation experiments, significantly improve strategy iteration efficiency, reduce the cost of reaching real users, and enhance the ability to predict behavior and assess risks before the promotion campaign goes live.

[0191] For example, such as Figure 6 As shown, Figure 6 This is a schematic diagram illustrating a scenario where a virtual user agent pool is processed. Figure 6 The left side of the diagram illustrates how, by executing S302, the system acquires historical user profile data across multiple dimensions, including basic characteristics, preference characteristics, occupational characteristics, educational characteristics, and geographical characteristics. Based on this high-dimensional data, unsupervised learning algorithms (such as K-Means and hierarchical clustering) are used to cluster users, generating multiple user groups with similar behaviors and characteristics. Each user group is abstracted as a high-dimensional vector, constituting... Figure 6 The "user group feature matrix" shown in the middle section serves as the basic semantic representation for individual generation. Then, executing S304-S308 yields the following result: Figure 6 The virtual user agent pool shown on the right includes multiple virtual user agents (agent 1, agent 2... agent n). These agents are derived from group characteristics and have been bound to the platform service model. Each agent can perform platform service interactions in the simulation environment. Figure 6This diagram illustrates the standard path for building virtual users, demonstrating a high degree of structure, universality, and scalability, and serving as a crucial infrastructure component of the sandbox platform service evaluation system. Leveraging this proxy pool, the system can simulate large-scale user behavior at low cost, providing reliable support for the testing, iteration, and evaluation of platform service promotion strategies.

[0192] In one feasible implementation, after forming a virtual user agent pool based on all the virtual user agents, the method further includes:

[0193] B2: In the sandbox platform service simulation environment, continuously collect the virtual decision-making results of each virtual user agent based on the historical platform service promotion activities, and compare the virtual decision-making results with the real user's historical behavior data to obtain behavior difference information;

[0194] In a sandbox platform service simulation environment, the virtual decision-making results of each virtual user agent based on historical platform service promotion activities are continuously collected. These virtual decision-making results are then compared with the behavioral data of real users in the same historical platform service promotion activities to obtain behavioral difference information reflecting the degree of deviation between their behaviors. Specifically:

[0195] Virtual decision results refer to the judgments and feedback responses made by virtual user agents after receiving information about a specific platform service promotion activity, such as whether to click, add to cart, or abandon the service.

[0196] During execution, the system parses historical user behavior data based on real platform service activity traffic. It then injects relevant platform service context information into virtual user agents matching these historical users (identified by determining the historical user persona, matching persona prompts, and thus the corresponding virtual user agent). This prompts the virtual user agents to respond behaviorally within the sandbox platform service simulation environment based on historical platform service promotion activities, forming a complete virtual decision result. Subsequently, the system calls upon historical online real user behavior data related to the platform service activity and constructs a behavioral comparison with the virtual decision result. It calculates behavioral difference information by comparing the virtual user agent's behavioral trajectory (virtual decision result) with the real user's behavioral trajectory (historical online real user behavior data) across multiple dimensions. These behavioral difference indicators may include, but are not limited to, platform service response probability difference, platform service path offset, operation latency, and selection distribution overlap. Based on this comparison, the system outputs a set of quantified behavioral difference information to characterize the current agent's fit in reproducing the real user's intentions.

[0197] This behavioral difference information serves as a crucial basis for subsequent prompt word optimization and virtual user agent adjustment, ensuring the high fidelity and evaluability of the simulated agent's behavioral performance. Through this process, dynamic evaluation and behavioral deviation control of the virtual user simulation capabilities can be achieved, thereby significantly improving the behavioral fidelity and evaluation reliability of the entire sandbox platform service simulation system.

[0198] B4: In the virtual user agent pool, the prompt words of the virtual user agent are adjusted based on the behavioral difference information to obtain the adjusted prompt words.

[0199] In this step, the system first identifies virtual user agents with low correlation to real user behavior based on the behavioral difference information output in the previous steps. Then, the system adjusts the agent's original persona prompts according to the type of behavioral deviation in the specific platform service scenario (e.g., overly aggressive response, hesitation, or no response at all). Prompt adjustment may include, but is not limited to, the following strategies:

[0200] Modify the description of the agent's personality or motivation, for example, change "tends to try new products" to "participates only after carefully evaluating the benefits";

[0201] Enhance the clarity of task preferences, for example, by replacing "price-sensitive" with "extremely sensitive to installment rates";

[0202] Supplement the historical behavioral context, such as adding supplementary descriptions like "the previous promotional activity did not result in a conversion";

[0203] Optimize the structure, order, or wording of prompts to guide the model to generate the expected behavioral logic more accurately.

[0204] The above-mentioned prompt word adjustment process can be achieved through expert strategy rules, batch generation via human-machine collaboration, or automatic iterative optimization combined with natural language models. The adjusted persona prompt words will be re-injected into the corresponding virtual user agent and used in subsequent sandbox platform service simulation tasks, forming new behavioral data for the next round of difference evaluation, thereby realizing an "adaptive evolution" mechanism for virtual user agent behavior.

[0205] By implementing this step, we can achieve dynamic correction and enhanced behavioral accuracy of the virtual user agent's behavior fitting ability, continuously improve its performance in simulating real user intentions and behavioral decisions, and thus enhance the behavioral credibility and strategy experimentation value of the entire platform service simulation evaluation system.

[0206] Optional, please see Figure 7 , Figure 7This is a flowchart illustrating a platform service operation decision-making process proposed in one or more embodiments of this specification. Specifically, the process involves using a large-scale platform service processing model to drive a reference virtual user agent to make platform service operation decisions based on the reference platform service promotion activities. This yields reference platform service decision results and an explanation of the reference agent's decision behavior. The explanation of the reference agent's decision behavior includes a natural language description, using the large-scale platform service processing model, of the subjective motivations, psychological activities, and cognitive logic behind the platform service operation decision results. This can be illustrated as follows:

[0207] S402: Determine the persona prompt information corresponding to the reference virtual user agent, and generate agent platform service behavior decision prompts based on the persona prompt information and the reference platform service promotion activities;

[0208] The persona-based cue words refer to pre-built and embedded behavioral role descriptions for virtual user agents, used to guide the large language model in generating response behaviors that conform to the virtual user's preset personality traits and behavioral preferences during the reasoning process. This cue word information is constructed based on historical user profile feature data, typically covering dimensions such as age, region, income level, education background, behavioral habits, risk preferences, and historical platform service participation records. After cue word engineering processing, it is transformed into structured or semi-structured natural language descriptions. For example: "I am a recent graduate from a second-tier city, I value cost-effectiveness, and I don't easily try new platform services."

[0209] The reference platform service promotion activities refer to platform service promotion tasks to be evaluated or simulated, such as marketing campaigns, retention tasks, and product promotions. These include activity types, reach methods, incentive mechanisms, promotion paths, and preset user action goals. This activity information is structured and contextually integrated with the aforementioned persona prompts.

[0210] In the specific processing, the system integrates the "persona prompt information" with the "platform service promotion task description" to generate "agent platform service behavior decision prompts" for input into the large language model by constructing prompt word templates, filling semantic slots, and adjusting language style. Optionally, the prompts can meet the following requirements: (1) express who the agent is (i.e., persona); (2) clarify what platform service task is being faced (i.e., the current promotion content); (3) guide "what decision to make" (such as whether to participate, which path to choose, and why).

[0211] For example, the generated prompt words could be:

[0212] "You are a young user with a monthly income of 6,000 yuan. You expressed concern about the high handling fees during the last installment promotion. This time, you have received an invitation to increase your credit limit. The details of the promotion are as follows... Are you willing to participate? Please state your decision and explain your reasons."

[0213] This method enables the construction of clearly structured, semantically complete, and behaviorally targeted prompts, providing semantic support for subsequent large-scale model behavioral reasoning. This drives the virtual user agent to complete simulation tasks, ensuring that its behavioral output is controllable and consistent with its persona. This step plays a crucial role in personality construction and context activation within the entire platform service agent behavior generation chain.

[0214] S404: Input the agent platform service behavior decision prompt into the platform service processing big model to make platform service operation decisions and obtain reference platform service decision results. Then, perform behavior interpretation processing on the reference platform service decision results to obtain reference agent decision behavior interpretations. Finally, output the reference platform service decision results and the reference agent decision behavior interpretations.

[0215] The agent platform service behavior decision prompts are input into the platform service processing model to drive the model to complete the platform service operation reasoning and obtain the reference platform service decision result. Based on the decision result, behavior interpretation processing is performed to generate the reference agent decision behavior interpretation. Finally, the reference platform service decision result and the reference agent decision behavior interpretation are output.

[0216] The platform service processing big model refers to a generative artificial intelligence model built on large-scale language modeling technology, which has the ability to understand tasks and reason about behavior. It can comprehensively understand the semantic information in the input prompts and deduce simulated platform service decision-making behaviors that conform to the context and role settings.

[0217] The decision prompts for agent platform service behavior are the natural language input content constructed in the aforementioned steps. They integrate the personalized persona information of a specific virtual user agent with the contextual information of the current platform service promotion activities, forming a complete problem context.

[0218] During execution, the system inputs the prompt words into the platform service processing model. The model performs semantic analysis and behavioral logic reasoning on the prompt word content, and generates the possible operation or decision result that the virtual user agent may perform in the current platform service context, which is recorded as the reference platform service decision result. The form of this decision result can be a structured output (such as participating in a loan, abandoning the application, or selecting option A) or a natural language statement, depending on the specific model configuration.

[0219] Subsequently, the control platform service processing big model performs behavioral interpretation processing on the aforementioned reference platform service decision results. This processing step aims to improve the interpretability of virtual user agent behavior. This can be achieved by extracting behavioral motivation statements based on the natural language reasoning path generated by the platform service processing big model itself; by summarizing the logical chain between input prompts and output results to form behavioral preference descriptions; and by combining with the auxiliary interpretation module to back-analyze the reference platform service decision results of the platform service processing big model's decisions, extracting labels such as attraction due to incentives, poor historical experience, or abandonment due to high barriers to entry.

[0220] Ultimately, the platform service processes the output of the large model, which includes two parts:

[0221] Reference platform service decision results: that is, the platform service operation judgments made by the virtual user agent in the simulation scenario;

[0222] Explanation of agency decision-making behavior: This refers to textual or labeled information that explains the motivations, judgment criteria, and behavioral styles behind the behavior.

[0223] This step ensures that every action of the virtual user agent in the sandbox platform service simulation environment not only has a clear outcome but also a traceable behavioral explanation, thereby improving the transparency and credibility of the entire simulation system during strategy evaluation, model debugging, and promotion optimization. This behavior generation and explanation linkage mechanism constitutes the core technical path to achieve decision-making, explainability, and verifiability under the drive of a large model.

[0224] This specification describes a method for constructing decision-making prompts for agent platform services, centered on persona-based prompts. These prompts are then input into a large-scale platform service processing model to drive virtual user agents to generate platform service operation behaviors and corresponding explanations. This enables personalized, context-aware platform service decision-making simulation for individual virtual users within a sandbox platform service simulation environment. This process not only ensures the targeted and diverse nature of virtual user behavior but also provides explainable behavioral motivations, contributing to improved semantic credibility in simulation evaluations and controllability in strategy optimization. It establishes a key mechanism for high-fidelity platform service responses from virtual users driven by a large-scale model.

[0225] Optionally, the specific implementation of adjusting the activity plan for the reference platform service promotion activity based on the reference platform service decision results and the interpretation of reference agent decision behavior, to obtain the target platform service promotion activity for the online platform service processing system, can be carried out in the following ways:

[0226] Based on the decision-making results of the reference platform service and the interpretation of the decision-making behavior of the reference agent, the characteristic factors of the reference platform service promotion activities are adjusted to obtain the target platform service promotion activities for the online platform service processing system.

[0227] Specifically, based on the aforementioned reference platform service decision results and reference agent decision behavior interpretation, several characteristic factors included in the reference platform service promotion activities are analyzed and adjusted to generate more adaptive target platform service promotion activities, and these target platform service promotion activities are used as optimization schemes to be launched into the online platform service processing system.

[0228] During the processing, the system analyzes the impact of this behavioral feedback information on multiple characteristic factors in the promotion activities of the reference platform services. Characteristic factors may include, but are not limited to:

[0229] Outreach mechanisms (such as SMS push notifications, in-site pop-ups, etc.);

[0230] Incentive content (such as coupon amount, exclusive benefit type, etc.);

[0231] Promotion process path (such as whether the participation process is simple, the confirmation steps, etc.);

[0232] Visual copywriting design (such as headline wording, introductory copy, etc.);

[0233] Targeted audience strategies (such as whether to accurately filter and match user profiles), etc.

[0234] Based on data metrics such as conversion rate, engagement rate, and preference distribution reflected in virtual user feedback, and combined with the results of behavioral interpretation semantic analysis, the system performs parameter fine-tuning or strategy replacement operations on the aforementioned promotion factors. For example, it may replace the benefits in high bounce rate activities with reward types that better align with user preferences, or remove and merge redundant steps in the process, thereby generating new activity strategy configurations.

[0235] The final output of the target platform service promotion campaign is an optimized version obtained after simulation evaluation and semantic feedback adjustment. It has higher consistency of behavioral expectations and execution efficiency, and is suitable for actual deployment or delivery in the subsequent online platform service processing system.

[0236] By following the steps above, the platform service promotion activities can be precisely optimized during the simulation phase, effectively avoiding the problems of mismatched audiences and performance deviations after launch, and significantly improving the foresight and refinement of the promotion strategy.

[0237] Optional, such as Figure 8 As shown below, the flowchart for another activity evaluation process is illustrated in the following diagram:

[0238] S502: Obtain a multi-dimensional activity feature set, and generate a reference platform service promotion activity for the online platform service processing system based on the feature combination of the multi-dimensional activity feature set.

[0239] In one feasible implementation, obtaining the reference platform service promotion activities for the online platform service processing system includes executing S502;

[0240] A multi-dimensional activity feature set refers to a collection of multiple key configuration dimensions that constitute the design scheme of a platform service promotion activity. This feature set is used to represent and abstract the strategy composition, delivery method, incentive structure, and audience matching dimensions of the platform service promotion activity. This feature set may include, but is not limited to, the following types:

[0241] The characteristics of the reach mechanism, such as in-site pop-ups, APP homepage entry, short link SMS, etc.;

[0242] Incentive strategy features, such as discount amount, discount threshold, and reward claiming method;

[0243] Visual and interactive design features, such as main title text, button text, and guide animations;

[0244] Promotion process characteristics, such as the number of steps involved, necessary operation nodes, and conversion path complexity;

[0245] Audience targeting features, such as differences between new and old users, geographic filtering, and behavioral tag matching;

[0246] Temporal rhythm characteristics, such as periodicity, duration, and holiday-related strategies.

[0247] The system extracts the aforementioned multi-dimensional activity features based on preset activity design templates or historical activity deployment data. These features are then structured and abstractly constructed into a multi-dimensional activity feature set. During the process of combining these features to form new platform service promotion activities, the system employs a combination generation algorithm to strategically match the aforementioned feature dimensions to complete the initial construction of a reference platform service promotion activity. Feedback from participating virtual user agents is observed, and the reference platform service promotion activity itself is iteratively optimized. The combination generation algorithm can be based on the following modes:

[0248] Rule combination: Construct activity prototypes based on preset rules (such as matching reach method with incentive intensity);

[0249] Reverse generation based on objectives, such as combining low-threshold incentives with prominent entry points on the homepage if the objective is to increase new customer registrations;

[0250] Random perturbations and A / B generation are used to generate a set of comparable active strategies for simulation testing.

[0251] Ultimately, the generated reference platform service promotion activity is a virtual promotion activity configuration scheme that conforms to the operating logic of the online platform service processing system. It is operable, quantifiable, and inferable, and is suitable for subsequent simulation of virtual user agent behavior and strategy evaluation in a sandbox environment.

[0252] Through the above methods, without affecting the real online user experience, a large number of platform service activity prototypes with strategic differences can be generated, and a closed-loop evaluation path can be formed with the simulation agent system, thereby effectively supporting the rehearsal, screening and iterative optimization of real platform service promotion activities.

[0253] S504: Configure a sandbox platform service simulation environment for the promotion activity of the reference platform service, wherein the sandbox platform service simulation environment includes a sandbox platform service processing system and multiple reference virtual user agents;

[0254] For details, please refer to the method steps in other embodiments of this specification, which will not be repeated here.

[0255] S506: In the sandbox platform service simulation environment, the sandbox platform service processing system is controlled to run the reference platform service promotion activities for each reference virtual user agent, and based on the platform service processing big model, the reference virtual user agents are driven to make platform service operation decisions based on the reference platform service promotion activities using persona prompt information to obtain reference platform service decision results and reference agent decision behavior explanations. The reference agent decision behavior explanations include natural language descriptions of the subjective motivations, behavioral decision psychological activities, and cognitive logic of decision behavior behind the platform service operation decision results using the platform service processing big model.

[0256] For details, please refer to the method steps in other embodiments of this specification, which will not be repeated here.

[0257] In one feasible implementation, the interpretation based on the reference platform service decision results and reference agent decision behavior is performed to adjust the characteristic factors of the reference platform service promotion activity, thereby obtaining the target platform service promotion activity for the online platform service processing system, as shown in S508-S512:

[0258] S508: Perform feature mapping processing on the reference platform service decision results and the reference agent decision behavior interpretation to obtain the contribution score of each feature factor in the multi-dimensional activity feature set;

[0259] The reference platform service decision results and the reference agent decision behavior interpretation are subjected to feature mapping processing to obtain the contribution score of each feature factor in the multi-dimensional activity feature set.

[0260] Among them, the reference platform service decision results refer to the platform service behavior decision data output by multiple reference virtual user agents in the sandbox platform service simulation environment during the execution of reference platform service promotion activities. The decision results may include structured behavior tags (such as participation, click rate, conversion path, etc.) and quantitative indicators (such as dwell time, incentive response rate, step completion rate, etc.).

[0261] The reference agent decision-making behavior explanation is a semantic explanation information generated by the platform service processing big model on the above-mentioned behavioral results. It is used to reveal the potential intentions, motivations or cognitive logic behind the virtual user behavior, such as "giving up participation due to excessively high rights threshold" or "bounced due to weak page copywriting".

[0262] Based on the constituent elements of the service promotion activities of the reference platform, the system constructs a multi-dimensional activity feature set. This feature set consists of multiple feature factors, each of which corresponds to key strategy configuration items in the promotion activity, such as activity reach methods, incentive strategy parameters, interface guidance elements, and process complexity level.

[0263] During the feature mapping process, the system maps and matches the aforementioned behavioral results with each feature factor. Through statistical modeling, causal inference analysis, or feature attribution algorithms based on model interpretability (such as SHAP, LIME, Permutation Importance, etc.), it assesses the marginal impact of each feature factor on user decision-making and quantifies its positive or negative contribution to the overall platform service outcome. This contribution is ultimately normalized into a set of numerical indicators, constituting the contribution score vector for each feature factor.

[0264] For example, in a certain reference activity, if multiple virtual user agents report that they "have the intention to give up due to too many process steps", the system will significantly increase the negative contribution score of the "process complexity" feature factor after interpreting and mapping this behavior. At the same time, if some users are triggered to convert due to "high incentive amount", the factor will receive a high positive score.

[0265] Through the above processing method, step S508 achieves accurate measurement of the impact of each component factor in the platform service promotion activities, providing clear data support and adjustment basis for the subsequent screening and local optimization of sensitive factors, and is operable, interpretable and technically feasible.

[0266] S510: Determine the sensitive feature factors whose contribution scores exceed a preset threshold, and perform local search optimization processing within the feature factor value space of the sensitive feature factors to obtain a reference feature factor combination that satisfies a predetermined local search index.

[0267] The sensitive feature factors whose contribution scores exceed a preset threshold are identified, and a local search optimization process is performed within the feature factor value space of the sensitive feature factors to obtain a reference feature factor combination that satisfies a predetermined local search index.

[0268] The contribution score, based on the quantitative indicator generated in step S508, measures the marginal impact of each activity characteristic factor on the decision-making outcome of the virtual user agent platform service. It is typically expressed as a continuous numerical value. The system presets a contribution score threshold to filter out characteristic factors that have a significant impact on behavioral decisions; these are called sensitive characteristic factors. This threshold can be set empirically or optimized using historical simulation results.

[0269] For each sensitive feature factor, the system further determines its corresponding feature factor value space, that is, the limited set of possible values ​​for the factor in the platform service activity configuration. For example, the incentive amount can be [5 yuan, 10 yuan, 20 yuan], the page entry position can be [homepage first screen, middle of function page, sidebar], and the process complexity can be expressed as the number of steps [2 steps, 3 steps, 4 steps]. These values ​​constitute the adjustable operation space of the factor in the current simulation environment.

[0270] Based on the aforementioned feature factor value space, the system employs a local search optimization algorithm to perform combined configuration search processing on multiple sensitive feature factors. This local search process can be implemented using one of the following algorithms:

[0271] Grid search: exhaustively searches through all combinations, suitable for situations with a small value space;

[0272] Bayesian optimization: By constructing surrogate functions and confidence intervals, it gradually approaches the optimal solution, which is suitable for simulation calls with high evaluation costs;

[0273] Genetic algorithms or evolutionary algorithms are suitable for selecting the optimal combination in a discrete combinatorial space.

[0274] Reinforcement learning-based local adjustment strategy: Gradually optimize behavioral feedback targets based on multiple rounds of simulation evaluation.

[0275] The optimization objective of local search is to satisfy or improve as much as possible a set of predetermined indicators, called local search indicators. These indicators may include, but are not limited to, virtual user conversion rate, reduced abandonment rate, shortened average completion time, and increased positive sentiment feedback ratio. During the simulation process, the system dynamically evaluates the simulation performance of each combination and selects the best feature combination scheme that satisfies or is better than the current baseline as the optimized reference feature factor combination.

[0276] Through the above methods, step S510 can systematically optimize key influencing factors, avoiding blind adjustments or manual guesswork, thereby improving the scientific nature and effectiveness of platform service promotion activities before launch, and laying a data-driven optimization foundation for subsequent structured reconstruction and formal deployment.

[0277] S512: Rewrite the structured description language script of the reference platform service promotion activity based on the combination of reference feature factors to form an optimized target platform service promotion activity for the online platform service processing system.

[0278] Based on the combination of reference feature factors, the structured description language script of the reference platform service promotion activity is rewritten to form an optimized target platform service promotion activity for the online platform service processing system.

[0279] The reference feature factor combination refers to the set of feature configurations determined in step S510 that performs optimally in the virtual user agent behavior simulation evaluation or meets the predetermined local search objectives. This combination reflects the activity strategy composition that best promotes the target platform service results (such as user conversion, engagement, and retention) in the current business context. Each feature factor corresponds to a dimension configuration item of the activity strategy, such as: the discount amount in the incentive strategy, the display position in the outreach method, and the number of operation steps in the process design.

[0280] To implement this optimization strategy in the online platform service processing system, the system needs to transform the combination of reference feature factors into a structured description language script that is executable and structurally clear. This script is used to represent the configuration logic of the platform service promotion activities and serves as the direct input for activity configuration and scheduling in the business system.

[0281] Structured description languages ​​can be JSON, YAML, DSL (domain-specific language), or enterprise-defined activity modeling syntaxes. They are used to abstractly describe various elements of platform service activities and typically include the following structural paragraphs:

[0282] Trigger condition definition (e.g., user behavior trigger, timed scheduling, promotion rules);

[0283] User segmentation and matching rules (e.g., expression of applicable audience characteristics);

[0284] Activity flow configuration (e.g., steps, flow nodes, and jump logic);

[0285] Incentive content and presentation (e.g., coupon type, amount range, display style);

[0286] Review and control parameters for gray-scale deployment (such as target indicators, launch time, experiment configuration, etc.).

[0287] In this step, the system matches the corresponding script template field according to each configuration parameter in the reference feature factor combination, and automatically fills the optimized factor value into the structured syntax node through the script generation module, thereby realizing the automatic rewriting of the original reference platform service promotion activity script.

[0288] For example, if the incentive amount in the reference feature factor combination is optimized to 20 yuan, the display position is optimized to "middle of the function page", and the process steps are optimized to "2 steps to reach", then the incentive configuration field in the structured script will be automatically updated to 20, the display module node will be updated to page.mid, and the process configuration field will be updated to step_count:2.

[0289] Through the above methods, step S512 achieves a seamless connection between simulation optimization results and actual deployment solutions, ensuring that the optimization strategy has direct execution capability and structural standard consistency, which helps to significantly improve the efficiency of platform service activities, strategy accuracy and system compatibility.

[0290] In this instruction manual, such as Figure 9 As shown, Figure 9 This is a schematic diagram illustrating a scenario for optimizing platform service promotion activities. Through the systematic processing flow of steps S502 to S512, a closed-loop process can be achieved during the platform service promotion activity construction phase, encompassing multi-dimensional feature collection, sandbox platform service simulation, collection of platform service decision results and agent decision behavior interpretations, and structured solution optimization. This solution constructs a feature factor space and combines it with the "platform service decision results and agent decision behavior interpretations" from virtual user agents for precise causal analysis. It identifies and optimizes key sensitive factors, ultimately generating a directly deployable target platform service promotion activity in the form of a structured language script. This significantly improves the intelligence of activity design, the rationality of parameter configuration, and the automation level of implementation, thereby effectively reducing promotion trial-and-error costs, shortening strategy iteration cycles, and enhancing the reliability of strategy verification before deployment.

[0291] The following will combine Figure 10 This manual provides a detailed description of the activity evaluation and processing device provided. It should be noted that... Figure 10 The activity evaluation processing device shown is used to execute this specification. Figures 1-9 The methods of the embodiments shown are illustrated only in connection with this specification for ease of explanation. For specific technical details not disclosed, please refer to this specification. Figures 1-9 The example shown.

[0292] Please see Figure 10This diagram illustrates the structure of the activity evaluation processing device described in this specification. The activity evaluation processing device 1 can be implemented as all or part of a user's electronic device through software, hardware, or a combination of both. According to some embodiments, the activity evaluation processing device 1 includes an activity acquisition module 11, an environment simulation module 12, an operation decision module 13, and an activity adjustment module 14, specifically used for:

[0293] Activity Acquisition Module 11 is used to acquire reference platform service promotion activities for the online platform service processing system;

[0294] Environment simulation module 12 is used to configure a sandbox platform service simulation environment for the promotion activities of the reference platform service. The sandbox platform service simulation environment includes a sandbox platform service processing system and multiple reference virtual user agents.

[0295] The operation decision module 13 is used to control the sandbox platform service processing system to run the reference platform service promotion activities for each reference virtual user agent in the sandbox platform service simulation environment, and to drive the reference virtual user agents to make platform service operation decisions based on the reference platform service promotion activities using persona prompt information based on the platform service processing big model, so as to obtain reference platform service decision results and reference agent decision behavior explanations. The reference agent decision behavior explanations include natural language descriptions of the subjective motivations, behavioral decision psychological activities and cognitive logic of decision behavior behind the platform service operation decision results using the platform service processing big model.

[0296] The activity adjustment module 14 is used to adjust the activity configuration features of the reference platform service promotion activity based on the reference platform service decision results and the reference agent decision behavior interpretation, so as to obtain the target platform service promotion activity for the online platform service processing system.

[0297] In one feasible implementation, the sandbox platform service simulation environment is configured for the reference platform service promotion activity. The sandbox platform service simulation environment includes a sandbox platform service processing system and multiple reference virtual user agents, comprising:

[0298] Construct a sandbox platform service processing system for the online platform service processing system, and determine multiple reference virtual user agents from the virtual user agent pool for the promotion activities of the reference platform service;

[0299] Configure a sandbox platform service simulation environment based on the multiple reference virtual user agents and the sandbox platform service processing system.

[0300] In one feasible implementation, the method further includes:

[0301] User clustering is performed on historical user profile feature information to obtain a user group feature matrix;

[0302] Based on the user group feature matrix, feature combination is performed to obtain multiple individual user feature vectors, and prompt word engineering is performed on the individual user feature vectors to obtain persona prompt word information;

[0303] The aforementioned character prompts are associated with the platform service processing model to generate multiple virtual user agents;

[0304] A virtual user agent pool is formed based on all the aforementioned virtual user agents.

[0305] In one feasible implementation, the step of combining features based on the user group feature matrix to obtain multiple individual user feature vectors includes:

[0306] Based on the user group feature matrix, user group transformation is performed to obtain multiple user subgroup features. A benchmark user feature vector is selected within each user subgroup feature. A random perturbation of a preset amplitude is applied to the benchmark user feature vector to obtain multiple derived user feature vectors. The derived user feature vectors are used to express the individual behavioral differences within a homogeneous user group.

[0307] Multiple individual user feature vectors are obtained by random feature fusion based on the baseline user feature vector and the derived user feature vector.

[0308] In one feasible implementation, after forming a virtual user agent pool based on all the virtual user agents, the method further includes:

[0309] In the sandbox platform service simulation environment, the virtual decision-making results of each virtual user agent based on the historical platform service promotion activities are continuously collected, and the virtual decision-making results are compared with the historical behavior data of real users to obtain behavior difference information.

[0310] In the virtual user agent pool, the prompt words of the virtual user agent are adjusted based on the behavioral difference information to obtain the adjusted personality prompt words.

[0311] In one feasible implementation, the platform service processing big model utilizes persona prompt information to drive the reference virtual user agent to make platform service operation decisions based on the reference platform service promotion activities, thereby obtaining reference platform service decision results and reference agent decision behavior explanations. The reference agent decision behavior explanations include a natural language description, using the platform service processing big model, of the subjective motivations, psychological activities, and cognitive logic behind the platform service operation decision results, including:

[0312] Determine the persona prompt information corresponding to the reference virtual user agent, and generate agent platform service behavior decision prompts based on the persona prompt information and the reference platform service promotion activities;

[0313] The agent platform service behavior decision prompts are input into the platform service processing big model to make platform service operation decisions and obtain reference platform service decision results. The reference platform service decision results are then processed to obtain reference agent decision behavior explanations. Finally, the reference platform service decision results and the reference agent decision behavior explanations are output.

[0314] In one feasible implementation, the step of adjusting the activity plan for the reference platform service promotion activity based on the reference platform service decision results and the interpretation of reference agent decision behavior to obtain a target platform service promotion activity for the online platform service processing system includes:

[0315] Based on the decision-making results of the reference platform service and the interpretation of the decision-making behavior of the reference agent, the characteristic factors of the reference platform service promotion activities are adjusted to obtain the target platform service promotion activities for the online platform service processing system.

[0316] In one feasible implementation, obtaining reference platform service promotion activities for the online platform service processing system includes:

[0317] Obtain a multi-dimensional activity feature set, and generate a reference platform service promotion activity for the online platform service processing system based on the feature combination of the multi-dimensional activity feature set;

[0318] The process of adjusting the characteristic factors of the reference platform service promotion activity based on the reference platform service decision results and the interpretation of reference agent decision behavior yields a target platform service promotion activity for the online platform service processing system, including:

[0319] The reference platform service decision results and the reference agent decision behavior interpretation are subjected to feature mapping processing to obtain the contribution score of each feature factor in the multi-dimensional activity feature set;

[0320] The sensitive feature factors whose contribution scores exceed a preset threshold are identified, and local search optimization processing is performed in the feature factor value space of the sensitive feature factors to obtain a reference feature factor combination that meets a predetermined local search index.

[0321] Based on the combination of reference feature factors, the structured description language script of the reference platform service promotion activity is rewritten to form an optimized target platform service promotion activity for the online platform service processing system.

[0322] It should be noted that the activity evaluation processing device provided in the above embodiments is only illustrated by the division of the above functional modules when executing the activity evaluation processing method. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the activity evaluation processing device and the activity evaluation processing method embodiments provided in the above embodiments belong to the same concept, and the implementation process is detailed in the method embodiments, which will not be repeated here.

[0323] The serial numbers in this specification are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0324] This specification also provides a computer storage medium capable of storing multiple instructions adapted to be loaded and executed by a processor as described above. Figures 1-9 The activity evaluation processing method described in the illustrated embodiment can be found in the following documentation for its specific execution process. Figures 1-9 The specific details of the illustrated embodiments will not be elaborated here.

[0325] This specification also provides a computer program product that stores at least one instruction, said at least one instruction being loaded and executed by the processor as described above. Figures 1-9 The activity evaluation processing method described in the illustrated embodiment can be found in the following documentation for its specific execution process. Figures 1-9 The specific details of the illustrated embodiments will not be elaborated here.

[0326] Please refer to Figure 11 This is a structural block diagram of an electronic device provided in an embodiment of this specification. The electronic device in this specification may include one or more of the following components: a processor 1010, a memory 1020, an input device 1030, an output device 1040, and a bus 1050. The processor 1010, memory 1020, input device 1030, and output device 1040 may be connected to each other via the bus 1050.

[0327] Processor 1010 may include one or more processing cores. Processor 1010 connects to various parts of the electronic device using various interfaces and lines, and performs various functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in memory 1020, and by calling data stored in memory 1020. Optionally, processor 1010 may be implemented using at least one hardware form of digital signal processing (DSP), field-programmable gate array (FPGA), or programmable logic array (PLA). Processor 1010 may integrate one or more of a central processing unit (CPU), graphics processing unit (GPU), and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the displayed content; and the modem handles wireless communication. It is understood that the modem may also not be integrated into processor 1010 and may be implemented separately through a communication chip.

[0328] The memory 1020 may include random access memory (RAM) or read-only memory (ROM). Optionally, the memory 1020 may include non-transitory computer-readable storage medium. The memory 1020 may be used to store instructions, programs, code, code sets, or instruction sets.

[0329] The input device 1030 is used to receive input instructions or data, and includes, but is not limited to, a keyboard, mouse, camera, microphone, or touch device. The output device 1040 is used to output instructions or data, and includes, but is not limited to, a display device and a speaker. In this embodiment, the input device 1030 can be a temperature sensor for acquiring the operating temperature of the electronic device. The output device 1040 can be a speaker for outputting audio signals.

[0330] In addition, those skilled in the art will understand that the structure of the electronic device shown in the above figures does not constitute a limitation on the electronic device. The electronic device may include more or fewer components than shown, or combine certain components, or have different component arrangements. For example, the electronic device may also include radio frequency circuits, input units, sensors, audio circuits, wireless fidelity (WIFI) modules, power supplies, Bluetooth modules, etc., which will not be described in detail here.

[0331] In the embodiments of this specification, the executing entity for each step can be the electronic device described above. Optionally, the executing entity for each step can be the operating system of the electronic device. The operating system can be Android, iOS, or other operating systems; this specification does not limit this.

[0332] exist Figure 11 In the electronic device, the processor 1010 can be used to call a program stored in the memory 1020 and execute it to implement the activity evaluation processing method as described in the various method embodiments of this specification.

[0333] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory, or random access memory, etc.

[0334] It should be noted that the information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, stored data, displayed data, etc.), and signals involved in the embodiments of this specification are all authorized by the user or fully authorized by all parties, and the collection, use, and processing of related data must comply with the relevant laws, regulations, and standards of the relevant countries and regions. For example, the platform service promotion activities, reference platform service decision results, and reference agent decision behavior interpretations involved in this specification were all obtained with full authorization.

[0335] The above-disclosed embodiments are merely preferred embodiments of this specification and should not be construed as limiting the scope of this specification. Therefore, any equivalent variations made in accordance with the claims of this specification shall still fall within the scope of this specification.

Claims

1. An activity evaluation processing method, the method comprising: performing user clustering on historical user portrait feature information to obtain a user group feature matrix; performing user group conversion based on the user group feature matrix to obtain a plurality of user subgroup features, selecting a reference user feature vector in each user subgroup feature, applying a random disturbance of a preset amplitude to the reference user feature vector to obtain a plurality of derived user feature vectors, the derived user feature vectors being used to express individual behavior differences within a homogeneous user group; performing random feature fusion based on the reference user feature vector and the derived user feature vectors to obtain a plurality of individual user feature vectors, performing prompt word engineering conversion on the individual user feature vectors to obtain human-assigned prompt word information, associating each of the human-assigned prompt word information with a platform service processing large model to generate a plurality of virtual user agents, and forming a virtual user agent pool based on all the virtual user agents; obtaining a reference platform service promotion activity for an online platform service processing system; configuring a sandbox platform service simulation environment for the reference platform service promotion activity, the sandbox platform service simulation environment including a sandbox platform service processing system and a plurality of reference virtual user agents determined from the virtual user agent pool; in the sandbox platform service simulation environment, controlling the sandbox platform service processing system to run the reference platform service promotion activity for each of the reference virtual user agents, and based on the platform service processing large model, driving the reference virtual user agents to make platform service operation decisions based on the reference platform service promotion activity to obtain reference platform service decision results and reference agent decision behavior explanations, the reference agent decision behavior explanations including natural language descriptions of decision behavior subjective reasons, behavior decision psychological activities, and decision behavior cognitive logic behind platform service operation decision results using the platform service processing large model; performing activity configuration feature adjustment on the reference platform service promotion activity based on the reference platform service decision results and the reference agent decision behavior explanations to obtain a target platform service promotion activity for the online platform service processing system; the method further comprising: continuously collecting virtual decision results of each virtual user agent based on historical platform service promotion activities in the sandbox platform service simulation environment, and comparing the virtual decision results with real user historical behavior data to obtain behavior difference information; in the virtual user agent pool, adjusting the human-assigned prompt word information of the virtual user agents based on the behavior difference information to obtain adjusted human-assigned prompt word information.

2. The method of claim 1, wherein configuring a sandbox platform service simulation environment for the reference platform service promotion activity, the sandbox platform service simulation environment including a sandbox platform service processing system and a plurality of reference virtual user agents determined from the virtual user agent pool, comprises: constructing a sandbox platform service processing system for the online platform service processing system, and determining a plurality of reference virtual user agents for the reference platform service promotion activity from a virtual user agent pool; configuring a sandbox platform service simulation environment based on the plurality of reference virtual user agents and the sandbox platform service processing system.

3. The method of claim 1, wherein the behavior difference information is obtained by comparing the virtual decision result with the real user historical behavior data.

4. The method of claim 1, wherein the reference platform service decision result and the reference agent decision behavior explanation are obtained by driving the reference virtual user agent to make platform service operation decisions based on the reference platform service promotion activity using the human-set prompt word information.

5. The method of claim 1, wherein the target platform service promotion activity for the online platform service processing system is obtained by adjusting the feature factors of the reference platform service promotion activity based on the reference platform service decision result and the reference agent decision behavior explanation.

6. The method of claim 5, wherein the reference platform service promotion activity for the online platform service processing system is obtained by combining features based on a multi-dimensional activity feature set.

7. The method of claim 5, wherein the target platform service promotion activity for the online platform service processing system is obtained by adjusting the feature factors of the reference platform service promotion activity based on the reference platform service decision result and the reference agent decision behavior explanation, including: mapping the reference platform service decision result and the reference agent decision behavior explanation to obtain the contribution score of each feature factor in the multi-dimensional activity feature set; determining a sensitive feature factor whose contribution score exceeds a preset threshold; and performing local search optimization processing in the feature factor value space of the sensitive feature factor to obtain a reference feature factor combination that satisfies a predetermined local search index. ​ ​ ​ ​ ​ ​ ​ Rewrite the structured description language script of the reference platform service promotion activity based on the reference characteristic factor combination to form an optimized target platform service promotion activity for the online platform service processing system.

7. An activity evaluation processing apparatus, the apparatus comprising: an activity acquisition module configured to acquire a reference platform service promotion activity for an online platform service processing system; an environment simulation module configured to configure a sandbox platform service simulation environment for the reference platform service promotion activity, the sandbox platform service simulation environment comprising a sandbox platform service processing system and a plurality of reference virtual user agents determined from a virtual user agent pool; an operation decision module configured to control the sandbox platform service processing system to run the reference platform service promotion activity for each of the reference virtual user agents in the sandbox platform service simulation environment, and to drive the reference virtual user agents to make platform service operation decisions based on the reference platform service promotion activity based on platform service processing large model using human-set prompt word information to obtain reference platform service decision results and reference agent decision behavior explanations, the reference agent decision behavior explanations comprising natural language descriptions of subjective motivations, behavioral decision-making psychological activities, and decision-making behavior cognitive logic behind platform service operation decision results using the platform service processing large model; an activity adjustment module configured to adjust activity configuration features of the reference platform service promotion activity based on the reference platform service decision results and the reference agent decision behavior explanations to obtain a target platform service promotion activity for the online platform service processing system; the apparatus is configured to: perform user clustering on historical user portrait feature information to obtain a user group feature matrix; perform user group conversion based on the user group feature matrix to obtain a plurality of user subgroup features, select a reference user feature vector within each user subgroup feature, and apply a random disturbance of a preset amplitude to the reference user feature vector to obtain a plurality of derived user feature vectors, the derived user feature vectors being used to express individual behavior differences within a homogeneous user group; perform random feature fusion based on the reference user feature vector and the derived user feature vectors to obtain a plurality of individual user feature vectors, perform prompt word engineering conversion on the individual user feature vectors to obtain human-set prompt word information, associate each of the human-set prompt word information with a platform service processing large model to generate a plurality of virtual user agents based on model agent correlation, and form a virtual user agent pool based on all the virtual user agents; continuously collect virtual decision results of each virtual user agent based on historical platform service promotion activities in a sandbox platform service simulation environment, and compare the virtual decision results with real user historical behavior data to obtain behavior difference information; in the virtual user agent pool, adjust the human-set prompt word information of the virtual user agents based on the behavior difference information to obtain adjusted human-set prompt word information.

8. A computer storage medium having stored thereon a plurality of instructions adapted to be loaded and executed by a processor to perform the steps of the method according to any one of claims 1 to 6.

9. A computer program product having stored thereon at least one instruction adapted to be loaded and executed by a processor to perform the steps of the method according to any one of claims 1 to 6.

10. An electronic device comprising: a processor and a memory; wherein the memory stores a computer program adapted to be loaded and executed by the processor to perform the steps of the method according to any one of claims 1 to 6. a processor and a memory; wherein the memory stores a computer program adapted to be loaded and executed by the processor to perform the steps of the method according to any one of claims 1 to 6.

Images