Strategy generation method and device based on multi-agent collaboration

Abstract

The invention provides a strategy generation method and device based on multi-agent cooperation, can be applied to the technical field of artificial intelligence and big data, and relates to application of a large model in a financial science and technology scene. The method comprises the steps that under the condition that user authorization or agreement is obtained, user behavior data are collected through a first intelligent agent; processing the user behavior data by using a first agent, and grouping the users to obtain user grouping features; acquiring market dynamic data by using a second intelligent agent; processing the market dynamic data by using a second agent to obtain market dynamic characteristics; conflict detection is carried out on the user grouping features and the market dynamic features, and in conflict detection, whether conflicts exist between the user grouping features and the market dynamic features is detected; and generating a compensation strategy in response to the conflict between the user grouping characteristics and the market dynamic characteristics, the compensation strategy being used for reducing the conflict between the user grouping characteristics and the market dynamic characteristics.

Description

Technical Field

[0001] This application relates to the fields of artificial intelligence and big data technologies, and to the application of large models in the fintech scenario. Specifically, it relates to a strategy generation method, device, electronic device, storage medium, and program product based on multi-agent collaboration. Background Art

[0002] In fields such as e-commerce and personalized recommendation, with the deep integration of big data technology and Internet of Things technology, user behavior data and market dynamic data have shown an exponential growth trend. In the prior art, user behavior data is usually collected in real time by a log collection system deployed at the front end of an e-commerce platform using message queue technology, covering all-link behavior information such as clickstream data, search keyword sequences, adding to cart, and payment on the web or mobile side; market dynamic data is obtained from public e-commerce platforms and industry information websites through web crawler technology, and combined with a structured database to store data such as commodity price fluctuations and category sales trends.

[0003] Currently, data processing systems mostly use distributed file systems and computing frameworks to build data storage and analysis platforms. However, in actual applications, since user behavior data and market dynamic data belong to different business systems, the interface protocols and data formats adopted by each system are different, and there is a lack of unified data governance standards, resulting in problems such as data loss and format incompatibility during the extraction, transformation, and loading of data, forming data islands. The existing processing methods mainly rely on a single data processing module or a serialized processing process, and do not adopt a multi-agent collaboration mechanism, so they cannot make full use of the correlation between data for parallel deep analysis, making it difficult to achieve real-time fusion analysis based on a stream processing engine.

[0004] At the same time, traditional user segmentation technologies mostly use traditional clustering algorithms to build static segmentation models through preset distance measurement formulas and clustering parameters. When facing sudden changes in market demand or abnormal fluctuations in user behavior caused by promotional activities, due to the lack of an adaptive decision-making ability of multi-agent collaboration, it is impossible to adjust the clustering strategy in real time according to dynamic data. Specifically, a single model is difficult to simultaneously take into account data feature extraction, model parameter optimization, and strategy generation, resulting in a lagging user portrait and making it difficult to support enterprises to quickly respond to market changes and formulate accurate marketing strategies. In addition, in the traditional strategy generation process, each business link (such as data cleaning, feature engineering, and strategy formulation) is independent, and there is no collaborative interaction between agents, and it is impossible to perform dynamic game and strategy optimization for a complex and changing market environment, making it difficult for enterprises to flexibly adjust their operation strategies in the competition. Summary of the Invention

[0005] In view of at least one aspect of the above problems, embodiments of this application provide a strategy generation method, apparatus, electronic device, storage medium, and program product based on multi-agent collaboration.

[0006] According to a first aspect of this application, a strategy generation method based on multi-agent collaboration is provided. The method includes: collecting user behavior data using a first agent upon obtaining user authorization or consent, wherein the user behavior data includes user behavior data on an e-commerce platform; processing the user behavior data using the first agent to segment the users and obtain user segmentation features; acquiring market dynamic data using a second agent, wherein the market dynamic data includes data representing at least one of commodity price trends, consumption trends, and market competition trends; processing the market dynamic data using the second agent to obtain market dynamic features; performing conflict detection on the user segmentation features and the market dynamic features, wherein the conflict detection detects whether there is a conflict between the user segmentation features and the market dynamic features; and generating a compensation strategy in response to the conflict between the user segmentation features and the market dynamic features, wherein the compensation strategy is used to reduce the conflict between the user segmentation features and the market dynamic features.

[0007] A second aspect of this application provides a strategy generation apparatus based on multi-agent collaboration. The apparatus includes: a behavior data acquisition module, configured to acquire user behavior data using a first agent upon obtaining user authorization or consent, wherein the user behavior data includes user behavior data on an e-commerce platform; a first agent module, configured to process the user behavior data using the first agent to segment the users and obtain user segmentation features; a dynamic data acquisition module, configured to acquire market dynamic data using a second agent, wherein the market dynamic data includes data representing at least one of commodity price trends, consumption trends, and market competition trends; a second agent module, configured to process the market dynamic data using the second agent to obtain market dynamic features; a conflict detection module, configured to perform conflict detection on the user segmentation features and the market dynamic features, wherein the conflict detection detects whether there is a conflict between the user segmentation features and the market dynamic features; and a strategy generation module, configured to generate a compensation strategy in response to a conflict between the user segmentation features and the market dynamic features, wherein the compensation strategy is used to reduce the conflict between the user segmentation features and the market dynamic features.

[0008] A third aspect of this application provides an electronic device comprising: one or more processors; and a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the method described above.

[0009] A fourth aspect of this application also provides a computer-readable storage medium having a computer program or instructions stored thereon, which, when executed by a processor, implement the steps of the above-described method.

[0010] The fifth aspect of this application also provides a computer program product, including a computer program or instructions that, when executed by a processor, implement the steps of the above-described method.

[0011] The method and apparatus provided in this application embodiment achieve real-time fusion analysis of customer behavior data and market dynamic data through a multi-agent collaborative system architecture, breaking the traditional data silo problem. When a conflict is detected, a compensation strategy is generated through an agent negotiation mechanism. Through collaborative reasoning and dynamic decision-making among agents, a complete closed loop of data acquisition, feature extraction, conflict detection, and strategy optimization is constructed, enabling the method to possess intelligent characteristics of self-awareness, self-learning, and self-optimization, significantly improving the competitiveness of enterprises in complex market environments. Furthermore, the multi-agent architecture supports horizontal scaling, and a single cluster can support multiple concurrent user analysis tasks, thereby effectively improving resource utilization. Attached Figure Description

[0012] The above-mentioned contents, other objects, features and advantages of this application will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0013] Figure 1 The diagram illustrates an application scenario of the policy generation method based on multi-agent cooperation according to an embodiment of this application.

[0014] Figure 2 This is a flowchart of a strategy generation method based on multi-agent cooperation according to an embodiment of this application;

[0015] Figure 3 This is an architecture diagram of the system to which the multi-agent collaborative policy generation method according to an embodiment of this application is applied;

[0016] Figure 4 This is a general flowchart of a strategy generation method based on multi-agent cooperation according to an embodiment of this application;

[0017] Figure 5 This is a flowchart of the steps involved in processing user behavior data using the streaming K-means algorithm;

[0018] Figure 6 This is a flowchart of the step of acquiring market dynamic characteristics using a second intelligent agent in the method according to an embodiment of this application;

[0019] Figure 7 This is a schematic diagram of the knowledge graph constructed in the step of acquiring market dynamic features using a second intelligent agent in the method according to an embodiment of this application;

[0020] Figure 8 This is a flowchart of the conflict detection step in the method according to an embodiment of this application;

[0021] Figure 9 This is a flowchart of the step of generating a compensation strategy in the method according to an embodiment of this application;

[0022] Figure 10 This schematically illustrates a structural block diagram of a policy generation apparatus based on multi-agent cooperation according to an embodiment of this application; and

[0023] Figure 11 A block diagram schematically illustrates an electronic device suitable for implementing a multi-agent cooperative policy generation method according to an embodiment of this application. Detailed Implementation

[0024] The embodiments of this application will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of this application. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of this application for ease of explanation. However, it will be apparent that one or more embodiments may be implemented without these specific details. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concepts of this application.

[0025] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0026] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0027] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0028] In the technical solution of this application, the user information (including but not limited to user personal information, user image information, user device information, such as location information) and data (including but not limited to data used for analysis, stored data, and displayed data) involved are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of related data all comply with relevant laws, regulations, and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entry points for users to choose to authorize or refuse.

[0029] In scenarios involving automated decision-making using personal information, the methods, devices, and systems provided in this application all offer users corresponding entry points for choosing to agree to or reject the automated decision-making results. If the user chooses to reject, the process proceeds to the expert decision-making stage. Here, "automated decision-making" refers to the activity of automatically analyzing and evaluating an individual's behavioral habits, interests, or economic, health, and credit status through computer programs, and then making a decision. Here, "expert decision-making" refers to the activity of making decisions by personnel who specialize in a particular field, possess specialized experience, knowledge, and skills, and have reached a certain level of professional expertise.

[0030] First, the following explanations and descriptions are provided for some of the expressions or technical terms used in this application.

[0031] In the field of artificial intelligence covered in this application, an intelligent agent represents an entity with perception, decision-making, and action capabilities, capable of operating autonomously in a specific environment and achieving goals through interaction with the environment and other intelligent agents. Intelligent agents emphasize autonomy, interactivity, and goal orientation. Large models are complex machine learning models trained on massive amounts of data, possessing powerful data analysis, feature extraction, and pattern recognition capabilities. For example, large language models can process and generate natural language text, performing exceptionally well in areas such as knowledge-based question answering and text generation. It should be understood that large models can empower intelligent agents. In the decision-making process, by incorporating large models, intelligent agents can leverage their powerful language understanding and knowledge reasoning capabilities to enhance their ability to handle complex tasks. Intelligent agents can serve as application vehicles for large models. While large models themselves do not possess autonomous decision-making and action capabilities, intelligent agents can make specific decisions and take actions based on the output of large models, combined with their own goals and the state of the environment.

[0032] Multi-agent collaboration refers to a technology that enables multiple agents to cooperate and communicate to achieve data sharing, task allocation, and joint decision-making.

[0033] Clustering algorithms are fundamental techniques in data mining and machine learning. They aim to divide samples in a dataset into different groups or clusters based on similarity, ensuring high similarity among data objects within the same cluster and significant differences between different clusters. K-means clustering is a classic clustering algorithm, and its general workflow is as follows: Initialization: Randomly select K data points as initial cluster centers; Assignment: Calculate the distance from each sample to each cluster center and assign the sample to the nearest cluster center; Update: Recalculate the mean of all samples within each cluster, using it as the new cluster center; Iteration: Repeat the assignment and update steps until the cluster centers no longer change significantly or the maximum number of iterations is reached.

[0034] Figure 1 The diagram illustrates an application scenario of the policy generation method based on multi-agent cooperation according to an embodiment of this application.

[0035] like Figure 1 As shown, application scenario 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, and a server 105. Network 104 serves as a medium for providing a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. Network 104 may include various connection types, such as wired or wireless communication links or fiber optic cables, etc.

[0036] Users can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 via the network 104 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social media platform software, etc. (for example only).

[0037] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be various electronic devices with displays and support web browsing, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0038] Server 105 can be a server that provides various services, such as a backend management server that supports websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103 (this is just an example). The backend management server can analyze and process data such as received user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices.

[0039] It should be noted that the multi-agent collaborative strategy generation method provided in this application embodiment can generally be executed by server 105. Correspondingly, the multi-agent collaborative strategy generation device provided in this application embodiment can generally be located in server 105. The multi-agent collaborative strategy generation method provided in this application embodiment can also be executed by a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105. Correspondingly, the multi-agent collaborative strategy generation device provided in this application embodiment can also be located in a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105.

[0040] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0041] The following will be based on Figure 1 The described scene, through Figures 2-9 The policy generation method based on multi-agent cooperation according to the embodiments of this application will be described in detail.

[0042] Some exemplary embodiments of this application provide a strategy generation method based on multi-agent collaboration. The method includes: collecting user behavior data using a first agent after obtaining user authorization or consent, wherein the user behavior data includes user behavior data on an e-commerce platform; processing the user behavior data using the first agent to segment the users and obtain user segmentation features; acquiring market dynamic data using a second agent, wherein the market dynamic data includes data representing at least one of commodity price trends, consumption trends, and market competition trends; processing the market dynamic data using the second agent to obtain market dynamic features; performing conflict detection on the user segmentation features and the market dynamic features, wherein the conflict detection detects whether there is a conflict between the user segmentation features and the market dynamic features; and generating a compensation strategy in response to the conflict between the user segmentation features and the market dynamic features, wherein the compensation strategy is used to reduce the conflict between the user segmentation features and the market dynamic features. In the method provided by the embodiments of this application, real-time fusion analysis of customer behavior data and market dynamic data is achieved through a system architecture based on multi-agent collaboration, breaking the traditional data silo problem. When a conflict is detected, a compensation strategy is generated through an agent negotiation mechanism. Through collaborative reasoning and dynamic decision-making among agents, a complete closed loop of data acquisition, feature extraction, conflict detection, and strategy optimization is constructed. This endows the method with intelligent characteristics of self-awareness, self-learning, and self-optimization, significantly enhancing the competitiveness of enterprises in complex market environments. Furthermore, the multi-agent architecture supports horizontal scaling; a single cluster can handle multiple concurrent user analysis tasks, thereby effectively improving resource utilization.

[0043] Figure 2 This is a flowchart of a policy generation method based on multi-agent cooperation according to an embodiment of this application. Figure 3 This is an architecture diagram of the system to which the multi-agent cooperative policy generation method according to an embodiment of this application is applied. Figure 4 This is a general flowchart of a strategy generation method based on multi-agent cooperation according to an embodiment of this application.

[0044] Reference Figure 3 The architecture of the system 300 used in the multi-agent collaborative policy generation method according to the embodiments of this application may include a data acquisition layer 310, a clustering and grouping layer 320, a policy generation layer 330, and a feedback optimization layer 340.

[0045] The data acquisition layer 310 serves as the system's data foundation, constructing a multi-agent-driven, cross-source heterogeneous data real-time access system. For example, the first agent, relying on a high-throughput real-time data stream pipeline and utilizing non-intrusive data tracking technology, can comprehensively and multi-dimensionally collect user browsing, adding to cart, and payment behaviors on e-commerce platforms, collecting dynamic behavioral data including but not limited to dwell time and browsing depth. Simultaneously, by deploying a federated learning platform, the first agent can securely integrate cross-platform third-party data, such as user activity data on social media. The second agent can efficiently connect to external data sources such as price monitoring interfaces and public opinion analysis platforms through an API (Application Programming Interface) gateway cluster, handling tens of thousands of API calls daily.

[0046] In the embodiments of this application, the first intelligent agent can be understood as a customer profiling intelligent agent, which can collect and integrate user behavior data and social attribute information in real time. By using clustering algorithms, especially streaming K-means and other algorithms adapted to dynamic data, it performs in-depth analysis of the collected data to generate dynamic user segmentation results. The second intelligent agent can be understood as a market analysis intelligent agent, which can obtain publicly available market data by calling public APIs and combine it with industry knowledge graphs to perform in-depth mining and analysis of market information, thereby enabling real-time monitoring of competitor strategies, public opinion hotspots, and changes in market trends; with the help of the semantic analysis and reasoning capabilities of knowledge graphs, the second intelligent agent can extract valuable market intelligence from massive amounts of data, providing an important basis for understanding market dynamics and adjusting marketing strategies.

[0047] The clustering and grouping layer 320 is a key component for the system to achieve deep data analysis. For example, the first agent can be equipped with a streaming K-means clustering engine, optimizing the traditional K-means algorithm to achieve real-time updates of user groups at a speed of over 100,000 data points per second, while reducing memory usage by 60%. Simultaneously, by introducing a reinforcement learning model, the first agent can dynamically adjust multiple user group labels and weights based on market feedback, efficiently improving the accuracy of identifying high-value users during promotional activities. The second agent can combine graph databases and natural language processing technology to construct an industry knowledge graph containing massive nodes and relationships, identifying market opportunities (such as emerging demands) and threats (such as competitor price reductions).

[0048] It should be noted that, in the embodiments of this application, the clustering algorithm may also employ other clustering algorithms or adaptive evolutionary clustering algorithms.

[0049] As the core hub of the system's intelligent decision-making, the strategy generation layer 330 achieves accurate decision-making through conflict detection and multi-agent negotiation mechanisms. For example, when the grouping results conflict with market trends, such as when the user group is "price-sensitive" but the market trend shows that premium demand is rising, the strategy generation layer 330 can trigger the multi-agent negotiation mechanism and perform the following steps: (1) Data sharing: The first agent provides user grouping characteristics such as user grouping labels and weights, and the second agent provides market dynamic characteristics; (2) Compensation strategy generation: Dynamic strategies are generated by combining the data or characteristics provided by multiple agents.

[0050] In the examples of this application, the multi-agent negotiation mechanism can represent a joint decision-making process between a first agent and a second agent to resolve conflicts between user segmentation strategies and market trends. During the negotiation process, the two agents can share data, exchanging user segmentation information and market trend analysis results; simultaneously, they evaluate existing rules to determine whether the segmentation strategy aligns with actual market conditions; and when conflicts are detected, they generate compensation strategies, such as adjusting user segmentation criteria or optimizing marketing strategies, to achieve dynamic optimization and balance in decision-making.

[0051] The feedback optimization layer 340 endows the system with continuous evolution capabilities, forming a complete data closed loop. For example, the feedback optimization layer 340 can build a multi-dimensional performance evaluation system, monitor multiple core indicators such as click-through rate, conversion rate, and average order value in real time, and feed back the execution effect (such as click-through rate and conversion rate) of the strategies generated by the strategy generation layer 330 to the clustering and grouping layer in real time, driving the iterative optimization of tag weights and clustering logic, thus forming a closed loop.

[0052] Combined with reference Figures 2 to 4 The strategy generation method based on multi-agent cooperation according to the embodiments of this application may include the following steps S210 to S260.

[0053] In step S210, with the user's authorization or consent obtained, the first intelligent agent collects user behavior data, which includes the user's behavior data on the e-commerce platform.

[0054] For example, the first intelligent agent 410 can achieve multi-dimensional data collection and integration through a distributed data tracking system architecture. Specifically, it is necessary to first obtain the user's authorization or consent. With the user's authorization or consent, the first intelligent agent uses data tracking technology to capture the user's browsing, adding to cart, and payment behaviors on the e-commerce platform in real time. For example, browsing behavior can be represented by multi-dimensional data including but not limited to page dwell time, jump path, and scroll depth; adding to cart behavior can be represented by operation sequences including but not limited to product selection order, quantity adjustment, and specification filtering; payment behavior can be represented by full-process information such as payment method selection, coupon usage, and order cancellation. The first intelligent agent 410 can also integrate third-party platform data based on a federated learning platform, or use privacy computing technologies such as homomorphic encryption and differential privacy computing to achieve secure collaboration of cross-platform data that is "usable but not visible." For example, when integrating social media data, the original data is encrypted and stored locally, and only the model gradient is uploaded through a secure aggregation protocol to participate in joint training. Under the premise of ensuring data privacy, the completeness of customer profiles is improved and the model training efficiency is increased.

[0055] For example, the first intelligent agent 410 may include a data collection module deployed on an e-commerce platform server cluster. Strictly adhering to data compliance principles, it initiates a full-link data collection process after obtaining explicit authorization or consent from the user. This data collection module uses a non-intrusive SDK (Software Development Kit) to collect comprehensive data on user browsing, adding to cart, and payment behaviors across multiple platforms, including apps, web applications, and mini-programs, as well as search records, favorites, and reviews.

[0056] In step S220, the first intelligent agent processes user behavior data and groups users to obtain user grouping characteristics.

[0057] For example, in step S220, user behavior data is processed using a clustering algorithm to group users into clusters, and user cluster labels and weights are added to each user cluster. The user cluster labels represent the behavioral characteristics of the user cluster, and the weights represent the proportion of users represented by the user cluster labels to the total number of users. In this step S220, through in-depth analysis of multi-dimensional user behavior data such as browsing and purchasing, user groups with similar behavioral patterns can be accurately identified, avoiding the drawbacks of traditional extensive segmentation. Assigning labels and weights to each cluster not only intuitively presents the behavioral characteristics of different user groups but also quantifies the proportion of group size, thereby enabling a rapid understanding of the user structure distribution.

[0058] In some exemplary embodiments, this step can first integrate and clean the data obtained in step S210. For example, in the data integration and cleaning stage, multi-source data standardization is first performed. For data from different sources such as event tracking data, API data, and federated learning platform data, the format is unified through a custom parser. Then, noise processing is performed. In the noise processing stage, a quality monitoring framework is integrated to achieve multi-dimensional detection: verifying field integrity, identifying behavioral anomalies based on the isolated forest algorithm, and detecting data drift using time series analysis. Then, the feature engineering module can extract features for user segmentation based on an automatic feature generation framework, such as, but not limited to, features like user activity and consumption frequency. Next, dynamic clustering is performed. In the dynamic clustering stage, a dual-engine collaborative architecture can be adopted, and a streaming K-means algorithm can be used to achieve real-time clustering. In the cluster center iteration stage, a gradient descent optimization algorithm can be introduced to improve the convergence speed. Finally, the clustering results can be output.

[0059] For example, the segmentation results may include user segmentation tags and weights. For instance, the user segmentation tags may include, but are not limited to, "high activity - low conversion", "price-sensitive - active users of promotions", "quality-seeking - new tasters", etc.

[0060] For example, "high activity - low conversion" is a dynamic user segmentation label output by the first intelligent agent, representing the typical behavioral pattern of a certain type of user. High activity indicates a high frequency of user activity on the platform (e.g., frequent browsing, adding to cart, clicking), while low conversion indicates a low actual conversion rate (e.g., payment conversion rate, order completion rate far below average). The generation method involves real-time clustering of user behavior data (e.g., click-through rate, add-to-cart rate, payment rate, etc.) using a streaming K-means algorithm, combined with dynamic adjustments based on time decay coefficients and label weights, ultimately segmenting user groups with similar behavioral patterns.

[0061] For example, a weight can represent the proportion of users categorized into a specific segment (e.g., "high activity - low conversion") within the current customer profile. For instance, if the "high activity - low conversion" segment comprises 30% of all users, and the total number of users is 1 million, then this segment would have approximately 300,000 users, corresponding to a weight of 0.3. This weight can be updated in real-time using a streaming K-means algorithm, reflecting the latest state of user behavior and market changes. For example, promotional activities may cause the proportion of this segment to rise or fall in the short term.

[0062] In other words, in some embodiments of this application, the clustering algorithm includes the streaming K-means algorithm. The streaming K-means clustering algorithm is an improved algorithm proposed to address the limitations of the traditional K-means clustering algorithm when processing real-time data streams. The main improvements are: incremental update mechanism: it does not need to store all historical data, but only maintains cluster centers and statistical information (such as the number of samples within a cluster, mean, and variance), and directly updates these statistics when new data arrives; time decay factor: it assigns higher weight to recent data, enabling the algorithm to quickly adapt to concept drift; micro-cluster structure: it merges similar points into micro-clusters, reducing memory usage and supporting online clustering.

[0063] Figure 5 This is a flowchart illustrating the steps involved in processing user behavior data using the streaming K-means algorithm. (Refer to...) Figure 5 Using clustering algorithms to process user behavior data and segment users can include steps S510 to S530.

[0064] In step S510, the central cluster is used to process periodic user behavior data to determine multiple clusters and the initial cluster center of each cluster. The periodic user behavior data includes historical user behavior data and / or initial batch user behavior data.

[0065] In step S520, after acquiring real-time user behavior data, edge computing nodes are used to process the real-time user behavior data in order to calculate the distance between the real-time user behavior data and the initial cluster centers of each cluster.

[0066] In step S530, based on the distance between the real-time user behavior data and the initial cluster centers of each cluster, the real-time user behavior data is assigned to the nearest cluster.

[0067] In the embodiments of this application, the streaming K-means algorithm can use edge computing nodes to process real-time data streams, while the central cluster is only responsible for periodic cluster merging and splitting. The throughput of a single node is >= 10,000 data entries / second, and the end-to-end latency is controlled within 90 seconds. In addition, elastic resource scheduling can be implemented, dynamically allocating computing resources according to the data stream. For example, when the flow rate is > 10,000 data entries / second, it can automatically expand to 3 computing nodes.

[0068] In the embodiments of this application, in the step of processing user behavior data using the streaming K-means algorithm, the number of clusters can be automatically adjusted according to changes in data distribution, that is, clusters can be merged and split.

[0069] For example, the cluster spacing threshold can be used as the minimum distance threshold to determine whether clusters need to be merged or split. In the Euclidean distance scenario, the threshold range can be 0.2-1.0 times the distance after data standardization. If the data is evenly distributed, it can be set based on the average distance between data points, for example, 1.5-2 times the average distance. If the data is sparsely distributed, the threshold should be reduced, for example, to 0.5-1 times the average distance. The dynamic adjustment strategy uses sliding window statistics and quantile method. The sliding window statistics dynamically adjust the threshold based on the average cluster spacing of the most recent N batches of data, while the quantile method uses the 90th quantile of historical cluster spacing as the threshold. For various business scenarios, if high-granularity clustering is required (e.g., precision marketing), the threshold can be reduced (e.g., 0.3-0.5 times the Euclidean distance), and if rapid response to market changes is required (e.g., competitor price fluctuations), the threshold can be appropriately relaxed (e.g., 0.8-1 times).

[0070] For example, data density metrics can be used to determine the sparsity of data within a cluster or the density threshold required for splitting. In standardized data scenarios, the density threshold can range from 5 to 20 points. Density is calculated by counting the number of data points per unit volume of a sphere within the cluster, with a minimum cluster size of 5-20 points. Recent data is given higher weight, and a time decay coefficient is incorporated into the density calculation. Furthermore, the density threshold is adjusted based on data flow rate; for example, the density requirement is lowered during high flow rates. In business scenarios, to identify high-value user groups (e.g., high-spending consumers), a higher density threshold can be used (e.g., a minimum cluster size of 15 points). To capture emerging market trends (e.g., niche demands), a lower density threshold can be used (e.g., a minimum cluster size of 5 points).

[0071] In the embodiments of this application, by utilizing the streaming K-means algorithm to process user behavior data, the matching with market dynamics can be improved, and at least one of the following technical effects can be achieved: (1) Real-time: Streaming processing supports minute-level updates of cluster centers, adapting to the immediacy of market changes (e.g., sudden price reductions by competitors). (2) Incremental learning: No need for full data recalculation, reducing end-to-end latency and ensuring that strategy generation is synchronized with market trends. (3) Dynamic adaptability: By using a time decay coefficient, recent market signals are strengthened (e.g., increased sensitivity during promotional periods), and the impact of outdated data is weakened. (4) Efficiency and resource optimization: Streaming computation has low complexity, making it suitable for high-throughput real-time data scenarios (e.g., user behavior streams), avoiding the batch processing bottleneck of traditional K-means. Furthermore, when the market shows an increase in premium demand, streaming K-means can quickly adjust user groups (e.g., adding a "quality-oriented" cluster), avoiding strategy lag. Compared with static clustering, streaming algorithms reduce the risk of strategy failure and improve the accuracy of dynamic marketing.

[0072] It's important to note that "premium demand" refers to consumers' willingness to pay extra for goods or services beyond their intrinsic value or the market average. In e-commerce marketing, this demand may stem from factors such as brand effect, product scarcity, and personalized services. Accurately identifying and meeting consumers' premium demand helps companies increase product added value, enhance market competitiveness, and achieve higher profit margins.

[0073] Return to reference Figure 2 In step S230, the second intelligent agent is used to acquire market dynamic data, wherein the market dynamic data includes data representing at least one of commodity price trends, consumption trends and market competition trends.

[0074] For example, in step S230, the second intelligent agent 420 can build an API call cluster, connect to the price monitoring interfaces of multiple mainstream e-commerce platforms, and support high-concurrency requests. It employs a sliding window sampling strategy to achieve real-time monitoring of competitor price fluctuations; it can also integrate a sentiment analysis engine through a natural language processing model, obtain product-related comments by calling social media APIs, and use a hybrid analysis method combining a BERT pre-trained model with a domain dictionary to improve sentiment classification accuracy and identify multiple types of consumer sentiment; it can also obtain industry research reports from public platforms and automatically extract key indicators using text extraction technology.

[0075] In step S240, the second intelligent agent processes market dynamic data to obtain market dynamic characteristics.

[0076] Figure 6 This is a flowchart of the step of acquiring market dynamic characteristics using a second intelligent agent in the method according to an embodiment of this application. Figure 7 This is a schematic diagram of the knowledge graph constructed in the step of acquiring market dynamic features using a second intelligent agent in the method according to an embodiment of this application.

[0077] Combined with reference Figure 2 , Figure 6 and Figure 7 Step S240 may further include steps S2410 to S2450.

[0078] In step S2410, a knowledge graph is constructed based on market dynamic data. The knowledge graph includes multiple nodes and multiple edges. The multiple nodes represent at least one of market entities, market strategies, and market events, and the multiple edges represent the relationships between market entities, market strategies, and market events.

[0079] Specifically, in step S2410, the second intelligent agent can construct a knowledge graph based on market dynamic data as the corpus through structured modeling. This knowledge graph can adopt a graph data structure, with nodes as knowledge units, covering market entities such as competitor strategies (e.g., differentiated pricing, channel optimization plans), user needs (e.g., functional preferences, service demands), and policies and regulations (e.g., industry access standards, regulatory rules); key elements such as market strategies (e.g., promotional activities, price adjustment mechanisms); and market events (e.g., new product launches, industry technological innovations); and edges as relationship carriers, quantifying the dependencies, influence paths, and intensity of effects among various elements through association weights and attribute annotations.

[0080] Taking the smartphone market as an example, refer to Figure 7 In the knowledge graph 700, exemplary nodes include: (1) Market entities: the “A series high cost-performance models” launched by mobile phone manufacturer A as a competitive strategy node 710; the demand of consumers for “long battery life and fast charging function” as a user demand node 720; the “5G communication equipment network access testing standard” issued by the Ministry of Industry and Information Technology as a policy and regulation node 730; (2) Market strategies: the promotional activity of “limited-time direct discount of 800 yuan on e-commerce platforms” of a certain competitive manufacturer B as a node 740; the price adjustment strategy of “trade-in subsidy mechanism” of a certain competitive manufacturer C as a node 750; (3) Market events: the “industry technology innovation” event triggered by the release of a new generation processor by a certain chip manufacturer D as a node 760; the press conference of a certain competitive manufacturer E as a node 770. Examples of edges include: between the event node 770 of a competitor's product launch and the user demand node 720 for "long battery life and fast charging function", there is a related edge that "triggers a change in demand", with a weight of 0.7 (indicating strong correlation) and attributes such as "time: Q3 2023, scope of impact: young user group" can be labeled; between the policy and regulation node 730 and the competitor's strategy node 710, there is a related edge that "compliance constraints", with a weight of 0.9 (indicating strong constraints) and attributes such as "effective date: 2023-01-01, compliance item: radio frequency performance indicators" can be labeled.

[0081] In step S2420, the relationships between market entities, market strategies, and market events are obtained based on the knowledge graph.

[0082] For example, in step S2420, the second intelligent agent initiates a deep relationship mining process based on the integration of publicly available API data (such as competitor price fluctuations and public opinion sentiment analysis results) and the aforementioned knowledge graph. The knowledge graph constructs a structured market relationship network with market entities (including competitor strategies, user needs, policies, and regulations) as nodes and inter-entity relationships as edges. For instance, the "competitor strategy → user need migration path" can visually demonstrate how competitor price reductions or feature upgrades drive user needs from "price sensitive" to "feature-first"; and by utilizing "public opinion hotspots → consumption trends," the intrinsic connection between the rising popularity of topics such as "healthy eating" on social media and the increasing demand for organic food can be captured.

[0083] For example, during the data mining process, graph neural networks and logical reasoning algorithms can be used to perform high-order correlation analysis on node features and edge attributes, systematically identifying causal chains, dynamic influence paths, and co-evolution patterns. Simultaneously, confidence assessment and verification mechanisms ensure that the mining results are both reliable and interpretable, accurately assisting in market trend prediction and competitive strategy optimization.

[0084] In step S2430, market dynamics data are processed using a time series model to obtain short-term market trends.

[0085] For example, in step S2430, the second agent can utilize time series analysis techniques to predict short-term market trends. For instance, a classic model and hybrid algorithm framework can be used to perform trend decomposition, seasonal adjustment, and outlier detection on market data, establishing a dynamic prediction model that includes trend, cyclical, and random components. Through rolling training and real-time data update mechanisms, the immediate impact of external factors such as holiday marketing and unexpected events on market indicators (e.g., sales volume, traffic) can be captured.

[0086] In step S2440, market dynamics data are processed using a large language model to obtain long-term market trends.

[0087] For example, in step S2440, a large language model can be used to conduct long-term market trend analysis. Utilizing the natural language processing capabilities of the large language model, topic modeling, sentiment analysis, and semantic association mining are performed on unstructured text data such as industry reports, news information, and user comments. Through knowledge distillation and transfer learning techniques, domain knowledge is injected into the model parameters to identify macro-trend patterns such as consumption upgrades and technological iterations. Taking the electronics industry as an example, by analyzing a corpus of user comments from social media over many years, the evolution of consumer demand for intelligent and lightweight products can be quantified to analyze long-term market trends.

[0088] In step S2450, market dynamic characteristics are obtained based on correlation, short-term market trends, and long-term market trends.

[0089] For example, in step S2450, the second agent can perform market feature fusion analysis. Based on the relationships in the knowledge graph, the short-term prediction results of the time series model, and the long-term trend judgment of the large language model, a multi-dimensional market analysis framework is constructed. For instance, principal component analysis and hierarchical clustering algorithms can be used to reduce the dimensionality and extract features from heterogeneous data, generating a panoramic market profile that includes dimensions such as competitive landscape, demand evolution, and technological evolution.

[0090] For example, the market dynamic features may include, but are not limited to: the relationships in the knowledge graph (e.g., competitor strategy → user demand migration path, "public opinion hotspot → consumption trend"), real-time market indicators (e.g., competitor price fluctuations, public opinion sentiment analysis results), and market trend analysis results (e.g., long-term market trends).

[0091] In this embodiment, a second intelligent agent is used to process market dynamic data, and market trends are accurately grasped through multi-model collaboration. The time series model can quickly capture short-term market fluctuations caused by factors such as holiday promotions and unexpected events; the large language model mines long-term market trends from massive amounts of unstructured text data. Finally, based on the correlation, short-term and long-term trends, the dynamic characteristics of the market are comprehensively obtained to form a comprehensive and three-dimensional market profile, which can flexibly respond to short-term market changes and grasp the direction in long-term development.

[0092] Return to reference Figure 2 In step S250, conflict detection is performed on user segmentation features and market dynamic features. In the conflict detection, it is detected whether there is a conflict between user segmentation features and market dynamic features.

[0093] In step S260, in response to the conflict between user segmentation characteristics and market dynamics characteristics, a compensation strategy is generated, wherein the compensation strategy is used to reduce the conflict between user segmentation characteristics and market dynamics characteristics.

[0094] Figure 8 This is a flowchart of the conflict detection step in the method according to an embodiment of this application. (See reference...) Figure 2 and Figure 8 Step S250 may further include steps S2510 to S2530. In these steps, the severity of the strategy conflict is measured by quantifying the degree of matching between the customer profile segmentation results and market dynamics.

[0095] In step S2510, the quantitative values ​​of user segmentation characteristics and market dynamic characteristics are obtained.

[0096] For example, the first intelligent agent can be used to quantify the user group labels and weights to form a group feature vector. Based on the weight distribution of the user group labels, for example, "price-sensitive" accounts for 45% and "quality-oriented" accounts for 30%, quantified values ​​of the weights are formed. The first intelligent agent can also be used to quantify market dynamic features such as relationships in the knowledge graph, real-time market indicators, and market trend analysis results to form a market feature vector. For example, the quantified values ​​of the market dynamic features may include, but are not limited to, quantified values ​​of market trend predictions, such as premium demand index and the influence of competitor strategies.

[0097] In step S2520, the quantified values ​​of user segmentation characteristics and market dynamic characteristics are normalized.

[0098] For example, the quantitative values ​​of the user group labels and weights are normalized to the same dimension as the quantitative values ​​of market dynamic characteristics such as the associations in the knowledge graph, real-time market indicators, and market trend analysis results, to ensure comparability.

[0099] In step S2530, a weighted dynamic loss function is used to process the quantified values ​​of the normalized user group characteristics and market dynamic characteristics to obtain the deviation.

[0100] In some exemplary embodiments, the weighted dynamic loss function includes dynamic weights, which are determined by the time decay coefficient and the service priority.

[0101] For example, a weighted dynamic loss function can be constructed using the weighted mean square error, which can be expressed by the following formula (1):

[0102] (1);

[0103] N represents the number of user groups, and i represents one of the N user groups;

[0104] : Indicates the weight of the i-th user group label (e.g., "quality-oriented" has a weight of 0.3);

[0105] : Represents a quantitative value corresponding to the market trend (e.g., premium demand index 0.8);

[0106] : Indicates dynamic weight, which is determined by the time decay coefficient and business priority (for example, recent market data has a weight of 0.7, and historical data has a weight of 0.3).

[0107] For example, the relationship between the time decay coefficient and the dynamic weight can be expressed by the following formula (2):

[0108] (2);

[0109] : Indicates the decay rate (e.g., 0.1); t: Indicates the interval between the data timestamp and the current time (e.g., in hours). This represents the base weight of the i-th user group.

[0110] For example, the KL deviation can be used to construct a weighted dynamic loss function, expressed by the following formula (3), to achieve dynamic calibration of the deviation. For example, if the market trend includes long-term forecasts (e.g., confidence level of consumption upgrade > 0.8), the KL deviation can be used to supplement the measurement of distribution differences.

[0111] (3).

[0112] Figure 9 This is a flowchart of the step of generating a compensation strategy in the method according to an embodiment of this application. (Referring to...) Figure 2 and Figure 9 Step S260 may further include steps S2610 to S2630.

[0113] In step S2610, in response to the deviation being less than a first deviation threshold, a first compensation strategy is generated, wherein in the first compensation strategy, the user group labels and weights remain unchanged, and only the first compensation scheme is generated.

[0114] In step S2620, in response to the deviation being between a first deviation threshold and a second deviation threshold, a second compensation strategy is generated, wherein the weights are adjusted in the second compensation strategy.

[0115] In step S2630, in response to the deviation being greater than the second deviation threshold, a third compensation strategy is generated, wherein the user group labels and weights are adjusted in the third compensation strategy, and a second compensation scheme is generated.

[0116] In the embodiments of this application, the first deviation threshold a1 is less than the second deviation threshold a2.

[0117] In this embodiment, compensation strategies can be generated in stages based on the degree of deviation between user segmentation characteristics and market dynamics, enabling precise decision optimization. When the deviation is low, user segmentation labels and weights are maintained, and only the first compensation scheme is generated to avoid over-adjustment affecting stability, while quickly responding to small market fluctuations. When the deviation is in the middle range, weights are adjusted to flexibly adapt to market changes and optimize resource allocation. When the deviation is high, user segmentation labels and weights are comprehensively adjusted, and a second compensation scheme is generated to deeply correct the segmentation strategy. This tiered compensation mechanism can respond to subtle market changes and adjust in a timely manner during periods of significant market fluctuations, enhancing the flexibility and adaptability of the strategy, effectively reducing the risks caused by the disconnect between the strategy and the market, and improving market competitiveness.

[0118] For example, in step S2620 or step S2630, adjusting the weights may include: extracting user segmentation labels that are most correlated with market dynamics features as first target adjustment labels; extracting user segmentation labels that conflict with market dynamics features as second target adjustment labels; and increasing the weight of the first target adjustment label and decreasing the weight of the second target adjustment label.

[0119] For example, in step S2620 or step S2630, adjusting the user grouping labels may include: extracting target market entities and target relationships from the knowledge graph based on the conflicting market dynamics features, wherein the target market entities and target relationships are market entities and relationships associated with the conflicting market dynamics features; converting the target market entities and target relationships into structured features; and injecting the structured features processed by privacy computing technology into the feature library of the first intelligent agent.

[0120] That is, in step S260, the conflict is classified according to the result of the conflict detection, for example, according to the calculated deviation. For example, the calculated deviation is compared with a predetermined first deviation threshold a1 and a second deviation threshold a2, respectively.

[0121] In some exemplary embodiments, if the calculated deviation is less than a first deviation threshold a1, this is considered a low deviation situation, meaning there is little conflict between user segmentation characteristics and market dynamics. This is determined to be a short-term fluctuation (e.g., holiday promotions), and the existing segmentation strategy is maintained. Only a first compensation plan is generated, which can be a lightweight compensation plan, such as, but not limited to, temporary discounts. As another example, if the calculated deviation is less than the first deviation threshold a1, the real-time suggestions given by the second agent can be prioritized, such as making temporary price adjustments.

[0122] In some exemplary embodiments, when the calculated deviation is between the first deviation threshold a1 and the second deviation threshold a2, this situation belongs to the medium deviation situation, that is, the conflict between user group characteristics and market dynamic characteristics is in the middle, which can trigger the adjustment of label weights. The grouping logic is dynamically optimized through the gradient normalization algorithm, for example, increasing the weight of the "quality preference" label.

[0123] In some exemplary embodiments, when the calculated deviation exceeds the second deviation threshold a2, this is considered a high deviation, indicating a significant conflict between user segmentation characteristics and market dynamics. This can be identified as a long-term trend (e.g., consumption upgrade), and the segmentation rules can be forcibly adjusted (e.g., adding a "high-end user group" cluster) and a compensation strategy can be generated (e.g., "discount coupons + brand story push"). For another example, when the calculated deviation exceeds the second deviation threshold a2, the compensation strategy can focus on reconstructing the segmentation strategy. The market analysis results generated by the second agent can be injected back into the feature library of the first agent, for example, by adding a "live streaming preference" tag, allowing the first agent to re-segment customers based on the new tag.

[0124] It should be noted that the classification model can be trained based on historical test data, and the values ​​of the first deviation threshold a1 and the second deviation threshold a2 can be dynamically set. For example, a1 = 0.2 × historical average deviation, a2 = 0.5 × historical maximum deviation.

[0125] For example, the market analysis results output by the second agent show an increase in demand for "quality-oriented" users, for example, a trend value of 0.75. However, the weight of this user group label in the first agent is only 0.3. The deviation is calculated according to the above formula (1) = In step S260, a decision is made if the deviation is 0.1418 > a1 (assuming a1 = 0.1) but less than a2 (assuming a2 = 0.7). In this case, the label weight adjustment can be triggered, for example, by increasing the "quality-oriented" weight to 0.6 through gradient normalization; and a compensation strategy is generated: push discount coupons to the original "price-sensitive" users, and push high-end products to the new "quality-oriented" users.

[0126] In other words, in the embodiments of this application, a compensation strategy is generated by jointly making decisions on the results output by the first intelligent agent and the second intelligent agent. The generated compensation strategy includes, but is not limited to, the following strategies: (1) Adjusting the priority of tags: for example, adjusting the priority of tags output by the first intelligent agent (e.g., increasing the weight of "quality preference"), and re-clustering users; (2) Generating a compensation scheme: for example, providing a limited-time discount to "price-sensitive" users, while pushing high-end products to "quality-oriented" users; (3) Dynamically updating the feature library: injecting the market analysis results output by the second intelligent agent (e.g., new features of competitors) back into the feature library of the first intelligent agent, prompting the first intelligent agent to consider the market analysis results when performing customer segmentation.

[0127] For example, the triggering conditions generated by each compensation strategy are illustrated below. It should be understood that the triggering conditions below are merely illustrative examples and not an exhaustive description of the triggering conditions generated by each compensation strategy.

[0128] For example, a compensation strategy for adjusting tag priorities could be triggered by conditions such as: the second agent detecting a long-term trend conflict (e.g., consumption upgrade lasting more than 6 months) and the calculated deviation exceeding a second deviation threshold. For instance, if the proportion of "price-sensitive" users is too high in a user segment, but market analysis shows a significant increase in "quality-oriented" demand.

[0129] For example, the triggering conditions for a compensation strategy that generates a compensatory solution may include: the second agent detecting short-term fluctuation conflicts (such as temporary demand shifts caused by holiday promotions) or deviations within a moderate range (e.g., 0.4 ≤ deviation ≤ 0.7). For instance, a sudden price reduction by a competitor causes a short-term mismatch between user segmentation tags and market trends, requiring immediate intervention to stabilize conversion rates.

[0130] For example, the triggering conditions for a compensation strategy that dynamically updates the feature library could include: the second agent recognizing structural market changes (e.g., competitors launching disruptive features) or the feature library lacking key dimensions (e.g., the "green consumption" tag not being covered). For instance, an industry knowledge graph might show a strong correlation between "environmental protection policies" and user needs, but the first agent's feature library lacks relevant fields.

[0131] In the embodiments of this application, when the triggering conditions of each compensation strategy are met, the corresponding compensation strategy can be generated in the following manner.

[0132] In generating a compensation strategy to adjust label priorities, the first step is conflict quantification, such as calculating the deviation between the clustering strategy and the market trend using a loss function. Secondly, if the deviation meets the triggering conditions, label selection can be performed, extracting the labels most correlated with the market trend (e.g., if the market trend is "quality-oriented," then labels like "quality preference" and "brand loyalty" are selected). Then, based on a gradient normalization algorithm, the contribution of each label to the deviation is calculated, prioritizing the adjustment of labels with higher contributions. Thirdly, weight adjustment is performed. In this step, for long-term trends, the weight of the target label can be directly increased (e.g., "quality preference" weight from 0.6 to 0.9), while the weight of conflicting labels can be decreased (e.g., "price sensitivity" from 0.8 to 0.5). During this process, dynamic decay can be considered, introducing a time decay coefficient (e.g., 5% decay per week) to ensure that weight adjustments automatically adapt to trend changes. In some embodiments of this application, by introducing a time decay coefficient and assigning higher weights to recent behaviors; setting key business time intervals and assigning higher weights to behaviors within the key time intervals, such as increasing the weight of the spending power tag during promotional periods; and using machine learning algorithms (such as gradient normalization and multi-task learning) to dynamically adjust the tag weights outside the key business time intervals, the tag priority can be dynamically adjusted.

[0133] For example, if the deviation is caused by the lack of "quality preference", the weight of the label is increased by 30% through a multi-task learning model, and streaming K-means clustering is re-executed.

[0134] In the process of generating compensation strategies for compensatory solutions, a predefined rule base can be pre-built. For example, the predefined rule base may include a mapping table of demand and strategy such as price-sensitive users → discount coupons and quality-oriented users → high-end product push (see Table 1 below). This predefined rule base supports precise matching and fuzzy matching. For example, if a user belongs to a single subgroup, a compensation solution can be generated according to the mapping table; if a user belongs to multiple subgroups, a compensation solution can be generated by weighting and superimposing strategies according to the mapping table.

[0135] Referring to Table 1, for single-tag rules, the triggering condition is that the user segmentation tag is singular and its weight is greater than a predefined weight threshold. For example, if a user segmentation tag is "price-sensitive," and its corresponding weight is greater than or equal to the predefined weight threshold of 0.7, it indicates that the user is price-sensitive. According to this mapping table, the corresponding compensation strategy is triggered, such as issuing a discount coupon (e.g., a discount of 20 yuan off for every 100 yuan spent). Similarly, if a user segmentation tag is "quality-oriented," and its corresponding weight is greater than or equal to the predefined weight threshold of 0.7, it indicates that the user is quality-oriented. According to this mapping table, the corresponding compensation strategy is triggered, such as pushing high-end products (e.g., a high-end product zone entry + targeted discounts). In single-tag rules, the generated compensation strategy does not involve multiple strategies stacked; a single strategy is triggered directly.

[0136] For multi-tag rules, the triggering condition is that the user segmentation tag has more than two tags and the weight of the user segmentation tag is less than the pre-defined weight threshold. For example, if a user segmentation tag is "price-sensitive" + "quality-oriented", its corresponding weights are "price-sensitive" 0.5 and "quality-oriented" 0.4. According to this mapping table, the corresponding compensation strategy is triggered, that is, the weighted superposition strategy is triggered, which can allocate resources according to the tag weight. For example, there is a 50% probability of issuing a discount coupon, a 40% probability of pushing high-end products, and the remaining 10% no action. In multi-tag rules, the generated compensation strategy can be superimposed with multiple strategies. For example, the compensation strategy can be generated according to the strategy superposition formula in Table 1.

[0137]

[0138] In the process of generating a compensation strategy for dynamically updating the feature library, key entities and relationships (e.g., "competitor's new feature → user demand migration") can be extracted from the knowledge graph output by the second agent and converted into structured features (e.g., adding a "feature attention" field, with dimensions including competitor feature mention frequency, user search hot words, etc.). Then, using privacy computing technology, the structured features are encrypted and injected into the feature library of the first agent. The input vector dimension of the streaming K-means algorithm of the first agent can then be updated to support new features participating in dynamic clustering. During the process of new features participating in dynamic clustering, the weights of new features can be initialized based on market trend confidence (e.g., when trend confidence > 0.8, the initial weight is set to 0.7). The importance of features is calibrated by combining historical data backtesting. For example, if market analysis results show that "live shopping" is becoming an emerging trend, a "live interaction frequency" feature can be added to the first agent and assigned an initial weight of 0.6, which can then be dynamically adjusted based on user behavior data.

[0139] In the embodiments of this application, after the new strategy, including the compensation strategy, is executed, the effect data is monitored in real time and fed back to the clustering layer. The impact of the strategy effect on the weights is quantified using a loss function. If the new strategy is effective, the label weight adjustment rules are solidified; if the new strategy is ineffective, a secondary negotiation is triggered, and a compensation strategy is generated again. In this step, the impact of the strategy effect on the weights can be quantified using a loss function. The loss function can be used to measure the deviation between the user group labels and their weights output by the first agent and the market dynamic features output by the second agent. After the new strategy is executed, the deviation is recalculated using the loss function to determine whether the weight adjustment is effective.

[0140] For example, a relationship exists in the knowledge graph: "Competitors launch live shopping features → Increased user interaction time → Increased impulse purchase rate." In this example, conflict detection can be performed: the first agent finds that the "live interaction frequency" tag has a low weight in the user segmentation, but the second agent shows that the relationship is strong. Further, a compensation strategy can be generated, for example, adding a "live streaming sensitivity" tag and adjusting the segmentation logic, while simultaneously pushing live streaming exclusive offers to highly interactive users. Then, feedback optimization can be performed: after the compensation strategy is implemented, if the impulse purchase rate increases, the relationship weight is solidified; if there is no effect, the relationship in the knowledge graph is adjusted in reverse.

[0141] Based on the above-described strategy generation method based on multi-agent cooperation, embodiments of this application also provide a strategy generation apparatus based on multi-agent cooperation. The following will be combined with... Figure 10 The device is described in detail.

[0142] Figure 10 A schematic block diagram of a policy generation apparatus based on multi-agent cooperation according to an embodiment of this application is shown. Figure 10 As shown, the multi-agent collaborative strategy generation device 1000 of this embodiment includes a behavior data acquisition module 1010, a first agent module 1020, a dynamic data acquisition module 1030, a second agent module 1040, a conflict detection module 1050, and a strategy generation module 1060.

[0143] The behavior data acquisition module 1010 is used to acquire market dynamic data using a second intelligent agent, wherein the market dynamic data includes data representing at least one of commodity price trends, consumption trends, and market competition trends. In one embodiment, the behavior data acquisition module 1010 can be used to perform step S210 described above, which will not be repeated here.

[0144] The first intelligent agent module 1020 is used to process user behavior data and segment users to obtain user segmentation characteristics. In one embodiment, the first intelligent agent module 1020 can be used to execute the step S220 described above, which will not be repeated here.

[0145] The dynamic data acquisition module 1030 is used to acquire market dynamic data using a second intelligent agent, wherein the market dynamic data includes data representing at least one of commodity price trends, consumption trends, and market competition trends. In one embodiment, the dynamic data acquisition module 1030 can be used to execute step S230 described above, which will not be repeated here.

[0146] The second agent module 1040 is used to process market dynamic data to obtain market dynamic characteristics. In one embodiment, the second agent module 1040 can be used to perform step S240 described above, which will not be repeated here.

[0147] The conflict detection module 1050 is used to perform conflict detection on user segmentation features and market dynamic features. In conflict detection, it detects whether there is a conflict between user segmentation features and market dynamic features. In one embodiment, the conflict detection module 1050 can be used to perform step S250 described above, which will not be repeated here.

[0148] The strategy generation module 1060 is used to generate a compensation strategy in response to a conflict between user segmentation characteristics and market dynamic characteristics. The compensation strategy is used to reduce the conflict between the user segmentation characteristics and market dynamic characteristics. In one embodiment, the strategy generation module 1060 can be used to execute step S260 described above, which will not be repeated here.

[0149] According to embodiments of this application, any and multiple modules among the behavior data acquisition module 1010, the first intelligent agent module 1020, the dynamic data acquisition module 1030, the second intelligent agent module 1040, the conflict detection module 1050, and the policy generation module 1060 can be combined into one module, or any one of these modules can be split into multiple modules. Alternatively, at least some of the functions of one or more of these modules can be combined with at least some of the functions of other modules and implemented in one module. According to embodiments of this application, at least one of the behavior data acquisition module 1010, the first intelligent agent module 1020, the dynamic data acquisition module 1030, the second intelligent agent module 1040, the conflict detection module 1050, and the policy generation module 1060 can be at least partially implemented as hardware circuits, such as field-programmable gate arrays (FPGAs), programmable logic arrays (PLAs), systems-on-a-chip, systems-on-a-substrate, systems-on-package, application-specific integrated circuits (ASICs), or any other reasonable means of integrating or packaging circuits, or implemented in hardware or firmware, or in any one of software, hardware, and firmware implementations, or in a suitable combination of any of these. Alternatively, at least one of the behavior data acquisition module 1010, the first intelligent agent module 1020, the dynamic data acquisition module 1030, the second intelligent agent module 1040, the conflict detection module 1050, and the policy generation module 1060 can be at least partially implemented as a computer program module, which can perform corresponding functions when the computer program module is run.

[0150] Figure 11 A block diagram schematically illustrates an electronic device suitable for implementing a multi-agent cooperative policy generation method according to an embodiment of this application.

[0151] like Figure 11 As shown, an electronic device 1100 according to an embodiment of this application includes a processor 1101, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1102 or a program loaded from a storage portion 1108 into a random access memory (RAM) 1103. The processor 1101 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 1101 may also include onboard memory for caching purposes. The processor 1101 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of this application.

[0152] RAM 1103 stores various programs and data required for the operation of electronic device 1100. Processor 1101, ROM 1102, and RAM 1103 are interconnected via bus 1104. Processor 1101 executes various operations of the method flow according to embodiments of this application by executing programs in ROM 1102 and / or RAM 1103. It should be noted that the programs may also be stored in one or more memories other than ROM 1102 and RAM 1103. Processor 1101 may also execute various operations of the method flow according to embodiments of this application by executing programs stored in said one or more memories.

[0153] According to embodiments of this application, the electronic device 1100 may further include an input / output (I / O) interface 1105, which is also connected to a bus 1104. The electronic device 1100 may also include one or more of the following components connected to the input / output (I / O) interface 1105: an input section 1106 including a keyboard, mouse, etc.; an output section 1107 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 1108 including a hard disk, etc.; and a communication section 1109 including a network interface card such as a LAN card, modem, etc. The communication section 1109 performs communication processing via a network such as the Internet. A drive 1110 is also connected to the input / output (I / O) interface 1105 as needed. A removable medium 1111, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 1110 as needed so that computer programs read from it can be installed into the storage section 1108 as needed.

[0154] This application also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of this application.

[0155] According to embodiments of this application, the computer-readable storage medium can be a non-volatile computer-readable storage medium, such as including but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of this application, the computer-readable storage medium may include ROM 1102 and / or RAM 1103 and / or one or more memories other than ROM 1102 and RAM 1103 described above.

[0156] Embodiments of this application also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code enables the computer system to implement the multi-agent cooperative strategy generation method provided in the embodiments of this application.

[0157] When the computer program is executed by the processor 1101, it performs the functions defined in the system / apparatus of this application embodiment. According to the embodiments of this application, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0158] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and may be downloaded and installed via the communication section 1109, and / or installed from the removable medium 1111. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0159] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 1109, and / or installed from the removable medium 1111. When the computer program is executed by the processor 1101, it performs the functions defined in the system of this application embodiment. According to the embodiments of this application, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0160] According to embodiments of this application, program code for executing the computer programs provided in the embodiments of this application can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can be executed entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0161] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0162] Those skilled in the art will understand that the features described in the various embodiments of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application.

Claims

1. A strategy generation method based on multi-agent cooperation, characterized in that, The method includes: With the user's authorization or consent, a first intelligent agent is used to collect user behavior data, wherein the user behavior data includes the user's behavior data on the e-commerce platform; The first intelligent agent is used to process the user behavior data and the users are grouped to obtain user grouping characteristics. The second intelligent agent is used to acquire market dynamic data, wherein the market dynamic data includes data representing at least one of commodity price trends, consumption trends, and market competition trends; The second intelligent agent is used to process the market dynamic data to obtain market dynamic characteristics; Conflict detection is performed on the user segmentation features and the market dynamic features, wherein the conflict detection detects whether there is a conflict between the user segmentation features and the market dynamic features; and In response to a conflict between the user segmentation characteristics and the market dynamics characteristics, a compensation strategy is generated, wherein the compensation strategy is used to reduce the conflict between the user segmentation characteristics and the market dynamics characteristics.

2. The method according to claim 1, characterized in that, The step of using the first intelligent agent to process the user behavior data and segment the users to obtain user segmentation features includes: The user behavior data is processed using a clustering algorithm to group the users and add user group labels and weights to each user group. The user group labels are used to represent the behavioral characteristics of the user group, and the weights are used to represent the proportion of the number of users represented by the user group labels to the total number of users.

3. The method according to claim 2, characterized in that, The clustering algorithm includes the streaming K-means algorithm.

4. The method according to claim 3, characterized in that, The process of using clustering algorithms to process the user behavior data and group the users includes: The central cluster is used to process periodic user behavior data to determine multiple clusters and the initial cluster center of each cluster, wherein the periodic user behavior data includes historical user behavior data and / or initial batch user behavior data; After acquiring real-time user behavior data, edge computing nodes are used to process the real-time user behavior data to calculate the distance between the real-time user behavior data and the initial cluster centers of each cluster; and Based on the distance between the real-time user behavior data and the initial cluster centers of each cluster, the real-time user behavior data is assigned to the nearest cluster.

5. The method according to any one of claims 1-4, characterized in that, The process of using the second intelligent agent to process the market dynamic data to obtain market dynamic characteristics includes: Based on the aforementioned market dynamic data, a knowledge graph is constructed, comprising multiple nodes and multiple edges. The nodes represent at least one of market entities, market strategies, and market events, and the edges represent the relationships between these entities, strategies, and events. Based on the knowledge graph, the relationships between market entities, market strategies, and market events are obtained.

6. The method according to claim 5, characterized in that, The process of using the second intelligent agent to process the market dynamic data to obtain market dynamic characteristics includes: The market dynamics data are processed using time series models to obtain short-term market trends. The market dynamics data is processed using a large language model to obtain long-term market trends; and Based on the aforementioned correlations, short-term market trends, and long-term market trends, the dynamic characteristics of the market are obtained.

7. The method according to any one of claims 1-4 and 6, characterized in that, The conflict detection of the user segmentation characteristics and the market dynamic characteristics includes: Obtain the quantitative values ​​of the user segmentation characteristics and the market dynamic characteristics; The quantified values ​​of the user segmentation characteristics and the market dynamic characteristics are normalized; and The weighted dynamic loss function is used to process the quantified values ​​of the normalized user group characteristics and the market dynamic characteristics to obtain the deviation.

8. The method according to claim 7, characterized in that, The weighted dynamic loss function includes dynamic weights, which are determined by the time decay coefficient and the business priority.

9. The method according to claim 7, characterized in that, The response to a conflict between the user segmentation characteristics and the market dynamics characteristics, generating a compensation strategy, includes: In response to the deviation being less than a first deviation threshold, a first compensation strategy is generated, wherein in the first compensation strategy, the user group label and the weight remain unchanged, and only a first compensation scheme is generated; In response to the deviation being between the first deviation threshold and the second deviation threshold, a second compensation strategy is generated, wherein the weights are adjusted in the second compensation strategy; In response to the deviation exceeding the second deviation threshold, a third compensation strategy is generated, wherein the user grouping label and the weight are adjusted, and a second compensation scheme is generated. Wherein, the first deviation threshold is less than the second deviation threshold.

10. The method according to claim 9, characterized in that, Adjusting the weights includes: Extract the user segmentation labels that have the highest correlation with the market dynamics characteristics, and use them as the first target adjustment labels; Extract user segmentation labels that conflict with the aforementioned market dynamics characteristics, and use these as second-target adjustment labels; and Increase the weight of the first target adjustment label and decrease the weight of the second target adjustment label.

11. The method according to claim 9 or 10, characterized in that, Adjusting the user segmentation tags includes: Based on the market dynamics that generate conflict, target market entities and target relationships are extracted from the knowledge graph, wherein the target market entities and target relationships are market entities and relationships associated with the market dynamics that generate conflict; Convert the target market entities and target relationships into structured features; and The structured features processed using privacy computing techniques are injected into the feature library of the first agent.

12. A strategy generation device based on multi-agent cooperation, characterized in that, The device includes: The behavior data collection module is used to collect user behavior data using a first intelligent agent when the user authorizes or agrees, wherein the user behavior data includes the user's behavior data on the e-commerce platform; The first intelligent agent module is used to process the user behavior data using the first intelligent agent, and to group the users to obtain user grouping characteristics. A dynamic data acquisition module is used to acquire market dynamic data using a second intelligent agent, wherein the market dynamic data includes data representing at least one of commodity price trends, consumption trends, and market competition trends; The second intelligent agent module is used to process the market dynamic data to obtain market dynamic characteristics. A conflict detection module is used to perform conflict detection on the user segmentation features and the market dynamic features, wherein the conflict detection detects whether there is a conflict between the user segmentation features and the market dynamic features; and A strategy generation module is used to generate a compensation strategy in response to a conflict between the user segmentation characteristics and the market dynamics characteristics, wherein the compensation strategy is used to reduce the conflict between the user segmentation characteristics and the market dynamics characteristics.

13. An electronic device, comprising: One or more processors; Memory, used to store one or more computer programs. The characteristic feature is that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 11.

14. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 11.

15. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 11.

Images