E-commerce intelligent operation monitoring and collaborative decision-making method based on big data analysis

Abstract

The present invention discloses an e-commerce intelligent operation monitoring and collaborative decision-making method based on big data analysis, including the following steps: S1, collecting multi-source heterogeneous data to construct an original data set; S2, performing data cleaning and characterization processing to generate a feature data set; S3, constructing a knowledge graph and dynamically updating the knowledge graph using a graph embedding model; S4, combining a neural relationship reasoning model and a diffusion model to generate potential causal paths and weights, and introducing a differentiable knowledge enhancement layer to generate causal enhancement features; S5, combining a multi-level anomaly detection mechanism to identify abnormal behaviors; S6, designing a dynamic game optimization model, determining the collaborative optimal solution under multi-objective conflicts through Nash equilibrium, and combining the Stackelberg game to process dominant and subordinate relationships to generate a collaborative optimization strategy; S7, displaying the monitoring status and collaborative optimization strategy through a visual interface. The present invention combines big data analysis, knowledge graphs and dynamic game technology to achieve intelligent operation monitoring and collaborative decision-making for e-commerce.

Description

Technical Field

[0001] The present invention relates to the field of e-commerce intelligent operation technology, and in particular to an e-commerce intelligent operation monitoring and collaborative decision-making method based on big data analysis. Background Art

[0002] With the rapid development of the e-commerce industry, the operation scale and complexity of e-commerce platforms are increasing. The generation of multi-source heterogeneous data such as massive user behavior, product data, logistics information, and market dynamics has put forward higher requirements for the operation monitoring and decision-making of e-commerce platforms. Traditional e-commerce operation monitoring and collaborative decision-making methods mainly rely on fixed rules or single-dimensional data analysis. Although they can play a certain role in specific scenarios, they are often unable to cope with multi-source data, complex causal relationships, and real-time optimization needs. Existing technologies have obvious deficiencies in the following aspects.

[0003] First, the real-time integration and dynamic modeling of multi-source heterogeneous data is still one of the difficulties of current technology. The data sources of e-commerce platforms are diverse, including transaction data, user behavior data, inventory data, and external data such as weather and social media hot spots. These data are not only in various formats, but also updated at different frequencies. The existing technology usually adopts the static data integration method, which lacks the ability to dynamically respond to real-time data and cannot meet the high requirements of e-commerce platforms for real-time performance and data accuracy. In addition, the existing data processing technology is not perfect in handling outliers, missing values, and redundant information in the data, resulting in uneven data quality, which affects the accuracy of subsequent analysis and decision-making.

[0004] Secondly, the ability to identify and reason about complex nonlinear causal relationships is limited. There are often complex nonlinear causal relationships between factors such as user behavior, product attributes, and market dynamics on e-commerce platforms. For example, the increase in sales due to promotional activities may be affected by a combination of factors such as user behavior habits, market competition intensity, and external environment. Existing causal reasoning technologies are usually based on a single model and lack the ability to jointly model multidimensional data, and cannot fully reveal hidden causal paths and weights. In addition, the interpretability of causal reasoning results is insufficient, making it difficult to provide clear decision support for actual operations.

[0005] Third, collaborative optimization under multi-objective conflicts is a core issue in e-commerce operations. Different departments of an e-commerce platform usually have their own operational goals. For example, the operations department hopes to increase sales, the inventory department focuses on inventory turnover, and the logistics department prioritizes delivery efficiency. There are often conflicts between these goals. For example, promotional activities may increase logistics pressure, while inventory optimization may affect the implementation effect of promotions. When solving multi-objective optimization problems, existing technologies usually use single-objective optimization or linear weighted methods based on static weights. It is difficult to balance the conflicting demands between multiple departments in a dynamic environment, and it is also impossible to dynamically adapt to real-time environmental changes.

[0006] In addition, the ability to actively generate abnormal scenarios and control strategies is another weak link in existing technologies. In e-commerce platforms, abnormal scenarios such as surges in demand and logistics delays have a significant impact on operations. However, most existing anomaly detection technologies remain at the passive monitoring stage and lack the ability to actively generate abnormal scenarios and simulate control strategies. This passive mode makes it difficult for the platform to identify potential problems in advance and formulate response strategies. When anomalies occur, only temporary remedial measures can be taken, affecting overall operational efficiency.

[0007] Finally, existing technologies are deficient in the visualization of real-time monitoring and collaborative decision-making. Although some platforms currently have basic data visualization capabilities, they are mostly limited to static reports and charts, making it difficult to dynamically display real-time data and decision results. In addition, the presentation of data analysis results is too complex, making it difficult for decision makers to quickly understand and take action, and lacking intuitive decision support for management.

[0008] Therefore, how to provide an e-commerce intelligent operation monitoring and collaborative decision-making method based on big data analysis is an urgent problem that technicians in this field need to solve. Summary of the invention

[0009] One purpose of the present invention is to propose an e-commerce intelligent operation monitoring and collaborative decision-making method based on big data analysis. The present invention combines big data analysis, dynamic knowledge graphs, causal reasoning and dynamic game optimization technology to realize the intelligent operation monitoring and precise collaborative decision-making of e-commerce platforms in complex and changeable environments. Through dynamic modeling of multi-source heterogeneous data, identification of nonlinear causal relationships, optimization of multi-objective conflicts and active detection and regulation of abnormal behaviors, the real-time performance, decision-making accuracy and collaborative efficiency of the system are significantly improved, while the explainability and transparency of the decision-making process are enhanced, providing all-round technical support for the intelligent operation of e-commerce platforms.

[0010] The e-commerce intelligent operation monitoring and collaborative decision-making method based on big data analysis according to an embodiment of the present invention includes the following steps:

[0011] S1, collect multi-source heterogeneous data, including internal data and external data, and construct the original data set;

[0012] S2. Perform data cleaning and characterization on the original data set, process missing values, outliers and redundant information, and extract time series features, product attribute features and user behavior features to generate a feature data set;

[0013] S3. Build a knowledge graph based on the feature data set, and dynamically update the knowledge graph using the graph embedding model to generate a dynamic knowledge graph representation;

[0014] S4. Using dynamic knowledge graph representation, combining neural relational reasoning model and diffusion model to generate potential causal paths and weights, and generating causal enhancement features by introducing differentiable knowledge enhancement layer;

[0015] S5. Based on the causal enhancement features, combined with the multi-level anomaly detection mechanism, abnormal behaviors are identified and anomaly detection results are generated;

[0016] S6. Design a dynamic game optimization model based on anomaly detection results, determine the optimal solution for collaboration under multi-objective conflicts through Nash equilibrium, and combine the Stackelberg game to deal with the dominant and subordinate relationship, dynamically optimize the cross-departmental collaboration path, and generate a collaborative optimization strategy;

[0017] S7. Realize intelligent e-commerce operations by displaying data monitoring status and collaborative optimization strategies in real time through a visual interface.

[0018] Optionally, the internal data includes transaction data, inventory data, user behavior data and logistics data, and the external data includes weather data, social media hot spots and market price trends.

[0019] Optionally, the S2 specifically includes:

[0020] S21. Process the missing values ​​in the original data set. For numerical data, use interpolation to complete the missing values. For categorical data, use the mode to fill the missing values.

[0021] S22, process the outliers in the original data set, mark the outliers in the numerical data by the 3 times standard deviation method, and set the data point to The original data set mean is , the standard deviation is , when the data point satisfies When , it is regarded as an outlier and replaced with the median; for outliers in categorical data, resampling and replacement are performed according to the category distribution probability;

[0022] S23. Process the redundant information in the original data set, reduce the dimension based on the principal component analysis method, and calculate the covariance matrix:

[0023] ;

[0024] in, represents the covariance matrix, represents the number of samples, Represents the feature matrix of the original data set;

[0025] Covariance matrix Perform eigenvalue decomposition and select eigenvectors with cumulative variance contribution rate of more than 95% to form the feature matrix after dimensionality reduction;

[0026] S24. Extract features from the time series data in the original data set. Suppose the time series data is ,in Represents the total length of the time series data, and the sliding window size is , the window moving step size is , the subsequence is ,in Indicates the first data points; calculate subsequence Time series characteristics:

[0027] ;

[0028] in, Represents a subsequence The time series characteristics of represents the mean of the subsequence, Indicates the first data points, represents the maximum value of the subsequence, Represents the minimum value of the subsequence;

[0029] S25, extracting features from the commodity attribute data, the commodity attributes include category, price, sales volume and rating, and the category attributes are encoded using one-hot encoding to generate commodity attribute features;

[0030] S26. Extract features from user behavior data, where the user behavior data includes browsing, clicking, collecting and purchasing operations, and are arranged in chronological order based on the behavior sequence. User behavior features are generated using sequence clustering analysis, and are integrated with time series features and product attribute features to generate a feature data set.

[0031] Optionally, the S3 specifically includes:

[0032] S31. Build an initial knowledge graph based on the feature data set, with products, users and external environment as node sets , with the relationship between nodes as the edge set , and calculate the relationship weight:

[0033] ;

[0034] in, Indicates Features and The relationship weight between features, Indicates Features, Indicates Features, Indicates Features and The correlation between features based on the strength of the relationship, Indicates Features and The temporal dependency of the context between features, and represents the weight coefficient;

[0035] S32. Use a multi-layer graph neural network to embed the nodes of the initial knowledge graph and calculate the initial embedding representation of each node:

[0036] ;

[0037] in, Indicates Layer Node The embedding representation of represents the activation function, Indicates The layer weight matrix, Indicates Layer Node The embedding representation of Representation Node The number of neighbor nodes, Representation Node The neighbor set of Indicates Layer Node Embedded representation of ;

[0038] S33, introduce the time dimension to dynamically update the initial knowledge graph. The knowledge graph at that time is , add nodes in the data and edge Update node collection and edge set ,in Indicates time The node set at Indicates time , and adjust the relationship weights:

[0039] ;

[0040] in, Indicates time Time Node and nodes The relationship weight between Indicates time Time Node and nodes The relationship weight between represents the exponential function, represents the time weight control parameter, Indicates a time interval;

[0041] S34, reasoning about the relationship between nodes through embedding results, calculating nodes and nodes Relationship inference score:

[0042] ;

[0043] in, Representation Node and nodes The relationship inference score, represents the hyperbolic tangent function, Indicates time Time Node The embedding representation of Indicates time Time Node Embedded representation of ;

[0044] S35. Generate and output dynamic knowledge graph representation.

[0045] Optionally, the S4 specifically includes:

[0046] S41. Using dynamic knowledge graph representation, extract the characteristic information of nodes and edges, including the relationship strength, relationship type and context dependency between nodes, and construct a preliminary causal relationship description;

[0047] S42. Combined with the neural relational reasoning model, node features, edge weights and relation types are jointly modeled to generate potential causal paths between nodes, and time dependencies are dynamically introduced in the path inference process;

[0048] S43. Use the diffusion model to globally optimize the potential causal path. By constructing the path representation in the high-dimensional feature space, analyze the multi-hop correlation between nodes, and perform multi-scale analysis on the local characteristics of nodes and edges, explore the hidden indirect causal relationship, and generate the optimized global causal path.

[0049] S44. In the optimized global causal path, dynamically assign path weights, and make multi-dimensional adjustments to the path weights based on the temporal characteristics, spatial distribution, and importance ranking of the causal path;

[0050] S45. Introduce a differentiable knowledge enhancement layer to dynamically optimize the global causal path weights and generate causal enhancement features:

[0051] Representing domain knowledge as an embedding vector, where the domain knowledge includes historical e-commerce data, user behavior patterns, and commodity associations, and using knowledge graph embedding technology to generate the embedding representation;

[0052] Through the multi-layer attention mechanism, the domain knowledge embedding and the node embedding of the global causal path are interactively calculated to measure the matching degree of knowledge to the path and dynamically adjust the path weight;

[0053] Combining the global path weight and domain knowledge matching degree, the paths are ranked in importance, the weight of paths with matching degrees higher than the set threshold is increased, and the weight of paths with matching degrees lower than the set threshold is weakened;

[0054] The parameters of the differentiable knowledge enhancement layer are optimized through gradient back propagation to achieve the fusion of domain knowledge and data-driven causal path optimization, and output the final causal enhancement features.

[0055] Optionally, the S5 specifically includes:

[0056] S51, extracting global features and local features based on causal enhancement features, and constructing a feature matrix for identifying global patterns and local correlations of abnormal behaviors;

[0057] S52, using a global anomaly detection method, by analyzing the overall distribution of the feature matrix, mining the anomalies in the data that deviate from the global distribution, and identifying abnormal behaviors in the global scope;

[0058] S53, based on the local anomaly detection method, conduct in-depth analysis of regional data, combine the relationship between nodes and local feature distribution, and identify abnormal behaviors in the region;

[0059] S54, introduce causal relationship analysis to verify the global and local anomaly detection results, and combine the weight and importance ranking of the causal path to determine the potential causal relationship of the abnormal behavior;

[0060] S55. Perform multi-dimensional fusion of global anomaly detection, local anomaly detection and causal verification results, comprehensively evaluate the importance and credibility of anomalies, and generate anomaly detection results.

[0061] Optionally, the S6 specifically includes:

[0062] S61. Determine the optimization target based on the anomaly detection results, extract the target parameters in e-commerce operations, including operating income, inventory turnover rate and logistics efficiency, and build a multi-objective optimization framework;

[0063] S62. Use dynamic game models to model the collaborative relationship among multiple departments, define each department as a game participant, clarify the strategy set and benefit function, and analyze the impact of different strategy combinations on the overall operational objectives;

[0064] S63. Solve the optimal solution of coordination under multi-objective conflicts through Nash equilibrium, ensure that each department cannot increase its benefits without changing its own strategy, and achieve interest balance and strategy stability among multiple departments;

[0065] S64. Combined with the Stackelberg game to deal with the dominant and subordinate relationship between departments, the core department is set as the leader and other departments as followers. The leading department has priority in formulating strategies, and the follower department adjusts its own strategy according to the leading strategy, thus optimizing the collaborative effect under the master-slave relationship.

[0066] S65. Dynamically optimize cross-departmental collaboration paths and generate collaborative optimization strategies based on the task dependencies between departments by analyzing the priorities of departmental tasks and resource allocation.

[0067] The beneficial effects of the present invention are:

[0068] First, the present invention realizes efficient data integration and dynamic modeling by collecting and cleaning multi-source heterogeneous data and combining it with the construction and update of dynamic knowledge graphs. The comprehensive processing of internal and external data enables the platform to obtain high-quality data sets in real time, and on this basis, extract time series features, product attribute features, and user behavior features. Through the representation and update of dynamic knowledge graphs, the platform can capture the complex correlations between data in a timely manner, laying a solid foundation for subsequent causal analysis and optimization decisions.

[0069] Secondly, the present invention has significant advantages in causal reasoning. By combining the neural relational reasoning model and the diffusion model, potential causal paths and their weights are dynamically generated to accurately identify the nonlinear causal relationship between user behavior, product attributes and market dynamics. In particular, the introduction of a differentiable knowledge enhancement layer integrates domain knowledge with data-driven methods, which not only improves the accuracy of causal path inference, but also enhances the interpretability and dynamic adaptability of causal reasoning. This causal enhancement feature provides a more reliable foundation for anomaly detection and strategy optimization.

[0070] In addition, the present invention solves the problem of comprehensive identification of global anomalies, local anomalies and causal anomalies through a multi-level anomaly detection mechanism. The combination of global and local anomaly detection fully considers the overall distribution of e-commerce platform data and the differences in local behavior patterns. Causal analysis further verifies the potential causal path of abnormal behavior, making the detection results more interpretable and credible. Through the multi-dimensional fusion of anomaly detection results, the present invention effectively improves the accuracy of anomaly detection and provides comprehensive support for the platform to cope with complex abnormal scenarios.

[0071] The present invention demonstrates outstanding capabilities in multi-objective optimization and collaborative decision-making. Based on the dynamic game optimization model, the optimal collaborative solution under multi-objective conflicts is determined through Nash equilibrium analysis, and the dominant and subordinate relationships between departments are processed in combination with the Stackelberg game to ensure the efficiency and stability of departmental collaboration. Dynamic optimization of cross-departmental collaboration paths further improves the rationality of task allocation and resource scheduling, enabling the platform to achieve global optimal operation under multi-objective conflict conditions.

[0072] Finally, the present invention provides the decision-makers with intuitive and easy-to-use operation monitoring and optimization tools through real-time monitoring and visualization technology. Dynamically displaying key indicators and collaborative optimization strategies, combined with the visualization of the cause-effect chain, transforms complex data analysis results into easy-to-understand decision recommendations. This efficient information transmission mechanism not only improves the management efficiency of the platform, but also enhances the transparency and scientific nature of the decision-making process. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0074] Figure 1 This is an overall flow chart of the e-commerce intelligent operation monitoring and collaborative decision-making method based on big data analysis proposed in the present invention. DETAILED DESCRIPTION

[0075] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0076] refer to Figure 1 , the e-commerce intelligent operation monitoring and collaborative decision-making method based on big data analysis includes the following steps:

[0077] S1, collect multi-source heterogeneous data, including internal data and external data, and construct the original data set;

[0078] S2. Perform data cleaning and characterization on the original data set, process missing values, outliers and redundant information, and extract time series features, product attribute features and user behavior features to generate a feature data set;

[0079] S3. Build a knowledge graph based on the feature data set, and dynamically update the knowledge graph using the graph embedding model to generate a dynamic knowledge graph representation;

[0080] S4. Using dynamic knowledge graph representation, combining neural relational reasoning model and diffusion model to generate potential causal paths and weights, and generating causal enhancement features by introducing differentiable knowledge enhancement layer;

[0081] S5. Based on the causal enhancement features, combined with the multi-level anomaly detection mechanism, abnormal behaviors are identified and anomaly detection results are generated;

[0082] S6. Design a dynamic game optimization model based on anomaly detection results, determine the optimal solution for collaboration under multi-objective conflicts through Nash equilibrium, and combine the Stackelberg game to deal with the dominant and subordinate relationship, dynamically optimize the cross-departmental collaboration path, and generate a collaborative optimization strategy;

[0083] S7. Realize intelligent e-commerce operations by displaying data monitoring status and collaborative optimization strategies in real time through a visual interface.

[0084] In this embodiment, the internal data includes transaction data, inventory data, user behavior data and logistics data, and the external data includes weather data, social media hot spots and market price trends.

[0085] In this implementation, S2 specifically includes:

[0086] S21. Process the missing values ​​in the original data set. For numerical data, use interpolation to complete the missing values. For categorical data, use the mode to fill the missing values.

[0087] S22, process the outliers in the original data set, mark the outliers in the numerical data by the 3 times standard deviation method, and set the data point to The original data set mean is , the standard deviation is , when the data point satisfies When , it is regarded as an outlier and replaced with the median; for outliers in categorical data, resampling and replacement are performed according to the category distribution probability;

[0088] S23. Process the redundant information in the original data set, reduce the dimension based on the principal component analysis method, and calculate the covariance matrix:

[0089] ;

[0090] in, represents the covariance matrix, represents the number of samples, Represents the feature matrix of the original data set;

[0091] Covariance matrix Perform eigenvalue decomposition and select eigenvectors with cumulative variance contribution rate of more than 95% to form the feature matrix after dimensionality reduction;

[0092] S24. Extract features from the time series data in the original data set. Suppose the time series data is ,in Represents the total length of the time series data, and the sliding window size is , the window moving step size is , the subsequence is ,in Indicates the first data points; calculate subsequence Time series characteristics:

[0093] ;

[0094] in, Represents a subsequence The time series characteristics of represents the mean of the subsequence, Indicates the first data points, represents the maximum value of the subsequence, Represents the minimum value of the subsequence;

[0095] S25, extracting features from the commodity attribute data, the commodity attributes include category, price, sales volume and rating, and the category attributes are encoded using one-hot encoding to generate commodity attribute features;

[0096] S26. Extract features from user behavior data, where the user behavior data includes browsing, clicking, collecting and purchasing operations, and are arranged in chronological order based on the behavior sequence. User behavior features are generated using sequence clustering analysis, and are integrated with time series features and product attribute features to generate a feature data set.

[0097] In this implementation, S3 specifically includes:

[0098] S31. Build an initial knowledge graph based on the feature data set, with products, users and external environment as node sets , with the relationship between nodes as the edge set , and calculate the relationship weight:

[0099] ;

[0100] in, Indicates Features and The relationship weight between features, Indicates Features, Indicates Features, Indicates Features and The correlation between features based on the strength of the relationship, Indicates Features and The temporal dependency of the context between features, and represents the weight coefficient;

[0101] S32. Use a multi-layer graph neural network to embed the nodes of the initial knowledge graph and calculate the initial embedding representation of each node:

[0102] ;

[0103] in, Indicates Layer Node The embedding representation of represents the activation function, Indicates The layer weight matrix, Indicates Layer Node The embedding representation of Representation Node The number of neighbor nodes, Representation Node The neighbor set of Indicates Layer Node Embedded representation of ;

[0104] S33, introduce the time dimension to dynamically update the initial knowledge graph. The knowledge graph at that time is , add nodes in the data and edge Update node collection and edge set ,in Indicates time The node set at Indicates time , and adjust the relationship weights:

[0105] ;

[0106] in, Indicates time Time Node and nodes The relationship weight between Indicates time Time Node and nodes The relationship weight between represents the exponential function, represents the time weight control parameter, Indicates a time interval;

[0107] S34, reasoning about the relationship between nodes through embedding results, calculating nodes and nodes Relationship inference score:

[0108] ;

[0109] in, Representation Node and nodes The relationship inference score, represents the hyperbolic tangent function, Indicates time Time Node The embedding representation of Indicates time Time Node Embedded representation of ;

[0110] S35. Generate and output dynamic knowledge graph representation.

[0111] In this implementation, S4 specifically includes:

[0112] S41. Using dynamic knowledge graph representation, extract the characteristic information of nodes and edges, including the relationship strength, relationship type and context dependency between nodes, and construct a preliminary causal relationship description;

[0113] S42. Combined with the neural relational reasoning model, node features, edge weights and relation types are jointly modeled to generate potential causal paths between nodes, and time dependencies are dynamically introduced in the path inference process;

[0114] S43. Use the diffusion model to globally optimize the potential causal path. By constructing the path representation in the high-dimensional feature space, analyze the multi-hop correlation between nodes, and perform multi-scale analysis on the local characteristics of nodes and edges, explore the hidden indirect causal relationship, and generate the optimized global causal path.

[0115] S44. In the optimized global causal path, dynamically assign path weights, and make multi-dimensional adjustments to the path weights based on the temporal characteristics, spatial distribution, and importance ranking of the causal path;

[0116] S45. Introduce a differentiable knowledge enhancement layer to dynamically optimize the global causal path weights and generate causal enhancement features:

[0117] Representing domain knowledge as an embedding vector, where the domain knowledge includes historical e-commerce data, user behavior patterns, and commodity associations, and using knowledge graph embedding technology to generate the embedding representation;

[0118] Through the multi-layer attention mechanism, the domain knowledge embedding and the node embedding of the global causal path are interactively calculated to measure the matching degree of knowledge to the path and dynamically adjust the path weight;

[0119] Combining the global path weight and domain knowledge matching degree, the paths are ranked in importance, the weight of paths with matching degrees higher than the set threshold is increased, and the weight of paths with matching degrees lower than the set threshold is weakened;

[0120] The parameters of the differentiable knowledge enhancement layer are optimized through gradient back propagation to achieve the fusion of domain knowledge and data-driven causal path optimization, and output the final causal enhancement features.

[0121] In this implementation manner, S5 specifically includes:

[0122] S51, extracting global features and local features based on causal enhancement features, and constructing a feature matrix for identifying global patterns and local correlations of abnormal behaviors;

[0123] S52, using a global anomaly detection method, by analyzing the overall distribution of the feature matrix, mining the anomalies in the data that deviate from the global distribution, and identifying abnormal behaviors in the global scope;

[0124] S53, based on the local anomaly detection method, conduct in-depth analysis of regional data, combine the relationship between nodes and local feature distribution, and identify abnormal behaviors in the region;

[0125] S54, introduce causal relationship analysis to verify the global and local anomaly detection results, and combine the weight and importance ranking of the causal path to determine the potential causal relationship of the abnormal behavior;

[0126] S55. Perform multi-dimensional fusion of global anomaly detection, local anomaly detection and causal verification results, comprehensively evaluate the importance and credibility of anomalies, and generate anomaly detection results.

[0127] In this implementation manner, S6 specifically includes:

[0128] S61. Determine the optimization target based on the anomaly detection results, extract the target parameters in e-commerce operations, including operating income, inventory turnover rate and logistics efficiency, and build a multi-objective optimization framework;

[0129] S62. Use dynamic game models to model the collaborative relationship among multiple departments, define each department as a game participant, clarify the strategy set and benefit function, and analyze the impact of different strategy combinations on the overall operational objectives;

[0130] S63. Solve the optimal solution of coordination under multi-objective conflicts through Nash equilibrium, ensure that each department cannot increase its benefits without changing its own strategy, and achieve interest balance and strategy stability among multiple departments;

[0131] S64. Combined with the Stackelberg game to deal with the dominant and subordinate relationship between departments, the core department is set as the leader and other departments as followers. The leading department has priority in formulating strategies, and the follower department adjusts its own strategy according to the leading strategy, thus optimizing the collaborative effect under the master-slave relationship.

[0132] S65. Dynamically optimize cross-departmental collaboration paths and generate collaborative optimization strategies based on the task dependencies between departments by analyzing the priorities of departmental tasks and resource allocation.

[0133] Embodiment 1:

[0134] In order to verify the feasibility of the present invention in implementation, the present invention is applied to the Double Eleven promotion of a large domestic e-commerce platform to analyze and optimize the effects of its operation monitoring and collaborative decision-making. In this scenario, the e-commerce platform faces many challenges, including a surge in demand caused by promotional activities, insufficient inventory, excessive logistics pressure, and insufficient collaboration efficiency among multiple departments. Traditional static monitoring and decision-making methods are difficult to meet the needs of real-time optimization, resulting in frequent inventory shortages, order processing delays and other problems during the event, which directly affects user experience and sales performance.

[0135] During the Double Eleven event, the system of the present invention first collected multi-source heterogeneous data, including internal transaction data, inventory data, user behavior data and logistics data, as well as external weather information, social media hot spots and market competition data. These data are transmitted to the big data platform in real time for cleaning and characterization processing to generate high-quality feature data sets. Through automatic processing of missing values ​​and outliers, combined with time series analysis and user behavior feature extraction, the platform constructs a comprehensive and accurate dynamic knowledge graph. The knowledge graph contains the complex relationships between goods, users, and the environment, and is dynamically updated using a graph embedding model to ensure the real-time and accuracy of data relevance.

[0136] Subsequently, the system generates potential causal paths and weights by combining the neural relational reasoning model and the diffusion model, and introduces a differentiable knowledge enhancement layer to optimize the path. For example, the analysis found that weather has a significant impact on the sales of certain products (such as umbrellas and cold-proof clothing), and the spread of social media hot spots has a strong correlation with the conversion rate of certain brand activities. Through the generation of causal enhancement features, the anomaly detection module can quickly identify abnormal behaviors during activities, such as logistics delays in some areas and insufficient inventory of certain hot-selling products.

[0137] The anomaly detection results are used as input to the dynamic game optimization model. The system handles the dominant and subordinate relationships between multiple departments through Nash equilibrium analysis and Stackelberg game optimization. For example, the platform's inventory department dynamically adjusts the product distribution strategy based on the real-time transportation capacity of the logistics department and the conversion rate forecast of user behavior to prioritize the supply of goods in high-demand areas. At the same time, the operation department collaborates with the advertising department to adjust the promotion efforts and advertising strategies in real time to reduce the order volume in the logistics bottleneck area and achieve the goal of global optimization.

[0138] The monitoring status and collaborative optimization strategies of the entire process are displayed in real time through a visual interface. For example, managers can view key indicators such as real-time sales, inventory status, and logistics efficiency through the dashboard, and quickly identify problems and take action based on cause-and-effect chain analysis.

[0139] In actual tests, the method of the present invention significantly improved the operational efficiency and user experience during the Double Eleven period. Taking a popular smartwatch on the e-commerce platform as an example, its sales data and inventory distribution in different regions were monitored before, during and after the event. The following table 1 is a specific data table:

[0140] Table 1 Comparison of smartwatch sales and inventory distribution data during the Double Eleven event

[0141]

[0142] In the above test data, through the method of the present invention, the platform realizes the rapid identification and dynamic adjustment of abnormal areas. For example, in response to the inventory shortage in East China, the system automatically allocates part of the inventory from South China to East China within 0.5 hours of the shortage, and shortens the shortage time to less than 2 hours with the cooperation of the logistics department. At the same time, the advertising department dynamically reduces the promotion efforts in East China, reduces the generation of new orders, and further alleviates inventory pressure. Compared with traditional operating methods, the method of the present invention significantly reduces the out-of-stock rate (from the traditional approximately 20% to 5%-12%), shortens the average delivery time (from 12 hours to 8.4 hours), and improves the overall user experience.

[0143] This embodiment proves that the present invention has excellent performance and significant effects in dealing with high-intensity operation scenarios of e-commerce platforms. It not only realizes the intelligence of the entire process, but also effectively solves the problems of data processing lag, insufficient causal reasoning capabilities and inefficient collaboration among multiple departments in traditional methods.

[0144] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. An e-commerce intelligent operation monitoring and collaborative decision-making method based on big data analysis, characterized in that: The steps include: S1, collect multi-source heterogeneous data, including internal data and external data, and construct the original data set; S2. Perform data cleaning and characterization on the original data set, process missing values, outliers and redundant information, and extract time series features, product attribute features and user behavior features to generate a feature data set; S3. Build a knowledge graph based on the feature data set, and dynamically update the knowledge graph using the graph embedding model to generate a dynamic knowledge graph representation; S4. Using dynamic knowledge graph representation, combining neural relational reasoning model and diffusion model to generate potential causal paths and weights, and generating causal enhancement features by introducing differentiable knowledge enhancement layer; S5. Based on the causal enhancement features, combined with the multi-level anomaly detection mechanism, abnormal behaviors are identified and anomaly detection results are generated; S6. Design a dynamic game optimization model based on anomaly detection results, determine the optimal solution for collaboration under multi-objective conflicts through Nash equilibrium, and combine the Stackelberg game to deal with the dominant and subordinate relationships, dynamically optimize the cross-departmental collaboration path, and generate a collaborative optimization strategy; S7. Realize intelligent e-commerce operations by displaying data monitoring status and collaborative optimization strategies in real time through a visual interface.

2. The e-commerce intelligent operation monitoring and collaborative decision-making method based on big data analysis according to claim 1 is characterized in that: The internal data includes transaction data, inventory data, user behavior data and logistics data, and the external data includes weather data, social media hot spots and market price trends.

3. The e-commerce intelligent operation monitoring and collaborative decision-making method based on big data analysis according to claim 1 is characterized in that: The S2 specifically includes: S21. Process the missing values ​​in the original data set. For numerical data, use interpolation to complete the missing values. For categorical data, use the mode to fill the missing values. S22, process the outliers in the original data set, mark the outliers in the numerical data by the 3 times standard deviation method, and set the data point to The original data set mean is , the standard deviation is , when the data point satisfies When , it is regarded as an outlier and replaced with the median; for outliers in categorical data, resampling and replacement are performed according to the category distribution probability; S23. Process the redundant information in the original data set, reduce the dimension based on the principal component analysis method, and calculate the covariance matrix: ； in, represents the covariance matrix, represents the number of samples, Represents the feature matrix of the original data set; Covariance matrix Perform eigenvalue decomposition and select eigenvectors with cumulative variance contribution rate of more than 95% to form the feature matrix after dimensionality reduction; S24. Extract features from the time series data in the original data set. Suppose the time series data is ,in Represents the total length of the time series data, and the sliding window size is , the window moving step size is , the subsequence is ,in Indicates the first data points; calculate subsequence Time series characteristics: ； in, Represents a subsequence The time series characteristics of represents the mean of the subsequence, Indicates the first data points, represents the maximum value of the subsequence, Represents the minimum value of the subsequence; S25, extracting features from the commodity attribute data, the commodity attributes include category, price, sales volume and rating, and the category attributes are encoded using one-hot encoding to generate commodity attribute features; S26. Extract features from user behavior data, where the user behavior data includes browsing, clicking, collecting and purchasing operations, and are arranged in chronological order based on the behavior sequence. User behavior features are generated using sequence clustering analysis, and are integrated with time series features and product attribute features to generate a feature data set.

4. The e-commerce intelligent operation monitoring and collaborative decision-making method based on big data analysis according to claim 1 is characterized in that: The S3 specifically includes: S31. Build an initial knowledge graph based on the feature data set, with products, users and external environment as node sets , with the relationship between nodes as the edge set , and calculate the relationship weight: ； in, Indicates Features and The relationship weight between features, Indicates Features, Indicates Features, Indicates Features and The correlation between features based on the strength of the relationship, Indicates Features and The temporal dependency of the context between features, and represents the weight coefficient; S32. Use a multi-layer graph neural network to embed the nodes of the initial knowledge graph and calculate the initial embedding representation of each node: ； in, Indicates Layer Node The embedding representation of represents the activation function, Indicates The layer weight matrix, Indicates Layer Node The embedding representation of Representation Node The number of neighbor nodes, Representation Node The neighbor set of Indicates Layer Node Embedded representation of ; S33, introduce the time dimension to dynamically update the initial knowledge graph. The knowledge graph at that time is , add nodes in the data and edge Update node collection and edge set ,in Indicates time The node set at Indicates time , and adjust the relationship weights: ； in, Indicates time Time Node and nodes The relationship weight between Indicates time Time Node and nodes The relationship weight between represents the exponential function, represents the time weight control parameter, Indicates a time interval; S34, reasoning about the relationship between nodes through embedding results, calculating nodes and nodes Relationship inference score: ； in, Representation Node and nodes The relationship inference score, represents the hyperbolic tangent function, Indicates time Time Node The embedding representation of Indicates time Time Node Embedding representation of ; S35. Generate and output dynamic knowledge graph representation.

5. The e-commerce intelligent operation monitoring and collaborative decision-making method based on big data analysis according to claim 1 is characterized in that: The S4 specifically includes: S41. Using dynamic knowledge graph representation, extract the characteristic information of nodes and edges, including the relationship strength, relationship type and context dependency between nodes, and construct a preliminary causal relationship description; S42. Combined with the neural relational reasoning model, node features, edge weights and relation types are jointly modeled to generate potential causal paths between nodes, and time dependencies are dynamically introduced in the path inference process; S43. Use the diffusion model to globally optimize the potential causal path. By constructing the path representation in the high-dimensional feature space, analyze the multi-hop correlation between nodes, and perform multi-scale analysis on the local characteristics of nodes and edges, explore the hidden indirect causal relationship, and generate the optimized global causal path. S44. In the optimized global causal path, dynamically assign path weights, and make multi-dimensional adjustments to the path weights based on the temporal characteristics, spatial distribution, and importance ranking of the causal path; S45. Introduce a differentiable knowledge enhancement layer to dynamically optimize the global causal path weights and generate causal enhancement features: Representing domain knowledge as an embedding vector, where the domain knowledge includes historical e-commerce data, user behavior patterns, and commodity associations, and using knowledge graph embedding technology to generate the embedding representation; Through the multi-layer attention mechanism, the domain knowledge embedding and the node embedding of the global causal path are interactively calculated to measure the matching degree of knowledge to the path and dynamically adjust the path weight; Combining the global path weight and domain knowledge matching degree, the paths are ranked in importance, the weight of paths with matching degrees higher than the set threshold is increased, and the weight of paths with matching degrees lower than the set threshold is weakened; The parameters of the differentiable knowledge enhancement layer are optimized through gradient back propagation to achieve the fusion of domain knowledge and data-driven causal path optimization, and output the final causal enhancement features.

6. The e-commerce intelligent operation monitoring and collaborative decision-making method based on big data analysis according to claim 1 is characterized in that: The S5 specifically includes: S51, extracting global features and local features based on causal enhancement features, and constructing a feature matrix for identifying global patterns and local correlations of abnormal behaviors; S52, using a global anomaly detection method, by analyzing the overall distribution of the feature matrix, mining the anomalies in the data that deviate from the global distribution, and identifying abnormal behaviors in the global scope; S53, based on the local anomaly detection method, conduct in-depth analysis of regional data, combine the relationship between nodes and local feature distribution, and identify abnormal behaviors in the region; S54, introduce causal relationship analysis to verify the global and local anomaly detection results, and combine the weight and importance ranking of the causal path to determine the potential causal relationship of the abnormal behavior; S55. Perform multi-dimensional fusion of global anomaly detection, local anomaly detection and causal verification results, comprehensively evaluate the importance and credibility of anomalies, and generate anomaly detection results.

7. The e-commerce intelligent operation monitoring and collaborative decision-making method based on big data analysis according to claim 1 is characterized in that: The S6 specifically includes: S61. Determine the optimization target based on the anomaly detection results, extract the target parameters in e-commerce operations, including operating income, inventory turnover rate and logistics efficiency, and build a multi-objective optimization framework; S62. Use dynamic game models to model the collaborative relationship among multiple departments, define each department as a game participant, clarify the strategy set and benefit function, and analyze the impact of different strategy combinations on the overall operational objectives; S63. Solve the optimal solution of coordination under multi-objective conflicts through Nash equilibrium, ensure that each department cannot increase its benefits without changing its own strategy, and achieve interest balance and strategy stability among multiple departments; S64. Combined with the Stackelberg game to deal with the dominant and subordinate relationship between departments, the core department is set as the leader and other departments as followers. The leading department has priority in formulating strategies, and the follower department adjusts its own strategy according to the leading strategy, thus optimizing the collaborative effect under the master-slave relationship. S65. Dynamically optimize cross-departmental collaboration paths and generate collaborative optimization strategies based on the task dependencies between departments by analyzing the priorities of departmental tasks and resource allocation.

Images