An economic management data analysis method and system for e-commerce

Abstract

The present invention discloses an economic management data analysis method and system for e-commerce, which relates to the technical field of data analysis. After constructing a dynamic graph, the heat scores of each node are calculated, and the node with the maximum heat score is highlighted in the dynamic graph. Each node is randomly assigned to multiple communities, and the modularity of each community is maximized and then iterated until the modularity converges. After that, the commodity community graph is displayed through the dynamic graph. The dynamic graph is divided into multiple time steps, an embedding representation is generated for the graph of each time step, the graph changes between adjacent time steps are calculated, the change trend of the competitive pattern in the dynamic graph is identified, and decision-making suggestions are generated based on the optimized dynamic graph. The analysis system maps information such as commodity association relationships and heat scores into time steps, dynamically reflects market changes, quantifies the change trend of the competitive pattern, provides a basis for quickly responding to market fluctuations, and provides accurate market response strategies based on the optimized dynamic graph.

Description

Technical Field

[0001] The present invention relates to the technical field of data analysis, and particularly to a method and system for economic management data analysis in e-commerce. Background Art

[0002] With the rapid development of e-commerce, there are a wide variety of products and complex user behaviors on the platform, and the market competition is becoming increasingly fierce. Enterprises not only need to attract users through innovative marketing strategies, but also need to improve operational efficiency, achieve cost control and maximize profits. In this context, the economic management data analysis system has gradually become an important tool for decision-making support in e-commerce enterprises.

[0003] The existing technologies have the following defects:

[0004] 1. The existing popularity evaluation methods usually rely on a single dimension such as sales volume and click-through rate, ignoring the collaborative relationship between products (such as co-click or keyword similarity). The popularity of a product may be affected by related products in the actual scenario (such as a promotion of a product may increase the popularity of other related products). Static evaluation cannot reflect this complex network effect, and long-tail products (with low sales volume or click-through rate but strong relevance) are often ignored in the existing models, making it impossible to comprehensively evaluate the market potential;

[0005] 2. In the face of sudden market changes (such as competitor promotions, new product launches), the existing technologies cannot quickly adjust strategies and are prone to losing short-term competitive advantages. The existing analysis models usually lack the attention and optimization ability for long-term trends, resulting in the strategy formulation being biased towards short-term interests and ignoring long-term sustainable development. There is currently a lack of effective modeling and analysis tool support for achieving a dynamic balance between short-term and long-term goals.

[0006] Based on this, the present invention proposes a method and system for economic management data analysis in e-commerce, which maps information such as product association relationships and popularity scores into time steps, dynamically reflects market changes, quantifies the changing trend of the competitive pattern, provides a basis for quickly responding to market fluctuations, and provides accurate market response strategies based on the optimized dynamic graph. Summary of the Invention

[0007] The purpose of the present invention is to provide a method and system for economic management data analysis in e-commerce to solve the deficiencies in the background art.

[0008] To achieve the above purpose, the present invention provides the following technical solution: A method for economic management data analysis in e-commerce, the analysis method comprising the following steps:

[0009] The analysis system obtains product data and user behavior data through an e-commerce platform, constructs a dynamic graph based on the product data and user behavior data, calculates the popularity scores of each node based on the PageRank algorithm, and highlights the node with the highest popularity score in the dynamic graph;

[0010] Each node is randomly assigned to multiple communities, and after maximizing the modularity of each community, iteration is performed until the modularity converges, and then the product community graph is displayed through the dynamic graph;

[0011] The dynamic graph is divided into multiple time steps, the DynGEM embedding algorithm is used to generate an embedding representation of the graph for each time step, the graph changes between adjacent time steps are calculated, the changing trend of the competitive pattern in the dynamic graph is identified, and decision-making suggestions are generated based on the optimized dynamic graph.

[0012] In a preferred embodiment, each node is randomly assigned to multiple communities, and after maximizing the modularity of each community, iteration is performed until the modularity converges, and then the product community graph is displayed through the dynamic graph, including the following steps:

[0013] In the initial state, each node is randomly assigned to multiple communities, and the modularity of the current community partition result is calculated;

[0014] In each round of iteration, each node is moved from the current community to the community where its neighbor nodes are located. When the number of iterations is equal to the threshold number of times, all community partition results are output, and the community partition result with the maximum modularity is selected for use.

[0015] In a preferred embodiment, the modularity of the current community partition result is calculated, and the expression is: , where is the modularity, is the total number of all edges in the dynamic graph, is the edge weight between node i and node j, is the number of edges of node i, is the number of edges of node j, is the indicator function. If node i and node j belong to the same community, is 1, otherwise is 0, is the sum of the purchase rates of all nodes in the community, is the maximum number of nodes in the community, , are adjustment coefficients, and , are all greater than 0.

[0016] In a preferred embodiment, the dynamic graph is divided into multiple time steps. The DynGEM embedding algorithm is used to generate an embedded representation for the graph at each time step, calculate the graph changes between adjacent time steps, and identify the changing trend of the competitive pattern in the dynamic graph, including the following steps:

[0017] The dynamic graph is divided into multiple time steps according to the data collection frequency of the e-commerce platform. The graph within each time step represents the nodes and user behavior data within that time step. Initial graph embedding is performed on the dynamic graph of the first time step to generate a low-dimensional vector representation of the nodes;

[0018] DynGEM generates an embedded representation for the graph of each subsequent time step through a recursive adjustment method, and performs iterative optimization according to the addition and deletion of nodes and the change of edge weights. Each node generates a vector of a fixed dimension, representing its characteristics within the current time step. The embedded vector reflects the characteristics of the node and its importance in the network;

[0019] The Euclidean distance or cosine similarity of the node embedding vectors in adjacent time steps is compared to evaluate the node changes. For the edges, the absolute difference or ratio of the weight changes is calculated. If the change amplitude of a certain node embedding is large, it indicates that the competitive position of this node has changed. If the weight of the edge between two nodes has changed significantly, it indicates that the association relationship between these two nodes has changed;

[0020] According to the change trajectory of the nodes in the embedding space, the popularity trend and community location of the nodes are predicted. Through time series analysis and embedded representation, the nodes with continuous growth or decline are identified, and the dynamic graph is adjusted in combination with the changing trend.

[0021] In a preferred embodiment, according to the change trajectory of the nodes in the embedding space, predicting the popularity trend and community location of the nodes includes the following steps:

[0022] In each time step, an embedded vector of the node is generated. The embedded vector reflects the popularity trend and community location of the node. The embedded representations of adjacent time steps are compared, and the change amplitude of each node is calculated. The expression is: , where is the change amplitude of node v, is the feature vector of node v at time t, is the feature vector of node v at time t + 1, represents the cosine distance;

[0023] A time series is constructed for the popularity score of each node to reflect its change in popularity over time. A time series model is used to predict the popularity score:

[0024] , where is the popularity score of node v at time t + 1, denotes a prediction function for predicting future popularity based on past popularity sequences. denotes the historical popularity score of node v from time 1 to t. denotes the historical popularity score of node v at the i-th time step.

[0025] According to the prediction results, classify the changes in node popularity. Within each time step, record the community to which the node belongs and analyze the community change trajectory: , where is the degree of community change of node v at time steps t and t + 1. denotes the cosine similarity function. denotes the set of communities to which node v belongs at time step t. denotes the set of communities to which node v belongs at time step t + 1.

[0026] In a preferred embodiment, calculate the popularity scores of each node based on the PageRank algorithm and highlight the node with the highest popularity score in the dynamic graph, including the following steps:

[0027] Assume there are N nodes in the dynamic graph. Set the initial popularity score of each node to be equal, and iteratively update the popularity score of each node according to the PageRank algorithm. The expression is:

[0028] , where is the popularity score of node i. is the damping factor, and , is the popularity score of node j. denotes the number of edges starting from node j. denotes the set of all nodes pointing to node i;

[0029] When the change in popularity score in any iteration is less than or equal to the change threshold, it is determined that the convergence condition is met. After outputting the popularity scores of all nodes, sort all nodes in descending order of popularity score to generate a popularity table, map the popularity table to the dynamic graph, and use different colors to mark different nodes.

[0030] In a preferred embodiment, the analysis system obtains product data and user behavior data through an e-commerce platform and constructs a dynamic graph based on the product data and user behavior data, including the following steps:

[0031] Obtain the category, sales volume, and price information of products, record the click behavior of users, including the number of co-clicks between products and the purchase path, remove abnormal or incomplete data, integrate the static product data with the dynamic user behavior data, and construct the global association relationship of products;

[0032] Each product corresponds to a node. The node attributes include sales volume, category, and price. Calculate the edge weights based on user behavior data, and obtain the information of the two nodes connected by the edge. The node information includes the number of times clicked by the same user and the text similarity. Normalize the number of times clicked by the same user and the text similarity so that the value ranges of the number of times clicked by the same user and the text similarity are mapped to [0, 1]. Sum the normalized number of times clicked by the same user and the text similarity to obtain the edge weight;

[0033] Construct an initial dynamic graph, where the nodes are products and the edges are association relationships, and the dynamic time steps are defined as different time windows.

[0034] An economic management data analysis system for e-commerce includes a graph construction module, a popularity ranking module, and a trend analysis module;

[0035] Graph construction module: Obtain product data and user behavior data through an e-commerce platform, and construct a dynamic graph based on the product data and user behavior data;

[0036] Popularity ranking module: Calculate the popularity scores of each node based on the PageRank algorithm, and highlight the node with the highest popularity score in the dynamic graph;

[0037] Trend analysis module: Each node is randomly assigned to multiple communities, and iteration is performed after maximizing the modularity of each community until the modularity converges. Then, display the product community graph through the dynamic graph. Divide the dynamic graph into multiple time steps, use the DynGEM embedding algorithm to generate an embedding representation for the graph of each time step, calculate the graph changes between adjacent time steps, identify the change trend of the competitive pattern in the dynamic graph, and generate decision-making suggestions based on the optimized dynamic graph.

[0038] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0039] The present invention constructs a dynamic graph based on product data and user behavior data, calculates the popularity scores of each node, and highlights the node with the highest popularity score in the dynamic graph. Each node is randomly assigned to multiple communities, and iteration is performed after maximizing the modularity of each community until the modularity converges. Then, display the product community graph through the dynamic graph. Divide the dynamic graph into multiple time steps, use the DynGEM embedding algorithm to generate an embedding representation for the graph of each time step, calculate the graph changes between adjacent time steps, identify the change trend of the competitive pattern in the dynamic graph, and generate decision-making suggestions based on the optimized dynamic graph. The analysis system maps information such as product association relationships and popularity scores into time steps, dynamically reflects market changes, quantifies the change trend of the competitive pattern, provides a basis for quickly responding to market fluctuations, and provides accurate market response strategies based on the optimized dynamic graph. Description of the Drawings

[0040] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings.

[0041] Figure 1 It is a flowchart of the method of the present invention. Specific embodiments

[0042] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0043] Embodiment 1: Please refer to Figure 1 As shown, in this embodiment, a method for analyzing economic management data for e-commerce is described. The analysis method includes the following steps:

[0044] The analysis system obtains product data and user behavior data through the e-commerce platform. The product data includes categories, sales volume, etc., and the user behavior data includes click associations between products. For example, after a user purchases product A, they may click on product B. A dynamic graph is constructed based on the product data and user behavior data. The nodes in the dynamic graph are products, and the edges are the association relationships between products (such as the number of co-clicks, similar keywords). Based on the PageRank algorithm, the popularity scores of each node are calculated, and the node with the highest popularity score is highlighted in the dynamic graph (for example, highlighted in a prominent color such as red). Each node is randomly assigned to multiple communities, and after maximizing the modularity of each community, iterations are performed until the modularity converges. Then, the product community graph is displayed through the dynamic graph. The dynamic graph is divided into multiple time steps, and the DynGEM embedding algorithm is used to generate an embedding representation for the graph at each time step. The changes in the graph between adjacent time steps are calculated to identify the changing trend of the competitive pattern in the dynamic graph, and decision-making suggestions are generated based on the optimized dynamic graph.

[0045] This application constructs a dynamic graph based on product data and user behavior data, calculates the heat scores of each node, and highlights the node with the highest heat score in the dynamic graph. Each node is randomly assigned to multiple communities, and after maximizing the modularity of each community, iteration is performed until the modularity converges. Then, the product community graph is displayed through the dynamic graph. The dynamic graph is divided into multiple time steps, and the DynGEM embedding algorithm is used to generate an embedding representation for the graph of each time step. The changes in the graph between adjacent time steps are calculated to identify the changing trend of the competitive pattern in the dynamic graph, and decision-making suggestions are generated based on the optimized dynamic graph. The analysis system maps information such as product association relationships and heat scores into time steps, dynamically reflects market changes, quantifies the changing trend of the competitive pattern, provides a basis for quickly responding to market fluctuations, and provides accurate market response strategies based on the optimized dynamic graph.

[0046] Embodiment 2: The analysis system obtains product data and user behavior data through an e-commerce platform. The product data includes categories, sales volume, etc., and the user behavior data includes click associations between products. For example, after a user purchases product A, they may click on product B. A dynamic graph is constructed based on the product data and user behavior data. The nodes in the dynamic graph are products, and the edges are the association relationships between products (such as the number of co-clicks, similar keywords), including the following steps:

[0047] Obtain static information such as the category, sales volume, and price of the product, record the click behavior of the user, such as the number of co-clicks and purchase paths between product A and product B, remove abnormal or incomplete data (such as product off-shelving or insufficient click records), and integrate the static product data with the dynamic user behavior data to construct the global association relationship of the product;

[0048] Each product corresponds to a node, and the node attributes include sales volume, category, price, etc. For the association relationship between products, calculate the edge weight according to the user behavior data, obtain the information of the two nodes connected by the edge, including the number of times clicked by the same user and the text similarity. Normalize the number of times clicked by the same user and the text similarity so that the value ranges of the number of times clicked by the same user and the text similarity are mapped to [0,1], and sum the normalized number of times clicked by the same user and the text similarity to obtain the edge weight;

[0049] Construct an initial dynamic graph, with nodes as products and edges as association relationships. The dynamic time step is defined as different time windows (such as days, weeks, months).

[0050] Based on the PageRank algorithm, calculate the heat scores of each node, and highlight the node with the highest heat score in the dynamic graph (for example, display it in a prominent color such as red), including the following steps:

[0051] Suppose there are N nodes in the dynamic graph, and the initial popularity scores of each node are set to be equal. The popularity scores of each node are iteratively updated according to the PageRank algorithm, and the expression is:

[0052] , where is the popularity score of node i, is the damping factor, and , is the popularity score of node j, represents the number of edges starting from node j, represents the set of all nodes pointing to node i (i.e., the nodes with edges pointing to node i);

[0053] When the change in popularity score in any iteration is less than or equal to the change threshold, it is determined that the convergence condition is met. After outputting the popularity scores of all nodes, all nodes are sorted from high to low according to the popularity scores to generate a popularity table. The popularity table is mapped into the dynamic graph, and different nodes are marked with different colors. Nodes with high popularity (such as product A) are marked in red, and nodes with low popularity (such as product C) are marked in gray.

[0054] The prominent display of high-popularity products (such as A) can help e-commerce platforms identify which products receive more attention in the market. Product A may be the focus of the current market and requires special attention;

[0055] For high-popularity products (such as product A): Strengthen marketing strategies: For example, increase the sales volume of product A through recommendation systems, discount promotions, etc. Pay attention to market responses: Monitor user feedback on product A and adjust strategies in a timely manner.

[0056] For low-popularity products (such as product C): Consider reducing inventory or optimizing promotion strategies to increase the exposure of product C. Further analyze the reasons for low-popularity products, which may be problems with pricing, description, or keyword optimization.

[0057] Each node is randomly assigned to multiple communities, and after maximizing the modularity of each community, iteration is performed until the modularity converges. Then, the product community graph is displayed through the dynamic graph, including the following steps:

[0058] In the initial state, each node is randomly assigned to multiple communities, which avoids dividing all products into fixed communities at the beginning, increases the flexibility of the algorithm, and provides room for subsequent optimization;

[0059] Calculate the modularity of the current community division result, and the expression is:

[0060] , where is the modularity, is the total number of all edges in the dynamic graph, is the edge weight between node i and node j, is the number of edges of node i, is the number of edges of node j, is an indicator function. If node i and node j belong to the same community, it is 1, otherwise it is 0, is the total purchase rate of all nodes in the community, is the maximum number of nodes in the community (i.e., after obtaining the number of nodes in each community, select the maximum number of nodes), , is a regulation coefficient, and , are both greater than 0;

[0061] By adjusting the community division of nodes, the connections within the community are made as many as possible, while the connections between communities are made as few as possible, and the number of nodes in the community is made as small as possible. The larger the modularity, the better the community structure, the stronger the association within the community, and the weaker the association between communities.

[0062] In each iteration, try to move each node from the current community to the community where its neighbor node is located, and select the operation that can maximize the modularity gain. When the number of iterations is equal to the number threshold, output all community division results, and select the community division result that maximizes the modularity for use.

[0063] 1) Relationship between the total purchase rate of all nodes in the community and modularity:

[0064] Assume that the purchase rate of each node (product) refers to the purchase probability or purchase frequency of the product on the e-commerce platform. In community division, the total purchase rate of all nodes in the community is the cumulative value of the purchase rates of all products, indicating the overall popularity of the products in this community.

[0065] The core of modularity is to measure the difference between the connection degree between nodes within the community and the connection degree between communities. If the purchase rate of the nodes (products) in a community is relatively high, it means that there may be a strong market attraction and consumer purchase intention among the products within this community, which will affect the edge weight within the community and the association strength between nodes.

[0066] Community with a high total purchase rate: When the total purchase rate of all products in a community is relatively high, it means that the products within this community may be more popular and have a higher user attention. Therefore, there will be more edge connections (such as co-purchasing, recommendation, etc.) between the nodes within the community, and the connections within the community are closer. The value of modularity depends on the density of the edge weights within the community. When the total purchase rate of a community is relatively high, the edge weights within the community increase, and the connections within the community become closer, thus increasing the modularity.

[0067] Communities with a low total purchase rate: On the contrary, if the total purchase rate of goods within a community is low, it indicates that the market attractiveness of these goods is weak, the correlation of goods within the community may be low, resulting in weak connections within the community. In this case, there may not be many connections between nodes in the community, and the modularity of the community may be low.

[0068] A community with a high total purchase rate usually means that the nodes within the community are more closely connected to each other, thus increasing the modularity. A community with a low total purchase rate may lead to weak associations between nodes within the community, thus reducing the modularity.

[0069] 2) Relationship between the maximum number of nodes in a community and modularity:

[0070] Communities with a small maximum number of nodes: If the maximum number of nodes is small, it indicates that all community partitions are balanced and there are no obvious overly large communities. The modularity is not penalized, and a higher modularity indicates that the current partition is more reasonable. Communities with a large maximum number of nodes: If there is one or more communities with a very large number of nodes, the modularity may be artificially high because overly large communities tend to contain a large number of connections, resulting in a high edge weight within the community. After introducing a penalty factor, communities with a large maximum number of nodes will be significantly penalized and the modularity will decrease, thus tending to a more reasonable partition.

[0071] An overly large single community: If a certain community contains most of the nodes (for example, the entire graph is partitioned into one community), although the original modularity may be high, it has little practical significance. After introducing a penalty factor, this kind of partition will result in an extremely low modularity, thus promoting a finer-grained partition.

[0072] Divide the dynamic graph into multiple time steps, use the DynGEM embedding algorithm to generate an embedding representation of the graph for each time step, calculate the graph changes between adjacent time steps, identify the change trend of the competitive pattern in the dynamic graph, and generate decision-making suggestions based on the optimized dynamic graph, including the following steps:

[0073] Divide the dynamic graph into multiple time steps according to the data collection frequency of the e-commerce platform (such as daily, weekly or monthly). The graph within each time step represents the commodity and user behavior data during that time period. Perform an initial graph embedding on the dynamic graph of the first time step to generate a low-dimensional vector representation of the nodes. The dimension of the embedding representation is configured by the system and adaptively selected according to the number of nodes or edges;

[0074] DynGEM generates embedded representations for the graph at each subsequent time step through a recursive adjustment method. The embedding algorithm iteratively optimizes according to the addition and deletion of nodes and changes in edge weights, maintaining the continuity of the embedded representations to ensure comparability between the new embeddings and those of the previous time step. Each node generates a vector of a fixed dimension, representing its features at the current time step. The embedding vectors reflect the characteristics of the products and their importance in the network (such as the association strength);

[0075] By comparing the Euclidean distance or cosine similarity of the node embedding vectors in adjacent time steps, the changes in nodes are evaluated. For edges (the association relationships between products), the absolute difference or ratio of the weight changes is calculated. If the change amplitude of a certain node embedding is large, it indicates that the competitive position of this product has changed significantly (such as an increase or decrease in popularity). If the weight of the edge between two products changes significantly, it indicates that the association relationship between these two products (such as the co-purchase trend of users) has changed significantly;

[0076] Based on the changes in nodes and edges across the entire network, a visualization chart of the changes in the global competitive landscape is generated. Newly emerging high-weight nodes or edges are analyzed to identify products that have recently risen rapidly and their competitors, find other nodes (products) that these products are close to in the embedding space, analyze potential competitors or substitutes, detect which communities (product categories) have the most intense competition through changes in the embedded representations, find edges or nodes that frequently change across multiple time steps, determine market hot products or associated product combinations, predict the popularity trends of products and their possible community positions based on the change trajectories of nodes in the embedding space, and identify products that are continuously growing or may decline through time series analysis and embedded representations.

[0077] Adjust the dynamic graph in combination with the change trends (such as redefining edge weights or the importance of nodes).

[0078] Highlight key nodes (such as rapidly rising products) and edges (such as emerging association relationships).

[0079] Visualize the final dynamic graph, intuitively showing the competitive landscape through visual elements such as colors and sizes.

[0080] Generate decision-making suggestions:

[0081] Suggestions for competitive strategies: For the key products of competitors, adjust the pricing and promotion strategies of one's own products. For rapidly emerging competitor products, formulate targeted marketing activities.

[0082] Optimize the product recommendation strategy: According to the dynamic graph embedding, optimize the recommendation algorithm and give priority to recommending products with low competition intensity but great potential.

[0083] Inventory and supply chain optimization: For products with rapidly increasing popularity, increase inventory in advance. For products with decreasing popularity, optimize inventory to reduce waste.

[0084] New product launch strategy: Based on the blank areas in the competitive landscape, explore potential market demands and design and launch new products.

[0085] Regular iteration: Repeat the above process regularly to dynamically track market changes. Generate real-time updated decision-making suggestions according to the new dynamic graph.

[0086] Predict the popularity trend and community location of nodes based on the change trajectories of nodes in the embedding space. Identify nodes with continuous growth or decline through time series analysis and embedding representation, including the following steps:

[0087] In each time step, generate the embedding vector of the node. The embedding vector reflects the popularity trend and community location of the node. Compare the embedding representations of adjacent time steps and calculate the change amplitude of each node. The expression is: , where is the change amplitude of node v, is the feature vector of node v at time t, is the feature vector of node v at time t + 1, represents the cosine distance. The trajectory with a large change amplitude of the node indicates a significant change in its role in the community or network;

[0088] Use a clustering algorithm to automatically divide according to the distribution of the node change amplitudes, which are divided into three categories: growth, stable, and decline. Construct a time series for the popularity of each node (such as the associated edge weight, click-through rate) to reflect its change in popularity over time. Use a time series model to predict the popularity: , where: is the popularity score of node v at time t + 1, represents the historical popularity scores of node v from time 1 to t, and each value represents the historical popularity score of the node at the i-th time step.

[0089] represents the prediction function, which is used to predict future popularity based on the past popularity sequence. It can be a time series model (such as ARIMA, LSTM, etc.) or other regression prediction methods. The prediction method belongs to the prior art. An example is as follows:

[0090] Collect the popularity data of each node v in the past several time periods, such as the past 24 hours or the past 7 days, to form a popularity time series , normalize the heat sequence, for example, compress the data into the [0,1] interval through Min - Max normalization, and construct training samples in a sliding window manner, for example, use a heat sequence of length t to predict the heat value at the next moment.

[0091] Construct an LSTM neural network structure:

[0092] Input layer: Accept the heat sequence of t time steps;

[0093] LSTM layer: Several hidden layer units can be set;

[0094] Output layer: Output a scalar .

[0095] Use historical data to train the model and minimize the error between the predicted heat score and the true heat score (such as the mean squared error MSE).

[0096] Input the latest t heat values into the LSTM model and output the predicted value. If it has been normalized, then denormalize to obtain the true heat value.

[0097] According to the prediction results, classify the changes in node heat: Continuously growing nodes: The heat increases significantly, and the future trend is upward. Continuously declining nodes: The heat continues to decrease and needs attention or optimization. Stable nodes: The heat has no significant change, and the influence is stable;

[0098] Within each time step, record the community to which the node belongs and analyze the community change trajectory: , where is the degree of community change of node v at time steps t and t + 1, represents the cosine similarity function, represents the set of communities to which node v belongs at time step t, represents the set of communities to which node v belongs at time step t + 1;

[0099] Identification of growing nodes: The heat rises significantly, the change amplitude of the embedding features is large, and it tends to be close to other high - heat nodes. Typical features: Many new associated edges, and the heat trend prediction shows an upward trend.

[0100] Identification of declining nodes: The heat drops significantly, and the embedding features tend to move away from the central community or hot products. Typical features: Losing associated edges, and the change amplitude of the embedding vector is relatively low.

[0101] Identification of stable nodes: Both the heat and the change amplitude of the community position are relatively small, and it continues to maintain within the current influence range.

[0102] Optimize node heat display: Use a more prominent color (such as red) for growing nodes to indicate that they are potential competition hotspots. Use gray or light colors for declining nodes to reduce their visual prominence.

[0103] Redefine community division: Adjust according to the predicted community locations, and move frequently transferred nodes from the original community to the new community. Ensure that the community division is more in line with reality by dynamically adjusting the modularity optimization function.

[0104] Update edge weights: Dynamically adjust the edge weights based on the predicted changes in node associations to highlight the associations between emerging products. For disappeared associations, weaken or remove the corresponding edges.

[0105] Visualize the adjusted dynamic graph:

[0106] Node size: Display according to the predicted heat (the higher the heat, the larger the node).

[0107] Edge weights: Adjust the edge width or transparency according to the predicted association strength.

[0108] Time series changes: Display frame by frame through the dynamic graph to show the dynamic changes of nodes and communities.

[0109] Example 3: An economic management data analysis system for e-commerce described in this example includes a graph construction module, a heat sorting module, and a trend analysis module;

[0110] Graph construction module: Obtain product data and user behavior data through an e-commerce platform. The product data includes categories, sales volume, etc., and the user behavior data includes click associations between products. For example, after a user purchases product A, they may click on product B. Based on the product data and user behavior data, construct a dynamic graph. The nodes in the dynamic graph are products, and the edges are the association relationships between products (such as the number of co-clicks, similar keywords). The dynamic graph is sent to the heat sorting module;

[0111] Heat sorting module: Calculate the heat scores of each node based on the PageRank algorithm, and prominently display the node with the highest heat score in the dynamic graph (for example, display it in a prominent color such as red). The once-optimized dynamic graph is sent to the trend analysis module;

[0112] Trend analysis module: Randomly assign each node to multiple communities, and perform iteration after maximizing the modularity of each community until the modularity converges. Then display the product community graph through the dynamic graph. Divide the dynamic graph into multiple time steps, use the DynGEM embedding algorithm to generate an embedding representation for the graph at each time step, calculate the graph changes between adjacent time steps, identify the changing trend of the competitive pattern in the dynamic graph, and generate decision-making suggestions based on the optimized dynamic graph.

[0113] The above formulas are all dimensionless and only take their numerical values for calculation. The formulas are obtained by collecting a large amount of data and performing software simulations to get a formula that is closest to the actual situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0114] In the description of this specification, the descriptions referring to terms such as "one embodiment", "example", "specific example", etc. mean that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in a suitable manner in any one or more embodiments or examples.

[0115] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the present invention to only the specific implementation manners. Obviously, many modifications and variations can be made according to the content of this specification. The present specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can well understand and utilize the present invention. The present invention is only limited by the claims and their full scope and equivalents.

Claims

1. An economic management data analysis method for e-commerce, characterized in that: The analysis method includes the following steps: The analysis system obtains product data and user behavior data through an e-commerce platform, constructs a dynamic graph based on the product data and user behavior data, calculates the popularity scores of each node based on the PageRank algorithm, and highlights the node with the highest popularity score in the dynamic graph; Each node is randomly assigned to multiple communities, and after maximizing the modularity of each community, iteration is performed until the modularity converges, and then the product community graph is displayed through the dynamic graph; The dynamic graph is divided into multiple time steps, the DynGEM embedding algorithm is used to generate an embedding representation of the graph for each time step, the graph changes between adjacent time steps are calculated, the changing trend of the competitive pattern in the dynamic graph is identified, and decision-making suggestions are generated based on the optimized dynamic graph.

2. The economic management data analysis method for e-commerce according to claim 1, wherein: Each node is randomly assigned to multiple communities, and after maximizing the modularity of each community, iteration is performed until the modularity converges, and then the product community graph is displayed through the dynamic graph, including the following steps: In the initial state, each node is randomly assigned to multiple communities, and the modularity of the current community partition result is calculated; In each round of iteration, each node is moved from the current community to the community where its neighbor node is located. When the number of iterations is equal to the number threshold, all community partition results are output, and the community partition result with the maximum modularity is selected for use.

3. The economic management data analysis method for e-commerce according to claim 2, wherein: Calculate the modularity of the current community partition result, and the expression is: , where is modularity, is the total number of all edges in the dynamic graph, is the edge weight between node i and node j, is the number of edges of node i, is the number of edges of node j, is an indicator function. If node i and node j belong to the same community, is 1, otherwise is 0, is the sum of the purchase rates of all nodes in the community, is the maximum number of nodes in the community, , are adjustment coefficients, and , are both greater than 0.

4. The economic management data analysis method for e-commerce according to claim 3, characterized in that: The dynamic graph is divided into multiple time steps, the DynGEM embedding algorithm is used to generate an embedding representation of the graph for each time step, the graph changes between adjacent time steps are calculated, the changing trend of the competitive pattern in the dynamic graph is identified, including the following steps: The dynamic graph is divided into multiple time steps according to the data collection frequency of the e-commerce platform. The graph within each time step represents the nodes and user behavior data within that time step. An initial graph embedding is performed on the dynamic graph of the first time step to generate a low-dimensional vector representation of the nodes; DynGEM generates an embedding representation of the graph for each subsequent time step through a recursive adjustment method, and performs iterative optimization according to the addition and deletion of nodes and the change of edge weights. Each node generates a vector with a fixed dimension, representing its characteristics within the current time step. The embedding vector reflects the characteristics of the node and its importance in the network; Compare the Euclidean distance or cosine similarity of the node embedding vectors in adjacent time steps to evaluate the node changes. For the edges, calculate the absolute difference or ratio of the weight changes. If the change amplitude of a certain node embedding is large, it indicates that the competitive position of this node has changed. If the weight of the edge between two nodes changes, it indicates that the association relationship between these two nodes has changed; According to the change trajectory of the nodes in the embedding space, predict the popularity trend and community location of the nodes. Through time series analysis and embedding representation, identify the nodes with continuous growth or decline, and adjust the dynamic graph in combination with the changing trend.

5. A method for analyzing economic management data for e-commerce according to claim 4, characterized in that: According to the change trajectory of the nodes in the embedding space, predict the popularity trend and community location of the nodes, including the following steps: At each time step, an embedding vector of the node is generated. The embedding vector reflects the heat trend and community location of the node. By comparing the embedding representations of adjacent time steps, the change amplitude of each node is calculated, and the expression is: , where is the change amplitude of node v, is the feature vector of node v at time t, is the feature vector of node v at time t + 1, represents the cosine distance; Construct a time series for the popularity score of each node to reflect its change in popularity over time, and use a time series model to predict the popularity score: , where is the popularity score of node v at time t+1, represents the prediction function, which is used to predict future popularity based on past popularity sequences, represents the historical popularity score of node v from time 1 to t, represents the historical popularity score of node v at the i-th time step; According to the prediction results, classify the heat changes of nodes. In each time step, record the communities to which the nodes belong and analyze the change trajectories of the communities: , where is the degree of community change of node v at time steps t and t + 1, represents the cosine similarity function, represents the set of communities to which node v belongs at time step t, represents the set of communities to which node v belongs at time step t + 1.

6. The economic management data analysis method for e-commerce according to claim 5, characterized in that: Calculate the popularity scores of each node based on the PageRank algorithm, and highlight the node with the highest popularity score in the dynamic graph, including the following steps: Suppose there are N nodes in the dynamic graph, and the initial popularity score of each node is set to be equal. Iteratively update the popularity score of each node according to the PageRank algorithm. The expression is: , where, is the popularity score of node i, is the damping factor, and = 0.85, is the popularity score of node j, represents the number of edges starting from node j, represents the set of all nodes pointing to node i; When the change in popularity score in any iteration is less than or equal to the change threshold, it is determined that the convergence condition is met. After outputting the popularity scores of all nodes, sort all nodes in descending order of popularity score to generate a popularity table, map the popularity table to the dynamic graph, and use different colors to mark different nodes.

7. The economic management data analysis method for e-commerce according to claim 6, characterized in that: The analysis system obtains product data and user behavior data through an e-commerce platform, and constructs a dynamic graph based on the product data and user behavior data, including the following steps: Obtain the category, sales volume, and price information of the products, record the user's click behavior, including the number of co-clicks between products and the purchase path, remove abnormal or incomplete data, integrate the static product data with the dynamic user behavior data, and construct the global association relationship of the products; Each product corresponds to a node, and the node attributes include sales volume, category, and price. Calculate the edge weight according to the user behavior data, obtain the information of the two nodes connected by the edge. The node information includes the number of times clicked by the same user and the text similarity. Normalize the number of times clicked by the same user and the text similarity so that the value ranges of the number of times clicked by the same user and the text similarity are mapped to [0, 1], and sum the normalized number of times clicked by the same user and the text similarity to obtain the edge weight; Construct an initial dynamic graph, where the nodes are products and the edges are association relationships, and the dynamic time step is defined as different time windows.

8. An economic management data analysis system for e-commerce, which is used to implement the analysis method described in any one of claims 1-7, and is characterized in that: It includes a graph construction module, a popularity sorting module, and a trend analysis module; Graph construction module: Obtain product data and user behavior data through an e-commerce platform, and construct a dynamic graph based on the product data and user behavior data; Popularity sorting module: Calculate the popularity scores of each node based on the PageRank algorithm, and highlight the node with the highest popularity score in the dynamic graph; Trend analysis module: Randomly assign each node to multiple communities, and perform iteration after maximizing the modularity of each community until the modularity converges. Then display the product community graph through the dynamic graph. Divide the dynamic graph into multiple time steps, use the DynGEM embedding algorithm to generate an embedding representation for the graph at each time step, calculate the graph changes between adjacent time steps, identify the changing trend of the competitive pattern in the dynamic graph, and generate decision-making suggestions based on the optimized dynamic graph.

Images