Method and system for real-time extraction of regional demand characteristics under foreign trade scenarios

Abstract

The application relates to the technical field of feature modeling analysis, and particularly discloses a regional demand feature real-time extraction method and system in a foreign trade scene, which comprises the following steps: acquiring foreign trade multi-source data of a target region; performing real-time calculation on the data based on a stream computing framework to generate regional dynamic feature parameters; constructing a regional demand graph according to the regional dynamic feature parameters; performing feature propagation and feature aggregation processing on the regional demand graph based on a graph neural network model to predict the demand change trend of each region; and identifying a target market region with a potential growth trend according to the demand change trend. The application realizes modeling and trend prediction of the correlation of market demands in different regions by constructing a multi-source data fusion analysis framework and establishing a regional demand conduction model in combination with a graph neural network, so that the potential growth market can be more accurately identified, and the accuracy of market demand prediction and the decision support capability are significantly improved.

Description

Technical Field

[0001] This invention relates to the field of feature modeling and analysis technology, specifically a method and system for real-time extraction of regional demand features in foreign trade scenarios. Background Technology

[0002] With the profound evolution of the global trade pattern and the widespread application of digital technology, foreign trade enterprises are facing an increasingly complex international market environment, requiring them to have timely and accurate perception and response to the dynamic demand of various regional markets around the world. Traditional foreign trade analysis primarily relies on lagging data sources such as customs statistics, periodic market research reports, and static statistical methods based on historical data. These methods are somewhat limited in addressing the challenges of modern foreign trade. First, significant time delays exist in data acquisition and processing, failing to capture instantaneous market fluctuations and sudden trend changes, causing companies to miss key business opportunities or struggle to mitigate potential risks in a timely manner. Second, the analytical framework overemphasizes structured data dimensions such as transaction amounts and quantities, neglecting the rich market signals contained in unstructured data such as social media sentiment, multilingual consumer reviews, competitor dynamic pricing, and marketing activities, resulting in incomplete information coverage and insufficient depth of insight. Finally, existing analyses often fragment regional markets, failing to effectively depict the inter-regional demand transmission effects caused by economic ties, cultural influences, and logistics networks, making it impossible for companies to predict the rapid rise of hot markets or the cross-regional spread of risks. Therefore, optimizing the existing data acquisition and perception architecture to improve data response speed and accuracy is the technical problem this invention aims to solve. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for real-time extraction of regional demand characteristics in foreign trade scenarios, so as to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] A method for real-time extraction of regional demand characteristics in foreign trade scenarios, the method comprising:

[0006] Acquire multi-source foreign trade data for the target region; wherein, the multi-source data includes trade data, market intelligence data, and competitive intelligence data;

[0007] The multi-source data is cleaned, deduplicated, and standardized in structure. The processed data is then computed in real time based on a streaming computing framework to generate regional dynamic feature parameters.

[0008] A regional demand map is constructed based on the aforementioned regional dynamic characteristic parameters; wherein, regional nodes represent target regions, and the connection relationships between regional nodes represent the demand correlation relationships between regions;

[0009] Based on a graph neural network model, feature propagation and feature aggregation are performed on the regional demand map to predict the demand change trend in each region.

[0010] Identify target market regions with potential growth trends based on the aforementioned demand change trends;

[0011] Obtain the actual business operation data corresponding to the prediction results, and optimize and update the graph neural network model and dynamic feature parameters based on the actual business operation data.

[0012] As a further aspect of the present invention: the step of acquiring multi-source foreign trade data of the target region includes:

[0013] Obtain customs trade statistics and cross-border e-commerce platform transaction data through application programming interfaces;

[0014] Obtain social media sentiment data and search engine trend data through web crawlers;

[0015] Obtain competitor pricing data and marketing activity data through a price monitoring system;

[0016] The acquired data is converted into a unified data format and then stored on the data processing platform.

[0017] As a further aspect of the present invention: the regional dynamic feature parameters include at least:

[0018] Short-term demand intensity index, product sentiment polarity value, and market competition indicators; the short-term demand intensity index is obtained by weighted calculation of sales growth rate, order volume change rate, customer repurchase rate, and search popularity value; the calculation process of the short-term demand intensity index is as follows:

[0019] ;

[0020] in, This is a short-term demand strength index. For sales growth rate, For order volume change rate, To increase customer repurchase rate, This is a normalized search popularity value. to The weighting coefficients are dynamically adjusted; to Based on historical prediction errors, an adaptive weighting method is used for dynamic adjustment, specifically through backpropagation updates of the loss function in the feedback optimization module.

[0021] As a further aspect of the present invention: the step of constructing a regional demand map based on the regional dynamic characteristic parameters includes:

[0022] Map each target region as a graph node;

[0023] Establish graph-edge connections based on historical trade relations, supply chain relations, and market linkages between regions;

[0024] Generate a region adjacency matrix based on the graph nodes and graph edge relationships;

[0025] The dynamic feature parameters of the region are used as the feature vectors of the corresponding graph nodes.

[0026] As a further aspect of the present invention: the step of performing feature propagation and feature aggregation processing on the regional demand map based on a graph neural network model to predict the demand change trend of each region includes:

[0027] Obtain the node feature matrix and region adjacency matrix from the regional demand map; the node feature matrix includes the short-term demand intensity index, product sentiment polarity value and market competition indicators for each region.

[0028] The set of neighboring nodes for each region node is determined based on the adjacency matrix, and the node feature matrix is ​​input into the graph neural network model for feature propagation calculation.

[0029] In each graph neural network layer, the features of the neighboring nodes of the target node are weighted and aggregated to obtain the neighborhood feature representation of the target node. The neighborhood feature representation is then fused with the target node's own features to generate an updated node representation vector.

[0030] The node representation vectors obtained by the multi-layer graph neural network are input into the prediction function to output the predicted demand value of the corresponding region within a preset time window;

[0031] The demand change trend for each region is determined based on the time-varying trend of the demand forecast.

[0032] As a further aspect of the present invention: the step of identifying target market areas with potential growth trends based on the demand change trends includes:

[0033] Time series variation analysis is performed on the demand forecast values ​​corresponding to each regional node to calculate the demand growth rate index and the demand fluctuation stability index.

[0034] The demand growth rate indicator is compared with a preset growth threshold, and regional nodes with a demand growth rate higher than the preset growth threshold are selected as candidate market regions.

[0035] A joint evaluation of the product sentiment polarity value and market competition indicators of the candidate market regions is conducted to calculate the regional market opportunity score.

[0036] Candidate market regions are ranked according to the market opportunity score, and regions with scores higher than a preset score threshold are identified as target market regions with potential growth trends.

[0037] As a further aspect of the present invention: the step of obtaining actual business operation data corresponding to the prediction result, and optimizing and updating the graph neural network model and dynamic feature parameters based on the actual business operation data includes:

[0038] Collect actual transaction data, order volume data, and market access data for the target area within the prediction time window;

[0039] The actual business operation data is compared with the corresponding demand forecast values ​​to calculate the demand forecast error;

[0040] A model loss function is constructed based on the demand prediction error, and the weight parameters of the graph neural network model are updated using the gradient descent algorithm.

[0041] At the same time, the weight coefficients in the short-term demand intensity index calculation formula are adjusted based on the model training results to generate an updated dynamic feature parameter calculation model.

[0042] The present invention also provides a real-time extraction system for regional demand characteristics in foreign trade scenarios, the system comprising:

[0043] The data acquisition module is used to acquire multi-source foreign trade data of the target region; wherein, the multi-source data includes trade data, market intelligence data, and competitive intelligence data;

[0044] The demand feature extraction module is used to perform data cleaning, data deduplication, and data structure standardization on the multi-source data, and to perform real-time calculations on the processed data based on a streaming computing framework to generate regional dynamic feature parameters.

[0045] The graph processing module is used to construct a regional demand map based on the regional dynamic feature parameters; wherein, regional nodes represent target regions, and the connection relationships between regional nodes represent the demand relationships between regions.

[0046] The graph recognition module is used to perform feature propagation and feature aggregation processing on the regional demand map based on the graph neural network model in order to predict the demand change trend of each region.

[0047] The region identification module is used to identify target market regions with potential growth trends based on the aforementioned demand change trends.

[0048] The recursive update module is used to obtain the actual business operation data corresponding to the prediction result, and optimize and update the graph neural network model and dynamic feature parameters based on the actual business operation data.

[0049] As a further aspect of the present invention: the data acquisition module includes:

[0050] The transaction data acquisition unit is used to acquire customs trade statistics and cross-border e-commerce platform transaction data through application programming interfaces.

[0051] The network data acquisition unit is used to acquire social media sentiment data and search engine trend data through web crawlers.

[0052] The activity data acquisition unit is used to acquire competitor price data and competitor marketing activity data through the price monitoring system;

[0053] The format conversion unit is used to convert the acquired data into a unified data format and then store it on the data processing platform.

[0054] As a further aspect of the present invention: the regional dynamic feature parameters include at least:

[0055] Short-term demand intensity index, product sentiment polarity value, and market competition indicators; the short-term demand intensity index is obtained by weighted calculation of sales growth rate, order volume change rate, customer repurchase rate, and search popularity value; the calculation process of the short-term demand intensity index is as follows:

[0056] ;

[0057] in, This is a short-term demand strength index. For sales growth rate, For order volume change rate, To increase customer repurchase rate, This is a normalized search popularity value. to The weighting coefficients are dynamically adjusted; to Based on historical prediction errors, an adaptive weighting method is used for dynamic adjustment, specifically through backpropagation updates of the loss function in the feedback optimization module.

[0058] Compared with existing technologies, the advantages of this invention are as follows: By constructing a multi-source data fusion analysis framework and combining it with a graph neural network to establish a regional demand transmission model, this invention achieves modeling and trend prediction of the correlation between market demand in different regions, thereby enabling more accurate identification of potential growth markets. Simultaneously, by introducing multi-dimensional indicators such as demand growth rate, sentiment polarity changes, and competition intensity to filter the prediction results, the accuracy of identifying high-potential market opportunities is improved. Furthermore, by constructing a feedback optimization architecture based on actual business data, the model parameters and feature weights are dynamically adjusted, enabling the system to continuously learn and optimize during actual operation, thereby significantly improving the accuracy of market demand forecasting and decision support capabilities. Attached Figure Description

[0059] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention.

[0060] Figure 1 A flowchart illustrating a method for real-time extraction of regional demand characteristics in foreign trade scenarios.

[0061] Figure 2 This is a structural diagram of a system for real-time extraction of regional demand characteristics in foreign trade scenarios. Detailed Implementation

[0062] To make the technical problems to be solved, the technical solutions, and the beneficial effects of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present invention and are not intended to limit the present invention.

[0063] Figure 1 The flowchart illustrates a method for real-time extraction of regional demand characteristics in a foreign trade scenario. In this embodiment of the invention, a method for real-time extraction of regional demand characteristics in a foreign trade scenario includes:

[0064] Step S100: Obtain multi-source foreign trade data for the target region; wherein, the multi-source data includes trade data, market intelligence data, and competitive intelligence data;

[0065] Foreign trade-related data for the target region is collected in real time from trade data sources, market intelligence sources, and competitive intelligence sources. Data collection is the foundation of the entire analysis process. To ensure the comprehensiveness and timeliness of the data, various methods can be used for data acquisition; for example, trade statistics reports can be downloaded manually from public websites on a regular basis, or batch data files can be received regularly through partnerships with data providers; in addition, scripts can be written to periodically capture public discussions on social media, and market intelligence can be obtained by subscribing to industry reports. The integration of these data sources aims to provide a multi-dimensional data foundation to comprehensively reflect the foreign trade activities, market sentiment, and competitive landscape of the target region; this process is essentially a data acquisition process, which acquires known data. Data acquisition requires permissions, and in the technical solution of this invention, data acquisition permissions are assumed to be already in place.

[0066] Step S200: Perform data cleaning, data deduplication, and data structure standardization on the multi-source data, and perform real-time calculations on the processed data based on a streaming computing framework to generate regional dynamic feature parameters;

[0067] After acquiring multi-source data, preprocessing is performed, including data cleaning, deduplication, and data structure standardization. These are standard data preprocessing procedures and will not be elaborated upon here. Then, streaming computing frameworks (such as Apache Flink or Spark Streaming) are used to compute three types of dynamic features in parallel: short-term demand intensity index, product sentiment polarity value, and market competition indicators. Data processing and feature calculation are crucial steps in transforming data into insights. For the collected raw data, preprocessing operations such as cleaning, deduplication, and format standardization are required to ensure data quality. Subsequently, different statistical or analytical methods can be used to calculate various features. For example, the short-term demand intensity index can be obtained by simply averaging historical sales and order volume; the product sentiment polarity value can be initially judged by keyword matching and preset rules; and the market competition indicator can be obtained by directly comparing the publicly available price information of different competitors.

[0068] Step S300: Construct a regional demand map based on the regional dynamic characteristic parameters; wherein, regional nodes represent target regions, and the connection relationships between regional nodes represent the demand correlation relationships between regions;

[0069] Different target regions are mapped to nodes in a graph structure, with each node corresponding to a specific market region, such as a country, region, or city. Then, connections between nodes are established based on trade relations, supply chain relationships, and market linkages between regions, forming edges in the graph structure. For example, if there is a strong trade relationship between two regions, a connecting edge can be established between their corresponding nodes; if two regions have upstream and downstream relationships in the supply chain system, a connecting edge can also be established; if two regions have a strong correlation in changes in market demand, a connecting relationship can also be established. After establishing the node and edge relationships, a corresponding regional adjacency matrix can be generated to represent the connections between regions.

[0070] Furthermore, the regional dynamic characteristic parameters obtained from the aforementioned calculations can be used as feature vectors for the corresponding graph nodes. For example, the node feature vectors may include short-term demand intensity index, product sentiment polarity value, and market competition indicators; thereby, a complete regional demand map can be constructed.

[0071] Step S400: Based on the graph neural network model, perform feature propagation and feature aggregation processing on the regional demand map to predict the demand change trend of each region;

[0072] After constructing the regional demand map, a graph neural network model can be used to predict changes in regional demand. First, the node feature matrix and adjacency matrix corresponding to the regional demand map are obtained. The node feature matrix represents the feature information of each regional node; the adjacency matrix represents the connection relationship between regional nodes. Then, the set of neighboring nodes of each regional node is determined based on the adjacency matrix. In each layer of the graph neural network model, the features of the neighboring nodes of the target node are weighted and aggregated to obtain the neighborhood feature representation of the target node. Subsequently, the neighborhood feature representation is fused with the target node's own features to generate an updated node representation vector. Through multi-layer graph neural network calculation, demand information between different regions can be gradually propagated to obtain node representations containing regional demand correlation information. Finally, the node representation vector output by the multi-layer graph neural network is input into a prediction function, such as a regression function or a fully connected network, to output the predicted demand value of each region within a preset time window. Based on the time change trend of the predicted demand value, the demand change trend of each region can be determined.

[0073] Step S500: Identify target market areas with potential growth trends based on the aforementioned demand change trends;

[0074] After obtaining the regional demand forecast results, target market regions with potential growth trends can be further identified. Specifically, time series analysis can be performed on the demand forecast values ​​corresponding to each regional node to calculate the demand growth rate index and the demand fluctuation stability index. The demand growth rate index reflects the speed of demand growth, while the demand fluctuation stability index measures the stability of demand changes. Subsequently, the demand growth rate index is compared with a preset growth threshold, and regional nodes with demand growth rates higher than the preset growth threshold are selected as candidate market regions. For candidate market regions, a joint evaluation can be conducted by combining product sentiment polarity value and market competition index, and a regional market opportunity score can be calculated. For example, a comprehensive score can be given to a region based on demand growth rate, market sentiment trend, and competition level. Finally, the candidate regions are ranked according to the market opportunity score, and regions with scores higher than a preset score threshold are identified as target market regions with potential growth trends.

[0075] Step S600: Obtain the actual business operation data corresponding to the prediction result, and optimize and update the graph neural network model and dynamic feature parameters based on the actual business operation data;

[0076] Step S500 has already completed the requirement assessment, identifying and outputting potential target market areas. Building upon this, step S600 is introduced, which is essentially a reverse recursive optimization architecture used to improve the accuracy of the prediction model. The model is continuously optimized using actual business operation data. Specifically, actual business operation data for the target area within the prediction time window is collected, such as actual transaction data, order volume data, and market access data. Subsequently, the actual business data is compared with the corresponding demand prediction values, and the demand prediction error is calculated. In one implementation, the prediction error can be measured by constructing a model loss function, such as the mean squared error function, and then the weight parameters of the graph neural network model are updated using a gradient descent algorithm to optimize the model's predictive ability. Furthermore, the weight coefficients in the short-term demand intensity index calculation formula can be adjusted based on the model training results to update the dynamic feature parameter calculation model. Through this feedback optimization architecture, a continuously learning demand prediction model can be formed, thereby continuously improving the accuracy of market demand prediction, thus making the entire solution more complete.

[0077] Regarding step S100, the step of obtaining multi-source foreign trade data for the target region includes:

[0078] Obtain customs trade statistics and cross-border e-commerce platform transaction data through application programming interfaces;

[0079] Obtain social media sentiment data and search engine trend data through web crawlers;

[0080] Obtain competitor pricing data and marketing activity data through a price monitoring system;

[0081] The acquired data is converted into a unified data format and then stored on the data processing platform.

[0082] The above content explains the data acquisition process. All data points are actually known data, as detailed below:

[0083] Trade data sources include customs systems and APIs of cross-border e-commerce platforms. Customs systems, within these data sources, refer to official databases or platforms established and maintained by customs authorities of various countries to record information such as import and export of goods, tariff collection, and trade statistics. Their role is to provide macro-level, authoritative data on trade flows, commodity structure, country of origin, and destination, serving as a crucial basis for understanding the fundamentals of regional trade. This can be achieved by connecting to government data open platforms to obtain publicly available customs statistics, or by collaborating with third-party data service providers that possess customs data interfaces to obtain more granular trade records.

[0084] The APIs of cross-border e-commerce platforms in trade data sources refer to application programming interfaces (APIs) that are publicly available from major cross-border e-commerce platforms. Their function is to obtain real-time micro-transaction dynamics such as product sales data, order information, user behavior data, inventory status, and product reviews within the platform, directly reflecting market demand and consumer preferences. This can be achieved by signing API call agreements with mainstream cross-border e-commerce platforms in the target region, obtaining authorization, and then pulling and synchronizing data.

[0085] Market intelligence sources include trend data from social media and search engines; specifically, social media refers to various online platforms that allow users to create, share content, and interact. Their role is to capture unstructured information such as consumer sentiment, product reputation, emerging topics, cultural preferences, and potential demand trends, reflecting market sentiment and user concerns. This can be achieved by obtaining public data streams through open APIs provided by social media platforms and combining them with natural language processing techniques for sentiment analysis and topic identification.

[0086] Search engine trend data in market intelligence refers to data such as keyword search volume, popularity index, and related search terms provided by search engines. Its purpose is to reveal changes in user interest and potential demand for specific products, services, or topics, and it is an important indicator for predicting market hotspots and demand surges. This can be achieved by utilizing official tools or APIs provided by search engines such as Google Trends and Baidu Index to obtain keyword search trend data.

[0087] Competitive intelligence sources include monitoring data on competitor product prices and marketing activities. Specifically, competitor product prices refer to pricing information of similar or substitute products sold by competitors in the market. Their function is to directly quantify the intensity of market competition and price elasticity, providing important reference for one's own product pricing strategy. This can be achieved by regularly capturing competitor product price information from major e-commerce websites and price comparison platforms, and tracking historical prices. Monitoring data on marketing activities in competitive intelligence sources refers to information on competitors' advertising, promotional activities, new product launches, brand promotion, and other marketing activities. Their function is to gain insight into competitors' market strategies and resource investment, and to predict their market trends and potential impact. This can be achieved by monitoring competitors' official websites, social media, press releases, industry exhibition information, and other public channels to collect information on their marketing activities.

[0088] The above content clarifies and details the composition of data sources, ensuring that the collection of foreign trade-related data is no longer limited to single or lagging information channels. The customs system provides macro-level, authoritative trade data, laying the foundation for market analysis; the APIs of cross-border e-commerce platforms supplement micro-level, real-time transaction data, capturing the ever-changing market demands; trend data from social media and search engines reveal potential market interests and sentiments from the perspectives of user behavior and public opinion; and monitoring data on competitor prices and marketing activities provides key information on the competitive dimension. This multi-source, heterogeneous data is collected in real time and serves as a unified and rich input basis for the parallel calculation of three types of dynamic characteristics in subsequent steps: short-term demand intensity index, product sentiment polarity value, and market competition indicators.

[0089] Regarding step S200, the regional dynamic characteristic parameters include at least:

[0090] Short-term demand intensity index, product sentiment polarity value, and market competition indicators; the short-term demand intensity index is obtained by weighted calculation of sales growth rate, order volume change rate, customer repurchase rate, and search popularity value; the calculation process of the short-term demand intensity index is as follows:

[0091] The short-term demand intensity index is derived by weighting sales growth rate, order volume change rate, customer repurchase rate, and search popularity. Specifically, the short-term demand intensity index aims to comprehensively reflect the demand activity and potential growth trend of the target regional market in the short term. Through a weighted combination of multi-dimensional indicators, this index can more comprehensively and dynamically capture changes in market demand. The index can be generated by cleaning and standardizing real-time sales data (such as transaction amount and sales volume), order management system data (such as the number of new orders and canceled orders), customer relationship management system data (such as the number of repeat buyers and purchase frequency), and search trend data from mainstream search engines or e-commerce platforms, and then summing these raw data according to preset weights.

[0092] The product sentiment polarity value is obtained through sentiment analysis of multilingual text in the target region. Specifically, the product sentiment polarity value quantifies the attitude tendency of consumers in the target region towards a specific product or service, i.e., whether it is positive, negative, or neutral. By processing multilingual text, the accurate capture of the true sentiment of consumers in different language markets is ensured, overcoming language barriers. This value can be obtained by processing multilingual text data from the target region collected from channels such as social media, e-commerce reviews, and forums using a pre-trained multilingual sentiment analysis model. After normalization, the sentiment score or classification result output by the model can be used to obtain the product sentiment polarity value.

[0093] The market competition indicators include market share volatility coefficient and price competitiveness difference; specifically, these indicators are used to assess the dynamic changes in the competitive landscape within a target market region. The market share volatility coefficient reflects the stability of market concentration or the competitive structure, while the price competitiveness difference directly measures a product's competitive advantage or disadvantage in terms of price. The market share volatility coefficient can be obtained by monitoring the sales or volume data of major competitors in the target region and calculating their market share change rate or standard deviation over time. The price competitiveness difference can be calculated by comparing the average selling price and promotional prices of one's own product with those of similar competing products in the target region, determining the price difference or relative price index. The price competitiveness difference can also be calculated in real-time based on a dynamic pricing model, incorporating factors such as cost and demand elasticity, to assess the relative price advantage compared to competitors.

[0094] The above content, by clarifying the specific calculation methods for the short-term demand intensity index, product sentiment polarity value, and market competition indicators, ensures the extraction of high-value dynamic characteristics from raw data. The weighted calculation method of the short-term demand intensity index comprehensively considers multiple dimensions of information, including sales, orders, customer loyalty, and market attention, enabling the index to comprehensively and dynamically reflect market demand activity. The product sentiment polarity value, obtained through multilingual text sentiment analysis, solves the challenge of capturing consumer emotions in cross-cultural and cross-language environments, making the assessment of product reputation and consumer attitudes more accurate. The introduction of market competition indicators makes dynamic monitoring of the market competition landscape possible, avoiding lagging judgments on the competitive situation.

[0095] The formula for calculating the short-term demand intensity index is as follows:

[0096] ;

[0097] in, This is a short-term demand strength index. For sales growth rate, For order volume change rate, To increase customer repurchase rate, This is a normalized search popularity value. to The weighting coefficients are dynamically adjusted; to Based on historical prediction errors, an adaptive weighting method is used for dynamic adjustment, specifically through backpropagation updates of the loss function in the feedback optimization module.

[0098] The present invention describes the process of determining the short-term demand intensity index, which quantifies the immediate demand level for foreign trade goods or services in a target region within a specific time period. Its concept lies in providing a comprehensive value that quickly reflects market activity and potential purchasing intentions by integrating multiple key indicators. This index can serve as the basic input for subsequent demand transmission model analysis and forecasting, and its accuracy directly affects the reliability of the overall forecast. The sales growth rate measures the change in sales revenue of a specific commodity or service within the target region over a certain period. It can be calculated by comparing the total sales revenue of the current period (e.g., this week, this month) with the total sales revenue of the previous comparable period (e.g., last week, last month) to obtain a percentage increase or decrease; or it can be compared with the sales revenue of the same period last year to eliminate the impact of seasonal fluctuations. The order volume change rate reflects the dynamic trend of the number of orders for goods or services within the target region. The calculation can be based on comparing the total number of orders received in the current period with the total number of orders in the previous comparable period, expressing the increase or decrease as a percentage; alternatively, it can be reflected by calculating the deviation of the average number of orders over a period of time from the latest order volume. Customer repurchase rate is an important indicator for measuring customer loyalty and product market attractiveness, representing the proportion of customers who make repeat purchases within a certain time period in a target region. It can be calculated by statistically analyzing the ratio of customers who made at least two purchases within a specific time window (e.g., the past 90 days) to the total number of customers making purchases during the same period; alternatively, it can be calculated by tracking the purchase frequency and intervals of a specific customer group. Normalized search popularity values ​​reflect the online attention paid by users in a target region to specific goods, services, or related topics. Due to the significant differences in data volume between different search engines or platforms, normalization is necessary to ensure comparability. Normalization methods can employ min-max normalization (scaling the values ​​to between 0 and 1) or Z-score standardization (converting the values ​​to a distribution with a mean of 0 and a standard deviation of 1) to eliminate differences in units and magnitudes. These weighting coefficients... to This is used to balance the relative importance of sales growth rate, order volume change rate, customer repurchase rate, and normalized search popularity in the calculation of the short-term demand intensity index. Its "dynamic adjustment" characteristic means that these weights are not fixed but can be updated in real-time or periodically based on market conditions, data feedback, or model optimization results. For example, they can be automatically adjusted using machine learning algorithms (such as reinforcement learning or regression models) based on historical prediction accuracy and actual business performance; or they can be adjusted manually or semi-automatically based on expert experience and preset rules.

[0099] The above content defines the short-term demand intensity index as a weighted sum of sales growth rate, order volume change rate, customer repurchase rate, and normalized search popularity value, and introduces dynamically adjusted weight coefficients to construct a comprehensive indicator that can accurately reflect market demand fluctuations in real time. In operation, the system first collects multi-dimensional data related to foreign trade in the target region in real time from trade data sources, market intelligence sources, and competitive intelligence sources. After preprocessing, this data is used to calculate the sales growth rate, order volume change rate, customer repurchase rate, and normalized search popularity value. These indicators characterize market demand from different perspectives, such as transaction behavior, customer loyalty, and market attention. Subsequently, the system uses preset or dynamically adjusted weight coefficients through a feedback architecture to weight and sum these indicators to obtain the short-term demand intensity index. This weighted summation method allows each indicator to contribute according to its actual importance to demand, avoiding the one-sidedness of a single indicator. Furthermore, the dynamic adjustment architecture of the weight coefficients allows the index to be continuously optimized based on market feedback data and actual business results, ensuring the index's real-time adaptability and accuracy.

[0100] Regarding step S300, the step of constructing a regional demand map based on the regional dynamic characteristic parameters includes:

[0101] Map each target region as a graph node;

[0102] Establish graph-edge connections based on historical trade relations, supply chain relations, and market linkages between regions;

[0103] Generate a region adjacency matrix based on the graph nodes and graph edge relationships;

[0104] The dynamic feature parameters of the region are used as the feature vectors of the corresponding graph nodes.

[0105] The above content describes the specific structure of graph data, mapping each region to graph nodes. The node feature vector is composed of the three types of dynamic features of that region. Specifically, mapping each region to graph nodes means abstracting different geographical regions (e.g., countries, provinces, cities) or economic regions (e.g., free trade zones, specific industrial clusters) in foreign trade scenarios into basic units in the graph structure. Each graph node represents an independent market entity. This abstraction method lays the structural foundation for subsequent analysis of complex relationships using graph theory and graph neural networks, enabling the mutual influence between regions to be modeled in a clear and structured manner. As one implementation method, a standard geocoding system or a custom region ID can be used as the unique identifier of the node, storing its corresponding geographic coordinates, administrative division information, and other metadata. In addition, multiple regions can be clustered according to trade characteristics, cultural background, or supply chain relationships, and each cluster can be treated as a node to effectively manage the complexity of the model while maintaining information integrity.

[0106] The node feature vector is composed of the three types of dynamic features of the region, namely, the short-term demand intensity index, product sentiment polarity value, and market competition index, which are calculated in real time in step S2. The product sentiment polarity value is analyzed using a pre-trained multilingual BERT model, and sentiment polarity correction is performed using a domain terminology database to improve the accuracy of cross-lingual sentiment recognition. These features are dynamically updated, reflecting the latest changes and trends in market demand in the target region in real time. Using these multi-dimensional, real-time dynamic features as attribute inputs for each graph node ensures that the demand transmission model can capture multifaceted information about regional market demand, thus providing a rich and accurate data foundation for subsequent analysis and prediction. For example, the normalized values ​​of these three types of features can be directly concatenated into a multi-dimensional vector as the input features of the node. Alternatively, a small neural network (such as a multilayer perceptron) can be used to perform nonlinear transformations and fusions on these three types of features to generate a more expressive node feature vector.

[0107] Learning inter-regional connections and influence weights based on historical data involves automatically identifying and quantifying the mutual influence paths and their strengths between different regions by analyzing past trade data, logistics data, market linkage events, and social media information dissemination. This learning architecture avoids the subjectivity and limitations of relying on human experience to define inter-regional relationships, enabling the model to adaptively capture the complex and constantly evolving demand transmission structures in the real world. For example, adjacency matrices between regions can be constructed using historical trade volumes, cross-border e-commerce transaction records, and supply chain dependence data, and connections and influence weights can be determined through statistical correlation analysis (such as Granger causality) or edge learning architectures in graph neural networks. Another approach is to identify causal relationships by analyzing historical market events (such as how policy adjustments or economic fluctuations in one region affect the market performance of other related regions) and assign corresponding weights based on the strength and duration of the event's impact.

[0108] When the change in demand characteristics of a certain region exceeds a threshold, the system predicts the future demand trend of related regions. Specifically, predicting the future demand trend of related regions when the change in demand characteristics of a certain region exceeds a threshold involves introducing an intelligent triggering architecture. Only when key characteristics of a region, such as its short-term demand intensity index, product sentiment polarity value, or market competition indicators, change significantly (i.e., the magnitude or direction of the change exceeds a preset threshold range), will the system initiate the prediction of the future demand trend of related regions connected to it. This architecture effectively avoids continuous and indiscriminate prediction of all regions, thereby saving computing resources and making predictions more targeted and timely, focusing on key market events that may have demand transmission effects. As a specific implementation method, a fixed threshold can be set, for example, triggering prediction when the short-term demand intensity index changes by more than 10% within 24 hours. Alternatively, a dynamic threshold can be used, for example, based on historical volatility or anomaly detection algorithms to determine whether the current change is an anomaly, thereby triggering prediction.

[0109] The aforementioned approach abstracts regions within a foreign trade scenario as graph nodes and uses real-time updated short-term demand intensity indices, product sentiment polarity values, and market competition indicators as node feature vectors, providing rich and dynamic input to the graph neural network. Based on this, the model can automatically learn and quantify the complex connections and influence weights between regions, including trade, supply chain dependencies, and information dissemination, using extensive historical data. This allows for the construction of a demand transmission model that reflects real market dynamics. When a significant change in the demand characteristics of a particular region is detected, the model can promptly trigger predictions to infer future demand trends in related regions. This method not only provides a clear structure and dynamic input for building the graph neural network but also, by learning the connections and influence weights between regions, enables the model to more accurately capture and predict demand transmission effects, overcoming the shortcomings of traditional models in node feature definition, relationship learning architecture, and prediction triggering conditions.

[0110] Regarding step S400, the step of performing feature propagation and feature aggregation processing on the regional demand map based on the graph neural network model to predict the demand change trend of each region includes:

[0111] Obtain the node feature matrix and region adjacency matrix from the regional demand map; the node feature matrix includes the short-term demand intensity index, product sentiment polarity value and market competition indicators for each region.

[0112] The set of neighboring nodes for each region node is determined based on the adjacency matrix, and the node feature matrix is ​​input into the graph neural network model for feature propagation calculation.

[0113] In each graph neural network layer, the features of the neighboring nodes of the target node are weighted and aggregated to obtain the neighborhood feature representation of the target node. The neighborhood feature representation is then fused with the target node's own features to generate an updated node representation vector.

[0114] The node representation vectors obtained by the multi-layer graph neural network are input into the prediction function to output the predicted demand value of the corresponding region within a preset time window;

[0115] The demand change trend for each region is determined based on the time-varying trend of the demand forecast.

[0116] Regarding step S500, the step of identifying target market areas with potential growth trends based on the demand change trend includes:

[0117] Time series variation analysis is performed on the demand forecast values ​​corresponding to each regional node to calculate the demand growth rate index and the demand fluctuation stability index.

[0118] The demand growth rate indicator is compared with a preset growth threshold, and regional nodes with a demand growth rate higher than the preset growth threshold are selected as candidate market regions.

[0119] A joint evaluation of the product sentiment polarity value and market competition indicators of the candidate market regions is conducted to calculate the regional market opportunity score.

[0120] Candidate market regions are ranked according to the market opportunity score, and regions with scores higher than a preset score threshold are identified as target market regions with potential growth trends.

[0121] In this embodiment, after obtaining the demand forecast results for each regional node, it is necessary to further identify target market regions with potential growth value from the forecast results. Since the forecast results only reflect demand trends, while enterprises also need to consider factors such as the stability of demand growth, changes in consumer sentiment, and the competitive environment when making market decisions, this solution introduces a multi-dimensional evaluation framework based on the demand forecast results to screen out high-potential regions that truly have market value.

[0122] Specifically, the system first performs time-series variation analysis on the demand forecasts corresponding to each regional node. Demand forecasts are typically in time-series form, such as recorded daily, weekly, or monthly. Therefore, by analyzing the changes in forecasts over multiple consecutive time windows, demand growth rate indicators and demand volatility stability indicators can be calculated. The demand growth rate indicator reflects the rate of demand growth, while the demand volatility stability indicator measures the stability of demand changes. For example, the growth rate can be obtained by differencing the forecast value series, or the degree of demand volatility can be measured by calculating the standard deviation or coefficient of variation of the time series.

[0123] After obtaining the demand growth rate indicator, the system compares it with a pre-set growth threshold. This growth threshold can be set based on historical market data, industry average growth levels, or system experience parameters. When the demand growth rate of a target region exceeds the growth threshold, it indicates a significant upward trend in demand within that region, and the system then selects that region as a candidate market area.

[0124] After identifying candidate market regions, the system further conducts a comprehensive evaluation of the market environment in those regions. In this embodiment, the evaluation is primarily based on a joint analysis of product sentiment polarity and market competition indicators. Product sentiment polarity reflects the overall emotional tendency of consumers in the target region towards the product. This indicator is typically obtained through sentiment analysis of text data such as social media comments, user reviews, and forum discussions, and its value usually ranges from negative to positive, representing the trend of changing emotions from negative to positive. Market competition indicators measure factors such as the number of competitors, market share, and intensity of marketing activities within the region.

[0125] In practice, a regional market opportunity score can be obtained by weighting the aforementioned indicators. For example, demand growth rate, sentiment polarity, and market competition indicators can be comprehensively calculated according to preset weights to obtain a comprehensive score reflecting the degree of market opportunity. The higher the score, the greater the market potential of the region.

[0126] The system then ranks the candidate market regions based on their regional market opportunity scores, identifying those with scores above a preset threshold as target market regions with potential for growth. This method allows for the selection of the most promising markets from multiple candidate regions.

[0127] Furthermore, in some preferred embodiments, the high-potential market opportunity may further satisfy at least one of the following conditions:

[0128] First, the demand intensity index ranks in the top 10% among similar regions and shows a continuous upward trend. Specifically, this can be achieved by comparing the demand intensity index of the target region with similar regions and calculating its percentile ranking among all similar regions. When the ranking is in the top 10% and continues to grow over multiple consecutive monitoring periods, the region can be considered to have high market demand activity and sustained growth potential.

[0129] Second, the product's emotional polarity value changes from negative to positive and the increase exceeds a preset threshold. This change indicates that consumers' overall attitude towards the product is gradually shifting from negative to positive, and this change is significant. For example, when the emotional polarity value rises from negative to positive and the change exceeds a threshold, it can be considered that market acceptance is rapidly increasing.

[0130] Third, identify new demand themes with low competitor coverage. New demand themes can be identified through theme modeling or keyword clustering analysis of text data such as user reviews, search queries, or industry reports. When a demand theme is not yet fully covered by existing competing products, the market can be considered a potential "blue ocean opportunity."

[0131] Through the above screening process, demand forecasting results can be further transformed into market opportunity information with practical decision-making value, thereby improving the accuracy and practicality of market identification.

[0132] Regarding step S600, the step of obtaining the actual business operation data corresponding to the prediction result, and optimizing and updating the graph neural network model and dynamic feature parameters based on the actual business operation data includes:

[0133] Collect actual transaction data, order volume data, and market access data for the target area within the prediction time window;

[0134] The actual business operation data is compared with the corresponding demand forecast values ​​to calculate the demand forecast error;

[0135] A model loss function is constructed based on the demand prediction error, and the weight parameters of the graph neural network model are updated using the gradient descent algorithm.

[0136] At the same time, the weight coefficients in the short-term demand intensity index calculation formula are adjusted based on the model training results to generate an updated dynamic feature parameter calculation model.

[0137] In one embodiment of the technical solution of this invention, to improve the accuracy and adaptability of the demand forecasting model, this solution introduces a feedback optimization architecture during system operation. The model is continuously updated using real business operation data. Specifically, after the system generates decision signals based on the demand forecasting results and pushes them to the business system, the system continuously tracks the actual business effects corresponding to these decision signals. The decision signals may include business operation instructions such as inventory adjustment suggestions, marketing campaign launch suggestions, or target market selection suggestions. When the business system executes the corresponding decisions, the resulting business data, such as sales revenue, order volume, customer visits, or conversion rates, is automatically recorded by the system.

[0138] To establish a correlation between forecast results and actual business outcomes, one feasible approach is for the system to generate a unique identifier for each decision signal and associate relevant business data with this identifier during subsequent business data collection. This creates a complete data link from forecast results to actual business outcomes. After collecting actual business operation data, the system compares and analyzes this data with the corresponding demand forecast values ​​to calculate the demand forecast error. The demand forecast error measures the degree of deviation between the forecast and actual results. For example, the forecast error can be represented by calculating the difference between the forecasted demand and the actual demand.

[0139] Subsequently, the system constructs a model loss function based on the prediction error to quantify the accuracy of the model's prediction results. In specific implementations, for continuous prediction data, the mean squared error function or the mean absolute error function can be used as the loss function; for demand trend classification prediction, the cross-entropy loss function or the F1 score can be used as the optimization objective.

[0140] After constructing the loss function, the system updates the parameters of the graph neural network model through optimization algorithms. The most common update scheme is to use the gradient descent algorithm and its variants to iteratively update the weight parameters in the graph neural network model based on the gradient information of the model parameters with respect to the loss function, thereby gradually reducing the value of the loss function and improving the prediction accuracy of the model.

[0141] In addition to updating the graph neural network model parameters, the above methods can also optimize the dynamic feature parameter calculation model. Specifically, in the calculation of the short-term demand intensity index, the weight parameters corresponding to each indicator have a significant impact on the final calculation result. Therefore, during model training, the system can adjust these weight parameters based on the optimization results. For example, when an indicator contributes little to the prediction result or causes a large prediction deviation, its weight can be appropriately reduced; conversely, its weight can be increased. Through these methods, the system can continuously update the dynamic feature parameter calculation model, making it more consistent with changes in the real market environment.

[0142] Figure 2 This is a structural block diagram of a real-time regional demand feature extraction system in a foreign trade scenario. In this embodiment of the invention, a real-time regional demand feature extraction system in a foreign trade scenario, system 10, includes:

[0143] The data acquisition module 11 is used to acquire multi-source foreign trade data of the target region; wherein, the multi-source data includes trade data, market intelligence data and competitive intelligence data;

[0144] The demand feature extraction module 12 is used to perform data cleaning, data deduplication and data structure standardization on the multi-source data, and to perform real-time calculation on the processed data based on the streaming computing framework to generate regional dynamic feature parameters.

[0145] The graphing module 13 is used to construct a regional demand map based on the regional dynamic characteristic parameters; wherein, regional nodes represent target regions, and the connection relationship between regional nodes represents the demand relationship between regions.

[0146] The graph recognition module 14 is used to perform feature propagation and feature aggregation processing on the regional demand map based on the graph neural network model in order to predict the demand change trend of each region.

[0147] The region identification module 15 is used to identify target market regions with potential growth trends based on the demand change trends.

[0148] The recursive update module 16 is used to obtain the actual business operation data corresponding to the prediction result, and to optimize and update the graph neural network model and dynamic feature parameters based on the actual business operation data.

[0149] Furthermore, the data acquisition module 11 includes:

[0150] The transaction data acquisition unit is used to acquire customs trade statistics and cross-border e-commerce platform transaction data through application programming interfaces.

[0151] The network data acquisition unit is used to acquire social media sentiment data and search engine trend data through web crawlers.

[0152] The activity data acquisition unit is used to acquire competitor price data and competitor marketing activity data through the price monitoring system;

[0153] The format conversion unit is used to convert the acquired data into a unified data format and then store it on the data processing platform.

[0154] Specifically, the regional dynamic feature parameters include at least:

[0155] Short-term demand intensity index, product sentiment polarity value, and market competition indicators; the short-term demand intensity index is obtained by weighted calculation of sales growth rate, order volume change rate, customer repurchase rate, and search popularity value; the calculation process of the short-term demand intensity index is as follows:

[0156] ;

[0157] in, This is a short-term demand strength index. For sales growth rate, For order volume change rate, To increase customer repurchase rate, This is a normalized search popularity value. to The weighting coefficients are dynamically adjusted; to Based on historical prediction errors, an adaptive weighting method is used for dynamic adjustment, specifically through backpropagation updates of the loss function in the feedback optimization module.

[0158] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for real-time extraction of regional demand characteristics in a foreign trade scenario, characterized in that, The method includes: Acquire multi-source foreign trade data for the target region; wherein, the multi-source data includes trade data, market intelligence data, and competitive intelligence data; The multi-source data is cleaned, deduplicated, and standardized in structure. The processed data is then computed in real time based on a streaming computing framework to generate regional dynamic feature parameters. A regional demand map is constructed based on the aforementioned regional dynamic characteristic parameters; wherein, regional nodes represent target regions, and the connection relationships between regional nodes represent the demand correlation relationships between regions; Based on a graph neural network model, feature propagation and feature aggregation are performed on the regional demand map to predict the demand change trend in each region. Identify target market regions with potential growth trends based on the aforementioned demand change trends; Obtain the actual business operation data corresponding to the prediction results, and optimize and update the graph neural network model and dynamic feature parameters based on the actual business operation data.

2. The method for real-time extraction of regional demand characteristics in foreign trade scenarios according to claim 1, characterized in that, The steps for obtaining multi-source foreign trade data for the target region include: Obtain customs trade statistics and cross-border e-commerce platform transaction data through application programming interfaces; Obtain social media sentiment data and search engine trend data through web crawlers; Obtain competitor pricing data and marketing activity data through a price monitoring system; The acquired data is converted into a unified data format and then stored on the data processing platform.

3. The method for real-time extraction of regional demand characteristics in foreign trade scenarios according to claim 1, characterized in that, The region dynamic feature parameters include at least: Short-term demand intensity index, product sentiment polarity value, and market competition indicators; the short-term demand intensity index is obtained by weighted calculation of sales growth rate, order volume change rate, customer repurchase rate, and search popularity value; the calculation process of the short-term demand intensity index is as follows: ； in, This is a short-term demand strength index. For sales growth rate, For order volume change rate, To increase customer repurchase rate, This is a normalized search popularity value. to The weighting coefficients are dynamically adjusted; to Based on historical prediction errors, an adaptive weighting method is used for dynamic adjustment, specifically through backpropagation updates of the loss function in the feedback optimization module.

4. The method for real-time extraction of regional demand characteristics in foreign trade scenarios according to claim 1, characterized in that, The step of constructing a regional demand map based on the regional dynamic characteristic parameters includes: Map each target region as a graph node; Establish graph-edge connections based on historical trade relations, supply chain relations, and market linkages between regions; Generate a region adjacency matrix based on the graph nodes and graph edge relationships; The dynamic feature parameters of the region are used as the feature vectors of the corresponding graph nodes.

5. The method for real-time extraction of regional demand characteristics in foreign trade scenarios according to claim 1, characterized in that, The steps of performing feature propagation and feature aggregation on the regional demand map based on the graph neural network model to predict the demand change trend in each region include: Obtain the node feature matrix and region adjacency matrix from the regional demand map; the node feature matrix includes the short-term demand intensity index, product sentiment polarity value and market competition indicators for each region. The set of neighboring nodes for each region node is determined based on the adjacency matrix, and the node feature matrix is ​​input into the graph neural network model for feature propagation calculation. In each graph neural network layer, the features of the neighboring nodes of the target node are weighted and aggregated to obtain the neighborhood feature representation of the target node. The neighborhood feature representation is then fused with the target node's own features to generate an updated node representation vector. The node representation vectors obtained by the multi-layer graph neural network are input into the prediction function to output the predicted demand value of the corresponding region within a preset time window; The demand change trend for each region is determined based on the time-varying trend of the demand forecast.

6. The method for real-time extraction of regional demand characteristics in foreign trade scenarios according to claim 5, characterized in that, The step of identifying target market areas with potential growth trends based on the demand change trends includes: Time series variation analysis is performed on the demand forecast values ​​corresponding to each regional node to calculate the demand growth rate index and the demand fluctuation stability index. The demand growth rate indicator is compared with a preset growth threshold, and regional nodes with a demand growth rate higher than the preset growth threshold are selected as candidate market regions. A joint evaluation of the product sentiment polarity value and market competition indicators of the candidate market regions is conducted to calculate the regional market opportunity score. Candidate market regions are ranked according to the market opportunity score, and regions with scores higher than a preset score threshold are identified as target market regions with potential growth trends.

7. The method for real-time extraction of regional demand characteristics in foreign trade scenarios according to claim 6, characterized in that, The step of obtaining the actual business operation data corresponding to the prediction result, and optimizing and updating the graph neural network model and dynamic feature parameters based on the actual business operation data, includes: Collect actual transaction data, order volume data, and market access data for the target area within the prediction time window; The actual business operation data is compared with the corresponding demand forecast values ​​to calculate the demand forecast error; A model loss function is constructed based on the demand prediction error, and the weight parameters of the graph neural network model are updated using the gradient descent algorithm. At the same time, the weight coefficients in the short-term demand intensity index calculation formula are adjusted based on the model training results to generate an updated dynamic feature parameter calculation model.

8. A real-time extraction system for regional demand characteristics in a foreign trade scenario, characterized in that, The system includes: The data acquisition module is used to acquire multi-source foreign trade data of the target region; wherein, the multi-source data includes trade data, market intelligence data, and competitive intelligence data; The demand feature extraction module is used to perform data cleaning, data deduplication, and data structure standardization on the multi-source data, and to perform real-time calculations on the processed data based on a streaming computing framework to generate regional dynamic feature parameters. The graph processing module is used to construct a regional demand map based on the regional dynamic feature parameters; wherein, regional nodes represent target regions, and the connection relationships between regional nodes represent the demand relationships between regions. The graph recognition module is used to perform feature propagation and feature aggregation processing on the regional demand map based on the graph neural network model in order to predict the demand change trend of each region. The region identification module is used to identify target market regions with potential growth trends based on the aforementioned demand change trends. The recursive update module is used to obtain the actual business operation data corresponding to the prediction result, and optimize and update the graph neural network model and dynamic feature parameters based on the actual business operation data.

9. The real-time extraction system for regional demand characteristics in foreign trade scenarios according to claim 8, characterized in that, The data acquisition module includes: The transaction data acquisition unit is used to acquire customs trade statistics and cross-border e-commerce platform transaction data through application programming interfaces. The network data acquisition unit is used to acquire social media sentiment data and search engine trend data through web crawlers. The activity data acquisition unit is used to acquire competitor price data and competitor marketing activity data through the price monitoring system; The format conversion unit is used to convert the acquired data into a unified data format and then store it on the data processing platform.

10. The real-time extraction system for regional demand characteristics in foreign trade scenarios according to claim 8, characterized in that, The region dynamic feature parameters include at least: Short-term demand intensity index, product sentiment polarity value, and market competition indicators; the short-term demand intensity index is obtained by weighted calculation of sales growth rate, order volume change rate, customer repurchase rate, and search popularity value; the calculation process of the short-term demand intensity index is as follows: ； in, This is a short-term demand strength index. For sales growth rate, For order volume change rate, To increase customer repurchase rate, This is a normalized search popularity value. to The weighting coefficients are dynamically adjusted; to Based on historical prediction errors, an adaptive weighting method is used for dynamic adjustment, specifically through backpropagation updates of the loss function in the feedback optimization module.

Images