Cross-border e-commerce commodity dynamic pricing method and system based on merchant multi-dimensional data

Abstract

The invention discloses a cross-border e-commerce commodity dynamic pricing method and system based on merchant multi-dimensional data, and relates to the technical field of cross-border e-commerce pricing, and the method comprises the steps: S1, multi-dimensional data collection: collecting structured data and unstructured data in a merchant operation process; by collecting and analyzing the unstructured data, the defect that an existing pricing model only depends on the structured data is overcome, the user price attitude and commodity value evaluation are accurately captured, the price elasticity coefficient is corrected by fusing the unstructured data and the structured data, the accuracy of price sensitivity calculation is improved, and the price sensitivity calculation efficiency is improved. Through data feedback and strategy adjustment after price adjustment, a dynamic closed loop is formed, it is ensured that a pricing strategy can adapt to user feedback and market changes in real time, the problems that an existing pricing method is inaccurate in pricing and not timely in strategy optimization are effectively solved, and finally the dual effects of merchant profit space guarantee and commodity market competitiveness improvement are achieved.

Description

Technical Field

[0001] This invention relates to the field of cross-border e-commerce pricing technology, specifically to a method and system for dynamic pricing of cross-border e-commerce products based on multi-dimensional data from merchants. Background Technology

[0002] Cross-border e-commerce refers to international business activities in which trading entities belonging to different customs territories use e-commerce platforms to complete product display, order placement, and electronic payment settlement, and achieve cross-border delivery and transaction loop through cross-border logistics. Its core value lies in breaking the dependence of traditional foreign trade on offline exhibitions and agents, lowering the entry barriers for merchants in the global market, and providing consumers with a wider selection of international products. It has become an important driver of global trade growth, covering various operating models such as B2B, B2C, and C2C, and involving multiple links such as product procurement, cross-border payment, customs clearance, international logistics, and after-sales service.

[0003] Dynamic pricing is one of the core strategies in cross-border e-commerce operations. It refers to merchants adjusting product prices in real time based on changes in the market environment to balance the relationship between "profit acquisition" and "market competitiveness." This involves raising prices appropriately when demand is high to increase revenue and lowering prices reasonably to stimulate sales when demand is low or inventory is piling up. At the same time, it involves adjusting its own strategies in response to changes in the prices of competing products to avoid customer churn or profit compression due to rigid pricing. In cross-border scenarios, the importance of dynamic pricing is further highlighted. The cost structure of cross-border products is complex, and each cost item may change with policy and market fluctuations. Different consumers have different price sensitivities and consumption preferences, so dynamic pricing needs to be adapted to the characteristics of regional markets.

[0004] However, existing cross-border e-commerce pricing technologies still have significant shortcomings in practice. They largely rely on structured transaction data such as historical sales volume and procurement costs to build pricing models, neglecting key information such as user price attitudes and product value evaluations contained in unstructured data such as user reviews and social media feedback. This results in pricing models failing to accurately capture changes in user price sensitivity and struggling to optimize pricing strategies in real time based on genuine user feedback. Consequently, issues arise such as excessively high pricing impacting sales volume and excessively low pricing squeezing profit margins. These technologies are unable to adapt to the diverse user needs and real-time market feedback requirements of cross-border e-commerce. Therefore, developing a dynamic pricing method and system for cross-border e-commerce products based on multi-dimensional merchant data is of great significance. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a dynamic pricing method and system for cross-border e-commerce products based on multi-dimensional merchant data. It can overcome the deficiency of existing pricing models that rely solely on structured data, achieve accurate capture of user price attitudes and product value evaluations, improve the accuracy of price sensitivity calculation, and form a dynamic closed loop through data feedback and strategy adjustments after price adjustments. This ensures that the pricing strategy can adapt to user feedback and market changes in real time, effectively solving the problems of inaccurate pricing and untimely strategy optimization in existing pricing methods, and ultimately achieving the dual effect of ensuring merchant profit margins and enhancing product market competitiveness.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a dynamic pricing method for cross-border e-commerce goods based on multi-dimensional merchant data, the method comprising the following steps: S1. Multi-dimensional data collection: Collect structured and unstructured data during the merchant's operation process. The structured data includes historical purchase prices of goods, discount response rates, procurement costs and inventory data. The unstructured data includes product reviews on the target e-commerce platform and product-related discussions on social media platforms. S2. Unstructured data parsing: The BERT semantic analysis model is used to process the unstructured data, extract users' sentiment towards product prices and their evaluation of product functional value, and quantify the two types of information into sentiment scores respectively. S3. Price sensitivity correction: The sentiment score is fused with the structured data to recalculate the price elasticity coefficient of the product and complete the dynamic correction of price sensitivity. S4. Dynamic pricing adjustment: Based on the revised price elasticity coefficient, combined with the merchant's preset profit target and inventory turnover needs, generate a real-time pricing plan and execute the price adjustment operation. S5. Price adjustment effect feedback loop: Within a preset time after the price adjustment operation is executed, unstructured data is collected and analyzed again to determine the user's acceptance of the new price. If the proportion of negative comments exceeds the preset threshold, the product price is adjusted back to the specified proportion of the original price and a matching promotional strategy is triggered. If the threshold is not exceeded, the new price is maintained.

[0007] Furthermore, in step S1, when collecting unstructured data, web crawling technology is used to selectively crawl product reviews from target e-commerce platforms, focusing only on reviews published within the last 30 days. Simultaneously, the product specifications and purchase location information for each review are recorded. Discussions on social media platforms containing product-specific tags or product name keywords are filtered out. The publication time and interaction volume of these discussions are collected concurrently, and the effective weight of the unstructured data is calculated using the following formula: ,in, For the effective weight of a single unstructured data point, This refers to the amount of interaction on social media, including likes, comments, and shares. This is the data freshness coefficient; the closer the publication time is to the collection time, the larger the coefficient value. This represents the total amount of unstructured data for this product during the same period. Set according to the active time periods of users in the target market. Obtain raw data directly from social media platform interfaces.

[0008] Furthermore, in step S2, when extracting users' sentiment towards product prices, the BERT semantic analysis model is used to specifically process unstructured data. The BERT semantic analysis model is first optimized using a bidirectional attention mechanism, and pre-trained by constructing a product-specific corpus to enhance the recognition accuracy of price-related expressions and functional evaluation terms in cross-border e-commerce scenarios. Simultaneously, entity linking technology is introduced to associate product specifications, accessories, and after-sales service information mentioned in comments and discussions with basic product data, reducing the interference of semantic ambiguity on sentiment recognition. Based on this, users' price sentiment is divided into three categories: positive, neutral, and negative, with different intensity levels corresponding to different score ranges. The sentiment probability distribution output by the BERT semantic analysis model is used to map the price sentiment score for each data point. Then, combined with the effective weights of individual unstructured data points, a comprehensive price sentiment score is calculated using the following formula: ,in, The overall score represents the price sentiment. To determine the effective number of unstructured data entries, For the first The price sentiment score for each data point ranges from -5 to +5. For the first Effective weights of each data point This represents the total amount of data remaining after filtering out invalid data within the data collection period. It is obtained by mapping the sentiment probability distribution output by the BERT model.

[0009] Furthermore, in step S2, when extracting user evaluations of the functional value of the product, the BERT semantic analysis model is used to process the unstructured data in different dimensions. The BERT semantic analysis model employs a multi-label classification optimization strategy in the functional value evaluation extraction stage, constructing sub-classifiers for the three core dimensions of logistics timeliness, packaging integrity, and usability. Each sub-classifier is fine-tuned using labeled corpora for the corresponding dimension to accurately identify the sentiment tendencies in the mixed-dimensional evaluation content. Simultaneously, semantic role labeling technology is introduced to extract degree adverbs and comparative words from the evaluation content and convert them into score adjustment coefficients. Based on the processing results of the BERT semantic analysis model, evaluation content for the three core dimensions is extracted, and positive and negative statements in each dimension are marked. When quantifying the functional value evaluation score, the final score is determined based on the proportion of positive statements in each dimension combined with the score adjustment coefficients. The higher the proportion of positive statements, the higher the score. The score weights for each dimension are pre-set according to the product category.

[0010] Furthermore, in step S3, when integrating sentiment scores and structured data, the structured data is first standardized to eliminate the dimensional differences between different data types. Then, the sentiment scores and the standardized structured data are assigned preset weights, and a comprehensive data is obtained through weighted calculation. Based on the comprehensive data, the price elasticity coefficient is recalculated. The corrected formula for calculating the price elasticity coefficient is as follows: ,in, This is the corrected price elasticity coefficient. The initial price elasticity coefficient, calculated based on structured data, relies on traditional structured data from merchant operations, excluding unstructured data such as user reviews and social media feedback. It is derived through analysis of this structured data, specifically by establishing a correlation between sales volume and price based on historical purchase price changes and user feedback on price changes reflected in discount response rates. This initial price elasticity coefficient serves as the foundation for subsequent adjustments using unstructured data. The overall score represents the price sentiment. The overall score for functional value evaluation Price sentiment influence coefficient The functional value impact coefficient. and Based on the sales changes of the product after price adjustments over the past 6 months, a linear regression model was used to determine the data, which was recalibrated quarterly. The fusion process employed multi-level processing: first, the sentiment score and structured data were aligned in terms of time and product dimensions; then, principal component analysis was used to extract key features of the two types of data to construct a fusion vector; and weights were dynamically allocated according to product category characteristics. The weights were optimized using gradient descent based on the pricing effect feedback over the past 3 months. In addition, the isolated forest algorithm was used to remove outliers in the sentiment score and structured data to ensure the overall data quality.

[0011] Furthermore, in step S4, when generating the real-time pricing scheme, the price adjustment range is first determined based on the corrected price elasticity coefficient. The upper and lower limits of the adjustment range are set according to the market price fluctuation range of the product category. Then, the minimum sellable price is calculated in combination with the procurement cost, and the target selling price is calculated in combination with the profit target. The target selling price calculation model comprehensively considers the corrected price elasticity coefficient. Merchant's preset profit margin Unit procurement cost Inventory turnover days The formula is: ,in, Target selling price, To determine the optimal inventory turnover days for the product, The maximum acceptable inventory turnover days for the product. hour, This means that the target selling price will not be lowered when inventory turnover is normal. As the inventory turnover days increase, this coefficient gradually decreases, and the target selling price decreases accordingly, stimulating sales and accelerating inventory turnover. If the target selling price is within the price adjustment range, the target selling price is used as the real-time price. If the target selling price exceeds the range, the boundary value of the range is used as the real-time price. If the target selling price is greater than the upper limit of the adjustment range, the upper limit of the adjustment range is used as the real-time price. If the target selling price is less than the minimum sellable price, the minimum sellable price is used as the real-time price, and the inventory warning mechanism is triggered at the same time.

[0012] Furthermore, in step S5, the preset duration is set to 48 hours, the threshold is adjusted according to the product sales cycle and market competition intensity, when the product price is adjusted back, the specified percentage of the original price is determined according to the current inventory turnover days of the product, and the triggered supporting promotional strategies include two types: free accessories with purchase and discounts on a specified amount. The execution duration and intensity of the promotional strategies are set according to the price adjustment range.

[0013] A dynamic pricing system for cross-border e-commerce products based on multi-dimensional merchant data, applicable to the aforementioned dynamic pricing method for cross-border e-commerce products based on multi-dimensional merchant data, the system includes: a data acquisition module, a data parsing module, a price sensitivity correction module, a pricing adjustment module, and a feedback closed-loop module; The data acquisition module is used to collect structured data and unstructured data. The structured data includes historical purchase prices of goods, discount response rates, procurement costs and inventory data. The unstructured data includes product reviews on the target e-commerce platform and product-related discussions on social media platforms. The data parsing module is equipped with the BERT semantic analysis model, which is used to parse unstructured data and extract price sentiment and functional value evaluation, and quantify them into sentiment scores. The price sensitivity correction module is used to integrate sentiment scores and structured data to recalculate the price elasticity coefficient; The pricing adjustment module is used to generate real-time pricing schemes and execute price adjustments based on price elasticity coefficients, profit targets, and inventory requirements. The feedback closed-loop module is used to collect and parse unstructured data after price adjustment, determine pricing acceptance, and perform operations such as price maintenance, pullback, or triggering promotional strategies.

[0014] Furthermore, the data acquisition module includes a structured data acquisition unit and an unstructured data acquisition unit. The structured data acquisition unit connects to the merchant's ERP system through an API interface to acquire structured data in real time and store it in a database. The database adopts a distributed storage architecture. The unstructured data acquisition unit has a built-in web crawler program with data filtering rules that can automatically filter duplicate unstructured data and discussion content unrelated to the product. The filtering rules are updated regularly according to changes in the data format of the target platform.

[0015] Compared with existing technologies, this method and system for dynamic pricing of cross-border e-commerce products based on multi-dimensional merchant data has the following advantages: This invention overcomes the shortcomings of existing pricing models that rely solely on structured data by collecting and analyzing unstructured data. It achieves accurate capture of user price attitudes and product value evaluations. By integrating unstructured and structured data to correct the price elasticity coefficient, it improves the accuracy of price sensitivity calculation. Through data feedback and strategy adjustments after price adjustments, a dynamic closed loop is formed, ensuring that the pricing strategy can adapt to user feedback and market changes in real time. This effectively solves the problems of inaccurate pricing and untimely strategy optimization in existing pricing methods. Ultimately, it achieves the dual effect of ensuring merchants' profit margins and enhancing product market competitiveness, while optimizing inventory turnover efficiency and adapting to the diverse operational needs of cross-border e-commerce.

[0016] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0018] Figure 1 A flowchart of a dynamic pricing method for cross-border e-commerce products based on multi-dimensional merchant data; Figure 2 A flowchart illustrating a dynamic pricing method for cross-border e-commerce products based on multi-dimensional merchant data; Figure 3 This is a schematic diagram of the structure of a dynamic pricing system for cross-border e-commerce products based on multi-dimensional data from merchants. Detailed Implementation

[0019] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0020] This invention provides a method and system for dynamic pricing of cross-border e-commerce goods based on multi-dimensional merchant data. It fully elaborates on the technical solution for achieving accurate and real-time pricing by combining multi-dimensional data. (See also...) Figure 1 , Figure 2 and Figure 3 The core content is as follows: The core steps of the dynamic pricing method are as follows: S1. Multi-dimensional Data Collection: Collect two types of data from merchant operations. Structured data includes historical purchase prices, discount response rates, procurement costs, and inventory data; unstructured data includes product reviews from the target e-commerce platform within the past 30 days, as well as discussions on social media platforms with product-specific tags or name keywords, while calculating the effective weight of each piece of unstructured data.

[0021] S2. Unstructured Data Parsing: Unstructured data is processed using the BERT semantic analysis model. User sentiment towards product prices is extracted, categorized into positive, neutral, and negative, and quantified into scores. User evaluations of product functional value are also extracted, focusing on three dimensions: logistics timeliness, packaging integrity, and usability. Scores are quantified based on the proportion of positive statements in each dimension and preset weights, ultimately yielding two types of sentiment scores.

[0022] S3. Price Sensitivity Correction: First, the structured data is standardized to eliminate differences in dimensions; then, the sentiment score and the standardized structured data are assigned preset weights, and a comprehensive data is obtained through weighted calculation; based on the comprehensive data, the price elasticity coefficient of the commodity is recalculated to complete the dynamic correction of price sensitivity.

[0023] S4. Dynamic Pricing Adjustment: Based on the revised price elasticity coefficient, a real-time pricing plan is generated by combining the merchant's preset profit target and inventory turnover needs. First, the price adjustment range is determined based on the coefficient. Then, the minimum sellable price is calculated based on the procurement cost, and the target selling price is calculated based on the profit target. If the target selling price is within the range, it is adopted directly; if it exceeds the range, the range boundary value is taken, and then the price adjustment operation is executed.

[0024] S5. Price Adjustment Effect Feedback Loop: Within 48 hours of the price adjustment, unstructured data is collected and analyzed again to determine user acceptance of the new pricing. If the proportion of negative comments stating "price is too high" exceeds a threshold, the price will be adjusted back to a specified percentage of the original price, triggering promotional strategies such as free accessories or discounts for purchases exceeding a certain amount; otherwise, the new pricing will be maintained.

[0025] The system comprises five core modules adapted to the aforementioned pricing methods. The data acquisition module is divided into structured and unstructured data acquisition units; the data parsing module, equipped with a BERT model, parses unstructured data and quantifies sentiment scores; the price sensitivity correction module integrates sentiment scores and structured data to recalculate the price elasticity coefficient; the pricing adjustment module generates a plan based on the coefficient, profit target, and inventory requirements and executes price adjustments; the feedback loop module assesses pricing acceptance after price adjustments and executes actions such as price maintenance, price correction, or triggering promotions.

[0026] Example 1 This embodiment applies to the pricing scenario of 3C products on a cross-border e-commerce platform, taking the platform's best-selling wireless Bluetooth headphones as an example. The platform's business covers multiple countries and regions. 3C products are generally characterized by rapid product updates and intense market competition, and users pay extremely high attention to the reasonableness of product prices and satisfaction with functional experiences. Traditional pricing methods rely solely on structured data such as historical sales volume and procurement costs, failing to capture users' true attitudes towards price and their feedback on functionality. This often leads to problems such as excessively high pricing causing declining sales, and excessively low pricing compressing profit margins. See also... Figure 1 , Figure 2 and Figure 3 Based on the dynamic pricing method and system of this invention, unstructured data such as user reviews and social media discussions can be integrated with traditional structured data to optimize pricing strategies in real time, adapt to the needs and competitive landscape of different regional markets, and achieve the dual goals of ensuring merchant profits and enhancing the market competitiveness of products.

[0027] The system's data acquisition module is activated, collecting relevant data for the wireless Bluetooth headset in two categories. The structured data acquisition unit connects to the merchant's ERP system via an API interface to obtain real-time data on the headset's historical purchase price, discount response rate, procurement cost, and inventory. All data is directly stored in a database using a distributed storage architecture, ensuring efficient data access and security. The unstructured data acquisition unit activates its built-in web crawler to selectively scrape product reviews for the headset from target e-commerce platforms according to preset rules. Only reviews posted within the last 30 days are retained, and the product specifications and purchase location information for each review are recorded. On social media platforms, discussions with hashtags specific to the headset or containing product name keywords are filtered, and the posting time and interaction volume of each discussion are collected simultaneously. After data collection, the effective weight of each piece of unstructured data is calculated using a formula: in, The effective weights representing a single piece of unstructured data. This represents the amount of interaction in social media discussions. This represents the data freshness coefficient; the closer the publication time is to the collection time, the higher the coefficient value, and it is set according to the active periods of users in the target market. This represents the total amount of unstructured data for that headphone during the same period.

[0028] The data parsing module calls the BERT semantic analysis model to perform deep processing on the collected unstructured data. The BERT semantic analysis model is optimized using a bidirectional attention mechanism and pre-trained using a corpus specifically for 3C products. This enhances the recognition accuracy of functional evaluation terms for 3C products such as "battery life," "noise cancellation effect," and "connection stability," as well as price-related expressions such as "cost-effectiveness," "high price," and "discount strength." Simultaneously, entity linking technology is introduced to associate product specifications such as "in-ear version" and "wireless charging case," as well as service information such as "after-sales warranty," mentioned in reviews with the basic data of the headphones. This avoids semantic ambiguity caused by unclear specifications in phrases like "this headphone charges slowly," further improving the accuracy of price sentiment scores for individual data points.

[0029] The first step is to extract users' sentiment towards the price of these headphones, categorizing sentiment into three types: positive, neutral, and negative. Positive sentiment corresponds to statements such as "reasonable price" and "high cost-performance ratio," negative sentiment corresponds to statements such as "excessive price" and "unreasonable pricing," and neutral sentiment corresponds to statements with no clear price stance. In the sentiment quantification stage, positive sentiment scores are set to positive values, negative sentiment scores to negative values, and neutral sentiment scores to 0, with different intensity of sentiment corresponding to different score ranges. Subsequently, a comprehensive price sentiment score is calculated using a formula: in, The overall score represents the price sentiment tendency. This represents the number of valid unstructured data entries, which is the total amount of data remaining after filtering out invalid data within the collection period. Representing the The price sentiment score for each data point is obtained by mapping the sentiment probability distribution output by the BERT model. Representing the Effective weights for each data point.

[0030] In the functional value evaluation extraction stage, the BERT semantic analysis model constructs sub-classifiers for three core dimensions of wireless Bluetooth headphones: logistics timeliness, packaging integrity, and usage performance. Each sub-classifier is fine-tuned using labeled corpora for the corresponding dimension. For example, the logistics timeliness dimension includes labeled data such as "delivery in 3 days" and "logistics delayed by one week," while the usage performance dimension includes labeled data such as "good noise cancellation" and "less than 2 hours of battery life." When processing mixed-dimensional evaluation content such as "fast logistics but packaging was crushed and deformed," the logistics timeliness sub-classifier can accurately identify positive evaluations, while the packaging integrity sub-classifier can identify negative evaluations, thus achieving multi-dimensional sentiment extraction.

[0031] The second step involves extracting user feedback on the headphones' functional value, focusing on three core dimensions: logistics efficiency, packaging integrity, and usability. Feedback content for each dimension is extracted, and positive and negative comments are marked for each item. In the functional value scoring stage, scores are determined based on the percentage of positive comments in each dimension; a higher percentage of positive comments corresponds to a higher score. The weighting of each dimension is pre-set according to the characteristics of 3C products, ultimately yielding a comprehensive functional value evaluation score. .

[0032] The price sensitivity correction module first standardizes the previously collected structured data to eliminate dimensional differences between different data types, ensuring data comparability and fusion. Then, a multi-level feature fusion process is initiated: The first layer is data-level fusion, aligning the standardized structured data and sentiment scores by time period and product specifications to ensure both types of data correspond to the same product specifications within the same time period, avoiding data misalignment. The second layer is feature enhancement fusion, extracting key features such as discount response rate fluctuation trends and inventory turnover acceleration / deceleration trends from the structured data through principal component analysis, and extracting core indicators such as daily price sentiment score fluctuations and average scores of functional dimensions from the sentiment scores, combining the two types of features to construct a fused feature vector. The third layer is dynamic weight allocation, initially setting weights based on the characteristics of 3C products, then combining the feedback data on the effects of pricing adjustments for this headphone and similar headphones over the past three months, optimizing the weight values ​​using a gradient descent algorithm, and finally determining the weights for this fusion. Simultaneously, the isolated forest algorithm is used to identify extreme outliers in the sentiment scores and abnormal fluctuation values ​​in the structured data, removing them to prevent abnormal data from affecting the overall data quality.

[0033] Next, the price sentiment index will be comprehensively scored. Functional value evaluation comprehensive score The standardized structured data and the standardized structured data are each assigned a preset weight, and a comprehensive data is obtained through weighted calculation. Based on this comprehensive data, the price elasticity coefficient of the headphones is recalculated according to the formula to achieve dynamic adjustment of price sensitivity. in, This represents the adjusted price elasticity coefficient. This represents the original price elasticity coefficient calculated based on structured data. Represents the price sentiment influence coefficient. The coefficient representing the functional value impact. and The coefficients were determined by fitting a linear regression model with data on sales changes following price adjustments over the past six months, and the model was recalibrated quarterly to ensure accuracy.

[0034] The pricing adjustment module uses the corrected price elasticity coefficient. Based on this as the core basis, and combined with the merchant's preset profit targets and inventory turnover needs, a real-time pricing scheme for the headphones is generated. Firstly, based on... The price adjustment range is determined, with the upper and lower limits set based on the market price fluctuation range of 3C wireless Bluetooth headphones. Then, the minimum sellable price is calculated based on the headphones' procurement cost to ensure the price covers the basic cost. Simultaneously, a target selling price is calculated based on the merchant's pre-set profit target. The target selling price calculation model comprehensively considers the adjusted price elasticity coefficient. Merchant's preset profit margin 、 Unit procurement cost 、 Inventory turnover days , the formula is: , where is the target selling price, is the optimal inventory turnover days of the product, is the maximum acceptable inventory turnover days of the product; if the target selling price is within the price adjustment range, the target selling price is directly used as the real-time pricing; if the target selling price exceeds the price adjustment range, the boundary value of the range is selected as the real-time pricing. If the target selling price is less than the lowest sellable price, the lowest sellable price is taken as the real-time pricing, and the inventory warning mechanism is triggered at the same time. After the plan is determined, the system automatically executes the price adjustment operation and synchronously updates the selling price of this headset on all target sales platforms.

[0035] Within the preset duration after the price adjustment operation is executed, the feedback closed-loop module starts the unstructured data collection and analysis process again, collects the latest reviews and social media discussion content of this headset on each platform, analyzes the data through the BERT semantic analysis model, and judges the acceptance degree of users for the new pricing. If the proportion of negative comments such as "too high price" exceeds the preset threshold, the product price is called back to a specified proportion of the original price. The specified proportion of the original price is determined according to the current inventory turnover days of this headset, and at the same time, a supporting promotion strategy is triggered: if the price adjustment range is between 5% and 10%, implement a promotion strategy such as "buying a headset and getting a portable storage bag + 6-month extended warranty service", and the promotion duration is set to 7 days; if the price adjustment range exceeds 10%, implement a promotion strategy such as "reduce 50 yuan for every 500 yuan spent + get a wireless charging base", and the promotion duration is set to 10 days, and the promotion intensity increases with the increase of the price adjustment range. For example, when the price adjustment range is 15%, the full reduction amount is increased to 80 yuan, and at the same time, an additional headset cleaning set is given; if the proportion of negative comments such as "too high price" does not exceed the preset threshold, the new pricing is maintained, and thus a complete dynamic pricing closed-loop process is completed.

[0036] To sum up, this embodiment realizes the deep integration of structured data and unstructured data by fully implementing the whole process of multi-dimensional data collection, unstructured data analysis, price sensitivity correction, dynamic pricing adjustment and feedback closed-loop of price adjustment effect. The corrected price elasticity coefficient is more in line with the real price sensitivity of users, and the generated pricing plan accurately matches the market characteristics and user needs of 3C products. Compared with the traditional pricing method, this embodiment effectively solves the problems of inaccurate pricing and untimely strategy optimization, significantly improves the market competitiveness of products while ensuring the profit space of merchants, and also optimizes the inventory turnover efficiency, providing a technical implementation plan that can be replicated and promoted for the dynamic pricing of 3C products on cross-border e-commerce platforms.

[0037] Embodiment 2 This embodiment applies to the pricing scenario of home furnishing products on a cross-border e-commerce platform, taking the platform's core product, a solid wood dining table, as an example. This platform focuses on the home furnishing vertical, with business covering multiple regions including Europe and Southeast Asia. Home furnishing products are characterized by high unit prices, long procurement cycles, and user decisions relying heavily on functional experience feedback. Traditional pricing methods only consider historical transaction prices, procurement costs, and inventory quantities, failing to capture timely user evaluations of product materials, ease of installation, and logistics experience. They also struggle to perceive price differences among users in different regions, often resulting in excessively high prices in some areas leading to inventory backlog, while excessively low prices in others negatively impact profits. See also Figure 1 , Figure 2 and Figure 3 Based on the dynamic pricing method and system of this invention, unstructured data such as user comments and social media discussions from multiple platforms can be integrated with structured data to adjust pricing strategies in different regions in a targeted manner, thereby achieving a balance between efficient inventory turnover and stable profit growth.

[0038] The system's data acquisition module is activated, collecting relevant data for the solid wood dining table in two categories. The structured data acquisition unit connects to the merchant's ERP system via an API interface to obtain real-time data on the dining table's historical purchase price, discount response rate, procurement cost, and inventory. This data is directly stored in a distributed database to ensure efficient access. The unstructured data acquisition unit initiates a web crawler to selectively collect product reviews of the solid wood dining table from target e-commerce platforms, retaining only reviews posted within the last 30 days, and recording the product specifications and purchase location information for each review. On social media platforms, discussions containing tags specific to the solid wood dining table or keywords related to the product name are filtered, and the posting time and interaction volume of each discussion are collected simultaneously. After collection, the effective weight of each piece of unstructured data is calculated using a formula: in, The effective weights representing a single piece of unstructured data. This represents the amount of interaction in social media discussions. This represents the data freshness coefficient. The closer the publication time is to the collection time, the higher the coefficient value, and it is set according to the active periods of users in the target region. This represents the total number of unstructured data points for the solid wood dining table during the same period.

[0039] The data parsing module calls the BERT semantic analysis model to process the collected unstructured data. The BERT semantic analysis model is optimized using a bidirectional attention mechanism and pre-trained by building a corpus specifically for home furnishing products. This enhances the recognition accuracy of functional evaluation terms for home furnishing products such as "solid wood material," "load-bearing capacity," and "installation difficulty," as well as price-related expressions such as "reasonable pricing," "low cost-performance ratio," and "insufficient discount." At the same time, entity linking technology is introduced to associate product specification information such as "1.8-meter size" and "walnut wood material" mentioned in the reviews, as well as service information such as "home delivery," with the basic data of the solid wood dining table. This avoids semantic ambiguity caused by the lack of specific dimensions in statements such as "this dining table is of good quality but expensive," and further improves the accuracy of the price sentiment score for individual data points.

[0040] The first step is to extract users' sentiment towards the price of this solid wood dining table, categorizing sentiment into three types: positive, neutral, and negative. Positive sentiment corresponds to statements such as "the price is reasonable" and "it's good value for money," negative sentiment corresponds to statements such as "the price is too high" and "it's not cost-effective," and neutral sentiment corresponds to statements with no clear price stance. When quantifying sentiment, positive sentiment scores are set to positive values, negative sentiment scores to negative values, and neutral sentiment scores to 0. Different intensities of sentiment correspond to different score ranges. Then, a comprehensive price sentiment score is calculated using a formula: in, The overall score represents the price sentiment tendency. This represents the number of valid unstructured data entries, which is the total amount of data remaining after filtering out invalid data within the collection period. Representing the The price sentiment score for each data point is obtained by mapping the sentiment probability distribution output by the BERT model. Representing the Effective weights for each data point.

[0041] In the functional value evaluation extraction stage, the BERT semantic analysis model constructs sub-classifiers for three core dimensions of solid wood dining tables: logistics timeliness, packaging integrity, and performance. Each sub-classifier is fine-tuned using labeled corpora for its corresponding dimension. When dealing with mixed-dimensional evaluations such as "the packaging is very secure, but the logistics is too slow," the packaging integrity sub-classifier can accurately identify positive evaluations, while the logistics timeliness sub-classifier can identify negative evaluations, achieving multi-dimensional sentiment extraction. The second step extracts user evaluations of the functional value of the solid wood dining table, focusing on the three core dimensions of logistics timeliness, packaging integrity, and performance. Evaluation content for each dimension is extracted, and positive and negative statements are marked for each item. When quantifying the functional value score, the score is determined based on the proportion of positive statements in each dimension's evaluation; the higher the proportion of positive statements, the higher the score. The weights of each dimension's score are pre-set according to the characteristics of home furnishing products, ultimately yielding a comprehensive functional value evaluation score. .

[0042] The price sensitivity correction module first standardizes the collected structured data to eliminate dimensional differences between different data types, ensuring data can be fused and calculated. Then, a multi-level feature fusion process is initiated: the first layer is data-level fusion, aligning the standardized structured data with sentiment scores by region and time period to ensure both types of data correspond to product data within the same region and time period, avoiding data misalignment; the second layer is feature enhancement fusion, extracting key features such as regional discount response rate differences and inventory turnover trends from the structured data through principal component analysis, and extracting core indicators such as regional price sentiment score differences and average scores for functional dimensions from the sentiment scores, combining the two types of features to construct a fused feature vector; the third layer is dynamic weight allocation, initially setting weights based on the characteristics of the home furnishing category, then combining the feedback data on the effects of pricing adjustments for this solid wood dining table and similar dining tables over the past three months, optimizing the weight values ​​using a gradient descent algorithm, and finally determining the weights for this fusion; simultaneously, the isolated forest algorithm is used to identify extreme outliers in the sentiment scores and abnormal fluctuations in the structured data, removing them to prevent abnormal data from affecting the overall data quality.

[0043] Next, the price sentiment index will be comprehensively scored. Functional value evaluation comprehensive score The standardized structured data and the data are each assigned a preset weight, and a comprehensive data is obtained through weighted calculation. Based on the comprehensive data, the price elasticity coefficient of the solid wood dining table is recalculated according to the formula to achieve dynamic correction of price sensitivity. in, This represents the adjusted price elasticity coefficient. This represents the original price elasticity coefficient calculated based on structured data. Represents the price sentiment influence coefficient. The coefficient representing the functional value impact. and The coefficients were determined by fitting a linear regression model based on sales data of the solid wood dining table after price adjustments over the past six months, and the model was recalibrated quarterly to ensure the accuracy of the coefficients.

[0044] The pricing adjustment module uses the corrected price elasticity coefficient. Based on this core principle, and combined with the merchant's pre-set profit targets and inventory turnover needs, a real-time pricing scheme for the solid wood dining table is generated. Firstly, based on... The price adjustment range is determined, with the upper and lower limits set based on the market price fluctuation range of solid wood dining tables in the home furnishing category. Then, the minimum sellable price is calculated based on the procurement cost of the solid wood dining table to ensure cost coverage. Simultaneously, a target selling price is calculated based on profit targets. The target selling price calculation model comprehensively considers the adjusted price elasticity coefficient. 1. Merchant preset profit margin 2. Unit procurement cost 3. Inventory turnover days , The formula is: , where is the target selling price, is the optimal inventory turnover days of the product, is the maximum acceptable inventory turnover days of the product; If the target selling price is within the price adjustment range, the target selling price is used as the real-time pricing; If the target selling price exceeds the price adjustment range, the boundary value of the range is selected as the real-time pricing. If the target selling price is less than the lowest sellable price, the lowest sellable price is taken as the real-time pricing, and the inventory warning mechanism is triggered simultaneously. After the plan is determined, the system automatically executes the price adjustment operation and synchronously updates the selling prices of the product on the sales platforms in each target region.

[0045] Within the preset duration after the price adjustment operation is executed, the feedback closed-loop module starts the unstructured data collection and analysis process again, collects the latest comments on the solid wood dining table on each platform and the discussion content on the social platform, and judges the acceptance degree of users for the new pricing by analyzing the data through the BERT model. If the proportion of negative comments such as "too high price" exceeds the preset threshold, the product price is adjusted back to a specified proportion of the original price. The specified proportion of the original price is determined according to the current inventory turnover days of the solid wood dining table, and the supporting promotion strategy is triggered simultaneously: If the price adjustment range is between 3% - 8%, execute the promotion strategy such as "Buy a solid wood dining table and get 2 matching dining chairs + a dining table maintenance set", and the promotion duration is set to 10 days; If the price adjustment range exceeds 8%, execute the promotion strategy such as "Reduce 150 yuan for every 1500 yuan spent + get professional on-site installation service", and the promotion duration is set to 14 days, and the promotion intensity increases with the increase of the price adjustment range; If the proportion of negative comments such as "too high price" does not exceed the preset threshold, the new pricing is maintained to complete a complete dynamic pricing closed-loop.

[0046] In summary, through the full-process implementation of multi-dimensional data collection, unstructured data analysis, price sensitivity correction, dynamic pricing adjustment and feedback closed-loop, this embodiment realizes the deep integration of structured and unstructured data of home furnishing category products. The corrected price elasticity coefficient is more in line with the price sensitivity of users in different regions, and the pricing plan accurately adapts to the characteristics of high unit price and long cycle of home furnishing products. Compared with the traditional pricing method, this embodiment effectively solves the problems of regional pricing imbalance and lagging strategy optimization, speeds up the inventory turnover while ensuring the merchant's profit, and also improves the consumption satisfaction of users in different regions, providing a practical technical solution for the dynamic pricing of home furnishing category products on cross-border e-commerce platforms.

[0047] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A dynamic pricing method for cross-border e-commerce goods based on multi-dimensional merchant data, characterized in that, The method includes the following steps: S1. Multi-dimensional data collection: Collect structured and unstructured data during the merchant's operation process. The structured data includes historical purchase prices of goods, discount response rates, procurement costs and inventory data. The unstructured data includes product reviews on the target e-commerce platform and product-related discussions on social media platforms. S2. Unstructured data parsing: The BERT semantic analysis model is used to process the unstructured data, extract users' sentiment towards product prices and their evaluation of product functional value, and quantify the two types of information into sentiment scores respectively. S3. Price sensitivity correction: The sentiment score is fused with the structured data to recalculate the price elasticity coefficient of the product and complete the dynamic correction of price sensitivity. S4. Dynamic pricing adjustment: Based on the revised price elasticity coefficient, combined with the merchant's preset profit target and inventory turnover needs, generate a real-time pricing plan and execute the price adjustment operation. S5. Price adjustment effect feedback loop: Within a preset time after the price adjustment operation is executed, unstructured data is collected and analyzed again to determine the user's acceptance of the new price. If the proportion of negative comments exceeds the preset threshold, the product price is adjusted back to the specified proportion of the original price and a matching promotional strategy is triggered. If the threshold is not exceeded, the new price is maintained.

2. The method for dynamic pricing of cross-border e-commerce goods based on multi-dimensional merchant data as described in claim 1, characterized in that, In step S1, when collecting unstructured data, web crawling technology is used to selectively crawl product reviews from target e-commerce platforms, focusing only on reviews posted within the last 30 days. Simultaneously, the product specifications and purchase location information for each review are recorded. Discussions on social media platforms containing product-specific tags or product name keywords are filtered out. The posting time and interaction volume of these discussions are collected concurrently. The effective weight of the unstructured data is calculated using the following formula: ,in, For the effective weight of a single unstructured data point, The amount of interaction in social media discussions. The data freshness coefficient. This represents the total amount of unstructured data for this product during the same period.

3. The method for dynamic pricing of cross-border e-commerce goods based on multi-dimensional merchant data as described in claim 1, characterized in that, In step S2, when extracting users' sentiment towards product prices, the BERT semantic analysis model is used to process unstructured data in a targeted manner. The BERT semantic analysis model is first optimized using a bidirectional attention mechanism, and pre-trained by constructing a product-specific corpus to enhance the recognition accuracy of price-related expressions and functional evaluation terms in cross-border e-commerce scenarios. Simultaneously, entity linking technology is introduced to associate product specifications, accessories, and after-sales service information mentioned in comments and discussions with basic product data. Based on this, users' price sentiment is divided into three categories: positive, neutral, and negative, with different intensity levels corresponding to different score ranges. The sentiment probability distribution output by the BERT semantic analysis model is used to map the price sentiment score for each data point. Then, combined with the effective weights of individual unstructured data points, a comprehensive price sentiment score is calculated using the following formula: ,in, The overall score represents the price sentiment. To determine the effective number of unstructured data entries, For the first Price sentiment score for each data point. For the first Effective weights for each data point.

4. The method for dynamic pricing of cross-border e-commerce goods based on multi-dimensional merchant data as described in claim 1, characterized in that, In step S2, when extracting user evaluations of the functional value of a product, the BERT semantic analysis model is used to process unstructured data in multiple dimensions. The BERT model employs a multi-label classification optimization strategy in the functional value evaluation extraction stage, constructing sub-classifiers for three core dimensions: logistics timeliness, packaging integrity, and usability. Each sub-classifier is fine-tuned using labeled corpora for its corresponding dimension to accurately identify the sentiment tendencies in mixed-dimensional evaluations. Semantic role labeling technology is also introduced to extract degree adverbs and comparative words from the evaluation content and convert them into score adjustment coefficients. Based on the processing results of the BERT semantic analysis model, evaluation content for the three core dimensions is extracted, and positive and negative statements in each dimension are labeled. When quantifying the functional value evaluation score, the final score is determined based on the proportion of positive statements in each dimension combined with the score adjustment coefficients. The higher the proportion of positive statements, the higher the score. The weights of each dimension's score are pre-set according to the product category.

5. The method for dynamic pricing of cross-border e-commerce goods based on multi-dimensional merchant data as described in claim 1, characterized in that, In step S3, when integrating sentiment scores and structured data, the structured data is first standardized, and then the sentiment scores and standardized structured data are assigned preset weights. A weighted calculation is then performed to obtain comprehensive data. Based on this comprehensive data, the price elasticity coefficient is recalculated. The revised formula for calculating the price elasticity coefficient is as follows: ,in, This is the corrected price elasticity coefficient. The original price elasticity coefficient is calculated based on structured data. The overall score represents the price sentiment. The overall score for functional value evaluation Price sentiment influence coefficient To determine the functional value impact coefficient, the fusion process employs multi-level processing. First, the sentiment score and structured data are aligned in terms of time and product dimensions. Then, key features of the two types of data are extracted through principal component analysis to construct a fusion vector. Simultaneously, weights are dynamically allocated according to product category characteristics. The weights are optimized through gradient descent based on pricing performance feedback over the past three months. Furthermore, the isolated forest algorithm is used to remove outliers from the sentiment score and structured data.

6. The method for dynamic pricing of cross-border e-commerce goods based on multi-dimensional merchant data as described in claim 1, characterized in that, In step S4, when generating a real-time pricing scheme, the price adjustment range is first determined based on the corrected price elasticity coefficient. The upper and lower limits of the adjustment range are set according to the market price fluctuation range of the product category. Then, the minimum sellable price is calculated in combination with the procurement cost, and the target selling price is calculated in combination with the profit target. The target selling price calculation model comprehensively considers the corrected price elasticity coefficient. Merchant's preset profit margin Unit procurement cost Inventory turnover days The formula is: ,in, Target selling price, To determine the optimal inventory turnover days for the product, The maximum acceptable inventory turnover days for the product. hour, This means that the target selling price will not be lowered when inventory turnover is normal. As the inventory turnover days increase, this coefficient gradually decreases, and the target selling price decreases accordingly, stimulating sales and accelerating inventory turnover. If the target selling price is within the price adjustment range, the target selling price is used as the real-time price. If the target selling price exceeds the range, the boundary value of the range is used as the real-time price. If the target selling price is greater than the upper limit of the adjustment range, the upper limit of the adjustment range is used as the real-time price. If the target selling price is less than the minimum sellable price, the minimum sellable price is used as the real-time price, and the inventory warning mechanism is triggered at the same time.

7. The method for dynamic pricing of cross-border e-commerce goods based on multi-dimensional merchant data according to claim 1, characterized in that, In step S5, the preset duration is set to 48 hours, the threshold is adjusted according to the product sales cycle and market competition intensity, and when the product price is adjusted back, the specified percentage of the original price is determined according to the current inventory turnover days of the product. The triggered supporting promotional strategies include two types: free accessories with purchase and discounts on purchases exceeding a specified amount. The execution duration and intensity of the promotional strategies are set according to the price adjustment range.

8. A dynamic pricing system for cross-border e-commerce goods based on multi-dimensional merchant data, applicable to the dynamic pricing method for cross-border e-commerce goods based on multi-dimensional merchant data as described in any one of claims 1-7, characterized in that... The system includes: a data acquisition module, a data parsing module, a price sensitivity correction module, a pricing adjustment module, and a feedback closed-loop module; The data acquisition module is used to collect structured data and unstructured data. The structured data includes historical purchase prices of goods, discount response rates, procurement costs and inventory data. The unstructured data includes product reviews on the target e-commerce platform and product-related discussions on social media platforms. The data parsing module is equipped with the BERT semantic analysis model, which is used to parse unstructured data and extract price sentiment and functional value evaluation, and quantify them into sentiment scores. The price sensitivity correction module is used to integrate sentiment scores and structured data to recalculate the price elasticity coefficient; The pricing adjustment module is used to generate real-time pricing schemes and execute price adjustments based on price elasticity coefficients, profit targets, and inventory requirements. The feedback closed-loop module is used to collect and parse unstructured data after price adjustment, determine pricing acceptance, and perform operations such as price maintenance, pullback, or triggering promotional strategies.

9. A dynamic pricing system for cross-border e-commerce goods based on multi-dimensional merchant data as described in claim 8, characterized in that, The data acquisition module includes a structured data acquisition unit and an unstructured data acquisition unit. The structured data acquisition unit connects to the merchant's ERP system through an API interface to acquire structured data in real time and store it in a database. The database adopts a distributed storage architecture. The unstructured data acquisition unit has a built-in web crawler program with data filtering rules that can automatically filter duplicate unstructured data and discussion content unrelated to the product. The filtering rules are updated regularly according to changes in the data format of the target platform.

Images