Cross-platform e-commerce intelligent product selection and automatic product listing methods and systems

Abstract

This invention discloses a cross-platform e-commerce intelligent product selection and automatic product listing method and system. This invention relates to the fields of data processing and artificial intelligence technology. The second tower is a product listing feasibility assessment sub-model based on logistic regression and a rule engine, used to calculate the compliance and attribute mapping success rate of products on the target platform. The outputs of the two towers are weighted and fused to generate an intelligent product selection list. For products in the intelligent product selection list, an intelligent content reconstruction process is initiated. A natural language processing model is used to perform multilingual translation and semantic polishing of product titles and descriptions. Image processing algorithms are used to perform adaptive cropping and watermark removal on product images, generating a product listing data package adapted to the target platform's specifications. The advantages of this invention are: through a closed-loop feedback mechanism, the product listing execution results (such as successful listing, insufficient inventory, and violation-related removal) are fed back to the product selection model in real time, enabling the system to dynamically perceive changes in the supply chain and platform rules, and continuously optimize subsequent product selection strategies.

Description

Technical Field

[0001] This invention relates to the fields of data processing and artificial intelligence technology, and in particular to a cross-platform e-commerce intelligent product selection and automatic product distribution method and system, specifically involving the application of traditional machine learning in e-commerce scenarios. Background Technology

[0002] With the booming development of global e-commerce, cross-platform operation has become the mainstream model for e-commerce sellers. Sellers not only need to sell on a single platform (such as a comprehensive e-commerce platform), but also need to distribute their products to emerging social e-commerce platforms or regional e-commerce platforms in order to obtain traffic dividends. In this process, product selection and product distribution are the two most crucial links.

[0003] Traditional product selection methods rely heavily on human experience or single data statistics tools, making it difficult to discover potential best-selling products from massive, multi-source, and heterogeneous cross-platform data. At the same time, traditional product distribution methods often involve simply copying and pasting selected products or uploading them to the target platform using basic migration tools, lacking intelligent adaptation to the target platform's algorithm rules, user preferences, and inventory status.

[0004] In the prior art, Chinese patent application number CN202511190103.3 (publication number CN 121032548 A) discloses a cross-border e-commerce intelligent product selection system and method based on multi-platform data fusion. This technology integrates product data from multiple platforms and constructs versioned feature snapshots. It adopts a time-consistent multi-snapshot version evaluation strategy to optimize the hyperparameter selection of the model, effectively improving the robustness and adaptability of the product selection model in non-stationary data environments. However, this prior art mainly focuses on the accuracy optimization and hyperparameter search of the prediction model in the "product selection" stage, and has the following technical defects:

[0005] First, its product selection logic is mainly based on sales trend prediction, and it does not take "feasibility of distribution" into the decision-making dimension, which may result in the selected products failing to be successfully listed due to copyright, category incompatibility or missing attribute mapping.

[0006] Second, the technology lacks an automated closed loop between product selection and distribution. After the product selection results are generated, manual or third-party tools are still needed to perform tedious content adaptation and distribution.

[0007] Third, the failure to utilize real-time feedback during the product distribution process (such as product launch failure rate and real-time inventory changes) to correct the product selection model in reverse resulted in the model being lagging when facing supply chain fluctuations.

[0008] Furthermore, during cross-platform data transmission and synchronization, network jitter and API interface limitations often lead to interruptions in product listing tasks or inconsistencies in inventory data. Although existing technologies such as CN 120186100 B propose an e-commerce management method based on big data analysis, which optimizes network transmission through dynamic slicing scheduling, they do not specifically optimize the transcoding of multimedia content (images, videos) and the semantic reconstruction of text (titles, descriptions) in cross-platform product listing scenarios.

[0009] Therefore, there is an urgent need for a cross-platform e-commerce intelligent product selection and automatic product distribution method and system that can deeply integrate intelligent product selection decision-making with automated product distribution execution, and can dynamically adjust in real time based on network status and execution feedback. Summary of the Invention

[0010] In view of the above-mentioned problems in the existing technology, the present invention proposes a cross-platform e-commerce intelligent product selection and automatic distribution method and system, which aims to realize intelligent processing of the entire process from market analysis and product selection to automatic content generation and cross-platform distribution through machine learning algorithms, and solve the problems of fragmented product selection and distribution, lagging data synchronization and low manual efficiency in the traditional model.

[0011] To achieve the above objectives, the present invention provides the following technical solution:

[0012] A cross-platform e-commerce intelligent product selection and automatic product listing method, which includes the following steps:

[0013] Step 1: Build a cross-platform heterogeneous data collection network to acquire real-time market performance data and attribute parameters of products from source platforms, as well as category rules and competitor data from target platforms, and clean and structure the collected data.

[0014] Step 2: Establish a dual-tower product selection decision model. The first tower is a best-selling prediction sub-model based on Gradient Boosting Decision Tree (GBDT) to predict the potential sales volume and profit margin of products. The second tower is a distribution feasibility assessment sub-model based on logistic regression and rule engine to calculate the compliance of products on the target platform and the success rate of attribute mapping. The outputs of the two towers are weighted and fused to generate an intelligent product selection list.

[0015] Step 3: For the products in the intelligent product selection list, initiate the intelligent content reconstruction process. Utilize Natural Language Processing (NLP) models to perform multilingual translation and semantic refinement of product titles and descriptions, and employ image processing algorithms to adaptively crop and remove watermarks from product images, generating a product listing data package adapted to the target platform's specifications.

[0016] Step 4: Implement automated product distribution and dynamic synchronization strategies. Based on the target platform's API rate limits and current network status, a dynamic sliding window mechanism is used to distribute product distribution data packets. During the distribution process, inventory and price changes on the source platform are monitored in real time. Once a data change on the source platform or a product distribution failure is detected, a differentiated update or circuit breaker mechanism is immediately triggered, and the execution result is fed back to the dual-tower product selection decision model for online learning and updates.

[0017] Furthermore, in the dual-tower product selection decision model, the best-selling product prediction sub-model adopts the XGBoost algorithm, and the input feature vector includes the cross-platform arbitrage index, sales growth acceleration, comment sentiment polarity score, and saturation of competing products in the same category; the product distribution feasibility assessment sub-model focuses on examining the semantic similarity between the source product attributes and the target platform category attributes, the brand infringement risk coefficient, and the matching degree of logistics size restrictions.

[0018] Furthermore, the automated product distribution and dynamic synchronization strategy introduces a reinforcement learning-based sending frequency control agent. This agent uses product distribution success rate and throughput as reward functions to dynamically adjust the number of concurrent requests to the target platform API, so as to maximize product distribution efficiency while avoiding triggering platform flow control restrictions.

[0019] This invention also provides a cross-platform e-commerce intelligent product selection and automatic product listing system, comprising:

[0020] The data acquisition and preprocessing module is used to perform cross-platform data capture, cleaning, outlier handling, and multi-source data fusion.

[0021] The intelligent decision center module is equipped with the dual-tower product selection decision model, which is responsible for filtering, scoring and sorting a large number of products and outputting a product selection list containing priority weights.

[0022] The content adaptation engine module integrates NLP translation services and computer vision processing units to automatically convert unstructured information of source products into standardized content that conforms to the SEO rules and visual specifications of the target platform.

[0023] The automatic distribution and synchronization module, which includes a task scheduling queue and a network transmission controller, is responsible for pushing the processed product data to the target platform and performing corresponding operations based on the feedback signal type: when the feedback signal is a request failure, a retry operation is performed; when the feedback signal is a data conflict, a rollback operation is performed; and when the feedback signal is insufficient inventory, an inventory lock operation is performed.

[0024] The present invention has the following beneficial effects:

[0025] 1. Achieved deep integration of product selection and distribution: By introducing a distribution feasibility assessment sub-model, it not only focuses on whether a product is "easy to sell" but also on "whether it can be distributed", effectively reducing the failure rate of product listing due to compliance issues or missing data.

[0026] 2. Improved operational efficiency and accuracy: By utilizing machine learning and automated processing technologies, manual data handling, translation, and image editing work has been replaced, significantly shortening the cycle from product discovery to product listing.

[0027] 3. Enhanced system robustness and adaptability: Through a closed-loop feedback mechanism, the results of product placement (such as successful product listing, insufficient inventory, and product removal due to violations) are fed back to the product selection model in real time, enabling the system to dynamically perceive changes in the supply chain and platform rules and continuously optimize subsequent product selection strategies. Attached Figure Description

[0028] Figure 1 Flowchart of intelligent product selection and automatic product listing method for cross-platform e-commerce;

[0029] Figure 2 This is a structural diagram of a dual-tower product selection decision-making model;

[0030] Figure 3 System architecture module diagram;

[0031] Figure 4 Flowchart for automated product distribution. Detailed Implementation

[0032] To make the objectives, technical solutions, and advantages of this invention clearer, the following description, in conjunction with specific embodiments and appendices, provides further details. Figure 1 -Appendix Figure 4 The present invention will be described in further detail below. It should be understood that the specific embodiments described herein are only for explaining the present invention and are not intended to limit the present invention.

[0033] Example 1

[0034] This embodiment focuses on illustrating the intelligent product selection decision-making process based on machine learning in this invention, namely, how to accurately identify products with both the potential to become bestsellers and the suitability for automated distribution from the massive data of the source platform.

[0035] A cross-platform e-commerce intelligent product selection and automatic product listing method first performs a data collection phase. The system is configured with a distributed crawler cluster to crawl data from mainstream cross-border e-commerce platforms (such as Amazon as the source platform and TikTok Shop as the target platform) using a combination of full and incremental data collection. The collected data dimensions include: basic information (title, SKU, image, category), dynamic transaction data (real-time price, sales volume in the past 7 days, inventory level), user feedback data (ratings, comment keywords), and policy data of the target platform (prohibited keywords, category attribute requirements). In the data preprocessing stage, the system uses the K-Means clustering algorithm to identify and remove outliers with abnormal prices (such as traffic-driving products priced at 0.01 yuan) or falsified sales data. It also uses a text similarity algorithm (such as Cosine Similarity) to deduplicate and merge duplicate links of the same product in different stores, constructing a unified product knowledge graph.

[0036] The core product selection process relies on a dual-tower product selection decision-making model.

[0037] The first tower is the "Bestseller Prediction Tower," which is built based on the Gradient Boosting Decision Tree (GBDT). In terms of feature engineering, this embodiment constructs several key derived features:

[0038] Cross-platform arbitrage index: Calculation formula is as follows ,in The average price of similar competing products on the target platform. The price is based on the original platform price. To estimate logistics costs.

[0039] Trend Breakout Factor: Using time series analysis, calculate the second derivative of a product's sales over the past 30 days to identify "dark horse" products with accelerating sales growth.

[0040] Competition Blue Ocean Level: Based on data from the target platform's search results page (SERP), this level measures the number of products and the percentage of top sellers for specific keywords. The lower the value, the less competition there is.

[0041] After being trained on historical data, the model outputs a score predicting the product's sales volume for the next week. .

[0042] The second tower is the "Product Feasibility Tower," which is built based on Logistic Regression and a rule engine. It aims to address the issue of neglecting execution difficulty in existing technologies. Its input features include:

[0043] (1) Attribute mapping completeness: Calculate whether the source product attributes (such as material and size) can be found in the target platform's standard attribute library. Products with a mapping rate of less than 80% will be downgraded.

[0044] (2) Infringement risk value: Integrating OCR text recognition technology and image fingerprint matching technology, it detects whether the product image contains unauthorized brand logos or text. Once a high-risk feature is found, the score for this item is reduced to zero.

[0045] (3) Logistics compliance: Based on the length, width, height and weight data of the goods, determine whether they exceed the logistics channel restrictions of the target platform.

[0046] The tower's estimated success rate score for product distribution. .

[0047] Ultimately, the system uses a weighted fusion formula. Calculate the total score, where The system uses an adjustable weighting coefficient (set to 0.6 in this embodiment) to sort the product pool in descending order based on the total score and extracts the top N products (e.g., the top 500) to generate a list of products to be distributed.

[0048] This process not only utilizes machine learning's ability to fit nonlinear market patterns, but also incorporates business rules to constrain its feasibility, overcoming the shortcomings of existing technology CN 121032548 A, which only focuses on prediction accuracy while ignoring the difficulty of implementation.

[0049] Example 2

[0050] This embodiment focuses on illustrating the intelligent content adaptation and automated product listing process in this invention, specifically how to transform the product selection list generated in Embodiment 1 into a product details page that can be sold on the target platform, while ensuring the stability of data synchronization.

[0051] After obtaining the intelligent product selection list, the system enters the content adaptation engine module.

[0052] For text information, the system calls a neural machine translation (NMT) model based on the Transformer architecture for multilingual conversion. Unlike general translation, this model has been fine-tuned with e-commerce corpus, which can accurately identify and retain core keywords (such as "Breathable" and "Anti-slip"). At the same time, a keyword embedding algorithm is introduced to naturally embed high-traffic keywords into the first 30 characters of the product title based on the hot search term list of the target platform, so as to improve SEO (Search Engine Optimization) weight.

[0053] For image information, the system uses deep learning image segmentation technology (such as U-Net) to automatically identify the product subject and background. For common Chinese promotional watermarks and graffiti advertisements in source platform (such as domestic e-commerce) images, it uses generative adversarial network (GAN) for intelligent repair and background redrawing to restore a clean white background image. In addition, according to the target platform's requirements for image size (such as 1:1 or 3:4), it uses saliency detection technology to ensure that the product subject is always in the center of the image during the cropping process, avoiding the loss of key information.

[0054] After the content reconstruction is completed, the data packets enter the automatic distribution and synchronization module.

[0055] Considering that the target platform's API usually has a call frequency limit, this embodiment designs an adaptive sending queue based on the token bucket algorithm. The system maintains a virtual token bucket, and the token generation rate is dynamically adjusted according to the "X-RateLimit-Remaining" field in the real-time API response header of the target platform.

[0056] The specific product distribution process is as follows:

[0057] Step S1: The task scheduler retrieves the delivery data packet from the queue and checks if there are enough tokens in the token bucket.

[0058] Step S2: If a token is available, initiate an API request to create a product;

[0059] If no token is available, the task is suspended and enters an Exponential Backoff waiting state.

[0060] Step S3: During data transmission, a differential synchronization mechanism is adopted. The system will first calculate the hash value of the local data packet and the existing data on the target platform, and only transmit the changed fields (such as only the price changes, without transmitting the image) to reduce bandwidth consumption. This is based on the idea of ​​dynamic scheduling due to bandwidth limitations in the existing technology CN 120186100 B and applied to the cross-platform product distribution scenario.

[0061] Step S4: Establish a real-time inventory synchronization heartbeat mechanism. The system polls the inventory status of the source platform every preset time interval (e.g., 5 minutes). If the inventory of the source platform is zero, the system immediately sends a "remove from shelves" or "set inventory to 0" instruction to the target platform through a high-priority queue to prevent overselling.

[0062] Step S5: Anomaly feedback loop. If the API returns an error code (such as "5003: Brand Unauthorized"), the system will not only record it in the log, but also analyze the error cause, generate negative sample labels, and send them back to the "product placement feasibility tower" in Example 1 in real time for online incremental training, so as to automatically avoid similar brands' products in the next product selection.

[0063] Through the above mechanism, this embodiment constructs a complete automated closed loop from product selection decision-making to execution and feedback optimization, which greatly improves the efficiency and stability of cross-platform product distribution.

Claims

1. A cross-platform e-commerce intelligent product selection and automatic product placement method, characterized in that, Includes the following steps: Construct a cross-platform heterogeneous data collection network to acquire real-time market performance data and attribute parameters of products from source platforms, as well as category rules and competitor data from target platforms, and clean and structure the collected data. A dual-tower product selection decision model is established, in which the first tower is a best-selling product prediction sub-model based on gradient boosting decision tree, used to predict the potential sales volume and profit margin of products; The second tower is a sub-model for assessing the feasibility of product listing based on logistic regression and rule engine, used to calculate the compliance of products on the target platform and the success rate of attribute mapping. The outputs from the two towers are weighted and merged to generate an intelligent product selection list; For products in the intelligent product selection list, initiate the intelligent content reconstruction process, use natural language processing models to perform multilingual translation and semantic polishing of product titles and descriptions, use image processing algorithms to perform size adaptive cropping and watermark removal on product images, and generate a product listing data package that is compatible with the target platform specifications. Implement automated product distribution and dynamic synchronization strategies, and distribute product distribution data packets using a dynamic sliding window mechanism based on the API rate limits of the target platform and the current network status; During the product distribution process, the inventory and price changes of the source platform are monitored in real time. When a change in the source platform data is detected, a differentiated update mechanism is triggered. When a failed product placement feedback is detected, a circuit breaker mechanism is triggered, and the execution result is fed back to the dual-tower product selection decision model for online learning and updates.

2. The cross-platform e-commerce intelligent product selection and automatic product placement method according to claim 1, characterized in that, The best-selling product prediction sub-model adopts the XGBoost algorithm, and the input feature vector includes cross-platform arbitrage index, sales growth acceleration, comment sentiment polarity score, and saturation of competing products in the same category. The formula for calculating the cross-platform arbitrage index is as follows: ； in, This indicates a cross-platform arbitrage index. This indicates the average price of similar competing products on the target platform. Indicates the price on the original platform. This indicates the estimated logistics costs.

3. The cross-platform e-commerce intelligent product selection and automatic product placement method according to claim 1, characterized in that, The input features of the product distribution feasibility assessment sub-model include: semantic similarity between the source product attributes and the target platform category attributes, brand infringement risk coefficient, and matching degree of logistics size restrictions; The brand infringement risk coefficient is obtained by integrating OCR text recognition technology and image fingerprint matching technology to detect whether the product image contains unauthorized brand logos and brand text.

4. The cross-platform e-commerce intelligent product selection and automatic product placement method according to claim 1, characterized in that, The formula for calculating the weighted fusion is as follows: ； in, This indicates the overall score of the product. This represents the estimated sales score output by the best-selling product prediction sub-model. This represents the predicted success rate score of the product distribution success rate output by the product distribution feasibility assessment sub-model. This represents the adjustable weighting coefficient.

5. The cross-platform e-commerce intelligent product selection and automatic product placement method according to claim 1, characterized in that, The automated product distribution and dynamic synchronization strategy introduces a reinforcement learning-based sending frequency control agent. This agent uses product distribution success rate and throughput as reward functions to dynamically adjust the number of concurrent requests to the target platform API, so as to maximize product distribution efficiency while avoiding triggering platform flow control restrictions.

6. The cross-platform e-commerce intelligent product selection and automatic product placement method according to claim 1, characterized in that, The data cleaning and structuring process includes: using the K-Means clustering algorithm to identify and remove outliers with abnormal prices or falsified sales data; using the text similarity algorithm to deduplicate and merge duplicate links of the same product in different stores; and constructing a unified product knowledge graph.

7. The cross-platform e-commerce intelligent product selection and automatic product placement method according to claim 1, characterized in that, The dynamic sliding window mechanism adopts an adaptive sending queue based on the token bucket algorithm. The system maintains a virtual token bucket, and the token generation rate is dynamically adjusted according to the remaining call count field in the real-time API response header of the target platform. During data transmission, a differential synchronization mechanism is adopted. The system first calculates the hash value of the local data packet and the existing data on the target platform, and only transmits the changed fields to reduce bandwidth consumption.

8. A cross-platform e-commerce intelligent product selection and automatic product listing system, used to implement the cross-platform e-commerce intelligent product selection and automatic product listing method according to any one of claims 1 to 7, characterized in that, include: The data acquisition and preprocessing module is used to perform cross-platform data capture, cleaning, outlier handling, and multi-source data fusion. The intelligent decision center module is equipped with a dual-tower product selection decision model, which is responsible for filtering, scoring and sorting massive amounts of products and outputting a selection list that includes priority weights. The content adaptation engine module integrates NLP translation services and computer vision processing units to automatically convert unstructured information of source products into standardized content that conforms to the SEO rules and visual specifications of the target platform. The automatic distribution and synchronization module, which includes a task scheduling queue and a network transmission controller, is responsible for pushing the processed product data to the target platform and performing corresponding operations based on the feedback signal type: when the feedback signal is a request failure, a retry operation is performed; when the feedback signal is a data conflict, a rollback operation is performed; and when the feedback signal is insufficient inventory, an inventory lock operation is performed.

9. The cross-platform e-commerce intelligent product selection and automatic product distribution system according to claim 8, characterized in that, The content adaptation engine module uses a neural machine translation model based on the Transformer architecture for multilingual conversion. This model has been fine-tuned with an e-commerce corpus and can identify and retain core keywords. At the same time, a keyword embedding algorithm is introduced to embed high-traffic keywords into product titles based on the target platform's hot search term list in order to improve search engine optimization ranking.

10. The cross-platform e-commerce intelligent product selection and automatic product distribution system according to claim 8, characterized in that, The content adaptation engine module uses deep learning image segmentation technology to automatically identify the product subject and background, uses generative adversarial networks to intelligently repair promotional watermarks in source platform images and redraw the background, and uses saliency detection technology to ensure that the product subject is always in the center of the image during the cropping process.

Images