An e-commerce intelligent precise transaction matching system and method based on big data

CN122656726APending Publication Date: 2026-08-28陈燕菲
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202610829785.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-06-10
Publication Date
2026-08-28

AI Technical Summary

Technical Problem

[0010]针对现有技术的不足,本发明提供了一种基于大数据的电商智能精准交易匹配系统及方法,具备多模态意图深度解析、双边效用动态均衡、实时供需盘口式撮合、匹配结果可解释、反馈闭环自适应优化的优点,解决了现有技术中意图理解单一、供需匹配不对称、缺乏动态供需感知、匹配结果黑箱化、反馈闭环缺失的技术问题

Benefits of technology

[0057] Compared with existing technologies, this invention provides an intelligent and precise e-commerce transaction matching system and method based on big data, which has the following beneficial effects:

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122656726A_ABST
    Figure CN122656726A_ABST
Patent Text Reader

Abstract

The application relates to the technical field of intelligent e-commerce, and discloses an e-commerce intelligent precise transaction matching system based on big data, which comprises an e-commerce intelligent precise transaction matching system, the e-commerce intelligent precise transaction matching system comprises a data acquisition layer, a data center layer, an intelligent algorithm engine layer, a core business module layer, an application service layer and a security risk control layer. The e-commerce intelligent precise transaction matching system and method based on big data perform deep fusion on multi-dimensional data such as user search words, clicked images, staying time and price comparison behaviors through a multi-modal intention analysis module (fusion of BERT+self-attention mechanism and ViT image recognition), output a structured intention constraint set, significantly improve the recognition ability of user implicit preferences, solve the problem of single intention understanding of the traditional recommendation system, and realize the dynamic balance of the interests of buyers and sellers and the value maximization competition game distribution of scarce goods through a bilateral utility calculation mechanism in the precise transaction matching engine.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of intelligent e-commerce technology, specifically to an intelligent and precise e-commerce transaction matching system and method based on big data. Background Technology

[0002] With the rapid development of e-commerce platforms, the variety of goods and the number of users have exploded. The current advantages are: the platform has accumulated massive amounts of user behavior data, multimodal information of goods and merchant operation records; big data storage and computing technology is becoming increasingly mature; deep learning models have made significant progress in the fields of natural language processing and image recognition; real-time stream processing frameworks (such as Kafka and Flink) can support data collection and processing in high-concurrency scenarios; and personalized recommendation systems have been widely used in e-commerce scenarios and have achieved certain conversion improvement effects.

[0003] Traditional recommendation systems based on keyword search or collaborative filtering are no longer sufficient to meet the increasingly complex transaction matching needs. Existing technologies mainly suffer from the following shortcomings:

[0004] 1. Limited understanding of intent: Relying solely on search terms or historical records, lacking the integration of multimodal behavioral data such as clicked images, dwell time, price comparison, and order of adding items to cart, making it difficult to capture users' implicit preferences;

[0005] 2. Supply and demand mismatch: The focus is often on the buyer's perspective, ignoring bilateral factors such as the seller's supply capacity, inventory, and profit, resulting in a low matching and transaction rate;

[0006] 3. Lack of dynamic supply and demand perception: It is impossible to perceive the depth of buying and selling in different price ranges in real time, and it lacks a dynamic matching and price adjustment mechanism similar to that of a securities order book.

[0007] 4. Matching results are opaque: The recommendation results lack interpretability, making it difficult for users and merchants to understand the matching criteria, which affects trust and adoption;

[0008] 5. Lack of feedback loop: Rejection or bargaining behavior fails to effectively reflect the intention constraints and supply profiles, making it difficult for matching strategies to adaptively optimize. Summary of the Invention

[0009] (a) Technical problems to be solved

[0010] To address the shortcomings of existing technologies, this invention provides an intelligent and precise e-commerce transaction matching system and method based on big data. It has the advantages of multimodal intent deep analysis, dynamic equilibrium of bilateral utility, real-time supply and demand matching, interpretable matching results, and adaptive optimization of feedback loop. It solves the technical problems of existing technologies, such as single intent understanding, asymmetrical supply and demand matching, lack of dynamic supply and demand perception, black box matching results, and lack of feedback loop.

[0011] (II) Technical Solution

[0012] To achieve the aforementioned goals of deep multimodal intent analysis, dynamic equilibrium of bilateral utility, real-time supply and demand matching, interpretable matching results, and adaptive optimization of feedback loop, this invention provides the following technical solution: an e-commerce intelligent and precise transaction matching system based on big data, comprising an e-commerce intelligent and precise transaction matching system, wherein the e-commerce intelligent and precise transaction matching system includes a data acquisition layer, a data middle platform layer, an intelligent algorithm engine layer, a core business module layer, an application service layer, and a security risk control layer;

[0013] The data acquisition layer is used for real-time collection and synchronization of e-commerce big data across the entire domain, supporting real-time data stream and offline batch data synchronization; covering user behavior data (search, click, dwell time, price comparison, order of adding to cart, etc.), multimodal product data (images, videos, attributes, inventory, price, sales volume), merchant capability data (qualification, fulfillment, response, price change range, delivery stability), market trend data (industry hotspots, competitor prices, regional consumption characteristics), fulfillment data (logistics timeliness, delivery range), and risk control data (credit, abnormal transactions);

[0014] The data platform layer is used for data governance, feature modeling, and unified storage; it includes: a data governance module, a multi-dimensional modeling module, and a hierarchical storage module.

[0015] The intelligent algorithm engine layer is used to perform multi-dimensional and multi-stage intelligent matching calculations and is the core computing power module of the system.

[0016] The core business module layer is used to encapsulate matching functions for specific business scenarios;

[0017] The application service layer is used to provide multi-terminal application services for different roles;

[0018] The security risk control layer is used to ensure the security, compliance and stability of the transaction matching process.

[0019] Furthermore, the data platform layer includes:

[0020] Data governance module: Relying on generative AI to clean, complete, deduplicatize, and standardize data, and correct incomplete product information;

[0021] Multi-dimensional modeling module: Constructs precise user profiles, product feature profiles, merchant supply capacity profiles, and market scenario profiles, breaking down 200+ refined feature tags;

[0022] Tiered storage module: It adopts a hybrid storage architecture of HDFS (offline data), Kafka (real-time streaming), and Redis (high-frequency features) to ensure high-concurrency read and write efficiency.

[0023] Furthermore, the intelligent algorithm engine layer includes:

[0024] Multimodal intent parsing module: integrates BERT + self-attention mechanism (text) and ViT (image), outputting structured intent constraints {constraint: value, weight: w, type: must / preference};

[0025] Product Dynamic Knowledge Graph Module: Constructs nodes such as products, stores, warehouses, logistics routes, and historical transaction prices, as well as edges such as substitution, combination, replenishment, and negotiation space, and introduces seller sincerity index (response rate, price change range, and delivery stability).

[0026] Precise Transaction Matching Engine: Pre-filtering quickly filters out 95% of irrelevant goods based on necessary constraints; Two-sided utility calculation, combining buyer utility (matching degree, price difference, logistics, reputation) and seller utility (profit, inventory clearance, user value, associated purchases), dynamically adjusting weight α, real-time supply and demand depth indexing, constructing price-demand elasticity curves by category, similar to a stock market order book, supporting deep matching of buy and sell orders and price fine-tuning suggestions; Conflict resolution mechanism, based on value maximization competitive game, comprehensively allocating scarce goods based on bidding, credit, and historical completion rate;

[0027] Large model validation engine: performs semantic understanding and matching validation for fuzzy search, colloquial needs, and niche products, replacing manual review;

[0028] Dynamic iterative optimization engine: Collects feedback data in real time and updates algorithm weights and feature models.

[0029] Furthermore, the core business module includes:

[0030] Personalized matching module: Outputs Top-N candidate transaction pairs, combined with personalized urgency index (historical purchase cycle, number of price comparisons) to improve conversion rate;

[0031] Merchant capacity matching module: Matches customers with inventory, production capacity and delivery capabilities to avoid overselling and delivery delays;

[0032] Cold Start Intelligent Matching Module: Solve the matching problem for new users / new products based on industry baselines and transfer learning;

[0033] Real-time hotspot matching module: captures seasons, holidays, and trending events, and dynamically adjusts the matching priority;

[0034] Two-way supply and demand matching module: Supports two-way intelligent matching of buyer demand posting and merchant supply push, adaptable to B2B and customized transactions;

[0035] Matching Result Ranking and Explanation Generator: Outputs natural language explanations (e.g., "This product is matched for you because: ① You are interested in 'thin and light laptops'; ② Same-city 2-hour delivery; ③ Price is at its lowest point in the last 30 days").

[0036] Bilateral game feedback loop: Record reasons for rejection (price / style / logistics / profit / inventory), update implicit constraint weights and supply profiles, and strengthen positive matching features.

[0037] Furthermore, the application service layer includes:

[0038] User-side: Personalized recommendations, intelligent search, and precise demand push;

[0039] Merchant side: Intelligent customer matching, precise traffic distribution, and inventory-appropriate recommendations;

[0040] Platform-side: Traffic control, transaction ecosystem optimization, and data-driven decision support.

[0041] Furthermore, the security risk control layer includes:

[0042] Real-time identification of fraudulent transactions, malicious order placement, and high-risk supply and demand matching behavior; interception of abnormal matching requests.

[0043] Encrypting and protecting user privacy and business data complies with data security compliance requirements.

[0044] By integrating transaction risk control, data security, and access control, a full-chain security system is built.

[0045] A method for intelligent and precise transaction matching in e-commerce based on big data, the operation steps are as follows:

[0046] Step S1: Obtain real-time interaction data from buyers and construct a multimodal intent constraint set. This involves acquiring user search terms, clicked images, dwell time, price comparison behavior, and shopping cart addition order data through the data acquisition layer. In the intelligent algorithm engine layer, BERT+self-attention mechanism is used to extract attributes such as brand, price range, and function from the text, and ViT is used to extract implicit preferences such as style, color, and material from the images. The resulting structured intent constraint set is output in the following format:

[0047] Step S2: Obtain seller's product information, inventory, delivery capabilities, and dynamic pricing range; construct a dynamic product knowledge graph and supply capacity profile.

[0048] Through the data acquisition layer and data middle platform layer, data such as product parameters, inventory, logistics routes, historical transaction price range, and store fulfillment capabilities are obtained; a knowledge graph containing nodes such as products, stores, warehouses, logistics, and historical prices is constructed, and substitution relationships, combination relationships, replenishment cycles, and negotiable price space are established; at the same time, the seller sincerity index (response rate, price change range, and delivery stability) is calculated to form a supply capacity profile.

[0049] Step S3: Based on real-time supply and demand depth index, perform liquidity stratification for commodity categories. In the intelligent algorithm engine layer, construct price-demand elasticity curves by commodity category to form a "buy-sell depth" structure similar to a securities order book.

[0050] Purchase depth: The number of users waiting within the current price range and their urgency;

[0051] Sell ​​order depth: the supply of goods in different price ranges;

[0052] The system automatically matches transactions based on the intersection of supply and demand, and dynamically suggests fine-tuning prices to improve the success rate of matching.

[0053] Step S4: Calculate the buyer's personalized urgency index for each candidate product. Based on the user's historical behavior data (such as historical purchase cycle, number of times the same product was compared after being collected, browsing frequency, and time before adding to cart but not purchasing), calculate the buyer's personalized urgency index for the current candidate product. This index is used for subsequent matching score weighting to improve the matching priority of users with high conversion intention.

[0054] Step S5: Calculate the seller's preference matching score for the buyer. Based on the merchant's supply capacity profile, calculate the seller's preference matching score for the buyer, including: whether the buyer is a high-repurchase user, the logistics cost of the region, historical bargaining behavior, user value score, etc.; at the same time, combine the seller's current inventory clearance needs and the possibility of related purchases to form a seller-side utility score.

[0055] Step S6: Perform bilateral utility-weighted matching to generate a candidate pair set, and reallocate the matching weights through an attention mechanism to output the final matching list and explanation.

[0056] (III) Beneficial Effects

[0057] Compared with existing technologies, this invention provides an intelligent and precise e-commerce transaction matching system and method based on big data, which has the following beneficial effects:

[0058] 1. This big data-based e-commerce intelligent and precise transaction matching system and method deeply integrates multi-dimensional data such as user search terms, clicked images, dwell time, and price comparison behavior through a multi-modal intent parsing module (integrating BERT + self-attention mechanism and ViT image recognition), outputting a structured intent constraint set, which significantly improves the ability to identify users' implicit preferences and solves the problem of single intent understanding in traditional recommendation systems.

[0059] 2. This big data-based e-commerce intelligent and precise transaction matching system and method, through the bilateral utility calculation mechanism (buyer utility + seller utility) and dynamic adjustment weight α in the precise transaction matching engine, combined with the supply and demand depth index (price-demand elasticity curve, similar to the securities order book structure), realizes the dynamic equilibrium of the interests of buyers and sellers and the competitive and cooperative game allocation to maximize the value of scarce commodities, effectively improving the transaction matching success rate and resource allocation efficiency.

[0060] 3. This big data-based e-commerce intelligent and precise transaction matching system and method outputs natural language matching reasons (such as keyword matching, same-city logistics, low price, etc.) through a matching result ranking and interpretation generator. It also records rejection reasons and updates implicit constraint weights and supply profiles by combining bilateral game feedback closed loop, thereby realizing the interpretability of matching results and adaptive optimization of strategies, enhancing the trust between users and merchants and the long-term iteration capability of the system. Attached Figure Description

[0061] Figure 1 This is a schematic diagram of the overall layered architecture of the system of the present invention;

[0062] Figure 2 This is a flowchart of the core method of the present invention. Detailed Implementation

[0063] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0064] Please see Figure 1 and Figure 2 A big data-based e-commerce intelligent and precise transaction matching system includes an e-commerce intelligent and precise transaction matching system, which comprises a data acquisition layer, a data middle platform layer, an intelligent algorithm engine layer, a core business module layer, an application service layer, and a security risk control layer.

[0065] The data acquisition layer is used for real-time collection and synchronization of e-commerce big data across the entire domain, supporting real-time data stream and offline batch data synchronization; covering user behavior data (search, click, dwell time, price comparison, add-to-cart order, etc.), multimodal product data (images, videos, attributes, inventory, price, sales volume), merchant capability data (qualification, fulfillment, response, price change range, delivery stability), market trend data (industry hotspots, competitor prices, regional consumption characteristics), fulfillment data (logistics timeliness, delivery range), and risk control data (credit, abnormal transactions);

[0066] The data middle platform layer is used for data governance, feature modeling, and unified storage; it includes: a data governance module, a multi-dimensional modeling module, and a hierarchical storage module.

[0067] The intelligent algorithm engine layer is used to perform multi-dimensional and multi-stage intelligent matching calculations and is the core computing power module of the system.

[0068] The core business module layer is used to encapsulate matching functions for specific business scenarios;

[0069] The application service layer is used to provide multi-terminal application services for different roles;

[0070] The security risk control layer is used to ensure the security, compliance and stability of the transaction matching process.

[0071] In the implementation of this case, the data platform layer includes:

[0072] Data governance module: Relying on generative AI to clean, complete, deduplicatize, and standardize data, and correct incomplete product information;

[0073] Multi-dimensional modeling module: Constructs precise user profiles, product feature profiles, merchant supply capacity profiles, and market scenario profiles, breaking down 200+ refined feature tags;

[0074] Tiered storage module: It adopts a hybrid storage architecture of HDFS (offline data), Kafka (real-time streaming), and Redis (high-frequency features) to ensure high-concurrency read and write efficiency.

[0075] Among them, the hierarchical storage module and the multi-dimensional modeling module work together to build a support system of "separation of hot and cold data + online feature service";

[0076] Specifically: the offline layer (HDFS) stores all historical behavior and profile snapshots for model training and batch feature calculation; the real-time streaming layer (Kafka) handles millisecond-level behavior access and intent parsing triggers; and the high-frequency feature layer (Redis) carries dynamic tags such as user short-term interest vectors, real-time product popularity, and seller sincerity index to support online inference requests from the matching engine.

[0077] Through the organic collaboration of three-layer heterogeneous storage, the system achieves millisecond-level feature reading, stable operation of TB-level offline computing, and real-time data end-to-end latency of less than 500ms in high-concurrency scenarios, significantly improving the response speed and scalable processing capabilities of the transaction matching system.

[0078] In the implementation case, the intelligent algorithm engine layer includes:

[0079] Multimodal intent parsing module: integrates BERT + self-attention mechanism (text) and ViT (image), outputting structured intent constraints {constraint: value, weight: w, type: must / preference};

[0080] Product Dynamic Knowledge Graph Module: Constructs nodes such as products, stores, warehouses, logistics routes, and historical transaction prices, as well as edges such as substitution, combination, replenishment, and negotiation space, and introduces seller sincerity index (response rate, price change range, and delivery stability).

[0081] Precise Transaction Matching Engine: Pre-filtering quickly filters out 95% of irrelevant goods based on necessary constraints; Two-sided utility calculation, combining buyer utility (matching degree, price difference, logistics, reputation) and seller utility (profit, inventory clearance, user value, associated purchases), dynamically adjusting weight α, real-time supply and demand depth indexing, constructing price-demand elasticity curves by category, similar to a stock market order book, supporting deep matching of buy and sell orders and price fine-tuning suggestions; Conflict resolution mechanism, based on value maximization competitive game, comprehensively allocating scarce goods based on bidding, credit, and historical completion rate;

[0082] Large model validation engine: performs semantic understanding and matching validation for fuzzy search, colloquial needs, and niche products, replacing manual review;

[0083] Dynamic iterative optimization engine: Collects feedback data in real time and updates algorithm weights and feature models.

[0084] Among them, the precise transaction matching engine, together with the product dynamic knowledge graph module and the large model verification engine, form a three-level linkage matching link of "pre-filtering - knowledge enhancement - semantic verification".

[0085] Specifically, the user intent constraint set first enters the pre-filter of the precise transaction matching engine, which quickly prunes based on necessary constraints (such as brand, price range, and delivery region), filtering out more than 95% of irrelevant products. Subsequently, the remaining candidate products are expanded in the knowledge graph (such as substitutes, bundled products, and products negotiable within the same store), and the seller sincerity index is introduced to weight and correct the matching score. For conversational queries or long-tail niche products, the large model verification engine takes over the matching results to perform semantic rearrangement and compliance verification, and sends the correction signal back to the dynamic iterative optimization engine to adjust the feature weights and knowledge graph edge relationships.

[0086] Through the collaboration and closed-loop feedback of the three-level modules, the technical effects of improving matching accuracy by more than 40%, increasing the exposure rate of long-tail products by 25%, and improving the conversion rate of fuzzy queries by 30% were achieved. At the same time, the operational efficiency of reducing manual review costs by 80% was realized.

[0087] In the implementation of this case, the core business modules include:

[0088] Personalized matching module: Outputs Top-N candidate transaction pairs, combined with personalized urgency index (historical purchase cycle, number of price comparisons) to improve conversion rate;

[0089] Merchant capacity matching module: Matches customers with inventory, production capacity and delivery capabilities to avoid overselling and delivery delays;

[0090] Cold Start Intelligent Matching Module: Solve the matching problem for new users / new products based on industry baselines and transfer learning;

[0091] Real-time hotspot matching module: captures seasons, holidays, and trending events, and dynamically adjusts the matching priority;

[0092] Two-way supply and demand matching module: Supports two-way intelligent matching of buyer demand posting and merchant supply push, adaptable to B2B and customized transactions;

[0093] Matching Result Ranking and Explanation Generator: Outputs natural language explanations (e.g., "This product is matched for you because: ① You are interested in 'thin and light laptops'; ② Same-city 2-hour delivery; ③ Price is at its lowest point in the last 30 days").

[0094] Bilateral game feedback loop: Record reasons for rejection (price / style / logistics / profit / inventory), update implicit constraint weights and supply profiles, and strengthen positive matching features.

[0095] Among them, the matching result sorting and interpretation generator and the bilateral game feedback loop constitute a closed-loop optimization link of "interpretable output-implicit feedback-weight correction", which is the intelligent decision-making center of the core business module layer.

[0096] Specifically, the matching result ranking and interpretation generator receives the candidate set output by the precise transaction matching engine. Taking into account multiple dimensions such as user personalized urgency index, merchant capacity adaptation score, and real-time hot spot weighting factor, it uses the LambdaMART ranking model to rearrange the candidate transaction pairs and simultaneously generates structured interpretation templates (such as "price at its lowest point in the last 30 days", "same-city 2-hour delivery", "this merchant's response rate is 98%"). After receiving the matching results, users or merchants can choose to accept, reject, or negotiate. The two-sided game feedback loop automatically captures the reasons for rejection (such as buyers rejecting due to logistics time, sellers rejecting due to low profits), and transforms the rejection signals into implicit constraint weights (such as buyers lowering their logistics tolerance, sellers raising their minimum profit margin). These weights are then fed back to the multi-dimensional modeling module to update user profiles and merchant supply capacity profiles, while strengthening the weight of positive conversion features in the ranking model.

[0097] Through a closed-loop mechanism of "explainable output + implicit feedback + weight adaptation", the system achieved a 25% reduction in match rejection rate, a 35% increase in user click acceptance rate, and a 28% increase in merchant satisfaction. At the same time, it achieved a match conversion rate of nearly 90% for cold start users and new products within 7 days, which is close to the level of mature users, and significantly shortened the "learning cold start period" of the matching system.

[0098] In the implementation of this case, the application service layer includes:

[0099] User-side: Personalized recommendations, intelligent search, and precise demand-based push notifications;

[0100] Merchant side: Intelligent customer matching, precise traffic distribution, and inventory-appropriate recommendations;

[0101] Platform-side: Traffic control, transaction ecosystem optimization, and data-driven decision support.

[0102] Among them, the application service layer, as the interaction interface between the system and end users, realizes the last mile transformation from algorithm capabilities to business value through the three-in-one architecture of "device-side adaptation + scene awareness + role permissions".

[0103] Specifically: On the user side, a unified multi-terminal access mechanism (Web / App / Mini Program) is implemented, calling the matching result sorting and interpretation generator of the core business module layer to display Top-N candidate transaction pairs in scenarios such as homepage recommendations, search results pages, and demand squares, along with natural language explanations; On the merchant side, intelligent customer matching pushes are received in real time through the merchant workbench, and replenishment suggestions and promotional strategies are automatically generated in conjunction with the inventory matching recommendation module. At the same time, the traffic precision distribution module dynamically adjusts the store exposure weight based on the seller's sincerity index and production capacity matching score; On the platform side, the operation middleware integrates a traffic control dashboard and a transaction ecosystem health dashboard, which can manually intervene in the supply and demand cross-regions of high-demand categories by adjusting prices. The data decision support module outputs supply and demand gap analysis reports, daily category liquidity reports, and merchant growth curves, providing quantitative basis for platform operation strategies;

[0104] Through three-terminal collaboration and scenario-based encapsulation, the platform achieved a 32% increase in click-through rate on the user side, a 45% increase in customer matching response rate on the merchant side, and a 28% increase in supply and demand matching efficiency on the platform side. At the same time, it achieved the operational goals of increasing the overall transaction conversion rate of the platform by 18% and optimizing the traffic distribution fairness index (Gini coefficient) by 23%.

[0105] In the implementation of this case, the security risk control layer includes:

[0106] Real-time identification of fraudulent transactions, malicious order placement, and high-risk supply and demand matching behavior; interception of abnormal matching requests.

[0107] Encrypting and protecting user privacy and business data complies with data security compliance requirements.

[0108] By integrating transaction risk control, data security, and access control, a full-chain security system is built.

[0109] Among them, the security risk control layer adopts a three-line defense architecture of "pre-interception - in-process encryption - post-interception auditing" throughout the entire transaction matching lifecycle;

[0110] Specifically: Beforehand, based on a real-time stream processing framework (Kafka+Flink), multi-dimensional risk characteristics are calculated for each matching request, including user device fingerprints, abnormal IP aggregation, high-frequency requests within a short period, and price deviation index. Combined with a rule engine and an isolated forest anomaly detection model, behaviors such as fraudulent transactions, malicious order brushing, and scalping are identified in real time. High-risk matching requests are intercepted in milliseconds and trigger manual review. During the process, user behavior data, merchant operation data, and intermediate matching results are encrypted using the TLS 1.3 protocol, and statically stored data is encrypted using AES-256. Sensitive information such as user identity, mobile phone number, and address is anonymized and tokenized to ensure that data is "usable but not visible" throughout the entire chain. Afterward, a complete operation log auditing system is established to record all behaviors such as data access, model calls, and matching result export. Combined with the RBAC (Role-Based Access Control) permission management model, fine-grained permission allocation under the principle of least privilege is achieved, and compliance audit reports are generated regularly.

[0111] Through the synergy and in-depth defense of three lines of defense, the system achieved an accuracy rate of over 99.5% in identifying abnormal requests in transaction matching scenarios, an interception response latency of less than 100ms, and a 100% data security compliance pass rate. At the same time, it reduced the rate of fraudulent transactions by 76% and malicious order-brushing behavior by 82%. It also passed the Level 3 Information Security Protection Scheme and GDPR / Personal Information Protection Law compliance certifications, building a solid security foundation for the stable operation of the system and user trust.

[0112] A method for intelligent and precise transaction matching in e-commerce based on big data, the operation steps are as follows:

[0113] Step S1: Obtain real-time interaction data from buyers and construct a multimodal intent constraint set. This involves acquiring user search terms, clicked images, dwell time, price comparison behavior, and shopping cart addition order data through the data acquisition layer. In the intelligent algorithm engine layer, BERT+self-attention mechanism is used to extract attributes such as brand, price range, and function from the text, and ViT is used to extract implicit preferences such as style, color, and material from the images. The resulting structured intent constraint set is output in the following format:

[0114] Step S2: Obtain seller's product information, inventory, delivery capabilities, and dynamic pricing range; construct a dynamic product knowledge graph and supply capacity profile.

[0115] Through the data acquisition layer and data middle platform layer, data such as product parameters, inventory, logistics routes, historical transaction price range, and store fulfillment capabilities are obtained; a knowledge graph containing nodes such as products, stores, warehouses, logistics, and historical prices is constructed, and substitution relationships, combination relationships, replenishment cycles, and negotiable price space are established; at the same time, the seller sincerity index (response rate, price change range, and delivery stability) is calculated to form a supply capacity profile.

[0116] Step S3: Based on real-time supply and demand depth index, perform liquidity stratification for commodity categories. In the intelligent algorithm engine layer, construct price-demand elasticity curves by commodity category to form a "buy-sell depth" structure similar to a securities order book.

[0117] Purchase depth: The number of users waiting within the current price range and their urgency;

[0118] Sell ​​order depth: the supply of goods in different price ranges;

[0119] The system automatically matches transactions based on the intersection of supply and demand, and dynamically suggests fine-tuning prices to improve the success rate of matching.

[0120] Step S4: Calculate the buyer's personalized urgency index for each candidate product. Based on the user's historical behavior data (such as historical purchase cycle, number of times the same product was compared after being collected, browsing frequency, and time before adding to cart but not purchasing), calculate the buyer's personalized urgency index for the current candidate product. This index is used for subsequent matching score weighting to improve the matching priority of users with high conversion intention.

[0121] Step S5: Calculate the seller's preference matching score for the buyer. Based on the merchant's supply capacity profile, calculate the seller's preference matching score for the buyer, including: whether the buyer is a high-repurchase user, the logistics cost of the region, historical bargaining behavior, user value score, etc.; at the same time, combine the seller's current inventory clearance needs and the possibility of related purchases to form a seller-side utility score.

[0122] Step S6: Perform bilateral utility-weighted matching to generate a candidate pair set, and reallocate the matching weights through an attention mechanism to output the final matching list and explanation;

[0123] The matching score is calculated by combining buyer utility (matching degree, price difference, logistics, reputation) and seller utility (profit, inventory clearance, user value, and associated purchases) with seller utility. The score is calculated as α × buyer utility + (1-α) × seller utility, where α is dynamically adjusted according to the new product period or promotional period.

[0124] When multiple high-quality buyers match the same scarce product, a value-maximizing competitive and cooperative game mechanism is activated, which allocates resources based on bidding, credit, and historical completion rate.

[0125] By using an attention mechanism to redistribute matching weights in the candidate set, we can avoid excessive concentration of popular products.

[0126] Output the Top-N candidate transaction pairs (buyer-seller-product) and generate natural language explanations (e.g., matching reasons include keyword matching, same-city logistics, low price, additional services, etc.).

[0127] The matching results (acceptance / rejection / negotiation) are collected and updated to the bilateral profiles and the global supply and demand index, forming a closed-loop iterative optimization.

[0128] In summary, this big data-based intelligent and precise e-commerce transaction matching system and method deeply integrates multi-dimensional data such as user search terms, clicked images, dwell time, and price comparison behavior through a multi-modal intent parsing module (integrating BERT + self-attention mechanism and ViT image recognition), outputting a structured intent constraint set. This significantly improves the ability to identify users' implicit preferences and solves the problem of single intent understanding in traditional recommendation systems.

[0129] Furthermore, by using the bilateral utility calculation mechanism (buyer's utility + seller's utility) and dynamically adjusting the weight α in the precise transaction matching engine, combined with the supply and demand depth index (price-demand elasticity curve, similar to the securities order book structure), the dynamic equilibrium of the interests of both buyers and sellers and the competitive and cooperative game allocation that maximizes the value of scarce commodities are achieved, effectively improving the transaction matching success rate and resource allocation efficiency.

[0130] Furthermore, by sorting and generating matching results and outputting natural language matching reasons (such as keyword matching, same-city logistics, low price, etc.), and combining bilateral game feedback loop to record rejection reasons and update implicit constraint weights and supply profiles, the interpretability of matching results and adaptive optimization of strategies are achieved, enhancing the trust between users and merchants and the long-term iteration capability of the system.

[0131] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0132] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A big data-based intelligent and precise e-commerce transaction matching system, comprising an intelligent and precise e-commerce transaction matching system, characterized in that: The e-commerce intelligent and precise transaction matching system includes a data acquisition layer, a data middle platform layer, an intelligent algorithm engine layer, a core business module layer, an application service layer, and a security and risk control layer. The data acquisition layer is used for real-time collection and synchronization of e-commerce big data across the entire domain, supporting real-time data stream and offline batch data synchronization; covering user behavior data (search, click, dwell time, price comparison, order of adding to cart, etc.), multimodal product data (images, videos, attributes, inventory, price, sales volume), merchant capability data (qualification, fulfillment, response, price change range, delivery stability), market trend data (industry hotspots, competitor prices, regional consumption characteristics), fulfillment data (logistics timeliness, delivery range), and risk control data (credit, abnormal transactions); The data platform layer is used for data governance, feature modeling, and unified storage; it includes: a data governance module, a multi-dimensional modeling module, and a hierarchical storage module. The intelligent algorithm engine layer is used to perform multi-dimensional and multi-stage intelligent matching calculations and is the core computing power module of the system. The core business module layer is used to encapsulate matching functions for specific business scenarios; The application service layer is used to provide multi-terminal application services for different roles; The security risk control layer is used to ensure the security, compliance and stability of the transaction matching process.

2. The e-commerce intelligent and precise transaction matching system based on big data according to claim 1, characterized in that: The data platform layer includes: Data governance module: Relying on generative AI to clean, complete, deduplicatize, and standardize data, and correct incomplete product information; Multi-dimensional modeling module: Constructs precise user profiles, product feature profiles, merchant supply capacity profiles, and market scenario profiles, breaking down 200+ refined feature tags; Tiered storage module: It adopts a hybrid storage architecture of HDFS (offline data), Kafka (real-time streaming), and Redis (high-frequency features) to ensure high-concurrency read and write efficiency.

3. The e-commerce intelligent and precise transaction matching system based on big data according to claim 1, characterized in that: The intelligent algorithm engine layer includes: Multimodal intent parsing module: integrates BERT + self-attention mechanism (text) and ViT (image), outputting structured intent constraints {constraint: value, weight: w, type: must / preference}; Product Dynamic Knowledge Graph Module: Constructs nodes such as products, stores, warehouses, logistics routes, and historical transaction prices, as well as edges such as substitution, combination, replenishment, and negotiation space, and introduces seller sincerity index (response rate, price change range, and delivery stability). Precise Transaction Matching Engine: Pre-filtering quickly filters out 95% of irrelevant goods based on necessary constraints; Two-sided utility calculation, combining buyer utility (matching degree, price difference, logistics, reputation) and seller utility (profit, inventory clearance, user value, associated purchases), dynamically adjusting weight α, real-time supply and demand depth indexing, constructing price-demand elasticity curves by category, similar to a stock market order book, supporting deep matching of buy and sell orders and price fine-tuning suggestions; Conflict resolution mechanism, based on value maximization competitive game, comprehensively allocating scarce goods based on bidding, credit, and historical completion rate; Large model validation engine: performs semantic understanding and matching validation for fuzzy search, colloquial needs, and niche products, replacing manual review; Dynamic iterative optimization engine: Collects feedback data in real time and updates algorithm weights and feature models.

4. The e-commerce intelligent and precise transaction matching system based on big data according to claim 1, characterized in that: The core business modules include: Personalized matching module: Outputs Top-N candidate transaction pairs, combined with personalized urgency index (historical purchase cycle, number of price comparisons) to improve conversion rate; Merchant capacity matching module: Matches customers with inventory, production capacity and delivery capabilities to avoid overselling and delivery delays; Cold Start Intelligent Matching Module: Solve the matching problem for new users / new products based on industry baselines and transfer learning; Real-time hotspot matching module: captures seasons, holidays, and trending events, and dynamically adjusts the matching priority; Two-way supply and demand matching module: Supports two-way intelligent matching of buyer demand posting and merchant supply push, adaptable to B2B and customized transactions; Matching Result Ranking and Explanation Generator: Outputs natural language explanations (e.g., "This product is matched for you because: ① You are interested in 'thin and light laptops'; ② Same-city 2-hour delivery; ③ Price is at its lowest point in the past 30 days"). Bilateral game feedback loop: Record reasons for rejection (price / style / logistics / profit / inventory), update implicit constraint weights and supply profiles, and strengthen positive matching features.

5. The e-commerce intelligent and precise transaction matching system based on big data according to claim 1, characterized in that: The application service layer includes: User-side: Personalized recommendations, intelligent search, and precise demand push; Merchant side: Intelligent customer matching, precise traffic distribution, and inventory-appropriate recommendations; Platform-side: Traffic control, transaction ecosystem optimization, and data-driven decision support.

6. The e-commerce intelligent and precise transaction matching system based on big data according to claim 1, characterized in that: The security risk control layer includes: Real-time identification of fraudulent transactions, malicious order placement, and high-risk supply and demand matching behavior; interception of abnormal matching requests. Encrypting and protecting user privacy and business data complies with data security compliance requirements. By integrating transaction risk control, data security, and access control, a full-chain security system is built.

7. A method for intelligent and precise e-commerce transaction matching based on big data, comprising the intelligent and precise e-commerce transaction matching system based on big data as described in claims 1-6, characterized in that: The operation steps are as follows: Step S1: Obtain real-time interaction data from buyers and construct a multimodal intent constraint set. This involves acquiring user search terms, clicked images, dwell time, price comparison behavior, and shopping cart addition order data through the data acquisition layer. In the intelligent algorithm engine layer, BERT+self-attention mechanism is used to extract attributes such as brand, price range, and function from the text, and ViT is used to extract implicit preferences such as style, color, and material from the images. The resulting structured intent constraint set is output in the following format: Step S2: Obtain seller's product information, inventory, delivery capabilities, and dynamic pricing range; construct a dynamic product knowledge graph and supply capacity profile. Through the data acquisition layer and data platform layer, data such as product parameters, inventory, logistics routes, historical transaction price ranges, and store fulfillment capabilities are obtained; a knowledge graph containing nodes such as products, stores, warehouses, logistics, and historical prices is constructed, and substitution relationships, combination relationships, replenishment cycles, and negotiable pricing space are established; at the same time, the seller's sincerity index (response rate, price adjustment range, and delivery stability) is calculated to form a supply capacity profile. Step S3: Based on real-time supply and demand depth index, perform liquidity stratification for commodity categories. In the intelligent algorithm engine layer, construct price-demand elasticity curves by commodity category to form a "buy-sell depth" structure similar to a securities order book. Purchase depth: The number of users waiting within the current price range and their urgency; Sell ​​order depth: the supply of goods in different price ranges; The system automatically matches transactions based on the intersection of supply and demand, and dynamically suggests fine-tuning prices to improve the success rate of matching. Step S4: Calculate the buyer's personalized urgency index for each candidate product. Based on the user's historical behavior data (such as historical purchase cycle, number of times the same product was compared after being collected, browsing frequency, and time before adding to cart but not purchasing), calculate the buyer's personalized urgency index for the current candidate product. This index is used for subsequent matching score weighting to improve the matching priority of users with high conversion intention. Step S5: Calculate the seller's preference matching score for the buyer. Based on the merchant's supply capacity profile, calculate the seller's preference matching score for the buyer, including: whether the buyer is a high-repurchase user, the logistics cost of the region, historical bargaining behavior, user value score, etc.; at the same time, combine the seller's current inventory clearance needs and the possibility of related purchases to form a seller-side utility score. Step S6: Perform bilateral utility-weighted matching to generate a candidate pair set, and reallocate the matching weights through an attention mechanism to output the final matching list and explanation.