Green live broadcast credible transaction environment construction method based on e-commerce data cross verification

Abstract

This invention discloses a method for constructing a trustworthy green live-streaming transaction environment based on cross-validation of e-commerce data, relating to the field of green live-streaming e-commerce transaction management technology. This method collects multi-source e-commerce data from the live-streaming end, platform, and third-party ends corresponding to the online transaction chain of live-streaming e-commerce. After standardized preprocessing, a dataset to be verified is generated. A cross-validation rule base covering four dimensions—entity qualifications, product green attributes, transaction chain, and live-streaming content—is invoked for verification. A credibility score is generated and corresponding risk levels are marked. Tiered control actions are executed, and process records and result data are stored on the blockchain for evidence. The rule base is continuously iterated and optimized to complete the construction of a trustworthy transaction environment. This invention can improve the credibility of green live-streaming transactions, has high accuracy in risk management, and the end-to-end evidence storage can provide support for rights protection and supervision. It can be implemented without significant modifications to the original transaction chain and is adaptable to the green transaction scenario needs of various live-streaming e-commerce platforms.

Description

Technical Field

[0001] This invention relates to the field of green live-streaming e-commerce transaction management technology, and in particular to a method for constructing a green live-streaming trusted transaction environment based on cross-validation of e-commerce data. Background Technology

[0002] With the growing popularity of green consumption concepts, live-streaming e-commerce, emphasizing low-carbon and environmentally friendly products, has become a significant development direction in the live-streaming e-commerce sector. More and more streamers and merchants are launching green-themed live-streaming rooms, showcasing various products labeled with green certifications and low-carbon attributes, and consumers' willingness to purchase green products continues to rise. However, along with the rapid development of green live-streaming e-commerce, corresponding issues of trust in transactions are becoming increasingly prominent. Problems such as some merchants fabricating green certification qualifications for their products, streamers exaggerating the green attributes of products during live streams, and a lack of transparency in transaction data are frequently occurring. These issues not only harm consumers' legitimate rights and interests but also hinder the long-term healthy development of the green live-streaming e-commerce industry.

[0003] Currently, mainstream live-streaming e-commerce transaction control solutions mainly fall into two categories. One is the platform-side qualification review and content supervision solution, which requires merchants to upload product qualifications and green certification materials for manual verification. Simultaneously, during the live stream, it uses keyword recognition and sensitive image blocking to screen for violations. This type of solution relies solely on a single platform data source for verification, failing to cross-reference the materials submitted by merchants with authoritative third-party certification data, after-sales feedback data, and logistics transaction data. This makes it prone to cases where forged certification materials slip through, and also struggles to identify implicit false advertising in the host's rhetoric and displayed content. The other type is the transaction evidence storage solution, which only stores transaction orders and payment records for live-streaming e-commerce. It doesn't cover the entire data chain from qualification verification and live stream display to risk management. In the event of a transaction dispute, it's difficult to obtain complete evidence, resulting in high costs for consumers seeking redress, and regulatory authorities also lack access to reliable data across the entire chain for routine supervision.

[0004] Existing regulatory schemes do not have appropriate verification logic for the special attributes of green products. There is a lack of unified judgment standards for verifying green attributes such as carbon footprint and environmental parameters of products. The regulatory standards of different platforms vary greatly, making it difficult for truly compliant green products to gain full trust from consumers. Instead, it is easy for bad money to drive out good, which makes it difficult to support the long-term standardized development of green live-streaming e-commerce. Summary of the Invention

[0005] The green live streaming trusted transaction environment construction method based on cross-validation of e-commerce data proposed in this invention can solve the problems mentioned in the prior art.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for constructing a green live-streaming trusted transaction environment based on cross-validation of e-commerce data, comprising the following steps: It collects multi-source e-commerce data from three sources across the online transaction chain: the livestreaming end, the platform end, and the third-party end. The livestreaming end data includes the streamer's identity and qualifications, audio and video of product display, green attribute parameters of products on the blockchain, and information from the product details page. The platform end data includes merchant qualification registration, transaction orders, after-sales complaints and rights protection, payment records, and logistics and delivery information. The third-party end data includes product carbon footprint certification reports, quality inspection reports, consumer reviews, and industry compliance and standardization information. Standardized preprocessing is performed on the collected multi-source e-commerce data, including unified data format conversion, duplicate data removal, missing value completion, compliance verification, and removal of invalid data; A multi-dimensional cross-validation rule base is constructed, covering four core validation dimensions: cross-validation of anchor and merchant qualifications and platform filing information, cross-validation of product green attribute parameters and third-party certification reports, cross-validation of transaction orders, payment records and logistics information, and cross-validation of live broadcast content and actual product parameters. Based on the rule base, the associated data in the dataset to be verified is cross-verified item by item to enter the transaction link, and a credibility score is generated. The subjects, products and transaction link nodes corresponding to the associated data with scores below the qualified threshold are marked with corresponding risk levels. Tiered control measures are implemented based on risk level: high-risk measures include limiting live stream traffic, removing products from shelves, and blocking transaction links; medium-risk measures include publicizing risk information, issuing pop-up notifications to consumers, and manually reviewing the content involved; and low-risk measures include regular re-inspections and dynamic tracking of related data. Construct a trusted ledger for the entire transaction chain, synchronously upload evidence to the blockchain to cross-verify the process and results, risk control actions, and transaction feedback records, and provide access-controlled query portals to consumers, regulatory authorities, and platform operators; We continuously collect data on consumer feedback, new risk cases, and updated regulatory standards, and dynamically adjust the rule base's verification logic, weight parameters, and qualification thresholds.

[0007] Furthermore, it also includes introducing a comprehensive and reliable quantitative calculation logic for the green attributes of products in the cross-validation process between the green attribute parameters of products and third-party certification reports. The calculation formula is as follows: ;in This indicates the overall credibility score of the product's green attributes. This represents the weighting coefficient corresponding to the third-party carbon footprint certification results. This represents the quantitative value of a third-party carbon footprint certification result. This indicates the weighting coefficient corresponding to the consistency verification between the materials displayed in the live broadcast and the certification parameters. This represents the quantitative value used for verifying the consistency of product materials. This represents the weighting coefficient corresponding to historical consumer evaluations related to green attributes. This represents the quantitative value of historical consumer evaluations related to green attributes.

[0008] Furthermore, it also includes introducing a quantitative calculation logic for the subject's risk level in the risk label generation process, with the calculation formula as follows: ;in This represents the quantitative value of the subject's risk level. This represents the weight coefficient corresponding to the currently unqualified cross-validation item. This represents the quantified value of the number of non-compliant items in the current cross-validation. This represents the weighting coefficient corresponding to the subject's historical risk records. This represents the quantified value of the subject's historical risk records. This represents the weighting coefficient corresponding to the transaction volume of the main entity over the past 30 days. This represents the quantitative value of the corresponding entity's transaction volume over the past 30 days.

[0009] Furthermore, during the multi-source e-commerce data collection process, for live streaming data, a combination of real-time frame interception and pre-trained multimodal semantic recognition models is used to extract structured information from the anchor's speech, product display content, and product details page. For platform data, a dedicated interface call is used to obtain anonymized transaction data, qualification filing data, and after-sales rights protection data. For third-party data, a consortium blockchain node real-time synchronization method is used to obtain tamper-proof certification reports, industry standards, and public evaluation data.

[0010] Furthermore, in the standardized preprocessing of multi-source e-commerce data, structured data is processed by using unified field mapping, outlier removal, and missing values ​​filled with the median value of the same dimension and type of data. Unstructured video, audio, and text data are converted into standardized structured feature data using a pre-trained multimodal feature extraction model. All preprocessed data undergoes secondary compliance verification, and the generated dataset to be verified can be directly input into the cross-validation process for processing.

[0011] Furthermore, in the process of constructing the multi-dimensional cross-validation rule base, the first step is to introduce the compliance specifications for live-streaming e-commerce, green product certification specifications, and transaction security specifications issued by regulatory authorities as basic rules. Then, customized category-specific rules are added by combining the platform's historical risk cases and the compliance requirements of third-party certification agencies. All rules are configured with flexibly adjustable weight parameters to support the adaptation logic of adjusting rules for different product categories and different types of live-streaming scenarios.

[0012] Furthermore, during the risk label generation process, a corresponding traceability certificate is generated for each item that fails cross-validation. The traceability certificate contains complete information in four dimensions: the specific data item that failed validation, the data source, the comparison standard, and the judgment basis. All traceability certificates are bound and stored one by one with the corresponding subject, product, and transaction node.

[0013] Furthermore, during the execution of tiered control actions, control actions for high-risk entities are directly triggered by the system to be executed automatically, which can quickly block the spread of risks without human intervention. Control actions for medium-risk entities are first pushed to the platform's review personnel for manual confirmation before execution. Control actions for low-risk entities are only marked in the background and do not affect the normal operation of the current transaction chain. All execution records of control actions are synchronously stored in the trusted evidence ledger.

[0014] Furthermore, the end-to-end trusted evidence storage ledger is built using a consortium blockchain architecture. The platform operator, regulatory authorities, and third-party certification bodies serve as consensus nodes on the consortium blockchain. All evidence storage data must be jointly confirmed by at least three consensus nodes before it can be stored on the blockchain. Consumers can query the end-to-end verification record of the corresponding transaction through the transaction order number. Regulatory authorities can obtain all the risk data and verification data to carry out regulatory work. Third-party certification bodies can only query the verification records related to the products they have certified.

[0015] Furthermore, during the dynamic iterative optimization of the cross-validation rule base, new risk case data, consumer feedback data, and regulatory update data are collected at fixed intervals. A deep reinforcement learning model is used to iteratively adjust the validation logic, weight parameters, and qualification threshold standards of the rule base. The adjusted rules are first tested in a small-scale gray-scale test to verify their effectiveness and rationality before being fully launched.

[0016] Compared with existing technologies, the beneficial effects of this invention are: This invention uses a multi-source data cross-verification method to cover the full-link transaction-related data verification of three types of data sources: live streaming end, platform end, and third-party end. It can realize multi-dimensional cross-verification of anchor and merchant qualifications, green attributes of products, transaction link data, and live streaming content. It solves the problem of insufficient accuracy of existing technologies that rely on a single data source for verification. It can effectively identify risky content such as false qualifications and false advertising, and improve the credibility of green live streaming transactions.

[0017] This invention combines risk-based hierarchical control with end-to-end trusted evidence storage to execute corresponding control actions for entities with different risk levels. While quickly intercepting high-risk transactions, it does not cause unnecessary interference to the normal transaction chain. All verification processes, handling records, and transaction feedback data are synchronously stored on the blockchain, and access controllable query portals are opened for different entities. This solves the problems of insufficient accuracy in risk handling and difficulty in tracing transaction data in existing technologies, and can provide reliable data support for consumer rights protection and regulatory authorities to carry out regulatory work.

[0018] The cross-validation rule base constructed by this invention supports dynamic iterative optimization. It can flexibly adjust the validation logic and adaptation standards according to newly added risk cases, consumer feedback, and regulatory requirements. It solves the problems of rigidity and insufficient adaptability of existing technical rules. It can adapt to the verification needs of different categories of green products and different types of live streaming scenarios. It can be deployed and implemented without making significant changes to the original live streaming e-commerce transaction chain. It can be promoted and applied on various platforms that carry out live streaming e-commerce business, providing stable technical support for the standardized development of green live streaming e-commerce. Attached Figure Description

[0019] Figure 1 This is a schematic block diagram of the method for constructing a green live streaming trusted transaction environment based on cross-validation of e-commerce data proposed in this invention. Figure 2 This is a schematic block diagram of the multi-source e-commerce data collection and preprocessing flowchart proposed in this invention; Figure 3 This is a flowchart of the multi-dimensional cross-validation and risk labeling process proposed in this invention; Figure 4 This is a flowchart illustrating the hierarchical control action execution proposed in this invention. Figure 5 This is a flowchart of the trusted evidence storage and rule base dynamic optimization process proposed in this invention. Detailed Implementation

[0020] 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.

[0021] Reference Figures 1 to 5 A method for constructing a green live-streaming trusted transaction environment based on cross-validation of e-commerce data, comprising the following steps: This involves collecting multi-source e-commerce data from three data sources across the online transaction chain: the livestreaming platform, the platform itself, and third-party platforms. Livestreaming platform data includes the streamer's identity and qualifications, audio and video data of product displays during the livestream, green attribute parameters of products on the blockchain, and product details page information. Platform data includes merchant qualification and registration information, transaction order data associated with the livestream, after-sales complaint and rights protection data, payment transaction information, and logistics and delivery information. Third-party platform data includes product carbon footprint certification reports issued by authoritative institutions, product quality inspection reports, publicly available consumer review data, and industry compliance and standardization information. Standardized preprocessing operations are performed on all collected multi-source e-commerce data, including unified conversion of data formats from different data sources, removal of duplicate data, completion of missing values, and verification of data compliance. Invalid data that does not conform to e-commerce data storage specifications or is irrelevant to the determination of transaction credibility is removed, and a standardized dataset to be verified is generated. A multi-dimensional cross-validation rule base is constructed, covering four core validation dimensions: cross-validation of the qualifications of the anchor and merchant with the platform's filing information, cross-validation of the green attribute parameters of the product with the third-party certification report, cross-validation of transaction orders with payment records and logistics information, and cross-validation of the live broadcast content with the actual parameters of the product. Each dimension is configured with corresponding validation logic, comparison standards, and weight parameters. Based on the completed cross-validation rule base, all related data in the dataset to be validated that are about to enter the live e-commerce online transaction link are cross-validated item by item, and a credibility score is generated for each related data item. The subjects, products and transaction link nodes corresponding to the related data whose credibility scores are lower than the preset qualified threshold are marked with the corresponding risk level. Based on the risk level, tiered control measures are implemented for the corresponding entities. High-risk levels are subject to control measures such as limiting live stream traffic, removing the products involved, and blocking the current transaction link. Medium-risk levels are subject to control measures such as full-link disclosure of risk information, pop-up reminders for consumers before transactions, and manual review of the content involved. Low-risk levels are subject to control measures such as regular re-inspection and dynamic tracking of related data. Construct a trusted evidence storage ledger covering the entire transaction chain, synchronously record all cross-validation processes and results, risk control action execution records, and post-transaction feedback records on the blockchain for evidence storage, and provide access controllable evidence storage information query portals to consumers, regulatory authorities, and platform operators. We continuously collect consumer feedback data, new risk case data, and regulatory update data generated throughout the entire transaction process. We dynamically adjust the verification logic, weight parameters, and qualification threshold standards of the cross-validation rule base to achieve continuous iterative optimization of the green live streaming trustworthy transaction environment.

[0022] This invention also includes a cross-validation process between product green attribute parameters and third-party certification reports, introducing a comprehensive and reliable quantitative calculation logic for product green attributes. The calculation formula is as follows: ;in This indicates the overall credibility score of the product's green attributes. This represents the weighting coefficient corresponding to the third-party carbon footprint certification result, with a value ranging from 0 to 1. This represents a quantitative value indicating the third-party carbon footprint certification result, ranging from 0 to 100. This represents the weighting coefficient corresponding to the consistency verification between the live-stream displayed materials and the certification parameters, with a value ranging from 0 to 1. This represents the quantitative value for verifying the consistency of product materials, ranging from 0 to 100. This represents the weighting coefficient corresponding to historical consumer evaluations related to green attributes, with a value ranging from 0 to 1. This represents the quantitative value of historical consumer evaluations related to green attributes, ranging from 0 to 100, and satisfying the following conditions: + + =1. All parameters are calculated using dimensionless quantification values. The calculation results can be directly used to determine the credibility of the green attributes of products, effectively reducing the subjective error of manual judgment and improving the accuracy of cross-validation results.

[0023] This invention also includes introducing a subject risk level quantification calculation logic in the risk label generation stage, with the calculation formula as follows: ;in This represents the quantitative value of the subject's risk level. This represents the weight coefficient corresponding to the currently unqualified cross-validation item. This represents the quantified value of the number of non-compliant items in the current cross-validation. This represents the weighting coefficient corresponding to the subject's historical risk records. This represents the quantified value of the subject's historical risk records. This represents the weighting coefficient corresponding to the transaction volume of the main entity over the past 30 days. This represents the quantitative value of the corresponding entity's transaction volume over the past 30 days. All parameters are calculated using dimensionless quantitative values, with the results ranging from 0 to 100. This directly matches the corresponding risk level, improving the accuracy of risk classification and avoiding issues of over- or under-control.

[0024] In this invention, during the multi-source e-commerce data collection process, real-time frame interception combined with a pre-trained multimodal semantic recognition model is used to extract structured information from the anchor's speech, product display content, and product details page for live streaming data. For platform data, dedicated interface calls are used to obtain anonymized transaction data, qualification filing data, and after-sales rights protection data. For third-party data, real-time synchronization of consortium blockchain nodes is used to obtain tamper-proof certification reports, industry standards, and public evaluation data. All data collection processes strictly comply with relevant personal information protection regulations, collecting only necessary data directly related to transaction credibility determination and not collecting additional consumer privacy data.

[0025] In this invention, during the standardized preprocessing of multi-source e-commerce data, structured data is processed using a unified field mapping, outlier removal, and missing values ​​are filled with the median value of the same type of data in the same dimension. Unstructured video, audio, and text data are converted into standardized structured feature data using a pre-trained multimodal feature extraction model. All preprocessed data undergoes secondary compliance verification to ensure that there is no sensitive privacy data that has not been desensitized and no redundant data that is irrelevant to the credibility determination. The generated dataset to be verified can be directly input into the cross-validation process for processing.

[0026] In this invention, during the construction of the multi-dimensional cross-validation rule base, the first step is to introduce the compliance specifications for live-streaming e-commerce, green product certification specifications, and transaction security specifications issued by regulatory authorities as basic rules. Then, customized category-specific rules are added by combining the platform's historical risk cases and the compliance requirements of third-party certification agencies. All rules are configured with flexibly adjustable weight parameters, supporting the adjustment of rule adaptation logic for different product categories and different types of live-streaming scenarios, ensuring that the rule base can adapt to the transaction verification needs of green live-streaming across all categories.

[0027] In this invention, during the risk label generation process, a corresponding traceability certificate is generated for each item that fails cross-validation. The traceability certificate contains complete information in four dimensions: the specific data item that failed validation, the data source, the comparison standard, and the judgment basis. All traceability certificates are bound and stored with the corresponding subject, product, and transaction node to ensure that risk issues are traceable and can be manually reviewed, avoiding unnecessary losses caused by misjudgment.

[0028] In this invention, during the execution of tiered control actions, control actions for high-risk entities are directly triggered by the system to be executed automatically, which can quickly block the spread of risks without human intervention. Control actions for medium-risk entities are first pushed to the platform's review personnel for manual confirmation before execution to avoid misjudgment affecting normal transactions. Control actions for low-risk entities are only marked in the background and do not affect the normal operation of the current transaction chain. The execution records of all control actions are synchronously stored in a trusted evidence ledger for easy traceability and verification in the future.

[0029] In this invention, the end-to-end trusted evidence storage ledger is built using a consortium blockchain architecture. The platform operator, regulatory authorities, and third-party certification bodies serve as consensus nodes on the consortium blockchain. All evidence storage data must be jointly confirmed by at least three consensus nodes before it can be stored on the blockchain. Consumers can query the end-to-end verification record of the corresponding transaction through the transaction order number. Regulatory authorities can obtain all risk data and verification data to carry out regulatory work. Third-party certification bodies can only query verification records related to the products they have certified, ensuring the immutability of the evidence storage data and the controllability of access permissions.

[0030] In this invention, during the dynamic iterative optimization of the cross-validation rule base, newly added risk case data, consumer feedback data, and regulatory update data are collected at fixed intervals. A deep reinforcement learning model is used to iteratively adjust the validation logic, weight parameters, and qualification threshold standards of the rule base. The adjusted rules are first tested in a small-scale gray-scale test to verify their effectiveness and rationality before being fully launched, ensuring that the rule base always adapts to the latest needs of live e-commerce transaction scenarios and continuously improves the protection capabilities of the trusted transaction environment.

[0031] The following two examples further illustrate the specific implementation of this system: This invention discloses a method for constructing a green live-streaming trusted transaction environment based on cross-validation of e-commerce data. Deployed on the backend server cluster of a live-streaming e-commerce platform, it connects to the platform's live-streaming management system, transaction management system, and evidence storage system. Simultaneously, it connects to the data systems of third-party certification bodies and regulatory departments through consortium blockchain nodes. This allows for deployment and operation without significant modifications to the existing live-streaming transaction chain. The implementation process of this invention will be further explained below using two different green live-streaming scenarios for various product categories.

[0032] The first embodiment is a green live-streaming trading scenario for fresh agricultural products, applied to a special live-streaming event for regional public brand organic vegetables to help farmers.

[0033] First, perform multi-source data collection operations. The live broadcast end uses the collection module deployed on the live broadcast edge node to pull the audio and video frames of the live stream in real time, and calls the pre-trained multi-modal semantic recognition model to extract the pesticide residue values, organic certification validity periods, and planting base information mentioned in the host's speech. At the same time, it grabs the detail page fields of the vegetables listed in the live broadcast room, the farmer support qualification documents submitted by the host, and the green product attribute parameters uploaded by the merchant; the platform end obtains the desensitized merchant organic planting base filing information, the transaction order data associated with this special session, the after-sales complaint data of consumers for the merchant's green products within the past 6 months, the payment flow information for the corresponding transaction, and the node trajectory data of the fresh food cold chain logistics through a dedicated interface call; the third-party end synchronizes in real time through the alliance chain node the organic certification report of this batch of vegetables issued by the national agricultural product quality and safety platform, the pesticide residue detection report issued by the local agricultural testing center, the consumer evaluation data of the same batch of vegetables on the public e-commerce platform, and the industry compliance specifications for fresh green products.

[0034] After the collection is completed, perform standardized preprocessing. For structured data, perform unified field mapping, convert organic certification fields from different sources into unified certification subject, certification batch, and validity period standard fields,剔除 duplicate uploaded test reports, fill in the missing planting base coordinate fields with the base address filed by the merchant, convert the unstructured audio and video recognition results into standardized feature quantization values, and perform secondary compliance verification to剔除 all sensitive data containing consumer privacy, and finally generate a dataset to be verified.

[0035] Next, call the multi-dimensional cross-verification rule library to perform verification. The basic rules of the rule library come from the current live e-commerce supervision specifications and green agricultural product certification specifications. The exclusive rules are set in combination with the characteristics of fresh food categories. Verify each item in four dimensions: host qualification and platform filing information, product green attributes and third-party certification reports, transaction orders and payment and logistics data, and live broadcast display content and actual product parameters. The verification result of each dimension corresponds to a preset weight value, and the credibility score of the corresponding associated data is obtained by weighting. During the verification process, it is found that the listed batch of a certain leafy vegetable is inconsistent with the batch certified by the third party, and the corresponding credibility score is lower than the preset qualified threshold. The system generates a corresponding traceability certificate, which includes the specific data items with batch discrepancies, the two data sources of the batch uploaded by the merchant and the batch certified by the third party, the comparison standard for batch matching, and the judgment basis of the green agricultural product certification management specification, and marks the corresponding medium risk level.

[0036] Subsequently, perform hierarchical control actions. The medium risk mark triggers a pop-up window prompt before the consumer places an order to indicate that the certification batch of this product is pending verification, and at the same time pushes the risk information to the platform auditors for manual review. After the review confirms that the batch was misfilled by the merchant during upload, the risk mark is removed after updating the batch information, and the control process does not affect the normal transactions of other compliant products.

[0037] All verification processes, review records, and consumer feedback data after transaction completion are synchronously stored in the consortium blockchain's evidence ledger. The platform, local agricultural regulatory authorities, and third-party certification bodies, acting as consensus nodes, jointly confirm the validity of the evidence data. Consumers can query the full-chain verification records of the corresponding transaction using the order number. Regulatory authorities can access all risk data for routine supervision, while third-party certification bodies can only query verification records related to the products they have certified. Subsequently, the system will continuously collect consumer feedback data from this special event and new regulatory standards for fresh and green products. A deep reinforcement learning model will be used to adjust the verification weights of the rule base, increasing the priority of batch matching verification items. The adjusted rules will undergo a 10% gray-scale test to verify their effectiveness before being fully launched.

[0038] This embodiment solves the problem of identifying discrepancies in certified batches under the original single-platform verification model by using multi-source data cross-validation. The risk classification and control logic does not affect the normal operation of agricultural assistance live streaming while investigating risks. The full-chain evidence storage data provides reliable support for the traceability and verification of regulatory authorities. The dynamically iterative rule base can be adapted to the category characteristics of fresh agricultural products, improving the transaction credibility of green agricultural product live streaming, preventing falsely certified products from crowding out the traffic of compliant products, and providing stable support for the standardized operation of agricultural assistance green live streaming. It can be deployed without making significant modifications to the original fresh food live streaming transaction chain and can be promoted and applied on various e-commerce platforms that carry out agricultural product live streaming business.

[0039] The second example is a green live-streaming transaction scenario for low-formaldehyde home building materials, applied to an environmentally themed live-streaming event for a custom wardrobe brand.

[0040] First, multi-source data collection is performed. The live streaming client uses a collection module deployed at the edge node to pull audio and video frames from the live stream in real time. It then calls a pre-trained multimodal semantic recognition model to extract information such as the environmental protection level of the boards, formaldehyde emission, and the proportion of recyclable materials mentioned in the host's speech. At the same time, it captures fields from the details page of the wardrobes listed in the live stream, the home furnishing industry qualification documents submitted by the host, and the green attribute parameters of the products uploaded by the merchants. The platform uses a dedicated interface to obtain the anonymized information of the merchant's board supplier registration, the transaction order data associated with the special session, the after-sales complaint data of consumers for the merchant's green products in the past 6 months, the payment flow information of the corresponding transactions, and the delivery and installation node data of large items. The third party uses a consortium blockchain node to synchronize in real time the environmental protection level certification report of the boards issued by the National Building Materials Testing Center, the carbon footprint accounting report, consumer evaluation data of the same series of wardrobes on public e-commerce platforms, and the industry compliance standards for green home building materials.

[0041] After the collection is completed, perform standardized preprocessing. For structured data, execute unified field mapping to convert environmental protection certification fields from different sources into unified certification entity, board batch, and formaldehyde emission standard fields. Eliminate duplicate uploaded test reports. Fill in the missing board outbound batch field with the batch information备案 by the supplier. Convert the unstructured audio and video recognition results into standardized feature quantization values. Perform secondary compliance verification to eliminate all sensitive data containing consumer privacy. Finally, generate a dataset to be verified.

[0042] Next, call the multi-dimensional cross-verification rule library to perform verification. The basic rules of the rule library come from the current live e-commerce supervision specifications and green building material certification specifications. The exclusive rules are set in combination with the characteristics of the home building materials category. Verify each item in four dimensions: the host qualification and platform备案 information, the green attributes of the product and the third-party certification report, the transaction order and payment and logistics data, and the live display content and the actual parameters of the product. The verification result of each dimension corresponds to a preset weight value, and the weighted value is used to obtain the credibility score of the corresponding associated data. During the verification process, it is found that the board environmental protection level of a certain wardrobe is marked as E0 level, but the actual formaldehyde emission of the third-party test report does not meet the E0 level standard, and the corresponding credibility score is lower than the preset qualified threshold. The system generates a corresponding traceability certificate, which includes the specific data items with inconsistent environmental protection levels, the two data sources of the merchant-marked parameters and the third-party test parameters, the comparison standard of the E0 level environmental protection standard, and the judgment basis of the green building material certification management specification, and marks the corresponding high-risk level.

[0043] Subsequently, perform hierarchical control actions. The high-risk mark directly triggers control actions such as taking off the shelves of the involved products, intercepting the pending payment orders, and restricting the live stream. The risk can be quickly blocked without manual intervention, and the control process does not affect the normal transactions of other compliant products.

[0044] All verification processes, off-shelf disposal records, and consumer feedback data after the transaction are synchronously stored in the alliance chain deposit ledger. The platform, market supervision department, and third-party certification agency, as consensus nodes, jointly confirm the validity of the deposit data. Consumers can query the full-link verification records of the corresponding transactions through the order number. The supervision department can obtain all risk data for regular supervision. The third-party certification agency can only query the verification records related to the products it certifies. Subsequently, the system continuously collects consumer feedback data for this special session and newly added home building materials environmental protection specifications, and uses a deep reinforcement learning model to adjust the verification weights of the rule library, increase the priority of the formaldehyde emission verification item, and the adjusted rules are first verified for effectiveness through a 10% session gray-box test and then fully launched.

[0045] This embodiment solves the problem of difficulty in identifying counterfeit environmental protection ratings of building materials under the original single-platform verification model by using multi-source data cross-validation. The high-risk automatic control logic can quickly block the spread of false advertising content and prevent more consumers from suffering losses. The full-chain evidence storage data provides credible support for consumer rights protection and law enforcement verification by regulatory authorities. The dynamically iterative rule base can adapt to the characteristics of home furnishing and building materials categories and the latest environmental protection standards, improving the credibility of green live streaming transactions for home furnishing and building materials. It prevents products with false environmental protection claims from crowding out the market space of compliant products, and can provide stable support for the standardized operation of green live streaming in the home furnishing category. It can be deployed without making significant modifications to the original home furnishing live streaming transaction chain and can be promoted and applied on various e-commerce platforms that carry out live streaming business of building materials and home furnishings.

[0046] Reference Figure 2 This section details the specific execution steps of multi-source e-commerce data collection and standardized preprocessing. In the data acquisition phase, the system adopts differentiated collection strategies for different data sources. For live-streaming data, the system utilizes real-time frame interception combined with multimodal semantic recognition technology to accurately extract the anchor's dialogue and product display features. For platform-side data, it obtains anonymized transaction and after-sales records through a dedicated encrypted interface. For third-party data, the system synchronizes authoritative certification reports and industry standards in real time through consortium blockchain nodes. After data aggregation, the system enters the standardized preprocessing stage. The system performs field mapping, outlier removal, and missing value imputation on structured data, while converting unstructured audio and video data into standard feature vectors using a pre-trained model. All data undergoes a second compliance check before output to thoroughly remove redundant and unanonymized sensitive information, ensuring that the final generated dataset for verification is clean, standardized, and fully compliant.

[0047] Reference Figure 3 The presentation highlights how the system utilizes multi-dimensional rules for cross-validation and generates risk markers. After preprocessing, the dataset enters the core validation engine and is distributed to four core validation dimensions for parallel comparison. These four dimensions comprehensively cover the entity's identity and qualifications, product green attribute certification, fund and logistics flow, and the realistic display effect of the live stream. The system performs rigorous cross-logic operations on the associated data based on the configured comparison standards and weight parameters for each dimension, quantifying and outputting a comprehensive credibility score for the product or entity. When the system determines that the score is below the preset safety threshold, it not only blocks the normal flow of the data but also clearly marks the corresponding entity, product, or transaction link node with a risk level. Simultaneously, the system automatically generates and stores traceability vouchers containing complete information such as data items, sources, and comparison basis, providing detailed evidence for subsequent manual review and accountability.

[0048] Reference Figure 4This demonstrates the system's business process for implementing refined control over entities at different risk levels after generating risk markers. To strike a balance between ensuring platform transaction security and maintaining normal business operations, the system employs a tiered response mechanism. Upon detecting a high-risk level, the system bypasses manual intervention, directly triggering the highest level of automated interception. This immediately limits traffic to the involved livestream room, removes infringing products, and cuts off the current transaction link, quickly preventing the risk from spreading. For medium-risk levels, the system first displays a pop-up warning and risk disclosure on the consumer's transaction interface, while simultaneously packaging and pushing the involved content to the backend for manual review and characterization by dedicated auditors. For low-risk levels, the system does not interfere with the current normal transaction process but instead adds the target entity to the backend monitoring list, preventing potential risks by increasing the frequency of periodic re-inspections and dynamically tracking its subsequent related data. All control actions are fully recorded for auditing purposes.

[0049] Reference Figure 5 This paper analyzes the reliable evidence storage and dynamic optimization mechanisms to ensure data credibility and the system's continuous evolution capabilities. In the evidence storage phase, the system adopts a consortium blockchain architecture, uniting platform operators, regulatory authorities, and third-party certification bodies as consensus nodes. Verification records, control logs, and feedback information in the transaction chain must be confirmed by multiple consensus nodes before being uploaded to the blockchain for solidification, thus ensuring the absolute immutability of data and providing multiple parties with clearly defined access points for traceability and querying. In the optimization phase, the system is not static but periodically collects newly added risk cases, consumer feedback, and changes in regulatory policies from across the network. The underlying deep reinforcement learning model processes this new data and autonomously adjusts the internal logic, weight allocation, and threshold standards of the cross-validation rule base. Updated rules undergo small-scale gray-scale testing to confirm their accuracy and stability before being rolled out to the entire network, ensuring that environmental protection capabilities keep pace with the times.

[0050] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for constructing a green live-streaming trusted transaction environment based on cross-validation of e-commerce data, characterized in that, Includes the following steps: Collect multi-source e-commerce data from three sources: the live streaming end, the platform end, and the third-party end of the online transaction chain of live e-commerce; Standardized preprocessing is performed on the collected multi-source e-commerce data, including unified data format conversion, duplicate data removal, missing value completion, compliance verification, and removal of invalid data; A multi-dimensional cross-validation rule base is constructed, covering four core validation dimensions: cross-validation of anchor and merchant qualifications and platform filing information, cross-validation of product green attribute parameters and third-party certification reports, cross-validation of transaction orders, payment records and logistics information, and cross-validation of live broadcast content and actual product parameters. Based on the rule base, the associated data in the dataset to be verified is cross-verified item by item to enter the transaction link, and a credibility score is generated. The subjects, products and transaction link nodes corresponding to the associated data with scores below the qualified threshold are marked with corresponding risk levels. Tiered control measures are implemented based on risk level: high-risk measures include limiting live stream traffic, removing products from shelves, and blocking transaction links; medium-risk measures include publicizing risk information, issuing pop-up notifications to consumers, and manually reviewing the content involved; and low-risk measures include regular re-inspections and dynamic tracking of related data. Construct a trusted ledger for the entire transaction chain, synchronously upload evidence to the blockchain to cross-verify the process and results, risk control actions, and transaction feedback records, and provide access-controlled query portals to consumers, regulatory authorities, and platform operators; We continuously collect data on consumer feedback, new risk cases, and updated regulatory standards, and dynamically adjust the rule base's verification logic, weight parameters, and qualification thresholds.

2. The method of claim 1, wherein the method is characterized by, This also includes the cross-validation process between product green attribute parameters and third-party certification reports, introducing a comprehensive and reliable quantitative calculation logic for product green attributes. The calculation formula is as follows: ;in This indicates the overall credibility score of the product's green attributes. This represents the weighting coefficient corresponding to the third-party carbon footprint certification results. This represents the quantitative value of a third-party carbon footprint certification result. This indicates the weighting coefficient corresponding to the consistency verification between the materials displayed in the live broadcast and the certification parameters. This represents the quantitative value used for verifying the consistency of product materials. This represents the weighting coefficient corresponding to historical consumer evaluations related to green attributes. This represents the quantitative value of historical consumer evaluations related to green attributes.

3. The method of claim 1, wherein the method is characterized by, This also includes introducing a quantitative calculation logic for the subject's risk level in the risk label generation process. The calculation formula is as follows: ;in This represents the quantitative value of the subject's risk level. This represents the weight coefficient corresponding to the currently unqualified cross-validation item. This represents the quantified value of the number of non-compliant items in the current cross-validation. This represents the weighting coefficient corresponding to the subject's historical risk records. This represents the quantified value of the subject's historical risk records. This represents the weighting coefficient corresponding to the transaction volume of the main entity over the past 30 days. This represents the quantitative value of the corresponding entity's transaction volume over the past 30 days.

4. The method of claim 1, wherein the method is characterized by, During the multi-source e-commerce data collection process, for live streaming data, a combination of real-time frame interception and pre-trained multimodal semantic recognition models is used to extract structured information from the anchor's speech, product display content, and product details page. For platform data, a dedicated interface call is used to obtain anonymized transaction data, qualification filing data, and after-sales rights protection data. For third-party data, a consortium blockchain node real-time synchronization method is used to obtain tamper-proof certification reports, industry standards, and public evaluation data.

5. The method of claim 1, wherein the method is characterized by, In the standardized preprocessing of multi-source e-commerce data, structured data is processed by using unified field mapping, outlier removal, and missing values ​​filled with the median value of the same type of data in the same dimension. Unstructured video, audio, and text data are converted into standardized structured feature data using a pre-trained multimodal feature extraction model. All preprocessed data undergoes secondary compliance verification, and the generated dataset to be verified can be directly input into the cross-validation process for processing.

6. The method of claim 1, wherein the method is characterized by, In the process of building a multi-dimensional cross-validation rule base, the first step is to introduce the compliance standards for live-streaming e-commerce, green product certification standards, and transaction security standards issued by regulatory authorities as basic rules. Then, customized category-specific rules are added by combining the platform's historical risk cases and the compliance requirements of third-party certification agencies. All rules are configured with flexibly adjustable weight parameters to support the adaptation logic of adjusting rules for different product categories and different types of live-streaming scenarios.

7. The method of claim 1, wherein the method is characterized by, During the risk label generation process, a corresponding traceability certificate is generated for each item that fails cross-validation. The traceability certificate contains complete information in four dimensions: the specific data item that failed validation, the data source, the comparison standard, and the judgment basis. All traceability certificates are bound and stored one by one with the corresponding subject, product, and transaction node.

8. The method of claim 1, wherein the method is characterized by, During the execution of tiered control actions, control actions for high-risk entities are directly triggered by the system to be executed automatically, which can quickly block the spread of risks without human intervention. Control actions for medium-risk entities are first pushed to the platform's review personnel for manual confirmation before execution. Control actions for low-risk entities are only marked in the background and do not affect the normal operation of the current transaction chain. All execution records of control actions are synchronously stored in the trusted evidence storage ledger.

9. The method of claim 1, wherein the method is characterized by, The end-to-end trusted evidence storage ledger is built using a consortium blockchain architecture. The platform operator, regulatory authorities, and third-party certification authorities serve as consensus nodes on the consortium blockchain. All evidence storage data must be jointly confirmed by at least three consensus nodes before it can be stored on the blockchain. Consumers can query the end-to-end verification record of the corresponding transaction through the transaction order number. Regulatory authorities can obtain all risk data and verification data for supervision. Third-party certification authorities can only query the verification records related to the products they have certified.

10. The method of claim 1, wherein the method is characterized by, During the dynamic iterative optimization of the cross-validation rule base, new risk case data, consumer feedback data, and regulatory update data are collected at fixed intervals. A deep reinforcement learning model is used to iteratively adjust the validation logic, weight parameters, and qualification threshold standards of the rule base. The adjusted rules are first tested in a small-scale gray-scale test to verify their effectiveness and rationality before being fully launched.

Images