Trust feedback driving method and system based on OID traceability data

Abstract

The present application relates to the technical field of artificial intelligence, in particular to a trust feedback driving method and system based on OID traceability data, the method comprising: recording consumer traceability verification behavior statistical information through product OID two-dimensional code; constructing a consumer individual trust portrait and a brand aggregated trust portrait, and generating a brand trust comprehensive evaluation index; inputting the trust index as a dynamic weight factor into an AI recommendation engine; dynamically optimizing the recommendation ranking of the brand in each scene; and also corresponding to a system comprising a product OID traceability interaction and behavior recording module, a consumer trust portrait construction module, a trust portrait and recommendation engine fusion module, an AI recommendation strategy optimization module, a data visualization and attribution analysis module, and a trust model adaptive optimization module. The present application upgrades OID traceability data from an anti-fake tool to an AI recommendation trust signal, and takes the trust behavior of the consumer end as a recommendation weight factor, thereby significantly improving the consistency of the AI recommendation result and the actual trust degree of the consumer.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a trust feedback-driven method and system based on OID traceability data. Background Technology

[0002] As generative AI becomes a primary entry point for users to obtain information and make purchasing decisions, AI recommendation engines such as Doubao, DeepSeek, and ChatGPT are playing an increasingly crucial role in brand exposure and product marketing. Current GEO (Generative Engine Optimization) technology mainly focuses on optimizing the semantic structure of brand content within large AI models and enhancing the authority of information sources. However, existing technologies suffer from the following fundamental shortcomings: AI Recommendations are disconnected from genuine trust: Current AI recommendations primarily rely on the semantic relevance of content and static authoritative ratings from sources, failing to perceive genuine trust verification behaviors by consumers, such as scanning codes for authenticity, tracing and inspection, and user confirmation. This leads to AI potentially recommending brands with well-written content that consumers don't trust, or missing out on brands that consumers highly trust but whose content optimization is insufficient.

[0003] Lack of a trust loop: An ideal marketing ecosystem should be a closed loop encompassing AI recommendations, consumer purchases / verification, trust feedback, and AI-optimized recommendations. However, this loop has not yet been established in current technologies. Consumers' trust behavior towards a brand cannot influence the AI's future recommendation weight for that brand, resulting in recommendation quality failing to continuously improve based on genuine user feedback.

[0004] Trust accumulation cannot be quantified: Brands invest heavily in improving product quality and transparency, but these trust assets cannot be quantified and utilized by AI recommendation engines, making it difficult for brands to obtain AI recommendation traffic that matches their true credibility. Summary of the Invention

[0005] The purpose of this invention is to provide a trust feedback-driven method and system based on OID traceability data, which transforms the OID traceability verification behavior of consumers into trust signals that can be received by AI recommendation engines, uses the most authentic trust behavior of consumers as recommendation weight factors, and then transforms it into AI recommendation traffic.

[0006] To achieve the above objectives, this invention provides a trust feedback-driven method based on OID traceability data, comprising the following steps: Step S1: Assign a product-level OID identification code to each smallest SKU of the brand's products and display it via a QR code; when consumers scan the QR code to view the product's full-chain traceability file, record the consumer's traceability behavior statistics. Step S1: Product OID traceability interaction design and consumer behavior statistics recording Assign a unique product-level OID identifier to each smallest stock unit (SKU) of the branded products, and display the OID identifier in the form of a QR code on product packaging, electronic tags or marketing materials; When a consumer scans the OID QR code using a mobile anonymization device, the consumer's anonymization device displays the product's full-chain traceability file, which includes at least two of the following information: production and processing parameters and quality inspection reports; third-party testing and certification certificates; warehousing and logistics trajectory and monitoring records; terminal sales store information; and raw material procurement source and batch information.

[0007] During the display of traceability information, statistical information on consumer interaction behavior on the traceability page is recorded, including at least three of the following types: Basic interaction records: QR code timestamp, product OID code, and anonymized device; Verify in-depth information: the total time consumers spend on the traceability page, the names of the traceability modules they view and the order in which they clicked, and whether they have fully viewed the entire chain of information; Active confirmation record: Statistics on the number of times consumers click interactive buttons such as confirm authenticity and trust recommendation; only behavioral events are recorded. Social dissemination records: Statistics on sharing behavior events of the source tracing results and the number of secondary QR code scans resulting therefrom. Only aggregated data of dissemination events are recorded, without associating social relationship identity information between sharers and visitors. All records are stored with anonymized devices and event IDs. Feedback and evaluation records: Only the rating values ​​submitted by consumers are retained for calculating trust scores.

[0008] All the above-mentioned behavioral statistics are anonymized during recording, with personal identification information replaced by an anonymous device identifier.

[0009] Step S2: Based on the traceability behavior statistics recorded in Step S1, construct individual consumer trust profiles and aggregated trust profiles at the brand / product level, and generate a comprehensive brand trust evaluation index. Based on the multi-dimensional traceability behavior statistics recorded in step S1, a weighted scoring model is used to construct individual consumer trust profiles and aggregated trust profiles at the brand / product level. The individual consumer trust profile for each consumer includes at least three of the following rating dimensions: Verification depth rating D i The score is calculated based on the completeness of the traceability records viewed by consumers. The more modules viewed and the longer the time spent on each module, the higher the score. The calculation formula is as follows: D i =( 1 / n )×Σ(w j × c ij ),in c ij Consumers i Have you checked the first one? j Each module w j For module j The weights; Verification frequency score F i Based on the cumulative number of scans by the same consumer for the same brand or product SKU within a unit of time, an exponential decay model is used to reflect the trend of trust accumulation. Actively confirm rating A i : Reverse correction calculation based on the number of positive actions such as consumers clicking to confirm authenticity and trust recommendations, and negative actions such as negative feedback; Social Communication Score Si: Calculated based on the number of consumer-shared traceability results and the resulting secondary QR code scanning visits. The communication influence is measured by the number of secondary QR code scans generated by the sharing event, without tracking the identity of specific visitors or their relationship with the sharer.

[0010] Loyalty rating L i It reflects brand loyalty based on the accumulated historical traceability behavior of consumers towards different products of the same brand; Among them, the aggregated trust profile at the brand / product level is obtained by aggregating and calculating the trust profiles of all individual consumers: Furthermore, in step S2, the brand trust comprehensive evaluation index is calculated by weighted aggregation of trust scores from all individual consumers, and the calculation formula is as follows: In the formula, M The total number of effective consumers. D i To verify depth rating, F i for Verification frequency score A i To proactively confirm the score, S i for Social communication score L i Rate loyalty a+b+c+d+e=1, and a, b, c, d, e ∈ (0, 1).

[0011] Step S3: Transmit the comprehensive brand trust evaluation index and the decomposed indicators of each dimension to the AI ​​recommendation feedback system through a standardized interface as feature input for the AI ​​recommendation engine; Furthermore, the verification depth score D i The formula for calculating the completeness of the traceability records accessed by consumers is as follows: D i =( 1 / n )×Σ( w j × c ij ), Where n is the total number of modules in the traceability archive. c ij Consumers i Have you checked the first one? j Each module w j For module j The weight.

[0012] Step S3: Integrate trust profile data with the GEO / AI recommendation engine The brand trust comprehensive evaluation index Tbrand and its decomposed indicators generated in step S2 are transmitted to the AI ​​recommendation feedback system in real time or near real time through a standardized API interface or data pipeline; the transmission process includes: Data aggregation and cleaning: Time window aggregation is performed on the raw QR code scanning behavior statistics to remove abnormal data; Feature mapping: Mapping trust metrics into feature vectors that the AI ​​recommendation engine can receive, wherein the feature vectors include: As a dynamic correction factor of Trustworthiness in the EEAT (Experience, Expertise, Authority, Trust) signal; As a new feature weight for the brand in GEO content optimization, it affects the semantic vector weighting in step S4. Data synchronization: Aggregated trust data is pushed to the online feature storage of the AI ​​recommendation engine via message queues or real-time data lakes for the recommendation model to read in real time.

[0013] Before transmitting trusted data to a third-party AI recommendation engine, privacy protection measures are applied to the individual consumer's behavioral statistics. Only the city features of the anonymized devices used to generate the geolocation heat map for QR code scanning are transmitted, and not for individual identification. The original QR code scanning records of individual consumers are not transmitted. Individual trust profiles are used only for internal weight calculations and do not leave the boundaries of this system. If individual-level characteristics need to be transmitted, explicit authorization from the consumer is required, and differential privacy noise reduction processing must be performed.

[0014] Step S4: The AI ​​recommendation engine uses the comprehensive trust evaluation index as a dynamic weighting factor, combines it with semantic relevance score and source authority score, to calculate the brand's final recommendation score; and dynamically optimizes the brand's recommendation ranking in various generative AI application scenarios based on the final recommendation score. Input the comprehensive trust evaluation index and trust feature vector transmitted in step S3 into the feature weight matrix of the AI ​​recommendation system to dynamically optimize the brand recommendation strategy in one or more of the following application scenarios: Dialogue-based recommendations: When consumers ask category recommendation questions on the AI ​​dialogue platform, The recommendation engine incorporates trust score weights into the ranking of candidate brands. T brand Brands exceeding the threshold will be given priority for inclusion in the recommended candidate list; Furthermore, the formula for calculating the final recommendation score in step S4 is as follows: Score final =λ 1 · Score semantic +λ 2 · Score authority +λ 3 · T brand _norm in Score semantic This is the content semantic relevance score, with a value range of [0, 1]. Score authority This represents the source authority score, with a value range of [0, 1]. T brand _norm The brand trust comprehensive evaluation index is standardized by Min-Max, with a value range of [0, 1]; λ1+λ2+λ3=1, and λ1, λ2, λ3∈(0, 1).

[0015] In one implementation, for newly launched brands or products with insufficient QR code sample size, i.e., M < minimum effective sample threshold (e.g., M < 100), one or a combination of the following cold start strategies are adopted: Use the industry average trust score for similar products / originating countries as the initial... T brand ; Trust weighting will not be enabled for now; λ3 will be set to 0, and the weights will be gradually increased after enough samples have been accumulated. Introduce certification endorsement as a substitute for points: such as obtaining organic certification, GAP certification, etc., to give basic trust points.

[0016] AI recommendation scenarios on e-commerce platforms: In AI recommendation modules such as "You May Like," "Similar Products," and "Bestseller Lists" on e-commerce platforms, highly trusted brands are given a weighted boost, breaking through the limitations of recommendations based solely on sales volume or click-through rates, and giving brands with excellent quality but low brand awareness an opportunity to gain exposure.

[0017] Localized recommendation scenario: By combining the distribution of city-level location information inferred from IP addresses based on consumers' scanning behavior, a scanning heat map is generated; when the number of consumers in a certain region scanning and verifying a certain brand is significantly higher than in other regions, the corresponding brand is given a regional weighting boost in the AI ​​recommendation results for that region, achieving a precise match from trust in the place of origin to priority recommendation based on the place of origin.

[0018] Social trust recommendation scenario: When someone in a consumer's social network scans and verifies a product and shares it, the recommendation weight of that product is enhanced in conversational AI recommendations, realizing personalized recommendations based on the transmission of social trust.

[0019] Step S5: Record the trust feedback closed-loop data, generate a visual analysis dashboard, and show the temporal correlation and attribution path between QR code verification behavior indicators and AI recommendation exposure conversion indicators.

[0020] The entire feedback loop data (from consumer scanning to AI recommendation strategy optimization, and then to recommendation effect feedback) is processed through data visualization to generate a brand trust feedback effect analysis dashboard. The dashboard displays at least two of the following: The correlation curve between the growth trend of QR code verification and the exposure conversion rate of AI recommendations demonstrates the traffic increase effect brought about by trust feedback. Heatmap comparing the distribution of consumer trust across regions / channels with the regional weighting effect of AI recommendations; The system provides a complete attribution path analysis, from consumer scanning to AI recommendation and ranking improvement, and finally to purchase conversion, supporting multi-touchpoint attribution models. Analysis of the contribution of different consumer groups, including high-frequency validators, social disseminators, and silent validators, to the improvement of AI recommendation weight; The return on investment (ROI) of trust feedback includes a comparison between the cost of QR code verification and the value of incremental traffic from AI recommendations.

[0021] Furthermore, the method for enhancing AI recommendations based on consumer trust fed back from OID traceability data also includes: Step S6: Using the attribution analysis data generated in step S5, a reinforcement learning framework is used to periodically and adaptively adjust the trust scoring model weights in step S2 and the recommendation fusion weights in step S4.

[0022] Step S6: Continuous learning and adaptive optimization of the trust model. Using the attribution analysis data generated in Step S5, the trust scoring model weights α, β, γ, δ, ε in Step S2 and the recommendation fusion weights λ1, λ2, λ3 in Step S4 are periodically and adaptively adjusted. A reinforcement learning framework is adopted to balance trying new weight combinations with maintaining the validated effective weights, thereby continuously improving the effect of trust feedback on the final marketing conversion rate.

[0023] In one embodiment, the reinforcement learning framework adopts the multi-armed slot machine (MAB) framework, and uses the attribution analysis data generated in step S5 to periodically and adaptively adjust the trust scoring model weights in step S2 and the recommendation fusion weights in step S4. Furthermore, the adaptive adjustment method includes: mapping multiple preset weight combination schemes to candidate arms in a multi-armed slot machine framework, using the additional conversion gain brought by trust feedback within a period as immediate reward feedback, and using the Thompson Sampling algorithm to dynamically update the posterior reward distribution of each arm; at the beginning of each period, sampling is performed from the posterior distribution of each arm, selecting the weight combination scheme with the largest sample value for execution, while randomly exploring other candidate schemes at a preset exploration rate ε to avoid getting trapped in local optima.

[0024] Furthermore, the localized recommendation in step S4 refers to generating a QR code heat map based on the distribution of city-level location information inferred from IP addresses based on consumer QR code scanning behavior. When the number of QR code verifications for a certain brand by consumers in a region exceeds a preset threshold of the total number of QR codes scanned in that region, the corresponding brand is given a regional weighted boost in the AI ​​recommendation results for that region.

[0025] Furthermore, the social trust recommendation in step S4 refers to, With the authorization and consent of both the first and second consumers, when the first consumer's sharing behavior triggers the second consumer's scanning behavior, the brand products corresponding to the sharing behavior will be given a higher weight in the recommendation results during the second consumer's subsequent category search.

[0026] Specifically, with the user's authorization, based on the sharing and propagation path tracking, when it is detected that the first consumer with the anonymized identifier has scanned and verified a product and shared the traceability results, and the second consumer with the anonymized identifier has completed the scan verification through the shared content, then in the second consumer's subsequent category search, the product will be given a weighted increase in the recommendation results; The sharing propagation path tracking only records the association between the sharing event ID and the anonymous identifier of the secondary scanning device. The sharing behavior and secondary scanning access only record the anonymized device and the sharing event ID, and do not acquire, store, or upload the user's address book, friend list, or any external social relationship data.

[0027] This invention also provides a consumer trust feedback AI recommendation enhancement system based on OID traceability data, comprising: Product OID traceability interaction and behavior recording module: used to assign OID identifiers to each product SKU, provide a traceability profile display interface, and record consumer scanning behavior statistics; Consumer Trust Profile Building Module: Used to build individual consumer trust profiles and aggregated brand / product trust profiles based on recorded behavioral statistics, and generate a comprehensive brand trust evaluation index; Trust profile and recommendation engine fusion module: used to transmit trust index and feature vector to AI recommendation feedback system through standardized interface; AI Recommendation Strategy Optimization Module: Used to dynamically adjust the recommendation ranking weight of brands in at least one of the following scenarios: dialogue recommendation, e-commerce recommendation, regional recommendation, and social recommendation, based on the trust index; Data visualization and attribution analysis module: used to generate a data analysis dashboard for a trust feedback loop; Trust Model Adaptive Optimization Module: Used to continuously adjust the weight parameters of the trust scoring model and the recommendation fusion model based on attribution feedback.

[0028] In one embodiment, the Thompson Sampling algorithm is used for weight updates: Each weight combination α, β, γ, δ, ε is considered as an arm, and its reward function is defined as the weekly conversion rate improvement rate. The posterior distribution is updated every 7 days, and the weight combination with the highest expected reward is selected for the next cycle. Set the exploration rate ε=0.1 to randomly explore new combinations with a 10% probability, thus avoiding getting trapped in local optima.

[0029] Furthermore, the aforementioned consumer interaction behavior statistics information includes at least three of the following: basic interaction records, verification depth information, active confirmation records, and social dissemination records.

[0030] Furthermore, the dialogue scenarios include dialogue recommendation scenarios, e-commerce recommendation scenarios, regional recommendation scenarios, and social recommendation scenarios.

[0031] This invention presents a trust feedback-driven method and system based on OID traceability data. It uses the OID traceability system as a data source for AI recommendation trust, and incorporates the most authentic and direct trust behavior from consumers as recommendation weight factors. This significantly improves the consistency between AI recommendation results and actual consumer trust levels. After scanning a code, consumers' trust behavior helps the brands they trust receive more recommendations, creating a positive incentive and increasing consumers' initiative and engagement in traceability interactions. Trust scores, based on code verification behavior, are less susceptible to interference from fake positive reviews, enhancing the robustness of the recommendation system. It also allows brands to see how the consumer trust behavior resulting from investments in traceability system construction and quality improvement translates into AI recommendation traffic and commercial conversion, thus forming a virtuous cycle of quality improvement, trust accumulation, traffic growth, and increased sales. Attached Figure Description

[0032] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0033] Figure 1 This is a flowchart of a trust feedback-driven method based on OID traceability data according to the present invention.

[0034] Figure 2 This is an architecture diagram of a trust feedback-driven system based on OID traceability data according to the present invention.

[0035] Figure 3 This is a recommendation logic diagram of a trust feedback-driven system based on OID traceability data according to the present invention.

[0036] Figure 4 This is a correlation curve between the growth of brand QR code verification and AI recommendation exposure conversion in a trust feedback driven system based on OID traceability data, as described in this invention. Detailed Implementation

[0037] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0038] Please see Figure 1 to Figure 4This invention provides a trust feedback-driven method and system based on OID traceability data. Specifically, a domestic ready-to-eat meal supply chain brand has deployed OID: coded as 1.2.156.47005.1014 and used the system of this invention from April to June 2026. The following is the complete implementation process and effect data.

[0039] Step S1 Implementation: Product OID Traceability Interaction Design and Behavior Recording The brand prints an OID QR code on each product package. Consumers can scan the code with WeChat to access a traceability mini-program to view the product's origin. Raw material source: Carrots and potatoes from an organic farm; photos of the farm and organic certification are attached. Production batch: March 15, 2025, production workshop cleanliness monitoring record. Quality Inspection Report: SGS Report of 466 Pesticide Residues Not Detected Logistics trajectory: Temperature-controlled warehouse, cold chain transportation, store shelf placement The system recorded consumer behavior statistics from April to June 2026: Total number of QR code scans: 128,452 Number of unique consumers: 62,547 Average dwell time: 47 seconds Percentage of users who viewed the entire browsing journey: 32.4% Number of times users actively clicked on trusted recommendations: 34,281 Shared on social media platforms: 12,743 times Step S2 Implementation: Consumer Trust Profile Construction and Scoring A weighted scoring model is adopted, and the dimensions are defined as follows: Verification depth rating D i The score is calculated based on the completeness of the traceability profile reviewed by consumers. This brand's traceability profile consists of six modules: raw material origin, production batch, quality inspection report, logistics trajectory, certification certificates, and company qualifications. The more modules reviewed and the longer the consumer spends on the profile, the higher the score. Calculation formula: D i =(1 / 6)×Σ(number of modules consulted)×time of stay normalization factor. The average value D=0.7, and the score after standardization is 70.

[0040] Verification frequency score F i The score is calculated based on the cumulative number of times a consumer scans the same brand's code within 3 months, using an exponential decay model. The score is 0.5 for the first scan, 0.8 for the second, and 1.0 for the third and subsequent scans. The average score F = 0.58, which is 58 points after standardization.

[0041] Actively confirm rating A i The score is calculated based on the following: +1 point for consumer click confirming authenticity, +2 points for trust recommendation, and -5 points for negative feedback. The average score A = 0.786, and the standardized score is 78.6.

[0042] Social Communication Score S i Based on sharing behavior, sharing to WeChat friends / Moments earns +2 points, and a second scan resulting in a second QR code earns an additional +1 point. The average score S = 0.60, which is 60 points after standardization.

[0043] It is important to emphasize that this rating does not access the sharer's friend list or social relationship graph; it only uses the frequency of the sharing behavior and the number of secondary scans as the rating criteria.

[0044] Loyalty rating L i Based on the same consumer's scanning records of different products from the brand, purchasing more than 2 SKUs indicates high loyalty. The average L = 0.53, and the standardized score is 53.

[0045] With weights set α=0.2, β=0.2, γ=0.3, δ=0.2, and ε=0.1, calculate the comprehensive brand trust evaluation index: =14+11.6+23.58+12+5.3=66.48 points (out of 100).

[0046] Additional heatmap of QR code scanning distribution: This brand has the highest QR code scanning density in East China (especially Shanghai and Hangzhou), accounting for 47% of the total QR code scanning.

[0047] Step S3 Implementation: Integration of Trust Profile and AI Recommendation Engine The system aggregates the trust data from the previous day every day at midnight. T brand =66.48 points, with breakdown indicators across various dimensions and regional heatmap data, which are pushed via API to the AI ​​recommendation engines of partner e-commerce platforms and the content recommendation interfaces of Doubao and DeepSeek.

[0048] Integration method: E-commerce recommendation: T brand As a quality score signal, it is input into the recommendation model along with sales volume and positive review rate (weighting 20%).

[0049] AI-powered recommendations: Using regional heat map data as regional preference features to answer regional questions such as where to buy safe organic baby food in Shanghai.

[0050] Step S4 Implementation: Dynamic Optimization of AI Recommendation Strategy Based on Trust Profile Before joining this system in April 2026, the brand was usually ranked below 20th in the list of recommended ready-to-eat meal supply chains on the JD.com platform, with limited exposure.

[0051] After integration, the system will dynamically adjust: In the "You May Also Like" section on JD.com's homepage for users in East China, this brand's recommended position rose from 25th to 8th (trust index weighting factor λ3=0.3).

[0052] When users in Shanghai asked Doubao for recommendations on ready-to-eat meal supply chains, this brand appeared as the second-ranked answer, down from fifth place or later. Example of recommendation score calculation: Score final =0.4 · Score semantic （0.75）+0.3 · Score authority （0.70）+0.3 · T brand _norm (0.66)= 0.30 + 0.21 + 0.20 = 0.71, which is higher than the average of 0.58 of similar competing products.

[0053] Step S5 Implementation: Trust Feedback Closed-Loop Data Visualization and Attribution Analysis Results after 3 months: QR code verification volume: increased from an average of 710 times per day in April to an average of 1,420 times per day in June, an increase of 100%; AI recommendation exposure across JD.com, Doubao, and DeepSeek combined increased from an average of 23,000 times per month in March to an average of 40,000 times per month in June, representing a growth of 73.9%. Purchase conversion rate: increased from 0.7% to 1.9%, with the conversion rate in trust-based recommendation scenarios reaching as high as 3.2%; In this scenario, consumers have completed OID traceability verification, establishing direct cognitive trust in the brand's product quality, thus exhibiting a higher purchase intention when receiving AI recommendations. This data difference verifies the effectiveness of the technical path of this invention, which transforms OID traceability verification behavior into trust signals for AI recommendations. Consumers who have undergone traceability verification have a 1.68 times higher response rate to recommendations than ordinary consumers (3.2% / 1.9%), proving that the injection of trust signals can indeed improve the commercial conversion efficiency of AI recommendations.

[0054] Attribution analysis shows that the trust feedback path, namely the additional exposure directly resulting from the increased trust weight in AI recommendations, contributed 41% of the total new exposure and 37% of the new sales.

[0055] Step S6 Implementation: Adaptive Optimization of the Trust Model Based on three months of data, the system automatically adjusted the weights of the trust scoring model using a multi-armed slot machine strategy: it found that the proactive confirmation score A contributed the most to the final conversion rate, so its weight γ was increased from 0.3 to 0.4; correspondingly, the weight β of the verification frequency F was reduced from 0.2 to 0.15. After the adjustment, T brand The correlation coefficient with conversion rate increased from 0.53 to 0.67, improving the model's prediction accuracy. Specific Implementation Example 2: Specific Implementation Example 2: Trust Feedback from a Certain Health-Preserving Wine This embodiment takes a certain health wine as an example. The manufacturer of this health wine has registered an OID mark with the National OID Registration Center, with the code: 1.2.156.40000.350006. This means that the behavior of consumers scanning the code to verify traceability enhances AI recommendation.

[0057] Step S1: OID Traceability Interaction and Behavior Collection for Health-Preserving Wine Each product package of this health tonic wine is printed with an OID QR code linked to the product's traceability code, and the coding follows the national OID identification system standard. Consumers can scan the code using platforms such as WeChat and Alipay to view the product's traceability information. The product authenticity verification results show that the first scan indicates it is genuine, while the second scan indicates it has already been verified.

[0058] Raw material source and origin information: Grains and medicinal materials from the planting base in Jinshan Town, Guyang County, with origin certification information attached; Production batch information includes: production date, production workshop, alcohol content, and net content; The quality inspection report is: Certificate of Product Inspection Conformity; The company's qualifications include: a production license and other wine production licenses, including blended wines, other distilled spirits, and other fermented wines; The system collects behavioral statistics: From April to June 2026, a total of 84,732 QR code scans were collected. Based on device ID hash calculations, the number of unique consumers is approximately 54,129. Average dwell time: 43 seconds A viewing session lasting ≥30 seconds and reviewing ≥4 information modules is considered a complete review of the traceability file, representing 36% of the total. The number of times the "confirm product authenticity" button was clicked was 46,231. The number of times suspicious activity was reported was 87. The number of times product traceability information was shared with friends and family was 7845. Percentage of consumers who scan the QR code ≥ 2 times per month: 15.6% Step S2: Building a Trust Profile for Health-Preserving Wine Consumer individual trust profile dimensions: Verification depth rating Di The breakdown of points for each section is as follows: 0.5 points for checking raw material source information, 0.5 points for checking production batch information, 1 point for checking quality inspection reports, 0.5 points for checking company qualifications, and 0.5 points for watching product culture / process videos, for a total of 3 points. Statistics show that consumers viewed an average of 2.4 modules and spent an average of 43 seconds on each module. The overall average score was calculated to be Di = 2.05 points. The standardized score, D, was 68 points. The standardization method is: D = 2.05 / 3 × 100 ≈ 68.3 points.

[0059] Verification frequency score Fi 1 point is awarded for scanning the QR code ≥ 2 times per month, and 2 points are awarded for scanning the QR code regularly for 3 consecutive months. Statistics show that 15.6% of consumers scan the QR code ≥ 2 times per month, and 8.2% scan the QR code regularly for 3 consecutive months. The average score is calculated to be Fi = 0.52 points. The standardized score is F = 52 points. The standardization method is: F = 0.52 / 1 × 100 = 52 points.

[0060] Actively confirm rating Hey Clicking to confirm authenticity adds 1 point, reporting suspicious activity deducts 5 points. Statistics show that with 46,231 confirmed authenticity actions and 87 suspicious activity reports, the average net score per consumer is 0.82 points. The standardized average score (A) is 82 points. The standardization method is: A = 0.82 / 1 × 100 = 82 points.

[0061] Social Communication Score Yes Sharing traceability information with friends and family earns +2 points, and generating a second scan earns an additional +1 point. Statistics show 7845 sharing behaviors, averaging 0.55 shares per consumer. The standardized score is S=55 points. The standardization method is: S=0.55 / 1×100=55 points, with a maximum score of 1 point as the baseline.

[0062] Category loyalty score Li For the same consumer scanning the barcodes of different series of health wine products multiple times, an additional 0.3 points are awarded for each additional SKU. Statistics show that the average number of SKUs scanned by consumers is 0.46, with an average Li of 0.46 points. Standardized to a percentage, L = 46 points. The standardization method is: L = 0.46 / 1 × 100 = 46 points, with a maximum score of 1 point as the baseline.

[0063] Weight settings: α=0.25, β=0.15, γ=0.30, δ=0.20, ε=0.10.

[0064] The overall brand trust rating index is calculated directly using the weighted average of all dimensions: T brand = α·D + β·F + γ·A + δ·S + ε·L = 0.25×68 + 0.15×52 + 0.30×82 + 0.20×55 + 0.10×46 = 17.0 + 7.8 + 24.6 + 11.0 + 4.6 = 65.0 points This score is significantly higher than the average of similar health wine brands, mainly due to the high rate of proactive confirmation and the depth of verification.

[0065] In addition, the system generated a heat map showing the geographical distribution of QR code scans: the health wine had the highest scan density in Inner Mongolia Autonomous Region, especially in Baotou and Hohhot, accounting for 43% of the total scans; followed by Beijing at 18% and Shanghai at 12%, showing the radiation effect of the place of origin and the concentrated characteristics of health consumers in first-tier cities.

[0066] Step S3: Integration of Trust Profile and AI Recommendation Engine Will T brand =65.0 points and all dimensional indicators are pushed in real time via standard API to: Baidu Health AI Assistant JD Health / Tmall Health AI Recommendation Engine Douyin e-commerce AI recommendation system Conversational AI platforms such as Doubao and DeepSeek Integration method: E-commerce recommendation: T brand =65.0 is used as a quality trust score signal, which is input into the recommendation model along with sales volume and positive review rate, with an initial weight of 20%.

[0067] AI-powered recommendations: Using regional QR code heatmap data as regional preference features to answer regional questions such as where to buy authentic health tonics in Inner Mongolia / Beijing.

[0068] Step S4: Dynamically optimize the AI ​​recommendation strategy based on trust profiles In health consultation scenarios: When consumers consult an AI health platform about which brand of health-preserving wine is good or seek recommendations for health-enhancing wines, the system identifies the consumer's geographical location, such as Inner Mongolia. Combined with the high QR code scanning density of the health-preserving wine in Inner Mongolia, a comprehensive trust evaluation index is generated. T brandThe fusion weight was dynamically increased from the default value of 0.30 to 0.35. After the adjustment, the ranking of this health wine in the recommendation list improved from 12th to 4th. AI response example: According to the verification data of 54,000 consumers scanning codes, Baotou Longju health wine ranked high in verification volume, with a total of 85,000 scans, complete traceability information, and a high rate of active consumer confirmation.

[0069] Step S5 Effect Analysis and Verification Experimental Design To verify the technical effectiveness of the method of the present invention, a control group and an experimental group were set up for comparative experiments: Before integration, serving as a control group: From January to March 2026, this health wine was not integrated into the trust feedback AI recommendation system, and was only recommended based on traditional sales volume and positive review rate ranking.

[0070] After integration, it served as an experimental group: From April to June 2026, this health-preserving wine was integrated into the system of this invention. T brand =65.0 points are injected into various AI recommendation engines in real time.

[0071] The external market environment was basically the same during the two experiments, making them comparable.

[0072] Comparison of core performance indicators The average monthly exposure of AI recommendations was 18,000 times in the control group and 60,000 times in the experimental group, an increase of 233.3%, indicating that the injection of trust profiles significantly enhanced the AI ​​recommendation system's push for this health wine.

[0073] Average ranking on the recommendation list: The control group's average ranking was 12th, while the experimental group rose to 4th, an improvement of 8 places, indicating a strong overall trust rating index. T brand The weighting effectively improved the presentation position of this health wine in the AI ​​recommendation results.

[0074] Purchase conversion rate after QR code verification: The control group was 3.2%, while the experimental group increased to 5.7%, an increase of 78.1%, proving that consumers reached through the trust feedback path have a significantly higher purchase intention than consumers reached through traditional recommendation channels.

[0075] The average monthly number of QR code verifications was 28,200 in the control group and increased to 60,000 in the experimental group, an increase of 112.8%, reflecting that the increase in exposure to AI recommendations effectively drove the growth of consumers' proactive QR code verification behavior.

[0076] Consumers' behavior of sharing traceability information: The control group shared 2,735 times per month, while the experimental group shared 7,845 times per month, an increase of 186.8%. This shows that after trust is established, consumers have a strong willingness to spread information spontaneously, forming a positive cycle of verification, trust, sharing, and re-verification.

[0077] Attribution analysis After accessing the system of this invention, the consumer behavior path changes significantly: The conversion rates at each stage of the AI-recommended exposure, QR code verification, and purchase conversion funnel have all improved: the QR code conversion rate in the exposure stage increased from 11.2% to 18.6%, and the purchase conversion rate in the QR code stage increased from 3.2% to 5.7%.

[0078] Those who fully reviewed the traceability profile were significantly more likely to make a purchase than those who did not, validating the positive impact of the depth of trust information on purchasing decisions.

[0079] AI recommendation consistency verification Verify the consistency between AI-recommended content and OID traceability information: 200 AI responses were randomly selected during the experiment. After manual review and NLP consistency testing, the information on health wines recommended by the AI, such as place of origin, production batch, and qualifications, was 97.3% consistent with the OID traceability records.

[0080] No cases have been found where the AI-recommended content contradicts the source information.

[0081] Step S6: Model Iteration and Continuous Optimization Data-driven adaptive weight adjustment During its 3-month operation period, the system continuously collected feedback data and automatically optimized the trust scoring model weights using a multi-armed slot machine strategy. The Pearson correlation coefficient (r) between the scores of each dimension and the final purchase conversion rate was calculated as follows: active confirmation score A (r=0.59, the largest contribution), verification depth D (r=0.47), social communication S (r=0.41), repeat purchase verification F (r=0.35), and category loyalty L (r=0.22, the lowest contribution).

[0082] The weights were adjusted based on the correlation coefficients: γ (proactive confirmation score weight) was increased from 0.30 to 0.38; ε (category loyalty score weight) was decreased from 0.10 to 0.08.

[0083] Comparison of model prediction accuracy before and after adjustment using the conversion prediction AUC value as an indicator: AUC before adjustment = 0.74, AUC after adjustment = 0.81, prediction accuracy improved by 9.5%.

[0084] Adjusted T brandRecalculate Adjusted weight configuration: α=0.25, β=0.15, γ=0.38, δ=0.14, ε=0.08.

[0085] T brand (Adjusted) = 0.25×68 + 0.15×52 + 0.38×82 + 0.14×55 + 0.08×46 = 17.0 + 7.8 + 31.16 + 7.7 + 3.68 = 67.34 points Adjusted T brand The correlation coefficient with conversion rate increased from 0.49 to 0.61, significantly enhancing the explanatory power of the model.

[0086] Long-term effect prediction Based on trend extrapolation from 3 months of experimental data, the following predictions are made after 12 months of access to the system of this invention: This health-preserving wine is expected to maintain a stable ranking in the top 3 on the AI ​​health platform. Brand Trust Assets T brand It is expected to improve to the 72-75 point range. Summarize This embodiment fully presents the entire process of how the health-preserving wine (OID code: 1.2.156.40000.350006) uses OID traceability data to feed back into AI recommendations. Experimental data demonstrates that the method of this invention can: Correlation curve between brand QR code verification growth and AI recommendation exposure conversion like Figure 4 As shown, there is a positive correlation between the increase in AI recommendation exposure, the increase in QR code verification, and the improvement in purchase conversion rate, which verifies the effectiveness of the trust feedback mechanism.

[0087] The above-disclosed embodiments are merely preferred embodiments of the present invention and should not be construed as limiting the scope of the invention. Those skilled in the art will understand that implementing all or part of the above-described embodiments and making equivalent changes in accordance with the claims of the present invention are still within the scope of the invention.

Claims

1. A trust feedback-driven method based on OID traceability data, characterized in that, Includes the following steps: Step S1: Assign a product-level OID identification code to each smallest SKU of the brand's products and display it via a QR code; when consumers scan the QR code to view the product's full-chain traceability file, record the consumer's traceability behavior statistics. Step S2: Based on the traceability behavior statistics recorded in Step S1, construct individual consumer trust profiles and aggregated trust profiles at the brand / product level, and generate a comprehensive brand trust evaluation index. Step S3: Transmit the comprehensive brand trust evaluation index and the decomposed indicators of each dimension to the AI ​​recommendation feedback system through a standardized interface as feature input for the AI ​​recommendation engine; Step S4: The AI ​​recommendation engine uses the comprehensive trust evaluation index as a dynamic weighting factor, combines it with semantic relevance score and source authority score, to calculate the brand's final recommendation score; and dynamically optimizes the brand's recommendation ranking in various generative AI application scenarios based on the final recommendation score. Step S5: Record the trust feedback closed-loop data, generate a visual analysis dashboard, and show the temporal correlation and attribution path between QR code verification behavior indicators and AI recommendation exposure conversion indicators.

2. The trust feedback-driven method based on OID traceability data as described in claim 1, characterized in that, In step S2, the brand trust comprehensive evaluation index is calculated by weighted aggregation of trust scores from all individual consumers, and the calculation formula is as follows: In the formula, M The total number of effective consumers. D i To verify depth rating, F i for Verification frequency score A i To proactively confirm the score, S i for Social communication score L i Rate loyalty α+β+γ+δ+ε=1, and α, β, γ, δ, ε ∈ (0, 1).

3. The trust feedback-driven method based on OID traceability data as described in claim 2, characterized in that, The verification depth score D i The formula for calculating the completeness of the traceability records accessed by consumers is as follows: D i =( 1 / n )×Σ( w j × c ij ), Where n is the total number of modules in the traceability archive, c ij Consumers i Have you checked the first one? j Each module w j For module j The weight.

4. The trust feedback-driven method based on OID traceability data as described in claim 1, characterized in that, The formula for calculating the final recommendation score in step S4 is as follows: Score final =λ 1 · Score semantic +λ 2 · Score authority +λ 3 · T brand _norm in Score semantic This is the content semantic relevance score, with a value range of [0, 1]. Score authority This represents the source authority score, with a value range of [0, 1]. T brand _norm The brand trust comprehensive evaluation index is standardized by Min-Max, with a value range of [0, 1]; λ1+λ2+λ3=1, and λ1, λ2, λ3∈(0, 1).

5. The trust feedback-driven method based on OID traceability data as described in claim 1, characterized in that, The localized recommendation in step S4 refers to generating a QR code heat map based on the city-level location information distribution inferred from the IP address of consumers' QR code scanning behavior. When the number of QR code verifications for a certain brand by consumers in a region exceeds a preset threshold of the total number of QR codes scanned in that region, the corresponding brand is given a regional weighted boost in the AI ​​recommendation results for that region.

6. The trust feedback-driven method based on OID traceability data as described in claim 1, characterized in that, The social trust recommendation in step S4 refers to: With the authorization and consent of both the first and second consumers, when the first consumer's sharing behavior triggers the second consumer's scanning behavior, the brand products corresponding to the sharing behavior will be given a higher weight in the recommendation results during the second consumer's subsequent category search.

7. The trust feedback-driven method based on OID traceability data as described in any one of claims 1 to 6, characterized in that, The trust feedback-driven method based on OID traceability data also includes: Step S6: Using the attribution analysis data generated in step S5, a reinforcement learning framework is used to periodically and adaptively adjust the trust scoring model weights in step S2 and the recommendation fusion weights in step S4.

8. A trust feedback-driven system based on OID traceability data, characterized in that, include: The product OID traceability interaction and behavior recording module is used to assign an OID identifier to each product's smallest inventory unit SKU, provide a traceability file display interface, and record consumer interaction behavior statistics on the traceability page in an anonymous manner. The consumer trust profile building module is used to build individual consumer trust profiles and brand / product aggregate trust profiles based on recorded consumer interaction behavior statistics, and generate a comprehensive brand trust evaluation index. The trust profile and recommendation engine fusion module is used to transmit the trust index and feature vector to the AI ​​recommendation feedback system through a standardized interface; The AI ​​recommendation strategy optimization module is used to dynamically adjust the recommendation ranking weight of brands in dialogue scenarios based on the trust index. The data visualization and attribution analysis module is used to generate a data analysis dashboard for the trust feedback loop. The Trust Model Adaptive Optimization Module is used to periodically and adaptively adjust the weights of the trust scoring model and the recommendation fusion weights using attribution analysis data and a reinforcement learning framework.

9. The trust feedback-driven system based on OID traceability data as described in claim 8, characterized in that, The aforementioned consumer interaction behavior statistics include at least three of the following: basic interaction records, verification depth information, active confirmation records, social dissemination records, and evaluation feedback records.

10. The trust feedback-driven system based on OID traceability data as described in claim 8 or 9, characterized in that, The dialogue scenarios include dialogue recommendation scenarios, e-commerce recommendation scenarios, regional recommendation scenarios, and social recommendation scenarios.

Images