E-commerce platform data processing system and method

Abstract

The invention provides an e-commerce platform data processing system and method. The method comprises the steps that browsing behavior data and commodity selection behavior data of a user in an e-commerce platform are collected, the browsing behavior data comprise page staying duration, rolling browsing rate and page switching frequency, and the commodity selection behavior data comprise carbon emission reduction attributes and green consumption identifiers of commodities selected by the user; and calculating a cognitive fatigue index of the user in a preset time window based on the browsing behavior data, comparing the cognitive fatigue index with a personalized threshold value, and when the cognitive fatigue index exceeds the threshold value, triggering a push self-suppression mechanism so as to reduce the push frequency, simplify the content complexity or delay the push time. By introducing a cognitive fatigue index and green consumption point dual-drive mechanism, green consumption is promoted while user experience is ensured, humanization and sustainability of e-commerce platform pushing strategies are realized, and the method has the beneficial effects of improving user satisfaction and optimizing a consumption structure.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of e-commerce data intelligent processing, and particularly relates to an e-commerce platform data processing system and method. BACKGROUND

[0002] With the rapid development of e-commerce, e-commerce platforms have become the main channel for users to obtain product information and complete transactions. Existing e-commerce platforms generally use data collection and recommendation algorithms to model and analyze user browsing behavior and purchase records to achieve personalized recommendations and precise marketing. Specifically, platforms commonly use collaborative filtering, deep learning, and other recommendation models to generate push content or recommendation lists based on user historical clicks, dwell time, and shopping cart operations, thereby improving user click-through rates and conversion rates.

[0003] However, most existing platforms only use click-through rates and conversion rates as the main optimization targets, lack effective evaluation and control of cognitive fatigue caused by long-time browsing, and are prone to push overload, reducing user experience. Secondly, existing recommendation strategies mainly focus on sales improvement, and have limited guiding effect on green consumption and low-carbon goods, making it difficult to meet the demand for optimizing consumption structure under the dual-carbon background. Therefore, we propose an e-commerce platform data processing system and method. SUMMARY

[0004] The purpose of the present application is to overcome the shortcomings of the prior art and provide an e-commerce platform data processing system and method to solve the technical problems mentioned in the background.

[0005] To achieve the above purpose, the present application provides the following technical solutions: An e-commerce platform data processing method, comprising the following steps: S1, obtaining user browsing behavior data in the e-commerce platform, the browsing behavior data including page dwell time, scrolling browsing rate and page switching frequency; and obtaining user product selection behavior data in the e-commerce platform, the product selection behavior data including carbon emission reduction attributes and corresponding green consumption identifiers selected by the user; S2, calculating a cognitive fatigue index of the user within a preset time window based on the browsing behavior data, the cognitive fatigue index being used to represent the cognitive load level of the user in the information receiving process; S3, triggering a push self-inhibition mechanism when the cognitive fatigue index exceeds a preset fatigue threshold, the push self-inhibition mechanism including reducing the frequency of product push, adjusting the complexity of push content, and delaying the time of the next round of product push; S4, when the user selects a product with carbon reduction attributes, generate corresponding carbon reduction points based on the green consumption identifier of the product, the carbon reduction points are used to represent the cumulative contribution value of the user in the green consumption behavior; S5, based on the carbon reduction points, dynamically adjust the recommendation weight in the product recommendation model, so that the products with green consumption attributes have higher priority in subsequent pushing; S6, while the pushing self-inhibition mechanism is effective, combined with the dynamic adjustment result of the recommendation weight, execute the product pushing optimization operation, so that the pushing frequency meets the cognitive bearing capacity of the user, and guides the user to shift to the direction of green consumption.

[0006] S1 is specifically: obtaining the browsing behavior data of the user on the e-commerce platform, the browsing behavior data includes page stay time, scrolling browsing rate and page switching frequency; obtaining the product selection behavior data of the user on the e-commerce platform, the product selection behavior data includes the carbon reduction attributes and the corresponding green consumption identifier of the product selected by the user; store the browsing behavior data and the product selection behavior data uniformly to form a user behavior data set, which provides input for subsequent calculation.

[0007] S2 is specifically: extracting browsing behavior data from the user behavior data set; in a preset time window, the page stay time, scrolling browsing rate and page switching frequency are standardized; calculate the cognitive fatigue index based on the standardized result, the cognitive fatigue index is used to represent the cognitive load level of the user in the information receiving process; store and pass the calculated cognitive fatigue index to the subsequent pushing self-inhibition triggering step.

[0008] S3 is specifically: judge whether the cognitive fatigue index exceeds the preset fatigue threshold; if it exceeds the fatigue threshold, trigger the pushing self-inhibition mechanism; the pushing self-inhibition mechanism includes reducing the product pushing frequency, adjusting the pushing content complexity and delaying the next round of product pushing time; record the execution state of the pushing self-inhibition mechanism and pass it to the fusion execution step.

[0009] S4 is specifically: extracting product selection behavior data from the user behavior data set; judge whether the product selection behavior data contains carbon reduction attributes and green consumption identifier; if it contains, generate corresponding carbon reduction points based on the carbon reduction attributes; accumulate and store the carbon reduction points to form a green consumption point set, and pass it to the recommendation weight dynamic adjustment step.

[0010] S5 specifically is: obtaining the green consumption credit set; adjusting the recommendation weight in the commodity recommendation model according to the credit value size of the green consumption credit set; making the commodity with the green consumption attribute obtain a higher recommendation priority in subsequent pushing; and delivering the adjustment result of the recommendation weight to the fusion execution step.

[0011] S6 specifically is: receiving the execution state from the pushing self-inhibition mechanism; receiving the adjustment result from the recommendation weight dynamic adjustment step; generating an optimized commodity pushing strategy by combining the execution state and the adjustment result when executing commodity pushing; and executing the final commodity pushing operation according to the optimized commodity pushing strategy, so that the pushing frequency meets the cognitive bearing capacity of the user and guides the user to shift to green consumption.

[0012] An e-commerce platform data processing system, comprising: A user behavior data acquisition module is configured to acquire browsing behavior data and commodity selection behavior data of a user on an e-commerce platform, wherein the browsing behavior data includes page stay duration, scrolling browsing rate and page switching frequency, and the commodity selection behavior data includes whether the user selects a commodity with a carbon emission reduction attribute and a green consumption identifier; A cognitive fatigue index calculation module is configured to calculate a user cognitive fatigue index based on the browsing behavior data within a preset time window, and compare the cognitive fatigue index with a personalized fatigue threshold; A pushing self-inhibition module is configured to trigger a pushing self-inhibition mechanism when the cognitive fatigue index exceeds the personalized fatigue threshold, wherein the pushing self-inhibition mechanism includes reducing commodity pushing frequency, adjusting pushing content complexity and delaying pushing time; A green credit generation module is configured to generate corresponding carbon emission reduction credits based on a green consumption identifier of a commodity when the user selects the commodity with the carbon emission reduction attribute, and store the carbon emission reduction credits in a user credit set; A recommendation weight adjustment module is configured to dynamically adjust a recommendation weight in a recommendation model according to the carbon emission reduction credits, so that a commodity with a green consumption attribute has a higher priority in recommendation; A fusion optimization execution module is configured to execute a multi-objective optimization pushing strategy by combining the dynamic adjustment result of the recommendation weight while the pushing self-inhibition mechanism is in effect, so as to generate a pushing result that meets the cognitive bearing capacity of the user and guides green consumption.

[0013] The present application has the following advantages: This invention collects user browsing behavior data and calculates a cognitive fatigue index, enabling real-time assessment of users' cognitive state. When the index exceeds a threshold, a push self-inhibition mechanism is triggered, effectively preventing information overload and improving user experience. When users select products with carbon reduction attributes, corresponding green consumption points are generated and incorporated into recommendation weight adjustments, giving green products higher priority in recommendation results and subtly guiding users towards low-carbon and environmentally friendly consumption. By integrating cognitive fatigue control and green consumption points, a multi-objective optimization model is constructed to improve the conversion rate of green products while considering user experience, achieving a balance between consumption efficiency and sustainability. This invention not only solves the problems of push overload and decreased user experience in existing technologies but also introduces a green consumption incentive mechanism, achieving a unity between protecting users' cognitive tolerance and guiding sustainable consumption, demonstrating significant practical value and social benefits. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of an e-commerce platform data processing method according to the present invention. Detailed Implementation

[0015] 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.

[0016] Example 1: As Figure 1 As shown, this embodiment provides a data processing method for an e-commerce platform, including the following steps: S1. User behavior data collection steps: Obtain user browsing behavior data on the e-commerce platform, including page dwell time, scrolling speed, and page switching frequency; simultaneously obtain user product selection behavior data on the e-commerce platform, including the carbon emission reduction attributes and corresponding green consumption labels of the products selected by the user.

[0017] S2. Cognitive fatigue index calculation steps: Based on the browsing behavior data, calculate the user's cognitive fatigue index within a preset time window. The cognitive fatigue index is used to characterize the user's cognitive load level during the information reception process.

[0018] S3. Push Self-Inhibition Triggering Step: When the cognitive fatigue index exceeds the preset fatigue threshold, the push self-inhibition mechanism is triggered. The push self-inhibition mechanism includes reducing the frequency of product pushes, adjusting the complexity of push content, and delaying the time of the next round of product pushes.

[0019] S4, green consumption credit generation step: when the user selects a product with carbon emission reduction attribute, generate corresponding carbon emission reduction credit based on the green consumption identification of the product, the carbon emission reduction credit is used to represent the cumulative contribution value of the user in the green consumption behavior.

[0020] S5, recommended weight dynamic adjustment step: based on the carbon emission reduction credit, dynamically adjust the recommended weight in the product recommendation model, so that the product with green consumption attribute obtains higher priority in subsequent push.

[0021] S6, push self-inhibition and green driving fusion execution step: while the push self-inhibition mechanism is effective, combine the dynamic adjustment result of the recommended weight, execute the product push optimization operation, so that the push frequency meets the cognitive bearing capacity of the user, and guides the user to shift to green consumption.

[0022] S1, user behavior data collection step: obtain the browsing behavior data of the user in the e-commerce platform, the browsing behavior data includes page stay time, scrolling browsing rate and page switching frequency; at the same time, obtain the product selection behavior data of the user in the e-commerce platform, the product selection behavior data includes the carbon emission reduction attribute and the corresponding green consumption identification of the product selected by the user; specifically including the following sub-steps: S110, browsing behavior data collection step: obtain the browsing behavior data of the user in the e-commerce platform, the browsing behavior data includes: page stay time T (unit: second); scrolling browsing rate R (unit: pixel / second); page switching frequency F (unit: times / minute).

[0023] The collection method includes: obtaining page stay and scrolling events through JavaScript buried point script on the web page; obtaining scrolling rate and page switching events through built-in SDK in mobile APP; the data format adopts JSON or Protobuf structured storage. The collection frequency is set to 1 second / time. Comparative experiments show that: 0.5 second sampling is more precise, but the redundant data volume increases by about 60%; 2 second sampling has delay compared with actual click behavior. 1 second sampling balances between precision and storage overhead, so it is selected as the default parameter. The data is uploaded to the cloud in real time through HTTPS or MQTT protocol, and the transmission delay of a single data is less than 100ms.

[0024] S120, product selection behavior data collection step: obtain the product selection behavior data of the user in the e-commerce platform, the product selection behavior data includes whether the user selects a product with carbon emission reduction attribute.

[0025] The source of carbon emission reduction attribute is: supplier provided life cycle carbon emission report; third party database (ISO14067 carbon footprint standard, GHG Protocol, PAS 2050); platform verifies green certification label of goods through API call. In order to ensure credibility, the carbon emission reduction attribute data is stored and certified (using block chain tamper-proof mechanism), ensuring the traceability of subsequent point generation and recommendation adjustment.

[0026] S130, user behavior data set generation step: storing the browsing behavior data and the commodity selection behavior data as a user behavior data set.

[0027] Storage and security measures include: storage architecture adopts distributed HDFS or time series database InfluxDB to support million-level user concurrent data writing; data is stored by AES-256 encryption, and anonymization processing is performed before writing to remove user real identity information and only keep user ID; when the network is disconnected, the data is saved in the local cache, the cache capacity ranges from 50MB to 200MB, and the cache is stored by encryption, and after the network is restored, the data is uploaded in batches; Abnormal data processing: if the collection rate is lower than 90%, the batch data is marked as invalid; if single point loss occurs, linear interpolation is used to complete.

[0028] Cross-terminal consistency processing: when a user logs in on multiple terminals (such as mobile phone and PC) at the same time, the user ID and timestamp are combined to process the data, and the timestamp alignment error is not more than 200ms, which ensures the unity of the behavior data set.

[0029] S2, cognitive fatigue index calculation step: based on the browsing behavior data, the cognitive fatigue index of the user in a preset time window is calculated, which is used to represent the cognitive load level of the user in the information receiving process. Specifically, the following sub-steps are included: S210, data extraction step: extracting browsing behavior data from the user behavior data set, the browsing behavior data includes: page stay time T (unit: second), collection accuracy ±0.1 second; scrolling browsing rate R (unit: pixel / second), collection accuracy ±2 pixel / second; page switching frequency F (unit: times / minute), collection accuracy ±0.05 times / minute.

[0030] Experimental verification shows that the above collection error has less than 2% influence on the final calculation result, and will not cause substantial deviation in fatigue determination.

[0031] S220, data standardization and smoothing step: within a preset time window of 60 seconds, the stay time, scrolling rate and page switching frequency are standardized by Z-score method:

[0032] where X is the original data value (e.g. page dwell time, scroll rate of a certain user), is the average value of the data sample set, is the standard deviation of the data sample set, used to measure the fluctuation range of the data; X s is the standardized value, i.e. the deviation of the data from the average level, the unit is eliminated, and the result is a pure number.

[0033] When processing user behavior data (such as calculating the cognitive fatigue index), a statistical "time window" needs to be selected, that is, how many seconds of data are taken continuously to calculate. Here, 30 seconds, 60 seconds and 90 seconds are tested respectively; the comparative experimental results show that: under the 30-second window, the Pearson correlation coefficient r is 0.7272; under the 60-second window, r is 0.87, which is the best; under the 90-second window, r is 0.74. Therefore, 60 seconds is selected as the default time window; the Pearson correlation coefficient r is an index for measuring the linear correlation between variables, with a value range of [-1, 1]. The closer r is to 1, the higher the correlation between the calculated index and the true result, and the better the effect. In order to reduce instantaneous fluctuations, the exponential weighted moving average (EWMA) smoothing is used:

[0034] X t represents the original data value at time t (such as the page dwell time or scroll rate of a certain second); S t is the smoothed value at time t, i.e. the result after weighted smoothing. λ is the smoothing coefficient, with a value range of 0<λ<1. The above is the exponential weighted moving average (EWMA) formula, which is used to weaken the random fluctuations of the data and highlight the trend. If λ is larger, the latest data X t has a higher weight, and the result is more sensitive; if λ is smaller, the historical data S t-1 has a higher weight, and the result is more stable; this step is used to smooth the user behavior data (dwell time, scroll rate, page switching frequency) to avoid fluctuations caused by instantaneous abnormalities.

[0035] S230, cognitive fatigue index calculation step: based on the standardized and smoothed data, calculate the cognitive fatigue index :

[0036] where T s is the standardized or smoothed value of the time dwell feature; R s is the standardized or smoothed value of the scroll rate feature; F sis the standardized or smoothed value of the page switching frequency feature. α, β, γ are the weight coefficients of each feature, satisfying α + β + γ = 1; CF(t) is the cognitive fatigue index at time t, which is used to measure the current information receiving load of the user. The three user behavior features are fused into a unified cognitive fatigue index according to the weight. By adjusting α, β, γ, the importance of different features can be emphasized (for example: if the page switching frequency has a greater impact on fatigue, the value of γ can be increased); the weight training is based on no less than 1200 user samples, covering different age groups (18-65 years old), regions (≥5 large areas) and device types (mobile phones, tablets, PCs); the weight calculation uses a combination of least squares regression and entropy weight method, and is verified by 10-fold cross-validation, with an average error of ≤5%; CF(t) calculated in this step is compared with the personalized threshold to determine whether to trigger the push self-inhibition mechanism.

[0037] Typical training result example: in the experimental sample, is 0.4, is 0.35, is 0.25; Implementation environment: mobile end: memory requirement about 20MB, CPU occupancy <5%, single calculation time <50ms; cloud: supporting 100,000 user concurrency per second, complexity O(n), calculation delay <100ms.

[0038] S240, fatigue threshold setting step: set the threshold for the calculated cognitive fatigue index:

[0039] CF th is the global cognitive fatigue threshold (baseline value) set by the system; k is the adjustment coefficient, usually in the range of 0 < k < 10, used to control the amplitude of personalized correction. Δu is the user individual difference factor. is the user's personalized cognitive fatigue threshold. On the basis of the global threshold CF th , it is dynamically corrected according to the individual difference Δu of the user, so as to obtain a personalized threshold that is more in line with the personal characteristics;

[0040] is the average page dwell time of a certain user. is the overall average page dwell time of all users. Δu is the deviation of the user relative to the overall average level. If Δu > 0, it means that the user is more tolerant of information quantity (longer dwell time) than the average user; if Δu < 0, it means that the user is more prone to fatigue (shorter dwell time). The difference between the behavior characteristics of the user and the average level of the group is measured to adjust the threshold individually. Global threshold CF th: System first sets a baseline threshold; calculates user difference factor Δu: get its tolerance by comparing the user with the overall average dwell time. Personalized adjustment: bring the difference factor into the first formula to get the personalized threshold . For example: global threshold CF th =100, a user's dwell time is 20% higher than the average (Δu=0.2), if k=0.3, then =100⋅(1+0.3⋅0.2)=106, indicating that the user's fatigue threshold is 6% higher than the baseline, more tolerant of push.

[0041] The platform re-trains the threshold model every quarter; re-training process: collect a new round of user data → refit weights → Recalculate the global threshold → Update the user personalized threshold library; After updating, confirm through AB test that the new threshold has no negative impact on user experience before going online.

[0042] S3, push self-inhibition triggering step: when the cognitive fatigue index exceeds the preset fatigue threshold, trigger the push self-inhibition mechanism, which includes reducing the frequency of commodity push, adjusting the complexity of push content, and delaying the next round of commodity push time. Specifically, the following sub-steps are included: S310, fatigue determination step: determine whether the cognitive fatigue index CF(t) exceeds the user's personalized threshold . If , maintain the regular push strategy; if , enter the push self-inhibition triggering process.

[0043] Experimental basis: In an experiment of 1000 users (covering different age groups, regions, and device types), the average subjective fatigue score of users increased by 26% without the self-inhibition mechanism; After setting the self-inhibition mechanism, the average increase was only 8%, with a significant difference (p<0.01).

[0044] S320, hierarchical self-inhibition step: according to the exceeding amplitude Trigger different levels of self-inhibition: Mild over-standard : push frequency decreases by 30%, push interval is extended to 1.3 times the original setting, typically 2 minutes.

[0045] Moderate over-standard : Push frequency decreases by 50%, push content is simplified (text ≤50 words, pictures ≤2, video ≤5 seconds, interactive elements ≤3), push is delayed by 5 minutes.

[0046] Severe over-standard : Suspend all push for 15 minutes, only allow safety prompt type messages. Experiment basis: In A / B test of 1000 users, when set delay for 5 minutes, user click rate decreased <3%, but fatigue score decreased 15%, determined as the best balance point.

[0047] S330, Recovery and Redundancy Protection Step: Recovery Mechanism: Recalculate CF(t) every 5 minutes. If , gradually remove self-inhibition, push frequency increases by 10% step, until it returns to normal level.

[0048] Threshold selection basis: Experiments show that when the recovery condition is set to 0.8 times the threshold, user experience is optimal, avoiding premature recovery (leading to fatigue repeatedly) and slow recovery (leading to information missing). Redundancy protection: If the user triggers severe overage multiple times within 1 hour, the system enters protection mode, suspending push for a maximum of 1 hour.

[0049] S340, Fault Tolerance and Priority Step: Priority Rules: If "green recommendation" and "push self-inhibition" are triggered at the same time, prioritize self-inhibition to avoid information overload. User-defined mode: The system provides three modes, parameters as follows:

[0050] Misjudgment detection: Set the probability formula for misjudgment:

[0051] The above formula is used to measure the error rate of the system in the judgment process; P err is the misjudgment probability, with a value range of [0,1]; N false is the number of false cases, i.e. the number of samples for which the system made a wrong judgment; N total is the total number of cases, i.e. the total number of all judgment samples of the system.

[0052] If the system has made N total judgments in total, of which N false are wrong, then the misjudgment rate is P err ; for example, a total of 1000 judgments, 50 errors, then: P err =50 / 1000=0.05, that is, a 5% misjudgment rate. When is greater than 0.05, the system reverts to the safe push frequency (every 10 minutes).

[0053] Experiment feedback loop: The platform re-evaluates the grading parameters every quarter, recalibrates the frequency reduction ratio and delay time with new user data. Before updating the parameters, small-scale AB testing is conducted to verify user experience indicators (click rate decrease <5%, fatigue score improvement >10%), and after verification, the parameters are promoted. Performance: Grading determination and recovery calculation delay <200ms, supporting real-time push scenarios.

[0054] S4, green consumption point generation step: when the user selects a product with carbon reduction attribute, the corresponding carbon reduction points are generated based on the green consumption identification of the product, and the carbon reduction points are used to represent the cumulative contribution value of the user in the green consumption behavior. Specifically, the following sub-steps are included: S410, product selection data extraction step: extract product selection behavior data from user behavior data set, including whether the user selects a product with carbon reduction attribute and green consumption identification.

[0055] S420, green attribute determination step: determine whether the product has carbon reduction attribute and green consumption identification: data source: supplier provided life cycle carbon emission report; emission factor verification: call third party database (ISO14067, GHG Protocol, PAS 2050), and preferentially refer to international authoritative data source (such as IPCC 2019 guide, Ecoinvent database); credible mechanism: green attribute data is written into block chain for storage, ensuring that it cannot be tampered with.

[0056] S430, green consumption point calculation step: if the product has carbon reduction attribute, calculate the carbon reduction amount and generate green consumption points , the formula is:

[0057] where GP is green consumption point (Green Points), the point awarded to the user; δ is the point conversion coefficient, ranging from [0.1, 1], determined by platform strategy or user research. C r is the carbon reduction of the product (Carbon reduction). After the user selects the green product, the corresponding green points GP are generated according to the carbon reduction C r of the product and the conversion coefficient δ.

[0058] where: The calculation formula is:

[0059] is the production link emission factor (kgCO2e / piece), taken from IPCC or Ecoinvent; is the product yield or quantity; is the transportation link emission factor (kg CO2e / km·piece); is the transportation distance (km); is the packaging link emission factor (kg CO2e / kg packaging material); is the packaging weight (kg); δ belongs to [0.1, 1] is the integral conversion coefficient, the initial value is 0.5, and is updated dynamically according to platform operation data and user research, and is re-evaluated every quarter to prevent integral inflation or insufficient incentives. For example: assuming that the production emission factor EF p =1.2 kg CO2e / piece; the quantity Q p =1, the transportation emission factor EF t =0.05 kg CO2e / km·piece, the distance D t =100 km; the packaging emission factor EF pk =0.1 kg CO2e / kg; the packaging weight W pk =0.2 kg; then C r =1.2×1+0.05×100+0.1×0.2=6.22 kg CO2e; if δ=0.5, then GP=0.5×6.22=3.11, indicating that the user can obtain 3.11 green points by purchasing the green product.

[0060] S440, integral storage, anti-cheating and performance explanation step: storage mechanism: user green points are stored in the integral set and are encrypted by AES-256; the annual integral upper limit is 1000 points, with two decimal places; after the annual cycle ends, all integrals are cleared and not carried over. Use explanation: integrals are used for recommendation weight adjustment, coupon redemption, and freight deduction, but cannot be cashed out, cannot be transferred, and business loopholes are avoided.

[0061] Anti-cheating mechanism: rate detection: if the integral growth rate of the user within 24 hours is >3 times the average of all users, it is marked as abnormal; statistical anomaly detection formula:

[0062] GP user is the green consumption integral value of a certain user; μ all is the average value of all user green consumption integrals; σ allis the standard deviation of all users' green consumption points. Z is the normalized score of the user's green points in the overall distribution (i.e., how many standard deviations away from the mean). In detail: Z≈0 indicates that the user's green points are close to the average level. Z>0 indicates that the user's points are higher than the average, and the larger the value, the higher the deviation. Z<0 indicates that the user's points are lower than the average, and the smaller the value, the higher the deviation. Abnormal detection: In practical applications, if |Z|>3, it is usually determined that the user's point distribution is abnormal (there may be cheating or extreme consumption behavior).

[0063] Performance indicators: the point calculation takes less than 10ms per time; the storage system supports millions of concurrent user writes with a delay of less than 100ms; the anti-cheating detection runs in a minute-level batch processing manner, with a false positive rate controlled within 2%. Experimental support: in an A / B test of 500 users, after introducing the green consumption point mechanism, the green product purchase rate increased by 18%, and user satisfaction increased by 12%.

[0064] S5, a recommendation weight dynamic adjustment step: based on the carbon emission reduction points, the recommendation weight in the product recommendation model is dynamically adjusted, so that the products with green consumption attributes obtain higher priority in subsequent pushing. Specifically, the following sub-steps are included: S510, a point acquisition step: acquiring a user green consumption point set, the point set including annual cumulative points GP and annual upper limit GP max =1000 points, and the points are kept to two decimal places.

[0065] S520, a green weight calculation step: adjusting the green product recommendation weight according to the green consumption points, and the calculation formula is:

[0066] The above formula is used to dynamically adjust the weight of green products in the recommendation system according to the user's green consumption points; W g is the green recommendation weight (adjusted weight); W0 is the initial recommendation weight, i.e., the basic weight of the system when the green consumption points are not considered. GP: the green consumption points obtained by the user; GP max is the annual upper limit of the points (e.g., 1000 points); λ is a regulation coefficient, usually in the range of [0.1, 0.5], used to control the influence of points on the recommendation weight.

[0067] When the user's points are low, GP / GP max is close to 0, and the green weight is close to the basic value W0; when the user's points increase, the green weight W gThe priority of green products in recommendations is increased in proportion. λ determines the sensitivity of points to weight increase: larger λ → weight increases faster; smaller λ → weight increases more gently. Based on experimental data from 2000 users covering 5 regions, 3 main age groups (18-30, 31-45, 46-65), Bayesian optimization and 5-fold cross-validation are used to ensure the universality of the results; ceiling mechanism: to prevent excessive green recommendations, set an upper limit W g < 2W0. For example: assuming initial weight W0=1.0, adjustment coefficient λ=0.3, user points GP=500; annual points upper limit GP max =1000; then

[0068] This means that the user's current green recommendation weight has increased by 15% compared to the base value, and green products will be pushed more in recommendations.

[0069] S530, recommendation fusion step: fuse green recommendation weight and collaborative filtering factor to generate final recommendation score:

[0070] The above formula is used to integrate traditional recommendation algorithm scores and green recommendation weights in e-commerce platforms to generate final recommendation scores; Score is the final recommendation score, used for sorting and pushing goods. W cf is the traditional recommendation score obtained by collaborative filtering or deep learning models (usually based on user historical behavior, similar user behavior). W g is the green recommendation weight (which has been dynamically adjusted according to green consumption points). θ is the fusion coefficient, ranging from [0.3, 0.7], used to control the relative weight of the two parts. When θ is large: traditional recommendation factor W cf dominates, the system is more inclined to meet the user's immediate interest. When θ is small: green weight W g dominates, the system is more inclined to guide users to green consumption. By adjusting θ, a balance between personalized interest recommendation and green consumption guidance is achieved. The recommendation model can use LSTM, Transformer or Graph Neural Network (GNN), and the green weight W g is input as an additional feature; experimental results: in a test of 2000 user samples, when θ=0.5, the click rate increases by 12%, and the green product purchase rate increases by 18%. For example: assuming W cf =0.8; W g =1.15, θ=0.5; then Score=0.975, the final recommendation score of this product is 0.975, and its ranking in the recommendation list will be improved accordingly.

[0071] Fairness boundary: green product recommendation ratio ≥ 5% to avoid complete neglect of green products; green product recommendation ratio ≤ 30% to avoid excessive green recommendations. Experimental basis: In a test of 500 users, when the green recommendation ratio exceeds 30%, user satisfaction decreases by about 10%; when the ratio is less than 5%, there is no significant increase in green product sales.

[0072] S540, recommendation result output and update mechanism step: result output: generate a recommendation list according to the final recommendation score Score and push it to the user terminal. Cold start processing: for new registered users, the default green weight ratio is set to 0.3; the cold start stage lasts for 3 days, during which user click and purchase data are collected in real time, and the green weight is gradually corrected so that it converges to the personalized level within 3 days. Anti-cheating and anti-bias mechanism: if abnormal growth of user points is detected (based on the anomaly detection formula Z>3), freeze the user's green points and the recommendation weight reverts to the default value W0. Update implementation details: batch update: process all users in batches through the Spark computing framework every day, with a single batch taking <10 minutes; real-time update: high-frequency user behavior data is written in real time through the Kafka streaming architecture with a delay of <5 seconds; the recommendation engine has a computational complexity of O(1), a single calculation takes <5ms, supports millions of concurrent users, and the overall response delay is <100ms.

[0073] S6, push self-inhibition and green drive fusion execution step: while the push self-inhibition mechanism is in effect, combine the dynamic adjustment results of the recommendation weight to perform product push optimization operations, so that the push frequency meets the user's cognitive capacity and guides the user to shift towards green consumption. Specifically, the following sub-steps are included: S610, self-inhibition state receiving step: receive state information from the push self-inhibition mechanism, including: push frequency adjustment value, push delay duration, push content complexity constraint. Data structure: transmitted in JSON format, typical fields include {freq_adj: -30%, delay: 300s, content_level: simple}; receive delay <50ms to ensure real-time performance.

[0074] S620, recommendation weight receiving step: receive the weight adjustment results from the recommendation weight dynamic adjustment step, including the green recommendation weight W g and the fusion score Score. Data source: generated by the recommendation engine (LSTM / Transformer / GNN); update frequency: high-frequency user real-time update (≤5 seconds), ordinary user daily batch update.

[0075] S630, multi-objective optimization step: build a multi-objective optimization model, with the objective function being:

[0076] This is a multi-objective optimization function to balance user cognitive load and green consumption conversion rate in e-commerce platform recommendation process; where ΔCF is the change rate of cognitive fatigue index, representing the increasing degree of user fatigue within a certain time window; GR is the green consumption conversion rate, i.e. the proportion of users clicking or purchasing green products; ω1, ω2 are weighting coefficients, satisfying ω1+ω2=1, used to balance the importance of "user experience protection" and "green consumption guidance".

[0077] Control cognitive fatigue: 1-ΔCF part in the objective function: if ΔCF increases (users become more and more tired), the value becomes smaller, and the optimization model will reduce the push to protect user experience; if ΔCF is small, it means that the user is still in an acceptable state, and the push can be appropriately increased.

[0078] Promote green consumption: GR part in the objective function: if the green product conversion rate is high, the optimization model will continue to increase green recommendation; if the conversion rate is low, the model will appropriately reduce green push to avoid affecting the overall experience. Weight balance: by adjusting ω1 and ω2, we can decide whether the system focuses more on "user experience" (cognitive fatigue) or "green guidance" (conversion rate).

[0079] Genetic algorithm solution: population size: 50-100, typical value=80; crossover probability: 0.7-0.9, typical value=0.8; mutation probability: 0.05-0.1, typical value=0.05; maximum iteration: 100; convergence condition: if there is no improvement or the improvement amplitude is less than 0.1% for 10 consecutive generations, the convergence is terminated; experimental results: in a sample of 2000 users, the typical convergence generation is 60, and the average calculation time is 1.8 seconds.

[0080] Reinforcement learning solution (small sample scenario): state space: {CF(t), GP, CTR, Score}; action set: {push frequency adjustment, push delay adjustment, recommendation weight correction}; Reward function: ; where CTR is the click-through rate, the higher the better, the reward increases. GR is the green consumption conversion rate, the higher the better, the reward increases; ΔCF is the change rate of cognitive fatigue index, the higher the user fatigue, the reward decreases. The system will reward those push strategies that can both improve click-through rate and green consumption, and reduce user fatigue.

[0081] A learning rate a e [0.01, 0.1] controls the step size of model parameter updates; a discount factor g e [0.8, 0.95], typically a = 0.05, g = 0.9; training rounds: 5000 rounds, implemented using TensorFlow or PyTorch framework; convergence criterion: stop if the reward function improves by less than 1% in 200 consecutive rounds.

[0082] S640, push execution and feedback step: push execution: generate push strategy according to optimization result, push content contains recommendation list, push time and green recommendation proportion. Cold start mechanism: the default green weight of new users is 0.3, combined with registration area and interest label to generate recommendations, and the real behavior of users is corrected gradually within 3 days. Feedback mechanism: feedback indicators include click rate CTR, purchase rate CR and green consumption proportion GR; data sampling frequency: more than 1000 user behavior data is aggregated every minute; data delay: less than 5 minutes into the optimizer. Update mechanism: adopt "dual track" update: batch update: through Spark batch processing of user full data every day, single batch time is less than 10 minutes; real-time update: high-frequency user data is written in real time through Kafka streaming architecture, with a delay of less than 5 seconds. Performance indicators: the single user calculation complexity of optimization engine is O(1), and the single calculation time is less than 5ms; distributed parallel supports 100,000 user optimization requests per second, and the overall response delay is less than 100ms. Exception bottom: if the optimization result is abnormal (such as weight collapse or green proportion imbalance), the system falls back to the default push strategy to ensure platform stability.

[0083] Embodiment two: the embodiment provides an e-commerce platform data processing system, comprising: A user behavior data acquisition module is configured to acquire browsing behavior data and product selection behavior data of users on an e-commerce platform, wherein the browsing behavior data includes page dwell time, scrolling browsing rate and page switching frequency, and the product selection behavior data includes whether the user selects a product with carbon emission reduction attribute and green consumption identification. A cognitive fatigue index calculation module is configured to calculate a user cognitive fatigue index based on the browsing behavior data within a preset time window, and compare the cognitive fatigue index with a personalized fatigue threshold. A push self-inhibition module is configured to trigger a push self-inhibition mechanism when the cognitive fatigue index exceeds the personalized fatigue threshold, wherein the push self-inhibition mechanism includes reducing product push frequency, adjusting push content complexity and delaying push time. A green credit generation module is configured to generate corresponding carbon emission reduction credits based on the green consumption identification of the product when the user selects a product with carbon emission reduction attribute, and store the carbon emission reduction credits in a user credit set. The recommendation weight adjustment module is configured to dynamically adjust a recommendation weight in the recommendation model according to the carbon emission reduction credit, so that a product with a green consumption attribute has a higher priority in the recommendation. The fusion optimization execution module is configured to execute a multi-objective optimization push strategy to generate a push result that meets the user's cognitive bearing capacity and guides green consumption, in combination with the dynamic adjustment result of the recommendation weight while the push self-inhibition mechanism is in effect.

[0084] The above formulas are all dimensionless values calculated, the formulas are obtained by collecting a large amount of data to simulate a formula of the most recent real situation, and the preset parameters and threshold values in the formulas are set by a person skilled in the art according to actual conditions.

[0085] The above embodiments can be realized wholly or partially by software, hardware, firmware or any other combination. When realized by software, the above embodiments can be realized wholly or partially in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are wholly or partially generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network or other programmable devices. The computer instructions can be stored in a computer readable storage medium or transferred from one computer readable storage medium to another, for example, the computer instructions can be transferred from one website, computer, server or data center to another by wired (for example, infrared, wireless, microwave, etc.) mode. The computer readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center and the like containing one or more available medium sets. The available medium can be a magnetic medium (for example, floppy disk, hard disk, magnetic tape), an optical medium (for example, DVD) or a semiconductor medium. The semiconductor medium can be a solid state disk.

[0086] Those skilled in the art can realize that the modules and algorithm steps of the examples described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. The skilled person can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0087] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system, device and module described above can refer to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0088] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the modules is only a logical function division, and actual implementation can have another division manner, for example, a plurality of modules or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the shown or discussed modules can be indirect coupling or communication connection through some interfaces, devices or modules, and can be electrical, mechanical or other forms.

[0089] The modules described as separate components can or can not be physically separated, and the components shown as modules can or can not be physical modules, which can be located in one place or distributed on a plurality of network modules. Part or all of the modules can be selected according to actual needs to achieve the purpose of the embodiment scheme.

[0090] In addition, the functional modules in each embodiment of the present application can be integrated in one processing module, or each module can be physically present alone, or two or more modules can be integrated in one module.

[0091] If the functions are realized in the form of software function modules and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application essentially or the part of the prior art or the part of the technical solutions can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a plurality of instructions for making a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the method described in each embodiment of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various program code storage media.

[0092] The above merely provides the specific implementation of the present application, but the protection scope of the present application is not limited to this. Any person skilled in the art can easily think of the changes or replacements within the technical range disclosed by the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0093] Finally, the above merely provides the preferred embodiments of the present application, but is not used to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. An e-commerce platform data processing method, characterized in that, Comprise the following steps: S1, obtain the user's browsing behavior data in the e-commerce platform, the browsing behavior data includes page stay time, scroll browsing rate and page switching frequency; Obtain the user's commodity selection behavior data in the e-commerce platform, the commodity selection behavior data includes the carbon emission reduction attribute and corresponding green consumption mark of the user's selected commodity; S2, based on the browsing behavior data, calculate the cognitive fatigue index of the user in the preset time window, the cognitive fatigue index is used to represent the cognitive load level of the user in the information receiving process; S3, when the cognitive fatigue index exceeds the preset fatigue threshold, trigger the push self-inhibition mechanism, the push self-inhibition mechanism includes reducing the commodity push frequency, adjusting the push content complexity and delaying the next round of commodity push time; S4, when the user selects the commodity with carbon emission reduction attribute, the corresponding carbon emission reduction points are generated based on the green consumption mark of the commodity, the carbon emission reduction points are used to represent the cumulative contribution value of the user in the green consumption behavior; S5, based on the carbon emission reduction points, dynamically adjust the recommendation weight in the commodity recommendation model, so that the commodity with green consumption attribute obtains higher priority in subsequent push.

2. The data processing method of an electronic business platform according to claim 1, characterized in that, Also include S6, when the push self-inhibition mechanism takes effect, combined with the dynamic adjustment result of the recommendation weight, execute the commodity push optimization operation, so that the push frequency meets the user's cognitive bearing capacity, and guides the user to shift to green consumption.

3. The data processing method of an electronic business platform according to claim 1, characterized in that, S1 specifically: Obtain the user's browsing behavior data in the e-commerce platform, the browsing behavior data includes page stay time, scroll browsing rate and page switching frequency; Obtain the user's commodity selection behavior data in the e-commerce platform, the commodity selection behavior data includes the carbon emission reduction attribute and corresponding green consumption mark of the user's selected commodity; The browsing behavior data and the commodity selection behavior data are stored uniformly to form a user behavior data set, which provides input for subsequent calculation.

4. The data processing method of an electronic business platform according to claim 1, characterized in that, S2 specifically: Extract the browsing behavior data from the user behavior data set; In the preset time window, the page stay time, scroll browsing rate and page switching frequency are standardized; Calculate the cognitive fatigue index based on the standardization result, the cognitive fatigue index is used to represent the cognitive load level of the user in the information receiving process; The calculated cognitive fatigue index is stored and transmitted to the subsequent push self-inhibition triggering step.

5. The data processing method of an electronic business platform according to claim 1, characterized in that, S3 specifically: Determine whether the cognitive fatigue index exceeds the preset fatigue threshold; If the fatigue threshold is exceeded, trigger the push self-inhibition mechanism; The push self-inhibition mechanism includes reducing the commodity push frequency, adjusting the push content complexity and delaying the next round of commodity push time; The execution state of the push self-inhibition mechanism is recorded and transmitted to the fusion execution step.

6. The data processing method of an electronic business platform according to claim 1, wherein, S4 specifically: Extract the commodity selection behavior data from the user behavior data set; Determine whether the commodity selection behavior data contains carbon emission reduction attribute and green consumption mark; If it contains, the corresponding carbon emission reduction points are generated based on the carbon emission reduction attribute; The carbon emission reduction integral is accumulated and stored to form a green consumption integral set, and is transmitted to a recommendation weight dynamic adjustment step.

7. The data processing method of an electronic business platform according to claim 1, wherein, S5 is specifically: Obtaining the green consumption integral set; According to the integral value of the green consumption integral set, adjusting the recommendation weight in the commodity recommendation model; Make the goods with green consumption attribute get higher recommendation priority in subsequent push; The adjustment result of the recommendation weight is transmitted to a fusion execution step.

8. The data processing method of an electronic business platform according to claim 2, characterized in that, S6 is specifically: Receive the execution state from the push self-inhibition mechanism; Receive the adjustment result from the recommendation weight dynamic adjustment step; In the execution of commodity push, combine the execution state and the adjustment result to generate an optimized commodity push strategy; According to the optimized commodity push strategy, execute the final commodity push operation, so that the push frequency meets the user's cognitive tolerance and guides the user to shift to green consumption.

9. An e-commerce platform data processing system based on the e-commerce platform data processing method of any one of claims 1-8, characterized in that, Including: User behavior data acquisition module, for acquiring user browsing behavior data and commodity selection behavior data on e-commerce platform, the browsing behavior data includes page stay time, scrolling browsing rate and page switching frequency, the commodity selection behavior data includes whether the user selects the commodity with carbon emission reduction attribute and green consumption identification; Cognitive fatigue index calculation module, for calculating user cognitive fatigue index based on the browsing behavior data in a preset time window, and comparing the cognitive fatigue index with a personalized fatigue threshold; Push self-inhibition module, for triggering push self-inhibition mechanism when the cognitive fatigue index exceeds the personalized fatigue threshold, the push self-inhibition mechanism includes reducing commodity push frequency, adjusting push content complexity and delaying push time; Green integral generation module, for generating corresponding carbon emission reduction integral based on the green consumption identification of the commodity when the user selects the commodity with carbon emission reduction attribute, and storing the carbon emission reduction integral in the user integral set; Recommendation weight adjustment module, for dynamically adjusting the recommendation weight in the recommendation model according to the carbon emission reduction integral, so that the commodity with green consumption attribute has higher priority in recommendation.

10. The e-commerce platform data processing system of claim 9, wherein, Also including: Fusion optimization execution module, for executing multi-objective optimization push strategy to generate push result meeting user cognitive tolerance and guiding green consumption while the push self-inhibition mechanism is effective, in combination with the dynamic adjustment result of the recommendation weight.

Images