Big data driven mall goods hierarchical optimization method and system

Abstract

The present application relates to the technical field of machine learning, in particular to a big data driven mall goods grading optimization method and system, comprising the following steps: extracting video and supermarket shopping application residence time, evaluating cross-domain interest complement and mutual exclusion, combining exchange period to generate time decay penalty, comprehensive mapping to generate target final grading score, and arranging and distributing display position to generate optimization control instruction. In the present application, the time sequence interaction characteristics of video application and shopping application are extracted to quantify the mutual exclusion and complementary relationship of cross-domain interest, and a time utility decay penalty mechanism is constructed by combining the virtual card exchange period, and then the multi-dimensional positive and negative evaluation data is comprehensively mapped to generate the grading score, etc. Technical means, realizing the dynamic and accurate grading and order optimization of the target card, playing the role of intelligently allocating page display position, dynamically reorganizing and rendering display resources, and effectively improving the platform goods circulation and exposure efficiency.

Description

Technical Field

[0001] This invention relates to the field of machine learning technology, and in particular to a big data-driven method and system for classifying and optimizing products in an online marketplace. Background Technology

[0002] Machine learning technology is a discipline that studies how computers simulate human learning behavior and acquire new knowledge. Its main content covers algorithm design and model training processes. Its primary purpose is to enable computers to automatically extract patterns and establish mapping relationships by inputting historical experience data to complete specific classification or prediction tasks. It is widely used in many industries such as image recognition, natural language processing, medical assistance, and intelligent manufacturing. Traditional e-commerce product tiering and optimization refers to the evaluation and screening of a collection of products on an e-commerce platform based on specific dimensions. This is used to match products to consumer needs and help operators optimize inventory structure. It typically uses existing technologies such as fixed rule filtering based on human experience and basic sales attribute statistics to achieve the goals of product category definition and ranking optimization.

[0003] Traditional e-commerce product tiering and selection (such as e-commerce cards and video platform membership cards) can match consumers with products they need and help operators optimize inventory structure through manual experience rules and basic sales statistics. However, when faced with concurrent user interactions across multiple applications and complex benefit conversion scenarios, it suffers from problems such as a single evaluation dimension, a rigid evaluation mechanism, and a lag in the allocation of display resources. Summary of the Invention

[0004] To achieve the above objectives, the present invention adopts the following technical solution: a big data-driven method for tiered and optimized selection of goods in an online marketplace, comprising the following steps: S1: Extract the start timestamp and close timestamp of the video platform application and the supermarket shopping application from the user's mobile terminal device interaction log data, and output the video platform application dwell time data and the supermarket shopping application dwell time data. S2: Based on the video platform application dwell time data and the supermarket shopping application dwell time data, extract the overlapping application time segments and the target application intermittent time span data. Based on the overlapping application time segments, perform time-series overlap feature evaluation to generate application cross-domain rights mutual exclusion evaluation quantity. Based on the target application intermittent time span data, perform frequency feature aggregation evaluation to generate application cross-domain rights complementary evaluation quantity. S3: Extract the virtual card sales order generation timestamp and redemption activation timestamp from the historical transaction status change record data, calculate the corresponding virtual card average redemption cycle data, compare it with the platform's financial turnover upper limit threshold, and generate the time utility decay penalty amount for a single product category. S4: Based on the application cross-domain rights complementarity assessment quantity, application cross-domain rights mutual exclusion assessment quantity, and single-category time utility decay penalty quantity, a unified mapping is performed to construct cross-domain rights positive gain assessment data, first negative deduction assessment data, and second negative deduction assessment data. Multi-dimensional feature deviation comprehensive mapping is performed on the cross-domain rights positive gain assessment data, first negative deduction assessment data, and second negative deduction assessment data to generate single-category final rating score data. S5: Extract the corresponding target product category based on the final grading score data of the single product category, sort the target product categories in descending order, allocate display positions, reorganize the page rendering display resources of the corresponding products, and generate big data-driven mall product grading and optimization control instructions.

[0005] As a further aspect of the present invention, during the extraction of the overlapping application time segments, a user interaction reference time axis with a 24-hour cycle is constructed. The start and end time nodes corresponding to the video platform application dwell time data are projected onto the user interaction reference time axis, and the start and end time nodes corresponding to the supermarket shopping application dwell time data are synchronously mapped onto the user interaction reference time axis. The overlapping application time segments covered by both the video platform application dwell time data and the supermarket shopping application dwell time data are extracted from the user interaction reference time axis. During the extraction of the target application intermittent time span data, the application intermittent time gap span data is calculated based on the end time node of the video platform application dwell time data and the start time node of the supermarket shopping application dwell time data. A seamless alternation judgment threshold is established by statistically analyzing historical user continuous operation behavior records. If the application intermittent time gap span data is lower than the seamless alternation judgment threshold, the corresponding application intermittent time gap span data is included in the target application intermittent time span data; if the application intermittent time gap span data is not lower than the seamless alternation judgment threshold, the corresponding application intermittent time gap span data is filtered out.

[0006] As a further aspect of the present invention, during the generation of the single-category time utility decay penalty, if the average redemption period data of the virtual coupon is greater than the upper limit threshold of the platform's financial turnover, it is determined to be in an overdue state, and the average redemption period data of the virtual coupon is extracted to obtain the average redemption period data of the overdue virtual goods. Based on the average redemption period data of the overdue virtual goods and the upper limit threshold of the platform's financial turnover, deviation features are extracted to obtain overdue time deviation data. The exponential decay function is used to map the capital utility decay features to generate the single-category time utility decay penalty. If the average redemption period data of the virtual coupons is not greater than the upper limit threshold of the platform's financial turnover, then the average redemption period data of the virtual coupons is given a zero utility depreciation state and a single-category time utility decay penalty is generated.

[0007] As a further aspect of the present invention, step S1 specifically comprises: S111: Collect user mobile terminal device interaction log data within the natural month cycle, extract the video platform application startup timestamp and video platform application shutdown timestamp, perform difference calculation operation on the above startup and shutdown timestamps, and obtain video platform application dwell time data; S112: Classify and filter user mobile terminal device interaction log data within the same statistical period, extract the supermarket shopping application startup timestamp indicating the start of the supermarket application behavior, and simultaneously extract the supermarket shopping application shutdown timestamp indicating the shutdown of the supermarket application behavior. Perform the same time span difference extraction calculation to generate supermarket shopping application dwell time data.

[0008] As a further aspect of the present invention, step S2 specifically includes: S211: Obtain the video platform application dwell time data and the supermarket shopping application dwell time data to construct a user interaction benchmark time axis, map the two together, extract the overlapping application time segments covered at the same time and calculate the time sequence overlap features, and generate the application cross-domain rights mutual exclusion evaluation quantity. S212: Based on the application dwell time data of the video platform application and the dwell time data of the supermarket shopping application, extract the application intermittent time gap span data between the nodes, establish a seamless alternation judgment threshold for historical continuous operation behavior, compare the values ​​and retain the span content below the threshold range, and establish target application intermittent time span data. S213: Based on the cross-domain rights and interests mutual exclusion evaluation quantity of the application and the intermittent time span data of the target application, cross-frequency feature aggregation and extraction are performed on all the intermittent time span data of the target application. Combined with sliding time window processing, all aggregated feature values ​​are mapped and evaluated. Feature complementarity difference operation transformation is performed to obtain the cross-domain rights and interests mutual exclusion evaluation quantity of the application.

[0009] As a further aspect of the present invention, step S3 specifically comprises: S311: Monitor historical transaction status change records, extract the virtual coupon sales order generation timestamp and corresponding virtual coupon redemption activation timestamp from the order dimension of the record data, extract the latency difference feature to obtain the redemption cycle duration data, and calculate the average redemption cycle data of a single category of virtual goods after summarizing and aggregating the cycle feature values. S312: Based on the historical annual financial turnover audit records of the mall, establish the upper limit threshold of the platform's financial turnover, compare and judge the average redemption cycle data of the single category of virtual goods with the upper limit threshold of the platform's financial turnover, extract the corresponding content of the average value of the cycle that is greater than the upper limit threshold of the platform's financial turnover, perform numerical deviation difference calculation on the two, and obtain the overdue time deviation data. S313: Based on the overdue time deviation data calculated above and the preset capital occupation depreciation parameters, perform capital utility value mapping decay calculation. If the average exchange cycle data of a single category of virtual goods is not greater than the platform's financial turnover upper limit threshold, then assign a zero utility depreciation state and perform corresponding penalty depreciation processing. Combine the characteristic values ​​of the quantile depreciation state and the penalty depreciation amount to generate the time utility decay penalty amount for a single category.

[0010] As a further aspect of the present invention, step S4 specifically comprises: S411: Based on the cross-domain equity complementarity assessment quantity and the cross-domain equity mutual exclusion assessment quantity, perform range standard dimension mapping on both and the single-category time utility decay penalty quantity, extract the complementary standard correlation index and the mutual exclusion standard correlation index, and perform positive feature fusion extraction calculation based on the complementary standard correlation index and the preset complementary gain weight data to obtain cross-domain equity positive gain assessment data. S412: For the attenuation standard correlation index, negative feature fusion extraction is performed based on the mutually exclusive standard correlation index and the preset mutually exclusive attenuation weight to generate the first negative deduction evaluation data. At the same time, negative feature numerical product fusion extraction is performed based on the attenuation standard correlation index and the preset time loss weight to establish the second negative deduction evaluation data. S413: Based on the cross-domain positive gain assessment data, the first negative deduction assessment data, and the second negative deduction assessment data, perform multi-dimensional feature deviation correlation mapping on the three together, perform grading numerical score merging, and obtain the final grading score data for a single product category.

[0011] As a further aspect of the present invention, step S5 specifically comprises: S511: Extract the final rating score data of the single product category and schedule the dynamic display position of the target product category. Arrange the target product categories in descending order according to the final rating score data of the single product category. Establish a high-level percentile threshold based on the capacity of the core recommendation position on the mall page and establish a low-level percentile threshold based on the mall's long-tail product hiding and containment strategy. If the final rating score data of the single product category is higher than the high-level percentile threshold, extract the corresponding target product category to generate high-potential product image display materials. If the final rating score data of the single product category is lower than the low-level percentile threshold, extract the corresponding target product category to generate long-term retention product image display materials. If the final rating score data of the single product category is between the high-level percentile threshold and the low-level percentile threshold, extract the corresponding target product category and return it to the original basic product display sequence. S512: Assign a core recommendation position on the first screen to the high-potential product image display material to obtain a first position allocation result; and assign a hidden downgraded recommendation display position area to the long-term retention product image display material to obtain a second position allocation result. S513: Based on the first position allocation result and the second position allocation result, uniformly execute the page rendering product image display resource corresponding merging and reorganization arrangement operation, and establish a big data-driven mall product hierarchical selection control instruction.

[0012] A big data-driven e-commerce platform product tiering and optimization system includes: The dwell time extraction module extracts the start timestamps and close timestamps of video platform applications and supermarket shopping applications from the user's mobile terminal device interaction log data, and outputs the dwell time data of video platform applications and supermarket shopping applications. The cross-domain rights assessment module extracts overlapping application time segments and target application intermittent time span data based on the video platform application dwell time data and the supermarket shopping application dwell time data. It performs time-series overlap feature assessment based on the overlapping application time segments to generate cross-domain rights mutual exclusion assessment quantity and performs frequency feature aggregation assessment based on the target application intermittent time span data to generate cross-domain rights complementary assessment quantity. The utility decay assessment module extracts the virtual card sales order generation timestamp and redemption activation timestamp from the historical transaction status change record data, calculates the average redemption cycle data of the corresponding virtual card, compares it with the platform's financial turnover upper limit threshold, and generates the time utility decay penalty amount for a single category. The rating score generation module performs a unified mapping based on the application cross-domain rights complementarity assessment quantity, application cross-domain rights mutual exclusion assessment quantity, and single-category time utility decay penalty quantity to construct cross-domain rights positive gain assessment data, first negative deduction assessment data, and second negative deduction assessment data. It then performs a multi-dimensional feature deviation comprehensive mapping on the cross-domain rights positive gain assessment data, first negative deduction assessment data, and second negative deduction assessment data to generate the final rating score data for the single category. The hierarchical optimization scheduling module extracts the corresponding target product category based on the final grading score data of the single product category, arranges the target product category in descending order, allocates display positions, reorganizes the page rendering display resources of the corresponding products, and generates big data-driven e-commerce product hierarchical optimization control instructions.

[0013] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, the temporal interaction features of video applications and shopping applications are extracted to quantify the mutual exclusion and complementarity of cross-domain rights. A time utility decay penalty mechanism is constructed by combining the virtual card redemption cycle. Furthermore, multi-dimensional positive and negative evaluation data are comprehensively mapped to generate a rating score. Through these technical means, the target card is dynamically and accurately rated and ranked, which plays a role in intelligently allocating page display positions, dynamically reorganizing rendering display resources, and effectively improving the efficiency of product circulation and exposure on the platform. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying 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.

[0015] Figure 1 This is a schematic diagram of the steps of the present invention; Figure 2 This is a flowchart illustrating the specific steps of S1 in this invention; Figure 3 This is a flowchart illustrating the specific steps of S2 in this invention; Figure 4 This is a flowchart illustrating the specific steps of S3 in this invention; Figure 5 This is a flowchart illustrating the specific steps of S4 in this invention; Figure 6 This is a flowchart illustrating the specific steps of S5 in this invention; Figure 7 This is a system module diagram of the present invention. Detailed Implementation

[0016] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0017] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0018] Please see Figure 1 This invention provides a big data-driven method for tiered and optimized selection of goods in an online marketplace, comprising the following steps: S1: Extract the start timestamp and close timestamp of the video platform application and the supermarket shopping application from the user's mobile terminal device interaction log data, and output the video platform application dwell time data and the supermarket shopping application dwell time data. S2: Based on the video platform application dwell time data and the supermarket shopping application dwell time data, extract the overlapping application time segments and the target application intermittent time span data. Based on the overlapping application time segments, perform time series overlap feature evaluation to generate application cross-domain rights mutual exclusion evaluation quantity. Based on the target application intermittent time span data, perform frequency feature aggregation evaluation to generate application cross-domain rights complementary evaluation quantity. A user interaction baseline timeline with a 24-hour cycle is constructed. The start and end time nodes corresponding to the video platform application dwell time data are projected onto the user interaction baseline timeline. The start and end time nodes corresponding to the supermarket shopping application dwell time data are synchronously mapped onto the user interaction baseline timeline. The overlapping application time segments in the user interaction baseline timeline that are simultaneously covered by the video platform application dwell time data and the supermarket shopping application dwell time data are extracted. Based on the termination time of video platform application dwell time data and the start time of supermarket shopping application dwell time data, calculate the application intermittent time interval span data. Establish a seamless alternation judgment threshold by statistically analyzing historical user continuous operation records. If the application intermittent time interval span data is lower than the seamless alternation judgment threshold, the corresponding application intermittent time interval span data is included in the target application intermittent time interval span data; if the application intermittent time interval span data is not lower than the seamless alternation judgment threshold, the corresponding application intermittent time interval span data is filtered out. S3: Extract the virtual card sales order generation timestamp and redemption activation timestamp from the historical transaction status change record data, calculate the corresponding virtual card average redemption cycle data, compare it with the platform's financial turnover upper limit threshold, and generate the time utility decay penalty amount for a single product category. If the average redemption period of virtual coupons exceeds the platform's financial turnover limit, it is determined to be in an overdue state. The average redemption period of virtual coupons is then extracted to obtain the average redemption period of overdue virtual goods. Based on the average redemption period of overdue virtual goods and the platform's financial turnover limit, deviation features are extracted to obtain overdue time deviation data. Fund utility decay feature mapping is then performed to generate a single-category time utility decay penalty. If the average redemption period of virtual coupons is not greater than the platform's financial turnover limit, the average redemption period of virtual coupons is assigned a zero utility loss state and a single-category time utility decay penalty is generated. S4: Based on the evaluation quantities of cross-domain rights complementarity, cross-domain rights mutual exclusion, and single-category time utility decay penalty, a unified mapping is performed to construct cross-domain rights positive gain evaluation data, first negative deduction evaluation data, and second negative deduction evaluation data. Multi-dimensional feature deviation is comprehensively mapped on the cross-domain rights positive gain evaluation data, first negative deduction evaluation data, and second negative deduction evaluation data to generate the final grading score data for single-category. S5: Extract the corresponding target product category based on the final grading score data of a single product category, sort the target product categories in descending order, allocate display positions, reorganize the page rendering display resources of the corresponding products, and generate big data-driven mall product grading and selection control instructions.

[0019] Video platform application dwell time data includes cumulative video playback time and live interaction time; supermarket shopping application dwell time data includes product browsing dwell time and checkout page dwell time; cross-domain rights mutual exclusion assessment metrics include attention competition shift loss and purchase intention interruption coefficient; cross-domain rights complementary assessment metrics include cross-scenario consumption demand continuity gain value and potential shopping interest synergy stimulation index; single-category time utility decay penalty metrics include capital occupation implicit cost loss and rights realization delay downgrade score; single-category final grading score data includes product display bidding reference benchmark value and product exposure ranking absolute weight; big data-driven mall product grading and selection control instructions include front-end slot layout dynamic adjustment parameters and server-side card refresh scheduling instructions.

[0020] Please see Figure 2 The specific steps of S1 are as follows: S111: Collect user mobile terminal device interaction log data within the natural month cycle, extract the video platform application startup timestamp and video platform application shutdown timestamp, perform difference calculation operation on the above startup and shutdown timestamps, and obtain video platform application dwell time data; The system directly calls the underlying event listening interface of the mobile terminal device system layer to collect user mobile terminal device interaction log data in real time within a natural month cycle of typically 30 days. The collected raw log data undergoes data cleaning and preprocessing. Invalid log entries with missing timestamps are removed using null value filtering rules, and Kalman filtering is used to handle abnormal signals of timestamp jumps. Subsequently, using process identifier matching, the system extracts the video platform application startup timestamp corresponding to the moment the target video platform application switches from the background to the foreground or the moment it is first launched. Simultaneously, it extracts the video platform application shutdown timestamp corresponding to the moment the target video platform application returns to the background or the moment the process completely terminates. A strict difference calculation operation is performed on the two extracted timestamp feature values. Specifically, by extracting the specific value of the video platform application shutdown timestamp and subtracting the specific value of the video platform application startup timestamp, the absolute dwell time of a single application session is derived. Then, the dwell time data of all single sessions within the natural month period are fully accumulated and summarized to obtain complete dwell time data of video platform applications. For example, for a certain terminal device, the video platform application start timestamp is extracted as 1682899200, and the corresponding video platform application close timestamp is 1682902800. Substituting 1682902800 and 1682899200 directly into the above difference calculation logic, the single session duration of 3600 is obtained. This calculation process aims to accurately characterize the user's consumption stickiness for video streaming media, thereby providing the underlying user behavior characteristics for the subsequent refined classification and selection of special big data driven mall products such as video platform membership cards. After summing 30 similar calculations in the current month, the dwell time data of video platform applications is obtained as 108000. By directly subtracting the underlying timestamps and strictly eliminating extremely short invalid sessions, the redundant duration interference caused by the device screen-off state is eliminated.

[0021] S112: Classify and filter user mobile terminal device interaction log data within the same statistical period, extract the supermarket shopping application startup timestamp representing the opening of the supermarket application behavior, and simultaneously extract the supermarket shopping application closing timestamp representing the closing of the supermarket application behavior. Perform the same time span difference extraction calculation to generate supermarket shopping application dwell time data. Within the same monthly statistical period, the user mobile terminal device interaction log data, which has already been cleaned and filtered in the previous step, is directly read. This batch of log data undergoes strict data classification and attribute filtering. By identifying the application's package name characteristics, log entries belonging to the supermarket shopping business category are separated and archived. Then, the supermarket shopping application startup timestamp, indicating the activation of the supermarket application, is extracted separately. Simultaneously, the supermarket shopping application shutdown timestamp, indicating the closure of the supermarket application, is extracted from the archived entries. Following the time span difference logic of the aforementioned steps, the same time span difference extraction calculation operation is performed. That is, by subtracting the corresponding supermarket shopping application startup timestamp from the specific value of the supermarket shopping application shutdown timestamp, the resident data of each supermarket shopping application is derived. Retention time data is used. For example, if the startup timestamp of a supermarket shopping app is detected as 1682906400 and the corresponding closing timestamp is 1682908200, the difference between 1682908200 and 1682906400 is calculated to obtain a single-time retention time of 1800. By independently capturing the behavior of this category of apps, the potential activity of users in daily shopping scenarios can be objectively assessed. This supports the prediction of conversion probability and hierarchical optimization scheduling of mall products driven by core big data, such as supermarket cards. After summarizing similar calculations of the full data for the month, the monthly retention time of supermarket shopping apps is found to be 54,000. Through precise isolation of time periods across app behaviors and independent feature extraction calculations, retention time pollution from other non-shopping apps is eliminated.

[0022] Table 1 User Terminal Device Interaction Time Record Table Video applications 1682899200 1682902800 3600 Supermarket Applications 1682906400 1682908200 1800 As shown in Table 1, by comparing the start and stop timestamp records of each application type, the difference extraction and calculation process of the duration of a single interaction was verified, providing an accurate underlying factual data foundation for subsequent cross-domain analysis.

[0023] Please see Figure 3 The specific steps of S2 are as follows: S211: Obtain video platform application dwell time data and supermarket shopping application dwell time data to construct a user interaction benchmark timeline, map the two together, extract the overlapping application time segments covered at the same time and calculate the time series overlap features to generate application cross-domain rights mutual exclusion evaluation quantity. The process involves obtaining the video platform application dwell time data and the supermarket shopping application dwell time data derived in the previous steps. These two types of data are then mapped onto a unified user interaction benchmark timeline. A comparison and mapping operation of their time spans is performed. By scanning the data coverage area on the benchmark timeline, overlapping application time segments where both are simultaneously active are rigorously extracted. Subsequently, a time-series overlap feature calculation is performed on these overlapping application time segments. Specifically, the quotient of the extracted overlapping application time segment's duration is divided by the total dwell time of the supermarket shopping application, thereby generating an application cross-domain rights mutual exclusion assessment quantity. For example, in the initial processing, the video platform application dwell time data might cover a certain evening period, and the supermarket shopping application's dwell time data might also cover a certain evening period. The data on the dwell time of shopping apps also covers this period. The duration of the overlapping application time segment between the two is extracted to be 16200. The total dwell time of supermarket shopping apps in this period is known to be 54000. Substituting 16200 and 54000 directly into the above quotient calculation logic, a cross-domain benefit mutual exclusion evaluation value of 0.3 is derived. This kind of cross-domain data correlation mapping reveals the competitive situation between video platform membership card benefits and supermarket card benefits in the allocation of user attention resources. This is a key preliminary qualitative assessment in the big data-driven mall product combination strategy to avoid homogeneous recommendations and achieve scientific hierarchical selection. By extracting the proportional quotient feature of the cross-application overlap duration, the degree of user attention dispersion under the state of multiple applications concurrently is accurately characterized.

[0024] S212: Based on the video platform application dwell time data and the supermarket shopping application dwell time data, extract the application intermittent time interval span data between nodes, establish a seamless alternation judgment threshold for historical continuous operation behavior, compare the values ​​and retain the span content below the threshold range, and establish the target application intermittent time span data. Based on the mapped video platform application dwell time data and supermarket shopping application dwell time data, the application interruption time gap span data between adjacent nodes on the baseline timeline is extracted. Historical continuous operation records from the past 6 months are retrieved simultaneously. Feature values ​​are extracted from the 80th percentile of the historical gap data distribution through statistical calculation, and a seamless alternation judgment threshold is established. The extracted application interruption time gap span data is rigorously compared with this seamless alternation judgment threshold. Invalid long-term interruption records exceeding the threshold are filtered out, and span content below the threshold is retained, thus establishing the final target application interruption time gap data. For example, in a specific operation sequence, the application interruption time gap span data between the end of the video platform application and the start of the supermarket shopping application is extracted to be 1500. Historical operation distribution statistics set the seamless alternation judgment threshold for this type of switching behavior to 1800. The 1500 and 1800 were directly compared and judged. Since 1500 is clearly below the threshold of 1800, the gap span value of 1500 was directly retained and established as the valid target application intermittent time span data. The above judgment operation filtered out irrelevant long interruptions, ensuring that the retained data can truly reflect the user's continuous flow between video and shopping scenarios. This helps to determine whether the user has the potential for cross-consumption of video platform membership cards and supermarket cards at the same time. It provides high-value continuity time sequence feature guarantee for big data-driven cross-domain hierarchical selection of mall goods. Through dynamic percentile distribution threshold judgment and elimination of inferior long interruption features, the data purity of cross-application operation behavior continuity analysis is guaranteed.

[0025] S213: Based on the cross-domain rights and interests mutual exclusion assessment quantity and the intermittent time span data of the target application, cross-frequency feature aggregation and extraction are performed on all intermittent time span data of the target application. Combined with sliding time window processing, all aggregated feature values ​​are mapped and evaluated. Feature complementarity difference operation is performed to transform and obtain the cross-domain rights and interests mutual exclusion assessment quantity. Based on the aforementioned steps, the application cross-domain rights mutual exclusion evaluation quantity of 0.3 and the extracted effective target application intermittent time span data are used to perform cross-frequency feature summation and aggregation extraction operations on all target application intermittent time span data. Combined with a 24-hour sliding time window processing mechanism, all aggregated feature values ​​are smoothed and mapped for evaluation. Subsequently, feature complementarity difference operations are performed for transformation. Specifically, the application cross-domain rights complementarity evaluation quantity is derived by subtracting the product of the application cross-domain rights mutual exclusion evaluation quantity and the preset mutual exclusion penalty weight from the sliding aggregated total value of the target application intermittent time span data. For example, within a specific 24-hour sliding time window, the summation and aggregation operation on multiple sets of target application intermittent time span data yields an aggregated total value of 4500. Simultaneously, the system sets a fixed mutual exclusion penalty weight of 2000. Using the cross-domain rights mutual exclusion evaluation quantity of 0.3 determined in the previous steps, and multiplying it with 2000, a penalty of 600 is obtained. Then, the aggregated total value of 4500 is compared with this penalty of 600 to derive a cross-domain rights complementarity evaluation quantity of 3900. Through this sliding evaluation system with a penalty mechanism, the complementary gain effect of video platform membership cards and supermarket cards as bundled related rights can be accurately quantified. This directly drives the subsequent weight calculation for the hierarchical selection and ranking of mall goods driven by these two core big data types. By introducing sliding window aggregation and dynamic correction and transformation based on feature differences using mutual exclusion factors, a refined quantitative expression of cross-application complementary behavior is achieved.

[0026] Please see Figure 4 The specific steps of S3 are as follows: S311: Monitor historical transaction status change records, extract the virtual coupon sales order generation timestamp and corresponding virtual coupon redemption activation timestamp from the order dimension of the record data, extract the latency difference feature to obtain the redemption cycle duration data, and calculate the average redemption cycle data of a single category of virtual goods after summarizing and aggregating the cycle feature values. Continuously monitor historical transaction status change records in the order database. From this batch of records, accurately extract the virtual coupon sales order generation timestamp at the order segment level, and the corresponding virtual coupon redemption activation timestamp for that single product category. Perform absolute latency difference feature extraction calculation on these two timestamps. Subtract the sales order generation timestamp from the redemption activation timestamp to obtain the single redemption cycle duration. Further, perform full summation and arithmetic averaging on the durations of all redemption cycles under that single product category to finally derive the average redemption cycle data for that single product category of virtual goods. For example, if the underlying interface monitoring reveals a virtual coupon sales order generation timestamp of 1683000000, and the corresponding redemption... The activation timestamp is 1683259200. A difference operation is performed between 1683259200 and 1683000000 to obtain a single redemption cycle duration of 259200. The redemption cycle durations of 100 orders within the same category are extracted and arithmetically averaged to calculate the average redemption cycle for a single category of virtual goods, resulting in a value of 250,000. The core virtual order targets analyzed here deeply encompass high-frequency, data-driven mall products such as video platform membership cards and supermarket cards. In-depth measurement of their entire lifecycle is the fundamental defense to ensure that the tiered selection strategy does not expire or lag. By directly focusing on the extraction and large-scale average aggregation of the order-level status change time difference, statistical bias interference caused by extreme redemption behaviors is eliminated.

[0027] Table 2 Virtual Card Transaction Record Table 78945612 1683000000 1683259200 259200 78945613 1683010000 1683251000 241000 As shown in Table 2, by extracting the core timestamps of multiple sets of virtual coupon orders and performing difference calculations, the acquisition mechanism of redemption cycle duration data was verified, providing detailed order-level data support for subsequent analysis of the average redemption cycle pattern.

[0028] S312: Based on the historical annual financial turnover audit records of the mall, establish the upper limit threshold of the platform's financial turnover, compare and judge the average redemption cycle data of single-category virtual goods with the upper limit threshold of the platform's financial turnover, extract the corresponding content of the average value of the cycle that exceeds the upper limit threshold of the platform's financial turnover, perform numerical deviation difference calculation on the two, and obtain the overdue time deviation data. A comprehensive analysis of the e-commerce platform's backend historical annual financial flow audit records spanning 12 months was conducted. The average fund transfer rate was extracted to establish a platform financial flow upper limit threshold. The average redemption cycle data for 250,000 single-category virtual goods calculated in the previous steps was directly compared with this platform financial flow upper limit threshold. When a specific condition was verified where the average redemption cycle data for single-category virtual goods exceeded the platform financial flow upper limit threshold, the corresponding average cycle value under this condition was extracted. A strict difference calculation was performed on the two values. By subtracting the platform financial flow upper limit threshold from the average redemption cycle data for single-category virtual goods, the overdue time deviation data was obtained. For example, based on the financial audit flow report statistics, the platform financial flow upper limit threshold was set to 172,800. The average redemption cycle data for single-category virtual goods was then subtracted from the platform financial flow upper limit threshold. The average redemption period data for virtual goods, 250,000, is compared with the threshold of 172,800. It is confirmed that 250,000 is significantly greater than the threshold of 172,800. This triggers the numerical deviation calculation logic, directly subtracting 250,000 from 172,800 to derive an overdue time deviation of 77,200. By accurately identifying overdue time deviations, the system can proactively warn against video platform membership cards or supermarket cards that encounter conversion bottlenecks and slow redemption. This prevents such stagnant, big data-driven mall goods from occupying core traffic for extended periods, laying a solid data anchor for implementing a healthy product grading, selection, filtering, and removal mechanism. By introducing a circulation upper limit threshold calibrated by financial audit data for over-limit difference judgment, the degree of financial liquidity obstruction caused by virtual goods backlog is accurately quantified.

[0029] S313: Based on the overdue time deviation data calculated above and the preset capital occupation depreciation parameters, perform capital utility value mapping decay calculation. If the average exchange cycle data of a single category of virtual goods is not greater than the platform's financial turnover upper limit threshold, then assign a zero utility depreciation state and perform corresponding penalty depreciation processing. Merge the characteristic values ​​and penalty depreciation amounts of the quantile depreciation states to generate a single category time utility decay penalty amount. Based on the overdue time deviation data of 77200 obtained through the above comparative calculation, and combined with the system's preset capital occupation depreciation parameter reflecting the unit capital occupation cost, the capital utility value mapping decay calculation logic of the product operation is executed. Simultaneously, a supplementary judgment branch is set: if the average redemption cycle data of the preceding single-category virtual goods is not greater than the platform's financial turnover upper limit threshold, then a zero utility depreciation state is directly assigned and a corresponding constant-level penalty conversion is executed. Finally, the actual generated quantile depreciation state characteristic value and the fixed penalty conversion amount are merged to generate the single-category time utility decay penalty amount. For example, in the actual execution process, the determined overdue time deviation data is 77200, and the preset capital occupation depreciation parameter is set to 0.05 according to the financial discounting rules. Since the preceding judgment did not trigger a zero-loss judgment branch below the threshold, there is no need to superimpose a fixed penalty calculation amount. Instead, the product of 77200 and the preset capital occupation loss parameter of 0.05 is directly multiplied and fused for calculation, resulting in a single-category time utility decay penalty amount of 3860. The introduction of this decay penalty amount parameter directly quantifies the opportunity cost of the time the goods are stuck and converts it into a negative impact factor for subsequent grading assessment. This ensures that when implementing cross-domain grading and selection for video platform membership cards and supermarket cards, those big data-driven mall goods with extremely low financial capital utility and slow turnover efficiency can be objectively downgraded. By setting the product decay conversion mapping of deviation difference and loss parameter, the implicit capital cost consumption of overdue unredeemed goods accumulated over time is accurately simulated.

[0030] Please see Figure 5 The specific steps of S4 are as follows: S411: Based on the cross-domain equity complementarity assessment quantity and the cross-domain equity mutual exclusion assessment quantity, perform range standard dimension mapping on both and the single-category time utility decay penalty quantity, extract the complementary standard correlation index and the mutual exclusion standard correlation index, and perform positive feature fusion extraction calculation based on the complementary standard correlation index and the preset complementary gain weight data to obtain the cross-domain equity positive gain assessment data. The aforementioned derived cross-domain equity complementarity assessment quantity, cross-domain equity mutual exclusion assessment quantity, and single-category time utility decay penalty quantity are extracted. A maximum-minimum-range standard dimension mapping operation is performed on the values ​​of these three dimensions to uniformly scale their numerical space to a unified standardized interval. The scaled complementary standard correlation index and mutual exclusion standard correlation index are then extracted. Subsequently, based on the complementary standard correlation index and the preset complementary gain weight data set by the business side, a product-form positive feature fusion extraction calculation is performed to obtain the cross-domain equity positive gain assessment data. For example, if the complementary assessment quantity is 3900 before mapping, after range dimension mapping compression, the extracted complementary standard correlation index is 0.8. Simultaneously, the same compression process is also performed on the extracted... The obtained mutual exclusion standard correlation index is 0.3, and the preset complementary gain weight data value reflecting the superposition effect of rights and interests is set to 0.6. The complementary standard correlation index of 0.8 and the gain weight data of 0.6 are directly subjected to product feature fusion extraction calculation, and the cross-domain rights and interests positive gain evaluation data of 0.48 is derived. This positive feature fusion extraction link is essentially a weighted incentive for the cross-scenario joint promotion efficiency of video platform membership cards and supermarket cards. It enables big data driven mall products with high exposure synergy value and cross conversion potential to accumulate first-mover advantage in the hierarchical selection system. The difference in the dimensions of multi-source heterogeneous parameters is smoothed out by the range standard dimension alignment operation, and the positive feedback incentive effect of cross-domain complementary characteristics is amplified by the weight product.

[0031] Table 3. Multidimensional Feature Dimension Mapping Standard Table Complementary rights and interests 3900 0.8 Mutually exclusive rights dimension 0.3 0.3 As shown in Table 3, by listing the original input data of different evaluation dimensions and the standardized mapping index generated after range mapping calculation, the core process of multidimensional dimension elimination transformation is demonstrated, which provides a standardized input basis for subsequent feature fusion.

[0032] S412: For the attenuation standard correlation index, negative feature fusion extraction is performed based on the mutually exclusive standard correlation index and the preset mutually exclusive attenuation weight to generate the first negative deduction evaluation data. At the same time, negative feature numerical product fusion extraction is performed based on the attenuation standard correlation index and the preset time loss weight to establish the second negative deduction evaluation data. For the attenuation standard correlation index extracted synchronously in the dimensional mapping stage, firstly, a direct multiplicative negative feature fusion extraction operation is performed based on the previously established mutually exclusive standard correlation index and the preset mutually exclusive attenuation weight reflecting the exclusivity property, generating the first negative deduction assessment data. Simultaneously, based on the aforementioned attenuation standard correlation index and the preset time depreciation weight reflecting the time-sensitive cost, a pure negative feature value product fusion extraction calculation is performed synchronously to independently establish the second negative deduction assessment data. For example, in the actual calculation process, the aforementioned mutually exclusive standard correlation index is obtained as 0.3, and the preset mutually exclusive attenuation weight is set to 0.4 according to the exclusivity strategy. The product fusion calculation of 0.3 and 0.4 yields the first negative deduction assessment data of 0.12. At the same time, the data obtained from the single product category... The decay standard correlation index derived from the time utility decay penalty mapping is 0.5. Based on the financial flow planning, the preset time depreciation weight is set to 0.5. The 0.5 and 0.5 are also multiplied and extracted to derive the second negative deduction assessment data of 0.25. Based on the intervention of the negative feature assessment mechanism with the dual dimensions of mutual exclusion and time depreciation, it can comprehensively block video platform membership cards and supermarket cards with single-scenario exclusivity or serious inventory depreciation risks. It ensures that the big data-driven mall product grading and selection process has a strict risk-avoidance capability and the function of preventing the spillover of inferior assets. Through the dual-line product negative feature extraction mechanism that separates the mutual exclusion dimension and the time depreciation dimension, the negative factors that hinder product conversion are analyzed in detail.

[0033] S413: Based on the cross-domain positive gain assessment data, the first negative deduction assessment data, and the second negative deduction assessment data, perform multi-dimensional feature deviation correlation mapping on the three together, perform grading numerical score merging, and obtain the final grading score data for a single product category. The system comprehensively summarizes the cross-domain benefit positive gain assessment data, the first negative deduction assessment data, and the second negative deduction assessment data generated from the pre-processing. It then performs a direct difference mapping operation on these three numerical indicators based on multi-dimensional feature deviation. Specifically, it uses the positive gain value as a base and sequentially subtracts the two negative deduction values ​​to complete the final grading score calculation, thereby obtaining accurate final grading score data for each product category. For example, the currently extracted cross-domain benefit positive gain assessment data value is 0.48, the first negative deduction assessment data value is 0.12, and the second negative deduction assessment data value is 0.25. The system then combines 0.48,... The three sets of values, 0.12 and 0.25, are simultaneously substituted into the above-mentioned multi-dimensional feature deviation merging and difference operation logic. The value of 0.12 is directly subtracted from 0.48, and then the value of 0.25 is subtracted. Finally, the final classification score data of the single category with a value of 0.11 is obtained. As an absolute judgment indicator for the multi-dimensional comprehensive strength of video platform membership cards and supermarket cards, it directly serves as the last judgment checkpoint for starting the big data-driven fully automatic classification and selection of mall goods and page position allocation. Through the direct difference merging and classification operation of positive and negative bidirectional features, it facilitates the extremely simplified single-value quantitative expression of the multi-dimensional conversion potential of product categories, simplifying the data flow complexity of subsequent decision scheduling.

[0034] Please see Figure 6 The specific steps of S5 are as follows: S511: Extract the final rating score data of a single product category and schedule the dynamic display position of the target product category. Arrange the target product categories in descending order according to the final rating score data of the single product category. Establish a high-level percentile threshold based on the capacity of the core recommendation position on the mall page and establish a low-level percentile threshold based on the mall's long-tail product hiding and containment strategy. If the final rating score data of a single product category is higher than the high-level percentile threshold, extract the corresponding target product category to generate high-potential product image display materials. If the final rating score data of a single product category is lower than the low-level percentile threshold, extract the corresponding target product category to generate long-term retention product image display materials. If the final rating score data of a single product category is between the high-level percentile threshold and the low-level percentile threshold, extract the corresponding target product category and return it to the original basic product display sequence. The final ranking score of 0.11 for each product category is directly extracted and used as a direct benchmark for scheduling the dynamic display position of target product categories. All target product categories participating in the evaluation are strictly sorted in descending order based on their respective final ranking scores. A high-level percentile threshold is set based on the maximum capacity of the core recommendation position on the current mall page, and a low-level percentile threshold is simultaneously established based on the mall's hidden containment and cleanup strategy for long-tail products. Multi-level conditional judgment operations are performed: if the final ranking score of a product category is higher than the high-level percentile threshold, the corresponding target product category is automatically extracted to generate high-potential product image display materials; if the final ranking score of a product category is lower than the low-level percentile threshold, the corresponding target product category is extracted to generate long-term retention product image display materials; if the data is in either of these ranges... Between these, the data is extracted and returned to the original basic product display sequence. For example, a high-level quantile threshold of 0.8 and a low-level quantile threshold of 0.2 are set. The previously derived score of 0.11 is compared with the thresholds of 0.8 and 0.2 respectively. It is determined that 0.11 is clearly lower than the low-level quantile threshold of 0.2, thus triggering the low-quantile response mechanism. The target product category corresponding to the score of 0.11 is directly extracted and output as long-term retention level product image display material. Thus, a closed loop is opened from the underlying cross-domain behavior of user interaction to the downgrading of product display positioning. This makes the multi-category hierarchical selection of video platform membership cards and supermarket cards no longer rely on extensive manual operation experience. It truly realizes an automated and refined distribution and scheduling mechanism for mall products based on big data driven by a strict quantile threshold tier judgment mechanism to achieve a high degree of automated hierarchical management of the product pool.

[0035] S512: Assign the first screen core recommendation position to the high-potential product image display material to obtain the first position allocation result; assign the hidden and downgraded recommendation display position area of ​​the interface to the long-term retention product image display material to obtain the second position allocation result. Based on the results of various display materials generated from the prior hierarchical judgment, a refined page position resource scheduling and configuration operation is performed. For high-potential product image display materials, the front-end page core interface is directly scheduled to allocate the core recommendation position on the first screen of the mall, thus obtaining a precise first position allocation result. For long-term retention product image display materials, the end resource allocation program is initiated to allocate them to a hidden, downgraded recommendation display position area located deep within the page, thus obtaining a second position allocation result. For example, in the actual execution of scheduling and allocation, since a product with a score of 0.11 in the previous stage has been judged and output as a long-term retention product image display material, the resource scheduling operation will not trigger the first screen resource allocation instruction for it, but will directly call the deep page's... The implicit display area location identifier directs the long-term stagnation level product image display material to the collapsed and hidden downgraded recommendation display area at the bottom of the interface. Finally, a second location allocation result confirming the downgraded allocation path is generated. Through this highly polarized slot flow scheduling mode, the mall management can ensure that core products such as video platform membership cards and supermarket cards with high-frequency redemption activity firmly occupy the high traffic position on the front screen. In this way, at the visual interaction display level, the core demand and commercial value conversion expectation of strictly implementing hierarchical selection of mall products driven by big data are accurately realized. By implementing drastically different screen position resource tilting and dispersion scheduling for product materials with different potential levels, the golden exposure resources of the first screen are maximized and the occupation of core traffic by slow-moving products is avoided.

[0036] S513: Based on the first position allocation result and the second position allocation result, uniformly execute the page rendering product image display resource corresponding merging and reorganization arrangement operation, and establish a big data-driven mall product hierarchical selection control instruction; The system comprehensively summarizes and confirms the specific location mapping coordinates and material binding relationships contained in the aforementioned first and second position allocation results. It then synchronously imports both sets of allocation result data into the page dynamic rendering engine processing process, uniformly executing the underlying operations of merging and reorganizing the corresponding product image display resources for page rendering. Using coordinate overlay update instructions, it splices and assembles the high-quality materials and long-cycle materials, after hierarchical scheduling, according to a new spatial topology. Based on the reorganized page view data model, it formally establishes and outputs big data-driven e-commerce product hierarchical optimization control instructions. For example, it performs the following: it takes the core first-screen coordinate parameters of the first position allocation result for high-quality products and the deep folding area coordinate parameters of the second position allocation result for long-cycle products with a score of 0.11. The fusion mapping process performs corresponding merging and splicing of material resources and reorganization and arrangement of view layers within the virtual rendering space. This is encapsulated into a set of big data-driven product hierarchical optimization control instructions that include update configuration parameters for all levels of product image and text slots on all pages. This instruction file is then directly sent to the front-end e-commerce client. The control instruction set generated by the fusion and reorganization serves as the "central nervous system" driving the refresh of product positions across the entire site. It fundamentally solves the technical pain points of fixed positions and slow update flow for e-commerce products driven by massive amounts of big data, such as video platform membership cards and supermarket cards, when displayed on the front end. This ensures that the multi-level hierarchical optimization decision-making conclusions can reach the end-user device interface. Through the integrated fusion of global allocation results and the reorganization of rendered view layers, it facilitates efficient and smooth refresh control of the front-end product presentation status.

[0037] Please see Figure 7 A big data-driven e-commerce product tiering and selection system includes: The dwell time extraction module is used to execute S1: extract the start timestamp and close timestamp of the video platform application and the supermarket shopping application from the user's mobile terminal device interaction log data, and output the dwell time data of the video platform application and the supermarket shopping application. The cross-domain rights assessment module is used to execute S2: Based on the video platform application dwell time data and the supermarket shopping application dwell time data, it extracts the overlapping application time segments and the target application intermittent time span data, performs time series overlap feature assessment based on the overlapping application time segments to generate application cross-domain rights mutual exclusion assessment quantity, and performs frequency feature aggregation assessment based on the target application intermittent time span data to generate application cross-domain rights complementary assessment quantity. The utility decay assessment module is used to perform S3: extract the virtual card sales order generation timestamp and redemption activation timestamp from the historical transaction status change record data, calculate the corresponding virtual card average redemption cycle data, compare it with the platform's financial turnover upper limit threshold, and generate the time utility decay penalty amount for a single product category. The rating score generation module is used to execute S4: Based on the application cross-domain rights complementarity assessment quantity, application cross-domain rights mutual exclusion assessment quantity, and single-category time utility decay penalty quantity, a unified mapping is performed to construct cross-domain rights positive gain assessment data, first negative deduction assessment data, and second negative deduction assessment data. Multi-dimensional feature deviation comprehensive mapping is performed on the cross-domain rights positive gain assessment data, first negative deduction assessment data, and second negative deduction assessment data to generate the final rating score data for the single category. The hierarchical optimization scheduling module is used to execute S5: extract the corresponding target product category based on the final grading score data of a single product category, sort the target product categories in descending order, allocate display positions, reorganize the page rendering display resources of the corresponding products, and generate big data-driven e-commerce product hierarchical optimization control instructions.

[0038] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of protection of the described technical solutions.

Claims

1. A big data-driven method for tiered and optimized selection of goods in an online marketplace, characterized in that: Includes the following steps: S1: Extract the start timestamp and close timestamp of the video platform application and the supermarket shopping application from the user's mobile terminal device interaction log data, and output the video platform application dwell time data and the supermarket shopping application dwell time data. S2: Based on the video platform application dwell time data and the supermarket shopping application dwell time data, extract the overlapping application time segments and the target application intermittent time span data. Based on the overlapping application time segments, perform time-series overlap feature evaluation to generate application cross-domain rights mutual exclusion evaluation quantity. Based on the target application intermittent time span data, perform frequency feature aggregation evaluation to generate application cross-domain rights complementary evaluation quantity. S3: Extract the virtual card sales order generation timestamp and redemption activation timestamp from the historical transaction status change record data, calculate the corresponding virtual card average redemption cycle data, compare it with the platform's financial turnover upper limit threshold, and generate the time utility decay penalty amount for a single product category. S4: Based on the application cross-domain rights complementarity assessment quantity, application cross-domain rights mutual exclusion assessment quantity, and single-category time utility decay penalty quantity, a unified mapping is performed to construct cross-domain rights positive gain assessment data, first negative deduction assessment data, and second negative deduction assessment data. Multi-dimensional feature deviation comprehensive mapping is performed on the cross-domain rights positive gain assessment data, first negative deduction assessment data, and second negative deduction assessment data to generate single-category final rating score data. S5: Extract the corresponding target product category based on the final grading score data of the single product category, sort the target product categories in descending order, allocate display positions, reorganize the page rendering display resources of the corresponding products, and generate big data-driven mall product grading and optimization control instructions.

2. The big data-driven method for tiered selection of e-commerce products according to claim 1, characterized in that: During the extraction of the overlapping application time segments, a user interaction benchmark timeline with a 24-hour cycle is constructed. The start and end time nodes corresponding to the video platform application dwell time data are projected onto the user interaction benchmark timeline. The start and end time nodes corresponding to the supermarket shopping application dwell time data are synchronously mapped onto the user interaction benchmark timeline. The overlapping application time segments covered by both the video platform application dwell time data and the supermarket shopping application dwell time data are extracted from the user interaction benchmark timeline. During the extraction of the target application intermittent time span data, the application intermittent time gap span data is calculated based on the end time node of the video platform application dwell time data and the start time node of the supermarket shopping application dwell time data. A seamless alternation judgment threshold is established by statistically analyzing historical user continuous operation behavior records. If the application intermittent time gap span data is lower than the seamless alternation judgment threshold, the corresponding application intermittent time gap span data is included in the target application intermittent time span data. If the application interruption time gap span data is not lower than the seamless alternation determination threshold, then the corresponding application interruption time gap span data is filtered out.

3. The big data-driven method for tiered selection of e-commerce products according to claim 1, characterized in that: In the process of generating the time utility decay penalty for a single product category, if the average redemption period data of the virtual coupon is greater than the upper limit threshold of the platform's financial turnover, it is determined to be in an overdue state, and the average redemption period data of the virtual coupon is extracted to obtain the average redemption period data of the overdue virtual goods. Based on the average redemption period data of the overdue virtual goods and the upper limit threshold of the platform's financial turnover, deviation features are extracted to obtain overdue time deviation data. The exponential decay function is used to map the capital utility decay features to generate the time utility decay penalty for a single product category. If the average redemption period data of the virtual coupons is not greater than the upper limit threshold of the platform's financial turnover, then the average redemption period data of the virtual coupons is given a zero utility depreciation state and a single-category time utility decay penalty is generated.

4. The big data-driven method for tiered selection of e-commerce products according to claim 1, characterized in that, The specific steps of S1 are as follows: S111: Collect user mobile terminal device interaction log data within the natural month cycle, extract the video platform application startup timestamp and video platform application shutdown timestamp, perform difference calculation operation on the above startup and shutdown timestamps, and obtain video platform application dwell time data; S112: Classify and filter user mobile terminal device interaction log data within the same statistical period, extract the supermarket shopping application startup timestamp indicating the start of the supermarket application behavior, and simultaneously extract the supermarket shopping application shutdown timestamp indicating the shutdown of the supermarket application behavior. Perform the same time span difference extraction calculation to generate supermarket shopping application dwell time data.

5. The big data-driven method for tiered selection of e-commerce products according to claim 1, characterized in that, The specific steps of S2 are as follows: S211: Obtain the video platform application dwell time data and the supermarket shopping application dwell time data to construct a user interaction benchmark time axis, map the two together, extract the overlapping application time segments covered at the same time and calculate the time sequence overlap features, and generate the application cross-domain rights mutual exclusion evaluation quantity. S212: Based on the application dwell time data of the video platform application and the dwell time data of the supermarket shopping application, extract the application intermittent time gap span data between the nodes, establish a seamless alternation judgment threshold for historical continuous operation behavior, compare the values ​​and retain the span content below the threshold range, and establish target application intermittent time span data. S213: Based on the cross-domain rights and interests mutual exclusion evaluation quantity of the application and the intermittent time span data of the target application, cross-frequency feature aggregation and extraction are performed on all the intermittent time span data of the target application. Combined with sliding time window processing, all aggregated feature values ​​are mapped and evaluated. Feature complementarity difference operation transformation is performed to obtain the cross-domain rights and interests mutual exclusion evaluation quantity of the application.

6. The big data-driven method for tiered selection of e-commerce products according to claim 1, characterized in that, The specific steps of S3 are as follows: S311: Monitor historical transaction status change records, extract the virtual coupon sales order generation timestamp and corresponding virtual coupon redemption activation timestamp from the order dimension of the record data, extract the latency difference feature to obtain the redemption cycle duration data, and calculate the average redemption cycle data of a single category of virtual goods after summarizing and aggregating the cycle feature values. S312: Based on the historical annual financial turnover audit records of the mall, establish the upper limit threshold of the platform's financial turnover, compare and judge the average redemption cycle data of the single category of virtual goods with the upper limit threshold of the platform's financial turnover, extract the corresponding content of the average value of the cycle that is greater than the upper limit threshold of the platform's financial turnover, perform numerical deviation difference calculation on the two, and obtain the overdue time deviation data. S313: Based on the overdue time deviation data calculated above and the preset capital occupation depreciation parameters, perform capital utility value mapping decay calculation. If the average exchange cycle data of a single category of virtual goods is not greater than the platform's financial turnover upper limit threshold, then assign a zero utility depreciation state and do not perform penalty conversion or set the penalty conversion amount to zero. Combine the characteristic values ​​of the quantile depreciation state and the penalty conversion amount to generate the single category time utility decay penalty amount.

7. The big data-driven method for tiered selection of goods in an online marketplace according to claim 1, characterized in that, The specific steps of S4 are as follows: S411: Based on the cross-domain equity complementarity assessment quantity and the cross-domain equity mutual exclusion assessment quantity, perform range standard dimension mapping on both and the single-category time utility decay penalty quantity, extract the complementary standard correlation index and the mutual exclusion standard correlation index, and perform positive feature fusion extraction calculation based on the complementary standard correlation index and the preset complementary gain weight data to obtain cross-domain equity positive gain assessment data. S412: For the attenuation standard correlation index, negative feature fusion extraction is performed based on the mutually exclusive standard correlation index and the preset mutually exclusive attenuation weight to generate the first negative deduction evaluation data. At the same time, negative feature numerical product fusion extraction is performed based on the attenuation standard correlation index and the preset time loss weight to establish the second negative deduction evaluation data. S413: Based on the cross-domain positive gain assessment data, the first negative deduction assessment data, and the second negative deduction assessment data, perform multi-dimensional feature deviation correlation mapping on the three together, perform grading numerical score merging, and obtain the final grading score data for a single product category.

8. The big data-driven method for tiered selection of e-commerce products according to claim 1, characterized in that, The specific steps of S5 are as follows: S511: Extract the final rating score data of the single product category and schedule the dynamic display position of the target product category. Arrange the target product categories in descending order according to the final rating score data of the single product category. Establish a high-level percentile threshold based on the capacity of the core recommendation position on the mall page and establish a low-level percentile threshold based on the mall's long-tail product hiding and containment strategy. If the final rating score data of the single product category is higher than the high-level percentile threshold, extract the corresponding target product category to generate high-potential product image display materials. If the final rating score data of the single product category is lower than the low-level percentile threshold, extract the corresponding target product category to generate long-term retention product image display materials. If the final rating score data of the single product category is between the high-level percentile threshold and the low-level percentile threshold, extract the corresponding target product category and return it to the original basic product display sequence. S512: Assign a core recommendation position on the first screen to the high-potential product image display material to obtain a first position allocation result; and assign a hidden downgraded recommendation display position area to the long-term retention product image display material to obtain a second position allocation result. S513: Based on the first position allocation result and the second position allocation result, uniformly execute the page rendering product image display resource corresponding merging and reorganization arrangement operation, and establish a big data-driven mall product hierarchical selection control instruction.

9. A big data-driven e-commerce product tiering and selection system, characterized in that: The system is used to implement the method according to any one of claims 1-8, comprising: The dwell time extraction module extracts the start timestamps and close timestamps of video platform applications and supermarket shopping applications from the user's mobile terminal device interaction log data, and outputs the dwell time data of video platform applications and supermarket shopping applications. The cross-domain rights assessment module extracts overlapping application time segments and target application intermittent time span data based on the video platform application dwell time data and the supermarket shopping application dwell time data. It performs time-series overlap feature assessment based on the overlapping application time segments to generate cross-domain rights mutual exclusion assessment quantity and performs frequency feature aggregation assessment based on the target application intermittent time span data to generate cross-domain rights complementary assessment quantity. The utility decay assessment module extracts the virtual card sales order generation timestamp and redemption activation timestamp from the historical transaction status change record data, calculates the average redemption cycle data of the corresponding virtual card, compares it with the platform's financial turnover upper limit threshold, and generates the time utility decay penalty amount for a single category. The rating score generation module performs a unified mapping based on the application cross-domain rights complementarity assessment quantity, application cross-domain rights mutual exclusion assessment quantity, and single-category time utility decay penalty quantity to construct cross-domain rights positive gain assessment data, first negative deduction assessment data, and second negative deduction assessment data. It then performs a multi-dimensional feature deviation comprehensive mapping on the cross-domain rights positive gain assessment data, first negative deduction assessment data, and second negative deduction assessment data to generate the final rating score data for the single category. The hierarchical optimization scheduling module extracts the corresponding target product category based on the final grading score data of the single product category, arranges the target product category in descending order, allocates display positions, reorganizes the page rendering display resources of the corresponding products, and generates big data-driven e-commerce product hierarchical optimization control instructions.

Images