A Supplier Profile Intelligent Generation System

Abstract

This invention discloses an intelligent supplier profile generation system, relating to the field of data management technology. It effectively verifies the adjustment effect of data retrieval requests by writing standardized raw data packets from suppliers into a distributed transaction log and performing segmented recording and pipeline execution monitoring based on globally unique transactions. Subsequently, the verified data is pushed to a message queue, and event-driven windows are used to detect data disorder and dynamically adjust window parameters, achieving reasonable partitioning and efficient management of message queue data. The partitioned data is then imported into a sliding window manager for window scheduling and monitoring. By calculating the global trend deviation factor of the supplier's business numerical indicator sequence, the window parameters are further optimized and adjusted to improve the accuracy and stability of data timing. Finally, based on a pre-trained supplier profile generation model, a structured profile that dynamically reflects the supplier's business characteristics is intelligently generated.

Description

Technical Field

[0001] This invention relates to the field of data management technology, and in particular to a supplier profile intelligent generation system. Background Technology

[0002] The intelligent generation of supplier profiles stems from the need for comprehensive, accurate, and dynamic control over supplier information in modern supply chain management. The vast amounts of heterogeneous time-series data generated by suppliers in procurement, fulfillment, and logistics—such as order volume, transaction amount, fulfillment cycle, and system interaction performance indicators—contain rich information on business status and behavioral patterns. Efficiently mining and transforming this data into actionable supplier profiles is key to improving supply chain transparency and intelligent management.

[0003] For example, announcement number CN114723492B discloses a method and device for generating enterprise profiles. This addresses the problem that existing enterprise profiles struggle to provide accurate intelligent information recommendations to users. The method involves generating a profile of the goods purchased by the purchasing enterprise based on its browsing and purchase history, and generating a profile of the purchasing enterprise based on publicly available information. It also involves identifying the profiles of related purchasing enterprises' goods purchased based on the product information published by suppliers; determining the set of purchasing enterprises corresponding to the profiles of related purchasing enterprises' goods, and the corresponding set of purchasing enterprise profiles; comparing the data in the set of purchasing enterprise profiles with a pre-set standard enterprise profile data table to obtain the corresponding level result of the purchasing enterprise profile, and determining the comprehensive score corresponding to the purchasing enterprise profile based on the level result; ranking the purchasing enterprises based on the comprehensive score; and pushing the purchasing enterprise information to suppliers based on the ranking result.

[0004] For example, publication number CN119168504A discloses a method for intelligent generation of enterprise profile tags. The method includes the following steps: obtaining industry standard reference data; constructing and processing a tag corpus based on the industry standard reference data to obtain a tag corpus; performing domain knowledge modeling based on the tag corpus to obtain a domain knowledge graph; analyzing the original enterprise data based on the domain knowledge graph to obtain a preprocessed enterprise dataset; performing data fusion and association on the preprocessed enterprise dataset based on the domain knowledge graph to obtain a unified enterprise dataset; extracting enterprise features from the unified enterprise dataset based on the domain knowledge graph to obtain enterprise feature vectors; and performing feature clustering analysis on the enterprise feature vectors to obtain an enterprise cluster graph.

[0005] However, in the process of implementing the inventive technical solution in the embodiments of this application, it was found that the above-mentioned technology has at least the following technical problems: Data acquisition environments often involve uncertainties such as network fluctuations, data disorder, and latency, resulting in inconsistent data quality and increasing the difficulty of generating supplier profiles. Summary of the Invention

[0006] This invention provides a supplier profile intelligent generation system, including a data writing and transaction log management module, which is used to write the standardized raw data packets of suppliers into a distributed transaction log, record them in segments according to globally unique transactions, monitor and verify the pipeline execution process, determine the adjustment of the pull request of the record segment in the data acquisition link, and verify the effect of the adjustment of the pull request of the record segment in the data acquisition link. The event time window detection and adjustment module is used to push the verified standardized raw data packets of the supplier to the message queue, detect the message queue according to the event time driving window, obtain the out-of-order data factor of the standardized raw data packets of the supplier within the event time driving window, determine the adjustment requirements of the window parameters of the event time driving adjustment, adjust the window parameters of the event time driving, and divide the standardized raw data packets of the supplier in the message queue. The sliding window management and trend deviation analysis module is used to import the standardized raw data packets of suppliers in the divided message queue into the sliding window manager for window scheduling and monitoring, obtain the global trend deviation factor of the supplier's business numerical indicator sequence within the sliding window, make a secondary determination of the adjustment needs of the window parameters driven by event time, and readjust the window parameters driven by event time. The supplier profile generation module is used to generate supplier profiles based on a pre-trained supplier profile generation model.

[0007] Furthermore, the pipeline execution process is monitored and verified by segmenting records based on the globally unique transaction. Specifically, the process involves obtaining the time difference between the initiation of the pull request and the return of the first data in each record segment of the supplier's globally unique transaction, the time of receiving the last data, the data writing completion time, and the actual amount of data transmitted.

[0008] The pipeline execution throughput factor for each record segment is obtained by analyzing the proportion of the time difference between the pull request and the return of the first data, the time limit for defining the first packet response, the time for receiving the last data and the maximum allowable transmission latency of batch data, the data write completion time and the maximum write time required for local storage write throughput, as well as the actual amount of data transmitted and the expected number of batches. The pipeline execution throughput factor for each record segment is used to quantitatively evaluate the degree to which the pipeline meets the throughput and latency requirements.

[0009] Furthermore, the process of adjusting the pull request of the record segment in the data acquisition link is as follows: extract the pipeline execution throughput factor of each record segment and compare it with the pipeline execution throughput factor threshold stored in the database. If the pipeline execution throughput factor of a record segment is higher than or equal to the pipeline execution throughput factor threshold, the pull request range of the record segment is located and the pull request is reconstructed. If the pipeline execution throughput factor of a record segment is lower than the pipeline execution throughput factor threshold, it is not necessary to locate the pull request range of the record segment.

[0010] Furthermore, the effect of adjusting the pull request for record segments in the data acquisition chain was verified. The specific process is as follows: After a reconstruction pull request, a verification and monitoring period is preset. During the verification and monitoring period, the number of reconstruction pull requests executed by the pipeline for each record segment after the reconstruction pull request is counted, and the threshold number of reconstruction pull requests executed by the pipeline for each record segment stored in the database is extracted. If the number of reconstruction pull requests executed by the pipeline for each record segment after the reconstruction pull request is higher than or equal to the number of reconstruction pull requests executed by the pipeline for each record segment after the reconstruction pull request, the supplementary pull logic is triggered, and the concurrency and sharding granularity of the data pull task for that segment are dynamically adjusted. If the number of reconstruction pull requests executed by the pipeline for each record segment after the reconstruction pull request is lower than the number of reconstruction pull requests executed by the pipeline for each record segment after the reconstruction pull request, the supplementary pull logic does not need to be triggered.

[0011] Furthermore, the out-of-order data factor of the standardized raw data packets of the supplier within the event time-driven window is obtained. The specific process is as follows: obtain the deviation value between the average time of the event recorded within the event time-driven window and its processing time to the system, which is recorded as the processing delay amount recorded within the event time-driven window, and obtain the frequency of out-of-order packets occurring per unit time and the maximum buffer waiting time generated by correcting the out-of-order packets.

[0012] The out-of-order data factor of the supplier's standardized raw data packets within the event-time driven window is obtained by performing a proportion analysis on the processing delay amount recorded within the event-time driven window, the maximum allowed ordered delay value preset in the database, the frequency of out-of-order packets occurring per unit time, the defined frequency of out-of-order packets occurring per unit time, and the maximum buffer waiting time caused by correcting out-of-order packets, and the maximum cache time limit stored in the database. Weighting coefficients and pipeline execution throughput factors for each record segment are introduced. The out-of-order data factor of the supplier's standardized raw data packets within the event-time driven window is used to quantitatively evaluate the timing consistency and pipeline reliability of the supplier's data flow within the event-time driven window.

[0013] Furthermore, the event-time driven window parameters are adjusted, and the standardized raw data packets of the suppliers in the message queue are divided. Specifically, the out-of-order data factors of the standardized raw data packets of the suppliers within the event-time driven window are extracted and compared with the out-of-order data factor range of the standardized raw data packets of the suppliers within the event-time driven window stored in the database. If the out-of-order data factor of the standardized raw data packets of the suppliers within the event-time driven window exceeds the out-of-order data factor range of the standardized raw data packets of the suppliers within the event-time driven window, then the event-time driven adjustment window parameters are adjusted, and the standardized raw data packets of the suppliers in the message queue are re-divided. If the out-of-order data factor of the standardized raw data packets of the suppliers within the event-time driven window does not exceed the out-of-order data factor range of the standardized raw data packets of the suppliers within the event-time driven window, then no adjustment of the event-time driven adjustment window parameters is required.

[0014] Furthermore, the global trend deviation factor of the supplier's business numerical indicator sequence within the sliding window is obtained. The specific process is as follows: obtain the average trend slope of the target supplier's key business metrics within the sliding window.

[0015] Extract the baseline slope of key business metrics within the sliding window stored in the database.

[0016] The absolute value of the difference between the average trend slope and the baseline slope of the target supplier's key business metrics within the sliding window is used as the global trend deviation factor for the supplier's business numerical indicator series within the sliding window.

[0017] Furthermore, to address the adjustment requirements of the event-time driven adjustment window parameters, the event-time driven window parameters are readjusted. The specific process is as follows: extract the global trend deviation factor of the supplier's business numerical indicator sequence within the sliding window and compare it with the global trend deviation factor threshold range stored in the database. If the global trend deviation factor of the supplier's business numerical indicator sequence within the sliding window exceeds the global trend deviation factor threshold range, then the window width and sliding step size are dynamically readjusted. If the global trend deviation factor of the supplier's business numerical indicator sequence within the sliding window does not exceed the global trend deviation factor threshold range, then the existing sliding window parameters remain unchanged.

[0018] Furthermore, the dynamic readjustment of the window width and sliding step size is triggered. The specific process is as follows: extract the global trend deviation factor of the supplier's business numerical indicator sequence within the sliding window and the upper limit of the threshold range of the global trend deviation factor. Subtract the global trend deviation factor of the supplier's business numerical indicator sequence within the sliding window from the upper limit of the threshold range of the global trend deviation factor to obtain the global trend deviation value of the supplier's business numerical indicator sequence within the sliding window.

[0019] The global trend deviation value of the supplier's business numerical indicator sequence within the sliding window is compared with the window length expansion ratio and sliding step size reduction ratio corresponding to each interval of the global trend deviation value of the supplier's business numerical indicator sequence within the sliding window stored in the database. The length expansion ratio and sliding step size reduction ratio of the sliding window are obtained, and the sliding window is readjusted based on the length expansion ratio and sliding step size reduction ratio.

[0020] Furthermore, a supplier profile is generated based on a pre-trained supplier profile generation model. The specific process is as follows: the supplier's data fragments are aligned according to event time and normalized, the supplier's behavioral feature vector is extracted, the supplier's behavioral feature vector is input into the supplier profile generation model, and a structured supplier profile result is output.

[0021] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: (1) This invention proposes a supplier profile intelligent generation system. By writing the standardized raw data packets of suppliers into a distributed transaction log, and performing segmented recording and monitoring and verification of the pipeline execution process based on globally unique transactions, the adjustment effect of data retrieval requests is effectively verified. Subsequently, the verified data is pushed to a message queue. The event time-driven window is used to detect data disorder and dynamically adjust window parameters to achieve reasonable division and efficient management of message queue data. The divided data is then imported into a sliding window manager for window scheduling and monitoring. By calculating the global trend deviation factor of the supplier business numerical indicator sequence, the window parameters are further optimized and adjusted to improve the accuracy and stability of data time sequence. Finally, based on the pre-trained supplier profile generation model, a structured profile that dynamically reflects the characteristics of supplier business is intelligently generated.

[0022] (2) By obtaining the pipeline execution throughput factor of each record segment, this invention helps to locate the record segment with substandard performance and reconstruct the corresponding pull request to optimize the data pull strategy. Furthermore, it determines whether the supplementary pull logic is triggered, dynamically adjusts the concurrency and sharding granularity of the data pull task to improve data acquisition efficiency and stability, realizes continuous monitoring and intelligent optimization of the supplier data acquisition pipeline, ensures data quality, transmission efficiency and timeliness, and provides a high-quality, stable and timely data foundation for the accurate generation of subsequent supplier profiles, significantly improving the real-time performance and reliability of the profiles.

[0023] (3) By obtaining the out-of-order data factor of the standardized raw data packets of suppliers within the event time-driven window, this invention helps to analyze the dynamic adjustment of the event time-driven window parameters and re-divide the standardized raw data packets of suppliers in the message queue, ensuring data timing quality and processing efficiency. It effectively guarantees the temporal integrity of supplier data streams and the stability of pipeline transmission, providing a high-quality, accurate, and time-consistent data foundation for subsequent intelligent generation of supplier profiles, and improving the accuracy and real-time response capability of profile generation.

[0024] (4) This invention achieves flexible and adaptive window management by precisely adjusting the sliding window parameters. This step effectively improves the trend capture capability and anomaly identification accuracy of supplier business data, ensures the temporal consistency and representativeness of the input data of the supplier profile generation model, enhances the accuracy and real-time response performance of intelligent profile generation, and thus promotes the scientific and intelligent development of supplier behavior analysis and decision support. Attached Figure Description

[0025] 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.

[0026] Figure 1 This is a schematic diagram of the system modules provided in an embodiment of the present invention; Figure 2 This is a flowchart of a supplier profile intelligent generation process provided by an embodiment of the present invention; Figure 3 This is a diagram of the data flow execution monitoring interface in the intelligent supplier profile generation system provided in this embodiment of the invention; Figure 4 This is a line graph showing the load trend in the intelligent generation of supplier profiles provided in this embodiment of the invention. Figure 5 This is a diagram of the data time sequence detection interface in the intelligent generation of supplier profiles provided in this embodiment of the invention. Detailed Implementation

[0027] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0028] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0029] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0030] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0031] 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.

[0032] like Figure 1 As shown, this embodiment of the invention provides a supplier profile intelligent generation system, including: The data writing and transaction log management module is used to write the standardized raw data packets from the supplier into the distributed transaction log, record them in segments according to globally unique transactions, monitor and verify the pipeline execution process, determine the adjustment of the pull request of the record segment in the data acquisition link, and verify the effect of the adjustment of the pull request of the record segment in the data acquisition link.

[0033] The event time window detection and adjustment module is used to push the verified standardized raw data packets of the supplier to the message queue, detect the message queue according to the event time driving window, obtain the out-of-order data factor of the standardized raw data packets of the supplier within the event time driving window, determine the adjustment requirements of the window parameters of the event time driving adjustment, adjust the window parameters of the event time driving, and divide the standardized raw data packets of the supplier in the message queue.

[0034] The sliding window management and trend deviation analysis module is used to import standardized raw data packets of suppliers in the divided message queue into the sliding window manager for window scheduling and monitoring, obtain the global trend deviation factor of the supplier's business numerical indicator sequence within the sliding window, determine the adjustment needs of the event time-driven window parameters, and readjust the event time-driven window parameters.

[0035] The supplier profile generation module is used to generate supplier profiles based on a pre-trained supplier profile generation model.

[0036] The pipeline execution process is monitored and verified by segmenting records based on the globally unique transaction. Specifically, the process involves obtaining the time difference between the initiation of the pull request and the return of the first data in each record segment of the supplier's globally unique transaction, the time of receiving the last data, the data writing completion time, and the actual amount of data transmitted.

[0037] The pipeline execution throughput factor for each record segment is obtained by analyzing the proportion of the time difference between the pull request and the return of the first data, the time limit for defining the first packet response, the time for receiving the last data and the maximum allowable transmission latency of batch data, the data write completion time and the maximum write time required for local storage write throughput, as well as the actual amount of data transmitted and the expected number of batches. The pipeline execution throughput factor for each record segment is used to quantitatively evaluate the degree to which the pipeline meets the throughput and latency requirements.

[0038] It should be noted that the pipeline throughput factor for each record segment is determined by the following specific analysis conditions: ; In the formula, U i X represents the pipeline throughput factor for the i-th record segment. 1i Y represents the time difference between initiating the fetch request for the i-th record segment and returning the first data record, X represents the time limit for defining the first packet response stored in the database, and Y represents the time difference between initiating the fetch request for the i-th record segment and returning the first data record. 2i Z represents the reception time of the last data in the i-th record segment, Y represents the maximum allowable transmission delay of the batch data stored in the database, and Z represents the maximum allowable transmission delay of the batch data stored in the database. 3i S represents the data write completion time of the i-th record segment, Z represents the maximum write time required for the local storage write throughput in the database, and S represents the maximum write time required for the local storage write throughput in the database. 4iLet S represent the actual amount of data transmitted in the i-th record segment, S represent the expected number of batches stored in the database, A1 represent the weight coefficient corresponding to the time difference between initiating the pull request and returning the first piece of data stored in the database, A2 represent the weight coefficient corresponding to the receiving time of the last piece of data stored in the database, A3 represent the weight coefficient corresponding to the data writing completion time stored in the database, A4 represent the weight coefficient corresponding to the actual amount of data transmitted stored in the database, and i represent the number of each record segment, i=1,2,3,...,n, where n represents the total number of record segments.

[0039] It's important to note that each record segment corresponds to an independent fetch task, with its time interval starting from the moment the client issues the fetch command and ending when all data in that batch is written to local storage. The time difference between initiating the fetch request and receiving the first data record refers to the difference between the time the fetch API call is initiated and the arrival time of the first record in that data segment returned by the source system, reflecting the performance of the first packet response. The last data reception time is the timestamp of the last record in the fetch task being completely downloaded to the local message queue or buffer (starting from the start time of the fetch request), used to measure the completion time of batch data transmission. This can intuitively reveal the performance bottlenecks of network transmission or source response in large data volume scenarios. If this time frequently approaches or exceeds the preset transmission latency threshold, it indicates a significant delay at the end of the data retrieval process, requiring a thorough investigation of network throughput or the stability of the source system's batch interfaces. The data write completion time refers to the final moment from the start of the retrieval task until all received data in that segment has undergone idempotent verification and been successfully written to the distributed transaction log or target storage (starting from the retrieval task start time). This marks the end of the entire retrieval, transmission, and write process for this record segment. When this time continuously increases or fluctuates significantly, it means that the backend write nodes may have encountered write bottlenecks, transaction rollbacks, or hotspot contention. The actual data volume transmitted corresponds to the total number of bytes or records successfully downloaded and committed within the corresponding time period. Through refined measurement of these three key time points and data volume parameters, the system can accurately characterize the performance of each record segment at the network, pipeline, and storage levels, providing data support for subsequent throughput analysis and adaptive optimization.

[0040] It's important to note that in distributed transaction logs, the four key parameters—the time difference between initiating a pull request and receiving the first data record, the time to receive the last data record, the data write completion time, and the actual amount of data transmitted—are closely interrelated. For example, a decrease in the first packet response time (the time difference between pulling and receiving the first data record) indicates an improvement in the initial throughput performance of the source system, which can often shorten the arrival delay of the last data record in subsequent batch transmissions. However, if the actual amount of data transmitted increases significantly at this time, it will introduce additional I / O load during the write completion phase, extending the data write completion time. Conversely, if the write layer has sufficient throughput, then... Even in high-data-volume scenarios, the write completion time remains stable, thereby reducing fluctuations in the first packet response and transmission latency. Furthermore, when the last data reception time approaches the network's maximum latency threshold, the first packet response time may be slightly affected by connection persistence if the packet size or parallelism is not optimized. Under this multi-dimensional coupling relationship, it is necessary to find the optimal balance between improving the first packet response speed, balancing transmission throughput, and optimizing the local write rate to ensure that each record segment can quickly start data retrieval, efficiently complete batch transmission and persistent writing, and ensure that the overall service level protocol of the pipeline meets the standards and resource utilization is optimized.

[0041] It should be noted that the weighting coefficients corresponding to the time difference between initiating a pull request and returning the first piece of data, the time of receiving the last piece of data, the time of data writing completion, and the actual amount of data transmitted are all stored in the database, and their values ​​are usually set between 0 and 1. For example, by constructing mapping tables between the time difference between initiating a pull request and returning the first piece of data, the time of receiving the last piece of data, the time of data writing completion, and the actual amount of data transmitted, and their corresponding weighting coefficients, the real-time detected time difference between initiating a pull request and returning the first piece of data, the time of receiving the last piece of data, the time of data writing completion, and the actual amount of data transmitted are input into the corresponding mapping tables in the database, thereby quickly obtaining the weighting coefficients corresponding to the time difference between initiating a pull request and returning the first piece of data, the time of receiving the last piece of data, the time of data writing completion, and the actual amount of data transmitted, respectively.

[0042] Specifically, the process of determining the adjustment of pull requests for record segments in the data acquisition chain is as follows: extract the pipeline execution throughput factor of each record segment and compare it with the pipeline execution throughput factor threshold stored in the database. If the pipeline execution throughput factor of a certain record segment is higher than or equal to the pipeline execution throughput factor threshold, then the pull request range of that record segment is located and the pull request is reconstructed. If the pipeline execution throughput factor of a certain record segment is lower than the pipeline execution throughput factor threshold, then it is not necessary to locate the pull request range of that record segment.

[0043] It should be noted that the specific process of refactoring the fetch request is as follows: extract the batch transmission latency percentage from the transaction log of the segment. If the batch transmission latency percentage is high, the number of parallel fetch threads will be increased by a fixed percentage (e.g., 20% of the initial parallelism) to fetch more shards within the same time window. At the same time, the original shard size will be reduced by the same percentage (e.g., reduced by 20%) to shorten the single network round trip and buffer time, and reduce the risk of timeouts and retries. Conversely, if the write latency percentage is too high or network resources are saturated, the system will adjust in the opposite way—reducing the parallelism and increasing the shard size accordingly to reduce the concurrent pressure on storage I / O and improve the efficiency of batch writes.

[0044] Specifically, the effect of adjusting the pull request for the record segment in the data acquisition chain is verified. The specific process is as follows: After a reconstruction pull request, a verification and monitoring period is preset. During the verification and monitoring period, the number of reconstruction pull requests executed by the pipeline for each record segment after the reconstruction pull request is counted, and the threshold number of reconstruction pull requests executed by the pipeline for each record segment stored in the database is extracted. If the number of reconstruction pull requests executed by the pipeline for each record segment after the reconstruction pull request is higher than or equal to the number of reconstruction pull requests executed by the pipeline for each record segment after the reconstruction pull request, the supplementary pull logic is triggered, and the concurrency and sharding granularity of the data pull task for that segment are dynamically adjusted. If the number of reconstruction pull requests executed by the pipeline for each record segment after the reconstruction pull request is lower than the number of reconstruction pull requests executed by the pipeline for each record segment after the reconstruction pull request, the supplementary pull logic does not need to be triggered.

[0045] It should be noted that if the pipeline throughput compliance factor after re-verification still fails to reach the preset threshold, it indicates that the system has not fully resolved the performance bottleneck of the corresponding record segment during the fetching, transmission, or writing process after the first adaptive adjustment. In this case, the system will further increase the adjustment magnitude, implementing more targeted dynamic adjustments to the processing concurrency and fragmentation granularity of the data segment. The direction and extent of the adjustment are mainly determined based on the processing stage where the bottleneck occurred and the degree of deviation of its corresponding key indicators. For example, if monitoring reveals that the first packet response time is consistently higher than the source system's first packet response SLA threshold, the number of concurrent fetches will be gradually reduced within a convergence threshold, generally controlled within 60% to 80% of the previous round of concurrency. The aim is to reduce the instantaneous request pressure on the target system, thereby achieving a higher first packet response success rate and link stability. Conversely, if the data tail packet reception time or local write time consistently exceeds the preset maximum transmission latency or write performance threshold, the data splitting granularity will be further reduced. This means compressing the batch size of a single fetch or write operation to 50%~70% of the original granularity, making each transmission or write task lighter and reducing risks such as buffer buildup, I / O blocking, or disk write conflicts. If a parameter is nearing its limit (e.g., concurrency has dropped to the system's lower limit), the system will prioritize adjusting another variable parameter to avoid resource waste or excessive suppression. It is important to emphasize that during the second round or higher-level parameter adjustments, the system does not perform single-dimensional extreme compression or expansion. Instead, it combines comprehensive indicators such as overall task completion rate, data integrity, system load, and retry count, adjusting gradually within a certain elastic range in an exponentially gradual manner. This ensures throughput stability while avoiding processing fluctuations caused by drastic adjustments. The final adjustment actions will also be logged for subsequent optimization algorithms to learn from and improve the overall acquisition pipeline's steady-state adaptability to complex data scenarios.

[0046] Specifically, the out-of-order data factor of the standardized raw data packets of the supplier within the event time-driven window is obtained. The specific process is as follows: obtain the deviation value between the average time of the event recorded within the event time-driven window and the processing time of its arrival in the system, which is recorded as the processing delay amount recorded within the event time-driven window, and obtain the frequency of out-of-order packets occurring per unit time and the maximum buffer waiting time generated by correcting the out-of-order packets.

[0047] It should be noted that the processing delay amount of records within the event-time driven window is obtained by comparing the time difference between the original event timestamp of each data record (i.e., the time when the data is generated or marked at the source) and the system reception time when it actually arrives at the system and is processed. Then, the delay time of all data records in the window is averaged to obtain the overall processing delay amount within the window. The frequency of out-of-order packets per unit time is obtained by counting the number of records that do not arrive in the current window according to the event time order, and the proportion of all records. Records are usually judged as out-of-order because their event time is later than that of a certain data in the window, but they arrive at the system earlier. The maximum buffer waiting time caused by correcting out-of-order refers to the maximum buffer waiting time introduced to maintain the event time order. It is the maximum delay in tracking out-of-order records being reordered and pushed for processing within the window, and is used to reflect the cost paid by the system for out-of-order processing.

[0048] The out-of-order data factor of the supplier's standardized raw data packets within the event-time driven window is obtained by performing a proportion analysis on the processing delay amount recorded within the event-time driven window, the maximum allowed ordered delay value preset in the database, the frequency of out-of-order packets occurring per unit time, the defined frequency of out-of-order packets occurring per unit time, and the maximum buffer waiting time caused by correcting out-of-order packets, and the maximum cache time limit stored in the database. Weighting coefficients and pipeline execution throughput factors for each record segment are introduced. The out-of-order data factor of the supplier's standardized raw data packets within the event-time driven window is used to quantitatively evaluate the timing consistency and pipeline reliability of the supplier's data flow within the event-time driven window.

[0049] It should be noted that the out-of-order data factor of the supplier's standardized raw data packets within the event time-driven window is analyzed under the following specific conditions: ; In the formula, E represents the out-of-order data factor of the supplier's standardized raw data packets within the event time-driven window. C1 represents the amount of processing delay recorded within the event-driven window, G represents the frequency of out-of-order packets per unit time, G1 represents the defined frequency of out-of-order packets per unit time, TJ represents the maximum buffer wait time for correcting out-of-order packets, TJ1 represents the maximum buffer time limit, and U represents the maximum buffer time limit for correcting out-of-order packets. iB1 represents the pipeline execution throughput factor of the i-th record segment, B2 represents the weight coefficient corresponding to the processing delay amount of records within the event time-driven window stored in the database, B3 represents the weight coefficient corresponding to the frequency of out-of-order packets occurring per unit time stored in the database, B4 represents the weight coefficient corresponding to the maximum buffer waiting time generated by correcting out-of-order packets stored in the database, and i represents the number of each record segment, i=1,2,3,...,n, where n represents the total number of record segments.

[0050] It should be noted that in the event-time driven window, whenever the amount of delayed processing increases, it means that more and more events are not being processed in time according to their original business time. This will force the system to extend the buffer waiting time for out-of-order correction to accommodate more late events. As the buffer waiting time is amplified, the frequency of out-of-order packets per unit time will often also increase, because more events are backed up in the waiting queue, and the order misalignment is more significant. When the frequency of out-of-order packets rises to a certain level, in order to maintain the overall system throughput, the pipeline execution monitoring will automatically adjust the concurrency or fragmentation strategy, thereby affecting the pipeline execution throughput factor of each record segment. The increase in the throughput factor, in turn, aggravates the latency of the write and transmission links, causing the amount of delayed processing to increase further.

[0051] It should be noted that the weighting coefficients corresponding to the processing delay amount, the frequency of out-of-order packets within a unit time, the maximum buffer waiting time caused by correcting out-of-order packets, and the pipeline throughput factor recorded within the event-driven window are all stored in the database, and their values ​​are typically set between 0 and 1. For example, by constructing mapping tables between the processing delay amount, the frequency of out-of-order packets within a unit time, the maximum buffer waiting time caused by correcting out-of-order packets, and the pipeline throughput factor of each record segment, the real-time detected processing delay amount, the frequency of out-of-order packets within a unit time, the maximum buffer waiting time caused by correcting out-of-order packets, and the pipeline throughput factor of each record segment are input into the corresponding mapping tables in the database. This allows for the rapid acquisition of the weighting coefficients corresponding to the processing delay amount, the frequency of out-of-order packets within a unit time, the maximum buffer waiting time caused by correcting out-of-order packets, and the pipeline throughput factor within the event-driven window.

[0052] Specifically, the event-time driven window parameters are adjusted, and the standardized raw data packets of the supplier in the message queue are divided. The specific process is as follows: the out-of-order data factors of the standardized raw data packets of the supplier within the event-time driven window are extracted and compared with the out-of-order data factor range of the standardized raw data packets of the supplier within the event-time driven window stored in the database. If the out-of-order data factor of the standardized raw data packets of the supplier within the event-time driven window exceeds the out-of-order data factor range of the standardized raw data packets of the supplier within the event-time driven window, then the event-time driven adjustment window parameters are adjusted, and the standardized raw data packets of the supplier in the message queue are re-divided. If the out-of-order data factor of the standardized raw data packets of the supplier within the event-time driven window does not exceed the out-of-order data factor range of the standardized raw data packets of the supplier within the event-time driven window, then the event-time driven adjustment window parameters do not need to be adjusted.

[0053] It should be noted that the out-of-order data factor of the standardized raw data packets of the supplier within the event time-driven window is a quantitative assessment of the temporal consistency of the supplier's data stream and the reliability of pipeline processing within an event time window. It can comprehensively reflect the integrity of the arrival order of data packets, the degree of out-of-order, the latency cost incurred by the system to ensure sequential consistency, and the overall throughput performance of the pipeline. It is a key indicator for measuring the quality of streaming data and identifying pipeline bottlenecks.

[0054] It should be noted that the specific process of adjusting the window parameters for event-time driven regulation is as follows: when it is detected that the out-of-order data factor of a certain supplier within the event-time driven window is higher than the preset out-of-order data factor range of the supplier's standardized raw data packets within the event-time driven window, the system will use the value of this factor as the adjustment basis, smoothly increasing the current window length by a coefficient and simultaneously decreasing the sliding step size. For example, if the out-of-order factor is 0.3 (i.e., 30%), the window length can be increased by 30% on the original basis, while the sliding step size can be decreased by 30%, so that the window covers more late or out-of-order data and slows down the window advancement speed; when When the out-of-order factor continues to rise, the system linearly increases the expansion ratio, but is limited by a set maximum adjustment limit (e.g., no more than 50%) to prevent excessive aggregation from causing computational delays. When the system detects that the window expansion ratio corresponding to the out-of-order data factor exceeds the preset maximum expansion limit, the system will no longer expand the window by the factor value, but will fix the window expansion ratio at the upper limit (e.g., 50%). Similarly, the sliding step compression ratio is also locked at the maximum compression range. At this time, no matter how much the factor rises further, the window parameters remain within this controlled boundary to avoid computational delays and resource overload caused by excessive window expansion. If the out-of-order factor falls back to the threshold range or below the lower edge of the range in subsequent monitoring periods, the system will automatically resume the smooth contraction according to the factor ratio, gradually adjusting the window length and sliding step back to the normal state, thereby achieving dynamic limiting of out-of-order pressure and smooth switching of resource utilization efficiency. Conversely, when it is detected that the out-of-order data factor of a certain supplier within the event time-driven window is lower than the out-of-order data factor range of the supplier's standardized raw data packets within the preset event time-driven window, the window length is smoothly reduced by the same proportion, the sliding step size is restored to normal, and the data processing rhythm is accelerated. This not only ensures sufficient accommodation of out-of-order data, but also restores system throughput and real-time performance in a timely manner after the out-of-order situation is alleviated.

[0055] Specifically, the global trend deviation factor of the supplier's business numerical indicator sequence within the sliding window is obtained. The specific process is as follows: obtain the average trend slope of the target supplier's key business metrics within the sliding window.

[0056] It's important to note that the average trend slope of key business metrics is obtained by linearly fitting a time series of a specific business indicator (such as daily order volume, daily transaction amount, or daily fulfillment rate) for the supplier within the current sliding window period. Specifically, the system first arranges the business indicator values ​​(e.g., daily, hourly) for each point in time within the window period, such as the total number of orders for the day, into a numerical sequence based on event time. Then, using event time as the x-axis and indicator value as the y-axis, it fits a best-fit line using least-squares linear regression. The slope of this best-fit line reflects the average trend rate of the indicator within the window period.

[0057] Extract the baseline slope of key business metrics within the sliding window stored in the database.

[0058] The absolute value of the difference between the average trend slope and the baseline slope of the target supplier's key business metrics within the sliding window is used as the global trend deviation factor for the supplier's business numerical indicator series within the sliding window.

[0059] like Figure 2 As shown, the adjustment requirements of the window parameters driven by the secondary judgment event time are to readjust the window parameters driven by the event time. The specific process is as follows: extract the global trend deviation factor of the supplier's business numerical indicator sequence within the sliding window, and compare it with the global trend deviation factor threshold range stored in the database. If the global trend deviation factor of the supplier's business numerical indicator sequence within the sliding window exceeds the global trend deviation factor threshold range, then the dynamic readjustment of the window width and sliding step size is triggered. If the global trend deviation factor of the supplier's business numerical indicator sequence within the sliding window does not exceed the global trend deviation factor threshold range, then the existing sliding window parameters remain unchanged.

[0060] It should be noted that the global trend deviation factor threshold range refers to a stable range defined by both the lower and upper limits, rather than a single threshold. When a supplier's global trend deviation factor falls within this range, it indicates that the current deviation of its business trend is within an acceptable range, and the system does not need to adjust the sliding window parameters. If the global trend deviation factor of the supplier's business numerical indicator sequence within the sliding window is higher than or equal to the global trend deviation factor threshold range (higher than the upper limit of the range), it indicates that the trend change is too drastic, and the window needs to be expanded and the step size reduced to capture business evolution over a longer time scale. If the global trend deviation factor of the supplier's business numerical indicator sequence within the sliding window is lower than the global trend deviation factor threshold range (lower than the lower limit of the range), it indicates that the trend change is too gradual or even falls into long-term linear continuation. In this case, it is suggested that managers appropriately restore a shorter window period and a larger sliding step size to improve detection sensitivity and processing efficiency.

[0061] Specifically, the dynamic readjustment of the window width and sliding step size is triggered. The specific process is as follows: extract the global trend deviation factor of the supplier's business numerical indicator sequence within the sliding window and the upper limit of the threshold range of the global trend deviation factor. Subtract the global trend deviation factor of the supplier's business numerical indicator sequence within the sliding window from the upper limit of the threshold range of the global trend deviation factor to obtain the global trend deviation value of the supplier's business numerical indicator sequence within the sliding window.

[0062] The global trend deviation value of the supplier's business numerical indicator sequence within the sliding window is compared with the window length expansion ratio and sliding step size reduction ratio corresponding to each interval of the global trend deviation value of the supplier's business numerical indicator sequence within the sliding window stored in the database. The length expansion ratio and sliding step size reduction ratio of the sliding window are obtained, and the sliding window is readjusted based on the length expansion ratio and sliding step size reduction ratio.

[0063] It should be noted that the specific process of readjusting the sliding window according to the length expansion ratio and the sliding step size reduction ratio is as follows: the current window length and sliding step size are linearly expanded and reduced according to the matching expansion ratio and sliding step size reduction ratio, respectively, to generate a new window time span and sliding step frequency.

[0064] Specifically, a supplier profile is generated based on a pre-trained supplier profile generation model. The process involves aligning and normalizing the supplier's data fragments according to event time, extracting the supplier's behavioral feature vector, inputting the supplier's behavioral feature vector into the supplier profile generation model, and outputting a structured supplier profile result.

[0065] It should be noted that the supplier's data fragments are precisely aligned according to event time, and the data for each indicator is normalized to eliminate the impact of differences in units and data fluctuations, thereby extracting a multi-dimensional time-series feature vector reflecting the supplier's comprehensive behavioral characteristics. This feature vector includes, but is not limited to, several key business indicators such as the supplier's transaction frequency, order amount change trends, fulfillment time distribution, frequency of abnormal events, data transmission latency, pull request response stability, data disorder factors and their dynamic changes. These feature vectors are formed into fixed-length vector representations through time series aggregation and statistical feature extraction techniques, comprehensively reflecting the supplier's business performance and data interaction characteristics. A pre-trained supplier profile generation model (a bidirectional long short-term memory network model) is employed. During training, multi-dimensional historical indicators such as supplier order fulfillment rate, transaction volatility coefficient, frequency of quality inspection anomalies, and data missing ratio are collected and strictly aligned to a uniform granularity (e.g., aggregated by day / week) along the time axis. The raw data is then converted into a time-series vector sequence using sliding window encoding. In the supervised training phase, a bidirectional long short-term memory network is used as the core architecture. Its bidirectional gating mechanism can simultaneously capture the forward evolution and backward dependencies of features (e.g., the impact weight of a three-quarter consecutive quarterly decrease in transaction volume on credit score). The model takes the time-series vector as input and passes through 7 hidden layers. After abstracting the features of each unit, a multi-task learning head is connected to the output layer: the main task outputs basic labels for the profile (e.g., credit scores A-D), while the auxiliary task simultaneously predicts the stability index (0-100 points) and risk level (low / medium / high). During training, an adaptive weight decay strategy is used to strengthen the learning of basic labels in the initial stage, and gradually increase the weight of the auxiliary task to 50% in the later stage, finally iterating and converging on labeled samples. The output is a structured supplier profile vector, with vector dimensions covering multiple dimensions of supplier performance indicators, such as business activity index, performance stability score, data integrity level, risk potential assessment, interaction response efficiency level, and abnormal event warning level.

[0066] like Figure 3 As shown, Figure 3 The interface for monitoring the data flow execution is shown at the top, displaying module labels such as data acquisition and monitoring center. The central area displays segmented processing logs in a timeline format, with each record including a timestamp accurate to milliseconds and the data acquisition status (such as "completed" or "abnormally interrupted"). The bottom statistics panel dynamically updates processing metrics, the right-hand bar chart compares the time taken at different stages, and the bottom floating notification box displays the current server load rate in real time.

[0067] like Figure 4 As shown, Figure 4This is a load trend line chart. The left vertical axis represents CPU / memory utilization, and the right vertical axis displays network throughput. The main view area contains four dynamic lines that record peak CPU utilization, stable memory usage, database query queue length, and network bandwidth fluctuations. The horizontal axis is a 24-hour time scale, with key time points marked for operational actions.

[0068] like Figure 5 As shown, Figure 5This is a data time sequence detection interface diagram. It ensures the reliability of profile indicators by dynamically analyzing the degree of time stamp disorder in multi-source data (such as contradictions in the order of events like orders, inventory, and transactions). The semi-circular progress bar on the left side of the interface intuitively reflects the current proportion of time sequence disorder in supplier-related data, such as logical conflicts like purchase order times being later than sales outbound times, and provides targeted suggestions below. The adjustment strategy comparison diagram on the right marks historical calibration operation trajectories, marks the current time nodes that need optimization with red circles, and defines a reasonable adjustment range to help maintenance personnel identify whether manual intervention is needed. The adjustment action record at the bottom shows the automatic correction operations executed by the system in real time, ensuring that the time data on which subsequent profile analysis relies conforms to the actual business sequence. The "Data Time Sequence Detection Interface Diagram" shown intuitively reflects whether there are sequence conflicts between event times and system reception times during the supplier data flow process, specifically corresponding to the "Event Time Window Detection and Adjustment Module" in the system. After the "Data Writing and Transaction Log Management Module" completes the verification and writing of the original data packet, this module immediately performs event time-driven sequence detection analysis on the data pushed to the message queue. The semi-circular progress bar on the left side of the diagram represents the out-of-order data factor of the supplier within the current event-time driven window. This is a key quantitative indicator derived by the system through analysis of indicators such as late arrivals, out-of-order packet frequency per unit time, and maximum buffer waiting time, combined with a database preset threshold range and throughput factor weighting. When this progress bar shows a high proportion, it may indicate a large number of data timing logic errors, such as purchase orders being later than the actual delivery time, thus threatening the reliability of subsequent profile generation. The "Adjustment Strategy Comparison Chart" on the right side of the diagram corresponds to the system's adaptive adjustment strategy for the "Event-Time Driven Window Parameters." The red circles in the diagram mark the high-incidence out-of-order time nodes that the system determines to be the focus of adjustment, while the gray area represents the new analysis range covered after the current window length and sliding step size are automatically expanded, in order to more completely capture late or out-of-order data. During this process, if the system repeatedly determines that the current supplier's out-of-order data factor is higher than the upper limit, it will trigger a linear expansion of the window length and a step size reduction mechanism to ensure sufficient buffer time to absorb out-of-order data, thereby performing order correction in subsequent windows. The "Adjustment Action Record" at the bottom of the figure comes from the combined effect of the system's "Sliding Window Management and Trend Deviation Analysis Module" and the event time-driven module. It records the adjustment behavior triggered by sudden trend changes or severe disorder in real time. For example, if the disorder factor is continuously higher than the upper limit of the interval for a certain period of time, the system will extend the window by 20%, compress the step size by 30%, and clearly record the key parameters before and after the adjustment.

[0069] It should also be understood that the memory in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDR SDRAM), enhanced synchronous DRAM (ESDRAM), synchronous linked DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0070] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. A 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, all or part of the flow or function according to the embodiments of the present invention is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. A computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. Available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media. Semiconductor media can be solid-state drives.

[0071] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0072] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0073] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0074] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0075] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0076] In the embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0077] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0078] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0079] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0080] The above are merely specific embodiments 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 the claims.

Claims

1. A supplier profile intelligent generation system, characterized in that, The system includes: The data writing and transaction log management module is used to write the standardized raw data packets of the supplier into the distributed transaction log, record them in segments according to the globally unique transactions, monitor and verify the pipeline execution process, determine the adjustment of the pull request of the record segment in the data acquisition link, and verify the effect of the adjustment of the pull request of the record segment in the data acquisition link. The process of segmenting and recording based on globally unique transactions to monitor and verify the pipeline execution process is as follows: Obtain the time difference between the initiation of the pull request for each record segment in the supplier's globally unique transaction and the return of the first piece of data, the time of receiving the last piece of data, the time of data writing completion, and the actual amount of data transmitted; The time difference between the initiation of the pull request and the return of the first data, the time limit for defining the response time of the first packet, the time for receiving the last data and the maximum allowable transmission latency of batch data, the data writing completion time and the maximum write time required for local storage write throughput, as well as the actual amount of data transmitted and the expected number of batches, are analyzed in proportion and weighted to obtain the pipeline execution throughput factor for each record segment. The pipeline execution throughput factor for each record segment is used to quantitatively evaluate the degree to which the pipeline meets the throughput and latency requirements. The event time window detection and adjustment module is used to push the verified standardized raw data packets of the supplier to the message queue, detect the message queue according to the event time driving window, obtain the out-of-order data factor of the standardized raw data packets of the supplier within the event time driving window, determine the adjustment requirements of the window parameters of the event time driving adjustment, adjust the window parameters of the event time driving, and divide the standardized raw data packets of the supplier in the message queue. The specific process for obtaining the out-of-order data factor of the supplier's standardized raw data packets within the event time-driven window is as follows: Obtain the deviation between the average time of events recorded within the event time-driven window and the processing time of events arriving in the system, and record it as the processing delay amount recorded within the event time-driven window. Also obtain the frequency of out-of-order packets occurring per unit time and the maximum buffer waiting time generated by correcting out-of-order packets. The processing delays recorded within the event-driven window are compared with the maximum allowed ordered delay value preset in the database, the frequency of out-of-order packets within a unit time and the defined frequency of out-of-order packets within a unit time, and the maximum buffer waiting time generated by correcting out-of-order packets and the maximum cache time limit stored in the database. These are then analyzed for proportion, and weighting coefficients and pipeline execution throughput factors for each record segment are introduced to obtain the out-of-order data factor of the supplier's standardized raw data packets within the event-driven window. The out-of-order data factor of the supplier's standardized raw data packets within the event-driven window is used to quantitatively evaluate the timing consistency and pipeline reliability of the supplier's data flow within the event-driven window. The sliding window management and trend deviation analysis module is used to import the standardized raw data packets of suppliers in the divided message queue into the sliding window manager for window scheduling and monitoring, obtain the global trend deviation factor of the supplier's business numerical indicator sequence within the sliding window, make a secondary determination of the adjustment needs of the window parameters driven by event time, and readjust the window parameters driven by event time. The specific process for obtaining the global trend deviation factor of the supplier's business numerical indicator sequence within the sliding window is as follows: Obtain the average trend slope of key business metrics for the target supplier within the sliding window; Extract the baseline slope of key business metrics within the sliding window stored in the database; The absolute value of the difference between the average trend slope and the baseline slope of the target supplier's key business metrics within the sliding window is used as the global trend deviation factor for the supplier's business numerical indicator sequence within the sliding window. The supplier profile generation module is used to generate supplier profiles based on a pre-trained supplier profile generation model.

2. The intelligent supplier profile generation system according to claim 1, characterized in that, The specific process for adjusting the pull request of the record segment in the data acquisition link is as follows: Extract the pipeline execution throughput factor of each record segment and compare it with the pipeline execution throughput factor threshold stored in the database. If the pipeline execution throughput factor of a record segment is higher than or equal to the pipeline execution throughput factor threshold, locate the pull request range of the record segment and reconstruct the pull request. If the pipeline execution throughput factor of a record segment is lower than the pipeline execution throughput factor threshold, it is not necessary to locate the pull request range of the record segment.

3. The intelligent supplier profile generation system according to claim 2, characterized in that, The specific process for adjusting the pull request effect of the record segment in the verification data acquisition link is as follows: After a reconstruction pull request, a verification and monitoring period is preset. During the verification and monitoring period, the number of reconstruction pull requests executed by the pipeline for each record segment after the reconstruction pull request is counted, and the threshold number of reconstruction pull requests executed by the pipeline for each record segment stored in the database is extracted. If the number of reconstruction pull requests executed by the pipeline for each record segment after the reconstruction pull request is higher than or equal to the number of reconstruction pull requests executed by the pipeline for each record segment after the reconstruction pull request, the supplementary pull logic is triggered, and the concurrency and sharding granularity of the data pull task for that segment are dynamically adjusted. If the number of reconstruction pull requests executed by the pipeline for each record segment after the reconstruction pull request is lower than the number of reconstruction pull requests executed by the pipeline for each record segment after the reconstruction pull request, the supplementary pull logic does not need to be triggered.

4. The intelligent supplier profile generation system according to claim 1, characterized in that, The specific process of adjusting the event-time driven window parameters and dividing the standardized raw data packets of the supplier in the message queue is as follows: Extract the out-of-order data factor of the supplier's standardized raw data packets within the event-time driven window, and compare it with the out-of-order data factor range of the supplier's standardized raw data packets within the event-time driven window stored in the database. If the out-of-order data factor of the supplier's standardized raw data packets within the event-time driven window exceeds the out-of-order data factor range of the supplier's standardized raw data packets within the event-time driven window, then adjust the window parameters of the event-time driven adjustment and re-divide the supplier's standardized raw data packets in the message queue. If the out-of-order data factor of the supplier's standardized raw data packets within the event-time driven window does not exceed the out-of-order data factor range of the supplier's standardized raw data packets within the event-time driven window, then it is not necessary to adjust the window parameters of the event-time driven adjustment.

5. The intelligent supplier profile generation system according to claim 1, characterized in that, The adjustment requirement for the window parameters of the secondary event time-driven adjustment involves readjusting the event time-driven window parameters. The specific process is as follows: Extract the global trend deviation factor of the supplier's business numerical indicator sequence within the sliding window and compare it with the global trend deviation factor threshold range stored in the database. If the global trend deviation factor of the supplier's business numerical indicator sequence within the sliding window exceeds the global trend deviation factor threshold range, then the window width and sliding step size are dynamically readjusted. If the global trend deviation factor of the supplier's business numerical indicator sequence within the sliding window does not exceed the global trend deviation factor threshold range, then the existing sliding window parameters remain unchanged.

6. The intelligent supplier profile generation system according to claim 5, characterized in that, The dynamic readjustment of the trigger window width and sliding step size is as follows: Extract the global trend deviation factor and the upper limit of the threshold range of the global trend deviation factor of the supplier's business numerical indicator sequence within the sliding window. Subtract the global trend deviation factor of the supplier's business numerical indicator sequence within the sliding window from the upper limit of the threshold range of the global trend deviation factor to obtain the global trend deviation value of the supplier's business numerical indicator sequence within the sliding window. The global trend deviation value of the supplier's business numerical indicator sequence within the sliding window is compared with the window length expansion ratio and sliding step size reduction ratio corresponding to each interval of the global trend deviation value of the supplier's business numerical indicator sequence within the sliding window stored in the database. The length expansion ratio and sliding step size reduction ratio of the sliding window are obtained, and the sliding window is readjusted based on the length expansion ratio and sliding step size reduction ratio.

7. The intelligent supplier profile generation system according to claim 1, characterized in that, The specific process of generating supplier profiles based on the pre-trained supplier profile generation model is as follows: The supplier's data fragments are aligned according to event time and normalized to extract the supplier's behavioral feature vector. The supplier's behavioral feature vector is then input into the supplier profile generation model to output a structured supplier profile result.

Images