A multi-platform e-commerce order aggregation method and device based on dynamic clustering rules

Abstract

This invention relates to the field of e-commerce data processing technology, specifically to a method and apparatus for multi-platform e-commerce order aggregation based on dynamic clustering rules. The technical solution includes: collecting order data from e-commerce platforms and performing standardized processing to obtain platform order data; extracting various clustering feature values ​​from a clustering feature system; matching these values ​​with rules in a dynamic clustering rule set to determine the candidate cluster to which the order belongs; calculating the comprehensive similarity between orders; determining whether the comprehensive similarity is greater than or equal to a preset similarity threshold, and assigning the order to that cluster; performing missed order detection for orders not clustered; and outputting the order results for the clusters. This significantly improves clustering accuracy, effectively reduces missed orders, and minimizes operating and transportation costs. It also significantly improves order processing efficiency, adapting to the needs of large-scale order processing. The system operates stably and supports efficient processing of massive amounts of data.

Description

Technical Field

[0001] This invention relates to the field of e-commerce data processing technology, specifically to a method and apparatus for aggregating multi-platform e-commerce orders based on dynamic clustering rules. Background Technology

[0002] With the rapid development of the e-commerce industry, multi-platform operation has become the mainstream model for e-commerce companies to expand market coverage and increase sales. More and more e-commerce companies are choosing to open stores on multiple platforms for multi-channel operation in order to expand market coverage and increase product sales.

[0003] Existing multi-platform order aggregation technologies mainly fall into two categories: one is manual aggregation, which relies on operations staff to manually collect and categorize orders from various platforms. This method is not only inefficient but also prone to human error, leading to missed or incorrect orders. When order volume surges, operating costs increase exponentially. The other is fixed-rule aggregation, which clusters orders by pre-setting matching rules for a single dimension (such as delivery address or order time). This method cannot adapt to dynamic changes in order data. For example, fluctuations in order volume during promotional activities, field adaptation after integrating new platforms, and adjustments to delivery areas can all lead to decreased clustering accuracy, resulting in a large number of valid orders not being accurately aggregated, thus causing missed orders. Missed orders not only affect customer experience but also trigger chain reactions such as duplicate deliveries and inventory backlogs, significantly increasing the company's operating and transportation costs. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a method for aggregating e-commerce orders across multiple platforms based on dynamic clustering rules. By constructing a dynamic clustering rule system, it achieves accurate and real-time aggregation of orders from multiple platforms, effectively reducing missed orders and lowering operating and transportation costs. The method for aggregating e-commerce orders across multiple platforms based on dynamic clustering rules includes the following steps:

[0005] S1. Collect order data from e-commerce platforms to obtain raw order data.

[0006] S2. Standardize the original order data to obtain platform order data.

[0007] S3. Extract the cluster feature values ​​from the cluster feature system from the platform order data.

[0008] S4. Match the clustering feature values ​​with the rules in the dynamic clustering rule set to determine the candidate cluster to which the order belongs.

[0009] S5. Calculate the comprehensive similarity between the cluster feature values ​​and the core orders within the candidate clusters.

[0010] S6. Determine whether the overall similarity is greater than or equal to the preset similarity threshold. If yes, then classify the order into the cluster; otherwise, proceed to S7.

[0011] S7. For orders that have not been clustered, perform a missing order detection to determine if any orders are missing. If so, execute S3 and perform supplementary clustering for the missing orders. If not, create a new cluster.

[0012] S8. Output the order results for the clusters.

[0013] Preferred: Original order data includes: order number, user ID, shipping address, product information, order time, and delivery requirements.

[0014] The preferred standardization process includes: field mapping, data cleaning, correction of outlier data, and format unification, resulting in standardized platform order data. Field mapping maps heterogeneous order fields from different platforms to a pre-defined unified field set. Data cleaning removes invalid order data. Outlier data is corrected. Format unification standardizes the time format, amount format, address hierarchy format, etc., of the order data to a pre-defined standard.

[0015] Preferably, the clustering features in the clustering feature system include static features and dynamic features.

[0016] Preferred static features include dimensions such as delivery address, product, and user.

[0017] Preferred dynamic features include order time and delivery demand dimensions.

[0018] Preferred methods for obtaining dynamic clustering rule sets include: initial clustering rule set construction and dynamic rule update mechanism.

[0019] Preferred method for constructing initial clustering rule set includes: using hierarchical clustering algorithm, taking standardized order data as samples, calculating the comprehensive similarity between samples based on clustering feature system, and then grouping orders with comprehensive similarity greater than or equal to a pre-set similarity threshold into a cluster and determining core orders, thereby generating initial clustering rules.

[0020] Preferably: the comprehensive similarity Where i is the cluster feature number in the cluster feature system, I is the total number of cluster features in the cluster feature system, i = 1, 2, ..., I; ω i It is the similarity weight of the cluster feature numbered i, c i It is the cluster feature value of cluster feature number i.

[0021] Preferably: the comprehensive similarity Where i is the cluster feature number in the cluster feature system, I is the total number of cluster features in the cluster feature system, i = 1, 2, ..., I; ω i It is the similarity weight of the cluster feature numbered i, c i This refers to the value of the cluster feature numbered i. i It is the scaling value of the influence of cluster feature number i.

[0022] Preferred: Affecting scaling value .

[0023] The preferred dynamic rule update mechanism includes: real-time collection of order aggregation effect feedback data and business scenario change data. Rule updates are triggered when order aggregation effect feedback data exceeds a preset threshold or when significant changes occur in the business scenario. A reinforcement learning algorithm is employed, with the optimization objectives of "minimizing the missed order rate" and "maximizing clustering efficiency," adjusting the similarity weights and similarity thresholds of each clustering feature to update the clustering rule set. Simultaneously, a rule verification mechanism is established to conduct offline verification and online pilot testing of the updated rules; official implementation begins after successful verification.

[0024] Preferred: Order aggregation effect feedback data includes missed order rate, clustering accuracy, and duplicate clustering rate.

[0025] Preferred data includes changes in business scenarios such as new platform access, promotional activities, and adjustments to delivery areas.

[0026] Preferred: The cluster results include the order list of each cluster, the number of orders within the cluster, and a summary of delivery information.

[0027] The present invention also provides a multi-platform e-commerce order aggregation device based on dynamic clustering rules, comprising:

[0028] Data Acquisition and Standardization Module: This module is used to collect raw order data from multiple e-commerce platforms through a distributed acquisition architecture, and to standardize the raw order data to obtain platform order data.

[0029] Clustering Feature Construction Module: Used to construct a multi-dimensional clustering feature system.

[0030] Dynamic rule management module: Used to build an initial clustering rule set, and to optimize the clustering rules in real time through a dynamic rule update mechanism to generate a dynamic clustering rule set.

[0031] The order clustering execution module extracts clustering feature values ​​from platform order data, matches these feature values ​​with rules in a dynamic clustering rule set, and determines the candidate cluster to which an order belongs. It calculates the comprehensive similarity between the clustering feature values ​​and the core orders within the candidate clusters. It then determines whether the comprehensive similarity is greater than or equal to a preset similarity threshold, performs missing order detection for orders not yet clustered, and performs supplementary clustering for these missing orders.

[0032] The results output and application module is used to output the order clustering results and apply them to order sorting, inventory allocation, delivery route planning and other processes.

[0033] Preferably, the device further includes a data storage module for storing original order data, platform order data, clustering feature values, dynamic clustering rule sets, clustering result data, and aggregation effect feedback data. It can adopt a distributed database architecture to support efficient storage and fast querying of massive order data.

[0034] Preferably, the device further includes a monitoring and early warning module, which is used to monitor key indicators in the order aggregation process in real time. When the indicator exceeds a preset threshold, an early warning is triggered and an anomaly report is generated.

[0035] The technical effects and advantages of this invention are as follows:

[0036] 1. Achieve dynamic and accurate clustering of orders across multiple platforms: Construct a multi-dimensional clustering feature system and combine it with reinforcement learning algorithms to achieve dynamic updates of clustering rules. This can adapt to dynamic changes in order data and adjustments to business scenarios, significantly improving clustering accuracy and effectively reducing missed orders.

[0037] 2. Reduce operating and transportation costs: Order aggregation enables batch processing of orders, reducing repetitive sorting and delivery operations; at the same time, clustering results can support accurate inventory allocation and delivery route optimization, reducing the risk of inventory backlog and transportation mileage, thereby maximizing the reduction of operating and transportation costs.

[0038] 3. Improve order processing efficiency: Adopt a distributed acquisition architecture to realize real-time acquisition of order data from multiple platforms, eliminate data heterogeneity through standardized processing, and combine dynamic clustering rules to realize rapid matching and clustering of orders, significantly improving order processing efficiency and adapting to the needs of large-scale order processing.

[0039] 4. Excellent scalability and stability: This invention supports cluster feature expansion and access to new platforms without modifying the core algorithm; the aggregation process is monitored in real time through a monitoring and early warning module to ensure stable system operation; at the same time, the data storage adopts a distributed architecture, supporting efficient processing of massive amounts of data. Attached Figure Description

[0040] Figure 1This is a flowchart illustrating a multi-platform e-commerce order aggregation method based on dynamic clustering rules proposed in this invention.

[0041] Figure 2 This is a flowchart illustrating the method for obtaining the dynamic clustering rule set in a multi-platform e-commerce order aggregation method based on dynamic clustering rules proposed in this invention.

[0042] Figure 3 This is a flowchart illustrating the initial clustering rule set construction method in a multi-platform e-commerce order aggregation method based on dynamic clustering rules proposed in this invention.

[0043] Figure 4 This is a flowchart illustrating the dynamic rule update mechanism in a multi-platform e-commerce order aggregation method based on dynamic clustering rules proposed in this invention.

[0044] Figure 5 This is a structural block diagram of a multi-platform e-commerce order aggregation device based on dynamic clustering rules proposed in this invention. Detailed Implementation

[0045] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the invention, and should not be construed as limiting the invention. Rather, embodiments of the invention include all variations, modifications, and equivalents falling within the spirit and scope of the appended claims.

[0046] Example 1

[0047] refer to Figure 1 This embodiment proposes a multi-platform e-commerce order aggregation method based on dynamic clustering rules. By constructing a dynamic clustering rule system, it achieves accurate and real-time aggregation of orders from multiple platforms, effectively reducing missed orders and lowering operating and transportation costs. The multi-platform e-commerce order aggregation method based on dynamic clustering rules includes the following steps:

[0048] S1. Collect order data from e-commerce platforms to obtain raw order data. This can be done through API calls, data pushes, and other methods to collect raw order data from multiple e-commerce platforms. For example, you can access three platforms via Taobao Open Platform API, JD.com Merchant API, and Pinduoduo Open Platform API, and then use a distributed collection architecture. Each platform is allocated an independent collection thread, and the collection frequency can be set according to actual needs. It can be evenly distributed, such as collecting once every 5 minutes, or it can be divided into time periods based on order frequency. For example, during peak hours (9:00-12:00, 20:00-23:00), the collection frequency is set to once per minute, and during off-peak hours, the collection frequency is set to once every 5 minutes. Raw order data is collected in real time and can include order number, user ID, shipping address, product information, order time, delivery requirements, etc., which can be adjusted according to actual needs.

[0049] S2. Standardize the raw order data to obtain platform order data. Standardization involves mapping fields, cleaning data, and unifying formats to obtain standardized platform order data. Field mapping maps heterogeneous order fields from different platforms to a predefined unified field set; for example, "receive_addr" for platform A and "shipping_address" for platform B. Data cleaning removes invalid order data, such as canceled orders, refunded orders, and orders with a field missing rate exceeding a predefined threshold. Abnormal data is corrected, such as incorrectly formatted phone numbers and incomplete address information. Format unification standardizes the time format, amount format, address hierarchy format, etc., of the order data to a predefined standard. For example, a unified set of fields can be constructed, which may include the order's unique identifier, user identifier, receiving province, receiving city, receiving district, receiving street, detailed address, latitude and longitude, product code, product category, order time, order status, delivery method, and whether expedited processing is required. The "receive_addr" field from Taobao, the "shipping_address" field from JD.com, and the "consignee_address" field from Pinduoduo are all mapped to the "shipping address" field. Cancelled orders, refunded orders, and orders with missing address fields are removed. Incorrectly formatted mobile phone numbers are corrected, such as by adding area codes and removing special characters. The order time is standardized to "yyyy-MM-dd HH:mm:ss", the amount is standardized to "yuan", and the shipping address is broken down into province, city, district, street, and detailed address levels. The latitude and longitude corresponding to the address are obtained through a map API.

[0050] S3. Extract the clustering feature values ​​from the platform order data. The clustering feature system is based on the business characteristics of e-commerce orders. The clustering features in the clustering feature system can include static features and dynamic features. Static features are inherent to the order and do not change with time or business scenarios, including the delivery address dimension (province, city, district, street, detailed address, latitude and longitude), product dimension (product category, product code, warehouse where inventory belongs), and user dimension (user ID, user level, membership type), etc. Dynamic features are features that change with time and business scenarios, including the order placement time dimension (order placement period, whether it is a holiday, whether it is a promotional period), order status dimension (pending payment, pending shipment, order accepted), and delivery demand dimension (whether it is expedited delivery, whether a delivery time is specified, delivery method), etc. Gathering these clustering features together constitutes the clustering feature system. For example, the order data includes dimensions such as delivery address (province: Zhejiang Province, city: Hangzhou City, district: Yuhang District, street: Wuchang Street, detailed address: Building 1, XX Community), product (product code: SP001, product category: clothing, warehouse: Hangzhou warehouse), and user (user ID: U001, user level: VIP2, membership type: monthly card member). It also includes dimensions such as order time (order time: 20:30, holiday: no, promotional period: yes (618 promotion)), order status (pending shipment), and delivery request (no expedited delivery, specified delivery time: 10:00-12:00 the next day, delivery method: express). Clustering feature values ​​are extracted from the platform's order data. For example, the feature values ​​for order A are (Wuchang Street, clothing, 20:30, express, 618 promotion), and the feature values ​​for order B are (Wuchang Street, clothing, 21:10, express, 618 promotion).

[0051] S4. Match the clustering feature values ​​with the rules in the dynamic clustering rule set to determine the candidate cluster to which the order belongs. For example, if the clustering feature values ​​of order A are (Wuchang Street, Clothing, 20:30, Express Delivery, 618 Promotion), it is determined to be a core order. When each cluster is generated, one or more core reference orders are automatically selected. The selection criteria are: when the number of orders in the cluster is greater than 3: select the order whose features best match the clustering rules (e.g., the address is at the geographic center of the orders in the cluster, the order time is at the median of the order time in the cluster); when the number of orders in the cluster is less than or equal to 3: select the first order added to the cluster as the core reference order. The clustering feature values ​​of order B are (Wuchang Street, Clothing, 21:10, Express Delivery, 618 Promotion). Orders A and B both match the rules of "same street + same product category + order time interval ≤ 2 hours + same delivery method + same promotional scenario", and are determined to be in the same candidate cluster. This is an initial comparison and matching, which can greatly reduce the amount of computation. Reference Figure 2The method for obtaining a dynamic clustering rule set can include two steps, namely: 1. Construction of the initial clustering rule set: refer to Figure 3 Hierarchical clustering algorithms can be used, with standardized order data as samples. Based on a clustering feature system, the comprehensive similarity between samples is calculated. Orders with a comprehensive similarity greater than or equal to a pre-set similarity threshold are grouped into a single cluster, and core orders are identified. For example, Euclidean distance can be used to calculate latitude and longitude similarity, edit distance to calculate address text similarity, and cosine similarity to calculate the similarity of each clustering feature. These are then compared with the similarity thresholds, but this method is complex. Alternatively, a comprehensive similarity can be calculated for comparison. Where i is the cluster feature number in the cluster feature system, I is the total number of cluster features in the cluster feature system, i = 1, 2, ..., I; ω i It is the similarity weight of the cluster feature numbered i, c i This refers to the cluster feature value of cluster feature i, which represents the similarity between the cluster feature value of cluster feature i and the cluster feature value of the core order with cluster feature i. Then, based on the overall similarity being greater than or equal to a preset similarity threshold, the same initial cluster is formed, thus generating initial clustering rules. For example, using standardized order data as samples, the overall similarity between samples is calculated (address similarity weight ω1=0.4, product similarity weight ω2=0.2, time similarity weight ω3=0.2, user similarity weight ω4=0.1, delivery request similarity weight ω5=0.1); the overall similarity threshold is set to 0.7, and then orders with an overall similarity greater than 0.7 are grouped into the same initial cluster, generating initial clustering rules, such as "same street + same product category + order time interval ≤ 1 hour + same delivery method" forming one cluster. This method is simple and fast to calculate, but the obtained overall similarity is easily overwritten and distorted by other data. The overall similarity can also be... Where i is the cluster feature number in the cluster feature system, I is the total number of cluster features in the cluster feature system, i = 1, 2, ..., I; ω i It is the similarity weight of the cluster feature numbered i, which can be obtained empirically, c i This is the cluster feature value of cluster feature i, which represents the similarity between the cluster feature value of cluster feature i and the cluster feature value of core order number i. i This is the influence scaling value for cluster feature number i. It can be obtained empirically or calculated. Of course, other algorithms are not excluded, but details will not be elaborated here. The comprehensive similarity calculated using this method can quickly separate orders with small differences in the similarity of a certain cluster feature, avoiding distortion caused by overwriting the calculated comprehensive similarity and thus improving the accuracy of clustering. (Reference) Figure 4 , Figure 2 The dynamic rule update mechanism involves real-time collection of order aggregation effect feedback data and business scenario change data. Order aggregation effect feedback data can include missed order rate, clustering accuracy, and duplicate clustering rate. Business scenario change data can include new platform integration, promotional activities, and delivery area adjustments. Rule updates are triggered when the missed order rate exceeds a preset threshold (e.g., 5%) or when a significant change occurs in the business scenario. A reinforcement learning algorithm is used, with the optimization objectives of "minimizing the missed order rate" and "maximizing clustering efficiency," adjusting the similarity weights and similarity thresholds of each clustering feature to update the clustering rule set. Simultaneously, a rule verification mechanism is established to conduct offline verification (using historical order data) and online pilot testing (selecting a subset of new orders). Once verification is successful, the rules are officially implemented. For example, real-time data collection of aggregation effect feedback is used. If the missed order rate rises to 6% during the 618 promotion (exceeding the preset threshold of 5%), a rule update is triggered. A reinforcement learning algorithm is used to adjust the similarity weights of clustering features (the time similarity weight ω3 is adjusted to 0.3, and the address similarity weight ω1 is adjusted to 0.35), the order interval threshold is adjusted to 2 hours, and "whether it is a promotion period" is added as a rule condition. The updated clustering rule is "same street + same product category + order interval ≤ 2 hours + same delivery method + same promotion scenario". Offline verification is performed using historical order data (missed order rate reduced to 1.8%), and 1,000 new orders from the Taobao platform are selected for online pilot testing (clustering accuracy reaches 98.2%). After successful verification, the rule is officially implemented. The state space of the reinforcement learning algorithm consists of clustering feature weights, similarity thresholds, and business scenario parameters; the action space consists of adjusting feature weights, modifying similarity thresholds, and adding / deleting clustering rules; the reward function is: reward value = α × (1 - missed order rate) + β × clustering efficiency - γ × rule complexity, where α, β, and γ are weight coefficients that can be adjusted according to business needs. Details are omitted here.

[0052] S5. Calculate the comprehensive similarity between cluster feature values ​​and core orders within candidate clusters. The comprehensive similarity calculation method uses the method mentioned in the rules of the dynamic clustering rule set. Comprehensive similarity is a core indicator for measuring the degree of aggregation and matching between orders across multiple platforms, used to determine whether different orders can be classified into the same cluster. The calculation logic, using weighted summation as an example, is as follows: Comprehensive Similarity... The comprehensive similarity score between new orders and core orders ranges from [0, 1]. A higher score indicates a higher order matching degree and is more suitable for aggregation. `i` represents the cluster feature number in the cluster feature system, and `I` represents the total number of cluster features in the cluster feature system. In this invention, `I` can be 5, encompassing five dimensions: address, product, time, user, and delivery requirements. The similarity weights of the cluster features are updated in real-time by the dynamic clustering rule set based on business scenarios and aggregation effects, and their sum is 1. i This refers to the cluster feature value of cluster feature i, with a value range of [0, 1]. Different cluster features use different calculation methods. Taking product similarity as an example, the calculation method is based on product category hierarchy + product code matching: if the product code of the new order is completely consistent with that of the core order, c2=1; if the product category is the same but the code is different, c2=0.7; if the major category is the same but the sub-category is different, c2=0.4; if the categories are completely different, c2=0. Of course, this is just a simple example. Address similarity can be calculated using a weighted average of latitude and longitude Euclidean distance + address text edit distance. Time similarity can be calculated using normalization. User similarity can be calculated based on user ID matching + membership level association. Delivery demand similarity can be calculated based on delivery method + delivery time requirement matching. Specific details will not be elaborated here. Comprehensive similarity can also be... Examples will not be given here.

[0053] S6. Determine whether the overall similarity is greater than or equal to the preset similarity threshold. If yes, then classify the order into the same cluster; otherwise, proceed to S7. For example, if the overall similarity between order A and order B is calculated to be 0.85 ≥ 0.7, then order A and order B will be classified into the same cluster.

[0054] S7. For orders that have not been clustered, perform a missed order detection to determine if they are missing. If so, execute S3 to re-extract the clustering feature values ​​from the platform's order data. Then, combine the order's historical clustering records and dynamic clustering rules to perform supplementary clustering for the missing orders. If not, create a new cluster and include the order in this new cluster. A missed order is defined as a valid order that, after multi-platform clustering, has not been successfully assigned to any cluster and has not triggered the supplementary clustering mechanism. This invention adopts a three-stage judgment process of "real-time clustering result verification → isolated order anomaly detection → historical data association verification" to ensure the accuracy of missed order identification. Specific details are not elaborated here. For example, using the isolated forest algorithm to identify unclustered order C (feature values: Wuchang Street, Clothing, 22:00, Express Delivery, 618 Promotion), combined with dynamic clustering rules, the comprehensive similarity between order C and the above clusters is calculated to be 0.75 ≥ 0.7, and order C is supplemented into that cluster.

[0055] S8. Output the order results for each cluster. The cluster results can include the order list for each cluster, the number of orders within the cluster, and a summary of delivery information. The clustering results are applied to order sorting, inventory allocation, and delivery route planning to achieve batch order processing, reduce duplicate deliveries, and lower operating and transportation costs. For example, output the order list for this cluster (Order A, Order B, Order C), the number of orders within the cluster (3 orders), and a summary of delivery information (all delivery addresses are in Wuchang Street, delivery method is express delivery, and the specified delivery time is 10:00-12:00 the next day). Push the clustering results to the warehouse management system to achieve batch sorting and packaging of the 3 orders; push them to the logistics management system to plan the optimal delivery route (one delivery covers the delivery addresses of 3 orders), reducing transportation mileage.

[0056] Example 2

[0057] refer to Figure 5 The present invention also provides a multi-platform e-commerce order aggregation device based on dynamic clustering rules, comprising:

[0058] Data Acquisition and Standardization Module: This module is used to collect raw order data from multiple e-commerce platforms through a distributed acquisition architecture, and to standardize the raw order data to obtain platform order data. Standardization processing can include field mapping, data cleaning, and format unification.

[0059] Clustering Feature Construction Module: Used to construct a multi-dimensional clustering feature system that includes static and dynamic features, and supports feature expansion;

[0060] Dynamic rule management module: Used to build an initial clustering rule set, and to optimize the clustering rules in real time through a dynamic rule update mechanism (combining reinforcement learning algorithm and rule verification mechanism) to generate a dynamic clustering rule set;

[0061] The order clustering execution module extracts clustering feature values ​​from platform order data, matches these feature values ​​with rules in a dynamic clustering rule set, and determines the candidate cluster to which an order belongs. It calculates the comprehensive similarity between the clustering feature values ​​and the core orders within the candidate clusters. It then determines whether the comprehensive similarity is greater than or equal to a preset similarity threshold, performs missing order detection for orders not yet clustered, and performs supplementary clustering for these missing orders.

[0062] The results output and application module is used to output the order clustering results and apply them to order sorting, inventory allocation, delivery route planning and other processes.

[0063] The device also includes a data storage module for storing raw order data, platform order data, clustering feature values, dynamic clustering rule sets, clustering result data, and aggregation effect feedback data. It can adopt a distributed database architecture to support efficient storage and fast querying of massive order data.

[0064] The device also includes a monitoring and early warning module, which is used to monitor key indicators (including missed order rate, clustering accuracy, data collection delay, and rule matching efficiency) in the order aggregation process in real time. When the indicators exceed the preset threshold, an early warning is triggered (such as SMS warning or system pop-up warning), and an anomaly report is generated.

[0065] It should be understood that the various forms of processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this invention disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this invention can be achieved, and this is not limited herein.

[0066] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for aggregating multi-platform e-commerce orders based on dynamic clustering rules, characterized in that, The multi-platform e-commerce order aggregation method based on dynamic clustering rules includes the following steps: S1. Collect order data from e-commerce platforms to obtain raw order data; S2. Standardize the raw order data to obtain platform order data; S3. Extract the cluster feature values ​​from the cluster feature system from the platform order data; S4. Match the clustering feature values ​​with the rules in the dynamic clustering rule set to determine the candidate cluster to which the order belongs; The method for obtaining the dynamic clustering rule set includes: initial clustering rule set construction and dynamic rule update mechanism; initial clustering rule set construction: using hierarchical clustering algorithm, standardized order data is used as samples, and the comprehensive similarity between samples is calculated based on the clustering feature system. Then, orders with a comprehensive similarity greater than or equal to a pre-set similarity threshold are grouped into a cluster and the core orders are determined, thereby generating the initial clustering rules; the dynamic rule update mechanism includes: real-time collection of order aggregation effect feedback data and business scenario change data; when the order aggregation effect feedback data exceeds the preset threshold or the business scenario undergoes significant changes, rule updates are triggered; reinforcement learning algorithm is used, with "minimizing the missed order rate" and "maximizing clustering efficiency" as optimization objectives, to adjust the similarity weight and similarity threshold of each clustering feature and update the clustering rule set; at the same time, a rule verification mechanism is established to conduct offline verification and online pilot testing of the updated rules, and officially activate them after successful verification; S5. Calculate the comprehensive similarity between the cluster feature values ​​and the core orders within the candidate clusters; The overall similarity Where i is the cluster feature number in the cluster feature system, I is the total number of cluster features in the cluster feature system, i = 1, 2, ..., I; ω i It is the similarity weight of the cluster feature numbered i, c i It is the cluster feature value of cluster feature number i, a i It is the scaling value of the influence of cluster feature number i; S6. Determine whether the overall similarity is greater than or equal to the preset similarity threshold. If yes, then classify the order into the cluster; otherwise, proceed to S7. S7. Perform a missing order detection on orders that have not been clustered. If a missing order is found, execute S3 and perform supplementary clustering for the missing orders. If not, create a new cluster. S8. Output the order results for the clusters.

2. The method for multi-platform e-commerce order aggregation based on dynamic clustering rules according to claim 1, characterized in that, Standardization processes include: field mapping, data cleaning, correction of abnormal data, and / or format standardization.

3. The method for multi-platform e-commerce order aggregation based on dynamic clustering rules according to claim 1, characterized in that, The clustering features in the clustering feature system include: delivery address dimension, product dimension, user dimension, order time dimension, and delivery demand dimension.

4. The method for multi-platform e-commerce order aggregation based on dynamic clustering rules according to claim 1, characterized in that, The order aggregation effect feedback data includes the missed order rate, clustering accuracy, and duplicate clustering rate.

5. The method for multi-platform e-commerce order aggregation based on dynamic clustering rules according to claim 1, characterized in that, Business scenario change data includes new platform access, promotional activities, and delivery area adjustments.

6. The method for multi-platform e-commerce order aggregation based on dynamic clustering rules according to claim 1, characterized in that, The clustering results include the order list for each cluster, the number of orders within the cluster, and a summary of delivery information.

7. A multi-platform e-commerce order aggregation device based on dynamic clustering rules, characterized in that, A multi-platform e-commerce order aggregation device based on dynamic clustering rules includes: Data Acquisition and Standardization Module: It is used to collect raw order data from multiple e-commerce platforms through a distributed acquisition architecture, and to standardize the raw order data to obtain platform order data; Clustering feature construction module: It is used to construct a multi-dimensional clustering feature system; Dynamic rule management module: It is used to build an initial clustering rule set, and optimize the clustering rules in real time through a dynamic rule update mechanism to generate a dynamic clustering rule set; Order clustering execution module: It is used to extract clustering feature values ​​from platform order data, match the clustering feature values ​​with rules in the dynamic clustering rule set to determine the candidate cluster to which the order belongs; calculate the comprehensive similarity between the clustering feature values ​​and the core orders in the candidate cluster; determine whether the comprehensive similarity is greater than or equal to the preset similarity threshold, perform missing order detection for orders that have not been clustered, and perform supplementary clustering for missing orders; The method for obtaining the dynamic clustering rule set includes: initial clustering rule set construction and dynamic rule update mechanism; initial clustering rule set construction: using hierarchical clustering algorithm, standardized order data is used as samples, and the comprehensive similarity between samples is calculated based on the clustering feature system. Then, orders with a comprehensive similarity greater than or equal to a pre-set similarity threshold are grouped into a cluster and the core orders are determined, thereby generating the initial clustering rules; the dynamic rule update mechanism includes: real-time collection of order aggregation effect feedback data and business scenario change data; when the order aggregation effect feedback data exceeds the preset threshold or the business scenario undergoes significant changes, rule updates are triggered; reinforcement learning algorithm is used, with "minimizing the missed order rate" and "maximizing clustering efficiency" as optimization objectives, to adjust the similarity weight and similarity threshold of each clustering feature and update the clustering rule set; at the same time, a rule verification mechanism is established to conduct offline verification and online pilot testing of the updated rules, and officially activate them after successful verification; The overall similarity Where i is the cluster feature number in the cluster feature system, I is the total number of cluster features in the cluster feature system, i = 1, 2, ..., I; ω i It is the similarity weight of the cluster feature numbered i, c i It is the cluster feature value of cluster feature number i, a i It is the scaling value of the influence of cluster feature number i; The results output and application module is used to output the order clustering results and apply them to the order sorting, inventory allocation, and delivery route planning processes.

Images