A store standard library matching system and method based on big data

Abstract

The application discloses a store standard library matching system and method based on big data, and belongs to the technical field of data matching. The application collects store data of all stores in a store standard library; calculates store characteristic values based on the store data; constructs a store topology structure by taking a single store as an independent topology node and taking the store characteristic values as a correlation basis; statistically analyzes historical inventory data of the stores corresponding to the topology nodes, and establishes inventory normal characteristic benchmarks of the stores; inputs real-time inventory data in the store data, the store topology structure and the inventory normal characteristic benchmarks into an artificial intelligence model, and judges whether the real-time inventory data has cross-store string recording abnormalities; when the cross-store string recording abnormalities exist, inventory string household abnormal data is constructed; the inventory string household abnormal data is traced back to the source in combination with the store topology structure and the store characteristic values, and a tracing positioning result is obtained; and the store standard library and the inventory data are corrected according to the tracing positioning result.

Description

Technical Field

[0001] This invention relates to the field of data matching technology, specifically a store standard library matching system and method based on big data. Background Technology

[0002] In the field of dealer store operation and management, with the expansion of offline store scale and the increasing number of various types of stores such as directly operated and franchised stores year by year, the geographical distribution of stores is becoming more and more widespread. Some stores have problems such as missing latitude and longitude coordinates and non-standard descriptions of basic information. Store inventory data is the core basis for dealers' refined operation, inventory scheduling, performance accounting and business decision-making. Its accuracy directly affects the dealers' operational efficiency and profitability. The accurate matching of the store standard library is a key prerequisite for ensuring the accurate attribution of inventory data and realizing centralized management of all stores.

[0003] Current technologies fail to calculate differentiated feature values ​​based on whether stores have latitude and longitude coordinates. They either uniformly use a single text feature or rely solely on spatial features, failing to consider the unique information characteristics of different stores. When inventory data is copied across stores, it is impossible to quickly filter candidate stores based on inter-store relationships, making it difficult to accurately trace the source of copied data. Furthermore, the lack of an inter-store correlation verification system exacerbates the probability of data copying and increases the difficulty of investigation. When establishing normal inventory characteristic benchmarks for stores, most methods only use simple statistical methods without exploring the temporal fluctuation characteristics of inventory data, resulting in significant deviations between the benchmarks and the actual inventory operation patterns of stores. Summary of the Invention

[0004] The purpose of this invention is to provide a store standard library matching system and method based on big data to solve the problems raised in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] Firstly, this application provides a store standard library matching method based on big data, including the following steps:

[0007] Collect store data from all stores in the standard store library; calculate store feature values ​​based on the store data;

[0008] Using each store as an independent topology node and store characteristic values ​​as the basis for association, a store topology structure is constructed; statistical analysis is performed on the historical inventory data of the stores corresponding to each topology node to establish a normal inventory characteristic benchmark for the stores.

[0009] The real-time inventory data, store topology, and normal inventory characteristics benchmarks from the store data are input into the artificial intelligence model to determine whether there are cross-store recording anomalies in the real-time inventory data.

[0010] When cross-store data entry anomalies occur, construct inventory cross-store anomaly data; for inventory cross-store anomaly data, combine store topology structure and store feature values ​​to trace the source and obtain the source location result; based on the source location result, correct the store standard library and inventory data.

[0011] In conjunction with the first aspect, in a first embodiment of the first aspect of this application, the calculation of store feature values ​​based on store data includes:

[0012] Based on store data, determine whether the store has latitude and longitude coordinates;

[0013] When a store does not have latitude and longitude coordinates, the keywords and character combination patterns of the store name are extracted, the hierarchical features and geographic identifier features of the detailed address are extracted, and the text features of the store are integrated. The text features are then standardized and quantified to form store feature values. Specifically, the quantification is based on the distinguishing ability, correlation matching adaptability and uniqueness of the text features.

[0014] Furthermore, the ability to distinguish is used to determine the degree to which a single text feature can distinguish different stores. The more a feature can reflect the differences between stores, the stronger the ability to distinguish. The ability to match and adapt is used to determine the degree to which a text feature adapts to subsequent store association matching and topology construction. The more a feature fits the association comparison requirements, the higher the adaptability. The degree of uniqueness is used to determine the degree to which a text feature corresponds to a single store. The less likely a feature is to be repeated with other stores, the higher the degree of uniqueness.

[0015] For each type of standardized text feature, a quantification value is assigned one by one based on the three criteria mentioned above:

[0016] For store name keywords, values ​​are assigned based on the keywords' distinguishability and uniqueness. Core exclusive keywords are assigned higher values, while general keywords are assigned lower values. The weighting of the assigned values ​​is adjusted based on the keywords' adaptability to subsequent association matching.

[0017] For the character combination patterns in store names, scores are assigned based on the uniqueness and distinguishability of the patterns. Unique and fixed character combinations are assigned higher scores, while common combination patterns are assigned lower scores. The scores are also adjusted based on the ease of matching and associating the patterns.

[0018] For address hierarchy features, values ​​are assigned based on the completeness and accuracy of the hierarchy. The more complete the hierarchy and the more accurate the address number, the higher the value is assigned. The suitability of stores is determined in the association matching based on the hierarchy features, and the assigned values ​​are supplemented and adjusted accordingly.

[0019] For address geographic identifier features, values ​​are assigned based on the uniqueness and accuracy of the geographic identifier. Specific geographic identifiers are assigned higher values, while general geographic identifiers are assigned lower values. At the same time, the scores are adjusted to take into account their practicality in subsequent association comparisons.

[0020] The quantitative assignment results of various text features are integrated, and the quantitative scores of various features are weighted and summed to form the quantitative feature value of a single store.

[0021] When a store has latitude and longitude coordinates, the spatial features corresponding to the store's latitude and longitude coordinates are extracted as the store's feature values.

[0022] In conjunction with the first aspect, in the second embodiment of the first aspect of this application, the step of constructing a store topology structure with each store as an independent topology node and store feature values ​​as the association basis includes:

[0023] Each store is treated as an independent topology node. A unique node identifier is assigned to each node and corresponds to the store ID in the store standard library. The store data and store feature values ​​of each store are associated with their respective topology nodes.

[0024] For any two stores, based on their store feature values, determine the similarity of their feature values. The closer the feature value quantification results are, the stronger the relationship between the two stores. Set a feature value similarity threshold. When the feature value similarity of two stores reaches or exceeds the threshold, it is determined that the two stores are related. When the threshold is not reached, it is determined that the two stores are not related and no relationship is established. For any two topological nodes that are determined to be related, construct a topological association edge between the nodes. The attributes of the association edge are determined by the similarity of the store feature values.

[0025] In conjunction with the first aspect, in the third embodiment of the first aspect of this application, the step of statistically analyzing the historical inventory data of stores corresponding to each topology node to establish a normal inventory characteristic benchmark for the stores includes:

[0026] For the historical inventory data of the store corresponding to each topology node, descriptive statistical algorithms are used for statistical analysis to form the basic characteristic indicators of the store's inventory.

[0027] Using time series analysis algorithms, the data on changes in inventory balance, inbound volume, and outbound volume are broken down into daily, weekly, and monthly time dimensions to analyze the fluctuation trend of inventory over time; the time correlation characteristics of inventory data are identified to determine whether there is periodicity and trend in inventory changes, and to determine the normal fluctuation range of inventory in different time periods; the fluctuation trend and the normal fluctuation range of inventory are integrated to form fluctuation characteristics; the fluctuation characteristics are combined with basic inventory characteristic indicators to form the normal characteristics of the inventory to be processed.

[0028] Cluster analysis algorithms are used to process the normal characteristics of the inventory to be processed, and the baseline of normal inventory characteristics is obtained.

[0029] In conjunction with the first aspect, in the fourth embodiment of the first aspect of this application, the step of performing statistical analysis on the historical inventory data of each store corresponding to each topology node using descriptive statistical algorithms to form basic inventory characteristic indicators for the store includes:

[0030] Calculate the mean and median of store inventory balances to obtain the concentration range of inventory; calculate the maximum, minimum, and range of inventory balances to obtain the fluctuation boundary of inventory; calculate the mean and fluctuation range of inbound, outbound, and inventory turnover rates to obtain the normal level of inventory turnover in a single store; calculate the frequency of various inventory operations; integrate the calculated data to form the basic inventory characteristic indicators of the store.

[0031] In conjunction with the first aspect, in the fifth embodiment of the first aspect of this application, the step of using a clustering analysis algorithm to process the normal characteristics of the inventory to be processed to obtain a baseline of normal inventory characteristics includes:

[0032] The normal characteristics of the unprocessed inventory of each store corresponding to each topology node are organized into sample data. Missing and abnormal features in the samples are filled in and removed to form a standardized feature sample set. Based on the standardized feature sample set, a clustering analysis algorithm is used to group all feature samples according to the similarity of inventory features between samples. Samples with basic inventory feature indicators higher than a set threshold are grouped into the same cluster. Abnormal feature samples within the cluster are removed.

[0033] The effective feature samples retained within the same cluster are aggregated as a whole, and the common features of all samples within the cluster are extracted to obtain the normal inventory feature benchmark.

[0034] In conjunction with the first aspect, in the sixth embodiment of the first aspect of this application, the step of inputting real-time inventory data, store topology, and normal inventory characteristic benchmarks from store data into an artificial intelligence model to determine whether there is cross-store data recording anomaly in the real-time inventory data includes:

[0035] Load the pre-trained gradient boosting tree classification model. The model is trained based on historical normal inventory data samples and historical cross-store abnormal data samples. The model contains hierarchical decision rules of multiple decision trees to distinguish between normal inventory data and cross-store abnormal data. Adjust the model's decision threshold and decision tree iteration depth according to the feature distribution of the current store topology and the update status of the normal inventory feature benchmark.

[0036] In the gradient boosting tree classification model, the first-layer decision tree verifies whether the deviation between the feature distribution of the current store topology and the normal inventory feature benchmark exceeds a reasonable range, filtering out suspected abnormal data whose features deviate from the benchmark; the second-layer decision tree verifies whether the features of the suspected abnormal data match the normal inventory feature benchmark of stores in the same cluster of the currently bound store; through the remaining decision trees, the operation time, device ID, and document information of the real-time data are cross-validated to ensure consistency with the regular business characteristics of the bound store; the model assigns corresponding weights to the judgment results of each decision tree and integrates the judgment conclusions of all decision trees; based on the integrated judgment conclusions, the model determines whether there is cross-store recording anomaly in the real-time inventory data.

[0037] In conjunction with the first aspect, in the seventh embodiment of the first aspect of this application, the step of constructing inventory cross-account abnormality data when cross-store cross-recording anomalies exist includes:

[0038] When cross-store data recording anomalies occur, extract real-time inventory data and the identifier of the currently incorrectly bound store topology node; label the anomaly type, the amount of abnormal data, the degree of deviation between real-time inventory characteristics and the normal inventory characteristics benchmark of the incorrectly bound store, and the characteristics that trigger the anomaly judgment.

[0039] In conjunction with the first aspect, in the eighth embodiment of the first aspect of this application, the step of tracing the source of abnormal inventory cross-store data by combining the store topology and store feature values ​​to obtain the source location result includes:

[0040] Based on the incorrectly bound store topology nodes, all topology nodes with associated edges to the incorrectly bound store topology nodes are selected in the store topology structure, and nodes in the same cluster are selected as traceability candidate nodes. The features in the abnormal inventory cross-store data are compared with the store feature values ​​corresponding to all traceability candidate nodes one by one. Specifically: when the candidate node store has no latitude and longitude, only the similarity between the features in the abnormal data and the text feature similarity between the traceability candidate node store is compared; when the candidate node store has latitude and longitude, the spatial distance correlation and text feature similarity between the features in the abnormal data and the candidate node store are compared.

[0041] Based on the feature value similarity comparison results, all source tracing candidate nodes are sorted, and the candidate node with the highest feature value similarity is selected as the source tracing and localization result.

[0042] Secondly, this application provides a store standard library matching system based on big data, including:

[0043] Store Feature Value Calculation Module: Includes: Store Data Acquisition Unit collects store data from all stores in the store standard library; Store Feature Value Calculation Unit calculates store feature values ​​based on the store data;

[0044] The normal inventory characteristic benchmark establishment module includes: a topology construction unit that uses a single store as an independent topology node and store characteristic values ​​as the basis for association to construct a store topology structure; and an inventory normal characteristic benchmark establishment unit that performs statistical analysis on the historical inventory data of the stores corresponding to each topology node to establish the normal inventory characteristic benchmark for the stores.

[0045] Cross-store recording anomaly detection module: including: Cross-store recording anomaly detection unit inputs real-time inventory data, store topology and normal inventory characteristics into artificial intelligence model to determine whether there is a cross-store recording anomaly in the real-time inventory data;

[0046] The data correction module includes: an abnormal data construction unit that constructs abnormal inventory data when cross-store cross-entry anomalies exist; a source tracing and positioning unit that traces the source of abnormal inventory data by combining the store topology and store feature values ​​to obtain the source tracing and positioning results; and a data correction unit that corrects the store standard library and inventory data based on the source tracing and positioning results.

[0047] Compared with the prior art, the beneficial effects of the present invention are:

[0048] 1. This invention adopts a differentiated feature value calculation method based on whether the store has latitude and longitude coordinates. For stores without latitude and longitude coordinates, the core text features of name and address are standardized and quantified. For stores with latitude and longitude coordinates, spatial features are relied upon. This takes into account the information characteristics of different stores and greatly improves the distinguishability and relevance of store feature values.

[0049] 2. This invention uses each store as an independent topological node and constructs a topological structure based on store characteristic values, realizing the transformation of store management from isolated management to associated management. It can quickly screen candidate stores for traceability through node association relationships, improve the efficiency of data tracing, and form an association verification mechanism between stores, reducing the probability of data tracing from the source.

[0050] 3. This invention employs a progressive processing approach using three algorithms: descriptive statistics, time series analysis, and cluster analysis. First, it mines the basic characteristics and time fluctuation characteristics of inventory. Then, it uses cluster analysis to remove abnormal samples and extract common characteristics. Combined with the store relationship optimization benchmark, it takes into account both the individuality of individual stores and the commonality of groups, so that the normal inventory characteristic benchmark is highly consistent with the actual operation pattern of stores. Attached Figure Description

[0051] Figure 1 This is a schematic diagram illustrating the steps of a store standard library matching method based on big data according to the present invention;

[0052] Figure 2 This is a system structure diagram of a store standard library matching system based on big data according to the present invention. Detailed Implementation

[0053] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0054] Example: Figures 1-2 As shown, the present invention provides a technical solution:

[0055] like Figure 1 As shown, this application provides a store standard library matching method based on big data, including the following steps:

[0056] Step S100: Collect store data for all stores in the standard store database; calculate store feature values ​​based on the store data;

[0057] Specifically, based on store data, determine whether the store has latitude and longitude coordinates;

[0058] When a store does not have latitude and longitude coordinates, the keywords and character combination patterns of the store name are extracted, the hierarchical features and geographic identifier features of the detailed address are extracted, and the text features of the store are integrated. The text features are then standardized and quantified to form store feature values. Specifically, the quantification is based on the distinguishing ability, correlation matching adaptability and uniqueness of the text features.

[0059] When a store has latitude and longitude coordinates, the spatial features corresponding to the store's latitude and longitude coordinates are extracted as the store's feature values.

[0060] In one specific embodiment, three typical stores from a certain distributor's standard store library were selected as experimental subjects, and the collected store data are as follows: Store 1 (no latitude and longitude): Store name is "Huimin Fresh Community Store", detailed address is "near a certain street in a certain district of a certain city in a certain province, around a certain business district", and the product category is fresh fruits and vegetables; Store 2 (with latitude and longitude): Store name is "Huimin Fresh Direct Store", detailed address is "a certain road section in a certain district of a certain city in a certain province", latitude and longitude coordinates are (a certain latitude coordinate, a certain longitude coordinate), and the product category is fresh fruits and vegetables; Store 3 (with latitude and longitude): Store name is "Huimin Fresh Franchise Store", detailed address is "another road section in a certain district of a certain city in a certain province", latitude and longitude coordinates are (another latitude coordinate, another longitude coordinate), and the product category is fresh fruits and vegetables.

[0061] After determining that Store 1 lacks latitude and longitude coordinates, the keywords of the store name were extracted as "Huimin Fresh Food, Community Store", and the character combination pattern was "Brand Name + Business Type Name + (Regional Identifier) ​​Store". The detailed address hierarchical features were extracted as "Province-City-District-Street", and the geographical identifier feature was "Business District". After integrating them into text features, they were standardized (standardizing the expression of regional identifiers, correcting address abbreviations, and removing redundant embellishments). The text features were then quantified based on their distinguishing ability, correlation matching adaptability, and uniqueness. Among them, the core street and the corresponding business district had high distinguishing ability and uniqueness, and good correlation matching adaptability. Finally, the store feature value of Store 1 was 0.82.

[0062] After determining that stores 2 and 3 have complete latitude and longitude coordinates, the spatial features corresponding to the latitude and longitude coordinates of the two stores are extracted as store feature values: the latitude and longitude coordinates of store 2 correspond to the spatial positioning features of its area, and the quantized store feature value is 0.76; the latitude and longitude coordinates of store 3 correspond to its exclusive spatial positioning features, and combined with the spatial differences between the coordinates and the surrounding stores, the quantized store feature value is 0.89. The spatial features can directly reflect the geographical location differences between the two stores.

[0063] Step S200: Construct a store topology structure with each store as an independent topology node and store characteristic values ​​as the basis for association; perform statistical analysis on the historical inventory data of the stores corresponding to each topology node to establish a normal inventory characteristic benchmark for the stores.

[0064] Specifically, each store is treated as an independent topology node, and a unique node identifier is assigned to each node, which corresponds to the store ID in the store standard library. The store data and store feature values ​​of each store are associated with their respective topology nodes.

[0065] For any two stores, based on their store feature values, determine the similarity of their feature values. The closer the feature value quantification results are, the stronger the relationship between the two stores. Set a feature value similarity threshold. When the feature value similarity of two stores reaches or exceeds the threshold, it is determined that the two stores are related. When the threshold is not reached, it is determined that the two stores are not related and no relationship is established. For any two topological nodes that are determined to be related, construct a topological association edge between the nodes. The attributes of the association edge are determined by the similarity of the store feature values.

[0066] Furthermore, for the historical inventory data of the stores corresponding to each topology node, descriptive statistical algorithms are used for statistical analysis to form basic inventory characteristic indicators of the stores.

[0067] Using time series analysis algorithms, the data on changes in inventory balance, inbound volume, and outbound volume are broken down into daily, weekly, and monthly time dimensions to analyze the fluctuation trend of inventory over time; the time correlation characteristics of inventory data are identified to determine whether there is periodicity and trend in inventory changes, and to determine the normal fluctuation range of inventory in different time periods; the fluctuation trend and the normal fluctuation range of inventory are integrated to form fluctuation characteristics; the fluctuation characteristics are combined with basic inventory characteristic indicators to form the normal characteristics of the inventory to be processed.

[0068] Cluster analysis algorithms are used to process the normal characteristics of the inventory to be processed, and the baseline of normal inventory characteristics is obtained.

[0069] Furthermore, the mean and median of the remaining inventory in each store are calculated to obtain the concentration range of the inventory; the maximum, minimum, and range of the remaining inventory are calculated to obtain the fluctuation boundary of the inventory; the mean and fluctuation range of the inbound volume, outbound volume, and inventory turnover rate are calculated to obtain the normal level of inventory turnover in a single store; the frequency of various inventory operations is calculated; and the calculated data are integrated to form the basic characteristic indicators of the store's inventory.

[0070] Furthermore, the normal characteristics of the unprocessed inventory of each store corresponding to each topology node are organized into sample data. Missing and abnormal features in the samples are filled in and removed to form a standardized feature sample set. Based on the standardized feature sample set, a clustering analysis algorithm is used to group all feature samples according to the similarity of inventory features between samples. Samples with basic inventory feature indicators higher than a set threshold are grouped into the same cluster. Abnormal feature samples within the cluster are removed.

[0071] The effective feature samples retained within the same cluster are aggregated as a whole, and the common features of all samples within the cluster are extracted to obtain the normal inventory feature benchmark.

[0072] In one specific embodiment, unique node identifiers are assigned to store 1, store 2, and store 3 respectively. Node 1 (corresponding to store ID: MD001), node 2 (corresponding to store ID: MD002), and node 3 (corresponding to store ID: MD003) associate the store data and feature values ​​of each store with the corresponding nodes (node ​​1 is associated with feature value 0.82, node 2 with 0.76, and node 3 with 0.89). A feature value similarity threshold of 0.75 is set, and the feature value similarity between any two stores is calculated: if the similarity between store 1 and store 2 is 0.78 (≥0.75), a relationship is determined; if the similarity between store 1 and store 3 is 0.72 (<0.75), there is no relationship; if the similarity between store 2 and store 3 is 0.77 (≥0.75), a relationship is determined. Topological connection edges are constructed for node 1 and node 2, and node 2 and node 3 respectively. The attributes of the connection edges are determined by similarity. The weight of the connection edge between node 1 and node 2 is 0.78, and the weight of the connection edge between node 2 and node 3 is 0.77, forming a store topology structure containing 3 nodes and 2 connection edges.

[0073] Historical inventory data for the past 30 days was collected from three stores and analyzed using descriptive statistical algorithms: Store 1 (node ​​1) had an average inventory balance of 850 kg, a median of 840 kg (concentrated range of 840-850 kg), a maximum of 1020 kg, a minimum of 680 kg, and a range of 340 kg (fluctuation boundary of 680-1020 kg); the average inbound volume was 120 kg / day, the average outbound volume was 115 kg / day, the average inventory turnover rate was 0.13 / day, and the frequency of inbound, outbound, and inventory count operations was 30 times, 28 times, and 3 times, respectively; the above data were integrated to form the basic inventory characteristic indicators of Store 1. Stores 2 and 3 are calculated using the same logic. Store 2's inventory balance is concentrated in the range of 780-790kg, with a fluctuation range of 620-950kg and an average daily inbound volume of 110kg. Store 3's inventory balance is concentrated in the range of 920-930kg, with a fluctuation range of 750-1100kg and an average daily inbound volume of 135kg. These factors form their respective basic inventory characteristic indicators.

[0074] Time series analysis was applied to the historical inventory data of three stores, breaking down the data into daily, weekly, and monthly segments: Store 1's daily inventory fluctuation range was ≤50kg, with higher outbound volumes on Mondays and Fridays (cyclical characteristic), and slightly higher inventory levels in the middle and latter part of each month (trend characteristic). The normal fluctuation range for daily inventory was determined to be 800-900kg, the weekly outbound volume fluctuation range to be 780-820kg, and the monthly inventory fluctuation range to be 820-880kg, which were then integrated to form fluctuation characteristics. This fluctuation characteristic was combined with the basic inventory characteristic indicators of Store 1 to form the normal characteristics of Store 1's pending inventory. Stores 2 and 3 were processed using the same logic: Store 2's daily inventory fluctuation range was 730-830kg, and Store 3's daily inventory fluctuation range was 870-980kg, each forming its own normal characteristics for pending inventory.

[0075] The normal characteristics of the pending inventory from the three stores were compiled into sample data. The two missing inventory counts for store 2 were supplemented, and 31 extreme outbound shipments (200 kg / day) from store 31 were removed, forming a standardized feature sample set. A threshold of 0.7 was set for the basic inventory characteristic indicators. Cluster analysis was used to group the samples. Since the basic inventory characteristic indicators of all three stores were above the threshold, they were grouped into the same cluster. After removing extreme abnormal samples from store 3 within the cluster, the common characteristics of the valid samples within the cluster were aggregated, ultimately obtaining the normal inventory characteristic benchmark for this cluster (the three stores): reasonable inventory balance range of 750-1000 kg, daily inbound volume of 90-140 kg, daily outbound volume of 85-130 kg, inventory turnover rate of 0.10-0.15 / day, and daily inventory fluctuation ≤60 kg. This benchmark was used as the basis for subsequent anomaly determination for the three stores.

[0076] Step S300: Input the real-time inventory data, store topology, and normal inventory characteristics benchmark from the store data into the artificial intelligence model to determine whether there is any cross-store recording anomaly in the real-time inventory data;

[0077] Specifically, a pre-trained gradient boosting tree classification model is loaded. The model is trained based on historical normal inventory data samples and historical cross-store abnormal data samples. The model contains hierarchical decision rules of multiple decision trees to distinguish between normal inventory data and cross-store abnormal data. The decision threshold and decision tree iteration depth of the model are adjusted according to the feature distribution of the current store topology and the update of the normal inventory feature benchmark.

[0078] In the gradient boosting tree classification model, the first-layer decision tree verifies whether the deviation between the feature distribution of the current store topology and the normal inventory feature benchmark exceeds a reasonable range, filtering out suspected abnormal data whose features deviate from the benchmark; the second-layer decision tree verifies whether the features of the suspected abnormal data match the normal inventory feature benchmark of stores in the same cluster of the currently bound store; through the remaining decision trees, the operation time, device ID, and document information of the real-time data are cross-validated to ensure consistency with the regular business characteristics of the bound store; the model assigns corresponding weights to the judgment results of each decision tree and integrates the judgment conclusions of all decision trees; based on the integrated judgment conclusions, the model determines whether there is cross-store recording anomaly in the real-time inventory data.

[0079] In one specific embodiment, a gradient boosting tree classification model, pre-trained based on 5000 historical normal inventory data samples and 800 historical cross-store abnormal data samples, is loaded. This model contains hierarchical decision rules for 10 decision trees. Combining the feature distribution of the current store topology (3 nodes, 2 associated edges) and the update status of the normal inventory feature benchmark (inventory balance 750-1000kg), the model's decision threshold is adjusted to 0.8, and the decision tree iteration depth is adjusted to 5 layers to ensure the model is adapted to the current store's inventory data judgment scenario.

[0080] Real-time inventory data was collected from Store 1 (Node 1): real-time inventory balance 1150kg, operation time Monday 10:00, device ID POS001, document number D20260209001. This real-time data, the store topology, and the baseline inventory characteristics were input into the model. The first-level decision tree verification revealed that Store 1's real-time inventory balance of 1150kg exceeded the baseline reasonable range (750-1000kg), with a deviation of 15%, exceeding the preset 10% reasonable deviation range. Therefore, this data was filtered as suspected abnormal data.

[0081] The second-layer decision tree verified the characteristics of the suspected abnormal data and found that its inventory balance characteristics matched the baseline of normal inventory characteristics of Store 2, which is in the same cluster as Store 1, with a 92% match (the baseline upper limit for Store 2 is 950kg, and although 1150kg exceeds the baseline for Store 2, the characteristic patterns are highly consistent). The remaining 8 decision trees were cross-validated: the operation time of Monday at 10:00 AM is consistent with the normal peak outbound time of Store 2, the device IDPOS001 is actually bound to Store 2 but was mistakenly entered into Store 1, and the delivery address in the document information points to the area of ​​Store 2, which is inconsistent with the normal business characteristics of Store 1. The model assigns weights to the judgment results of each decision tree (weight 0.2 for the first layer, weight 0.3 for the second layer, and weight 0.05 for each of the remaining trees). After integration, the probability of anomaly judgment is 0.88, which is higher than the judgment threshold of 0.8. Finally, it is determined that the real-time inventory data has a cross-store recording anomaly (data from Store 2 was entered into Store 1).

[0082] Step S400: When cross-store data entry anomalies occur, construct inventory cross-store anomaly data; for inventory cross-store anomaly data, trace the source by combining the store topology and store feature values ​​to obtain the source location result; based on the source location result, correct the store standard library and inventory data.

[0083] Specifically, when cross-store data recording anomalies occur, real-time inventory data and the identifier of the currently incorrectly bound store topology node are extracted; the anomaly type, abnormal data volume, deviation of real-time inventory characteristics from the normal inventory characteristics benchmark of the incorrectly bound store, and the characteristics that trigger anomaly judgment are marked.

[0084] Furthermore, based on the incorrectly bound store topology nodes, all topology nodes with associated edges to the incorrectly bound store topology nodes are selected in the store topology structure, and nodes in the same cluster are selected as traceability candidate nodes. The features in the abnormal inventory cross-store data are compared with the store feature values ​​corresponding to all traceability candidate nodes one by one. Specifically, when the candidate node store has no latitude and longitude, only the similarity between the features in the abnormal data and the text feature similarity between the traceability candidate node store is compared; when the candidate node store has latitude and longitude, the spatial distance correlation and text feature similarity between the features in the abnormal data and the candidate node store are compared.

[0085] Based on the feature value similarity comparison results, all source tracing candidate nodes are sorted, and the candidate node with the highest feature value similarity is selected as the source tracing and localization result.

[0086] In one specific embodiment, it is known that a cross-store data entry anomaly exists (data from store 2 is entered into store 1). The real-time inventory data is extracted (inventory balance 1150kg, operation time Monday 10:00, device ID POS001, document number D20260209001). The topology node identifier of the currently incorrectly bound store is node 1 (MD001). The data entry anomaly type is labeled as "inbound data entry," the abnormal data volume is 1150kg, and the deviation of the real-time inventory characteristics from the normal inventory characteristics benchmark of the incorrectly bound store 1 is 15%. The characteristics that trigger the anomaly judgment are "inventory balance exceeds the benchmark limit, device ID does not match the bound store, and document address does not match the store's regular business area."

[0087] Based on the incorrectly bound node 1, node 2 (with an edge weight of 0.78) with a related edge is selected from the store topology. Combined with the same clustering setting, nodes 2 and 3 are selected as candidate nodes for tracing. The features in the abnormal inventory cross-store data (text features include "Huimin Fresh Food, corresponding surrounding area", spatial features correspond to the latitude and longitude region of store 2) are compared one by one with the feature values ​​of the candidate nodes: Node 2 (with latitude and longitude) shows a spatial distance correlation of 0.88, a text feature similarity of 0.94, and a comprehensive similarity of 0.92; Node 3 (with latitude and longitude) shows a spatial distance correlation of 0.65, a text feature similarity of 0.74, and a comprehensive similarity of 0.72.

[0088] Based on the feature similarity comparison results, the candidate nodes are ranked as follows: Node 2 (similarity 0.92) > Node 3 (similarity 0.72). Node 2 is selected as the source location result, meaning that the correct store to which this abnormal data belongs is store 2 (MD002) corresponding to node 2. Based on this source location result, the store standard library and inventory data are corrected: the inventory data corresponding to document number D20260209001 is adjusted from store 1 (MD001) to store 2 (MD002), the store ID and topology node binding relationship of this data in the store standard library are updated, and the real-time inventory balance of store 1 and store 2 is corrected to ensure that the inventory data matches the store standard library.

[0089] like Figure 2 As shown, this application provides a store standard library matching system based on big data, including:

[0090] Store Feature Value Calculation Module: Includes: Store Data Acquisition Unit collects store data from all stores in the store standard library; Store Feature Value Calculation Unit calculates store feature values ​​based on the store data;

[0091] The normal inventory characteristic benchmark establishment module includes: a topology construction unit that uses a single store as an independent topology node and store characteristic values ​​as the basis for association to construct a store topology structure; and an inventory normal characteristic benchmark establishment unit that performs statistical analysis on the historical inventory data of the stores corresponding to each topology node to establish the normal inventory characteristic benchmark for the stores.

[0092] Cross-store recording anomaly detection module: including: Cross-store recording anomaly detection unit inputs real-time inventory data, store topology and normal inventory characteristics into artificial intelligence model to determine whether there is a cross-store recording anomaly in the real-time inventory data;

[0093] The data correction module includes: an abnormal data construction unit that constructs abnormal inventory data when cross-store cross-entry anomalies exist; a source tracing and positioning unit that traces the source of abnormal inventory data by combining the store topology and store feature values ​​to obtain the source tracing and positioning results; and a data correction unit that corrects the store standard library and inventory data based on the source tracing and positioning results.

[0094] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for matching a store standard library based on big data, characterized in that, Includes the following steps: Collect store data from all stores in the standard store database; Store characteristic values ​​are calculated based on store data, including: Based on store data, determine whether the store has latitude and longitude coordinates; When a store does not have latitude and longitude coordinates, the keywords and character combination patterns of the store name are extracted, the hierarchical features and geographic identifier features of the detailed address are extracted, and the text features of the store are integrated. The text features are then standardized and quantified to form store feature values. Specifically, the quantification is based on the distinguishing ability, correlation matching adaptability and uniqueness of the text features. When a store has latitude and longitude coordinates, the spatial features corresponding to the store's latitude and longitude coordinates are extracted as the store's feature values. Using each store as an independent topology node and store feature values ​​as the basis for association, a store topology structure is constructed, including: Each store is treated as an independent topology node. A unique node identifier is assigned to each node and corresponds to the store ID in the store standard library. The store data and store feature values ​​of each store are associated with their respective topology nodes. For any two stores, based on their store feature values, determine the similarity of their feature values. The closer the feature value quantification results are, the stronger the relationship between the two stores. Set a feature value similarity threshold. When the feature value similarity of two stores reaches or exceeds the threshold, it is determined that the two stores are related. When the threshold is not reached, it is determined that the two stores are not related and no relationship is established. For any two topological nodes that are determined to be related, construct a topological relationship edge between the nodes. The attributes of the relationship edge are determined by the similarity of the store feature values. Statistical analysis is performed on the historical inventory data of the stores corresponding to each topology node to establish a baseline for normal inventory characteristics of the stores, including: For the historical inventory data of the store corresponding to each topology node, descriptive statistical algorithms are used for statistical analysis to form the basic characteristic indicators of the store's inventory. Using time series analysis algorithms, the data on changes in inventory balance, inbound volume, and outbound volume are broken down into daily, weekly, and monthly time dimensions to analyze the fluctuation trend of inventory over time; the time correlation characteristics of inventory data are identified to determine whether there is periodicity and trend in inventory changes, and to determine the normal fluctuation range of inventory in different time periods; the fluctuation trend and the normal fluctuation range of inventory are integrated to form fluctuation characteristics; the fluctuation characteristics are combined with basic inventory characteristic indicators to form the normal characteristics of the inventory to be processed. Cluster analysis algorithms are used to process the normal characteristics of the inventory to be processed, and the baseline of normal inventory characteristics is obtained. The real-time inventory data, store topology, and normal inventory characteristics benchmarks from the store data are input into the artificial intelligence model to determine whether there are cross-store recording anomalies in the real-time inventory data. When cross-store data entry anomalies occur, construct inventory cross-store anomaly data; for inventory cross-store anomaly data, combine store topology structure and store feature values ​​to trace the source and obtain the source location result; based on the source location result, correct the store standard library and inventory data.

2. The method for matching a store standard library based on big data according to claim 1, characterized in that, The historical inventory data for each store corresponding to each topology node is statistically analyzed using descriptive statistical algorithms to form basic inventory characteristic indicators for the stores, including: Calculate the mean and median of store inventory balances to obtain the concentration range of inventory; calculate the maximum, minimum, and range of inventory balances to obtain the fluctuation boundary of inventory; calculate the mean and fluctuation range of inbound, outbound, and inventory turnover rates to obtain the normal level of inventory turnover in a single store; calculate the frequency of various inventory operations; integrate the calculated data to form the basic inventory characteristic indicators of the store.

3. The method for matching a store standard library based on big data according to claim 1, characterized in that, The process of using a clustering analysis algorithm to process the normal characteristics of the inventory to obtain a baseline of normal inventory characteristics includes: The normal characteristics of the unprocessed inventory of each store corresponding to each topology node are organized into sample data. Missing and abnormal features in the samples are filled in and removed to form a standardized feature sample set. Based on the standardized feature sample set, a clustering analysis algorithm is used to group all feature samples according to the similarity of inventory features between samples. Samples with basic inventory feature indicators higher than a set threshold are grouped into the same cluster. Abnormal feature samples within the cluster are removed. The effective feature samples retained within the same cluster are aggregated as a whole, and the common features of all samples within the cluster are extracted to obtain the normal inventory feature benchmark.

4. The method for matching a store standard library based on big data according to claim 1, characterized in that, The process involves inputting real-time inventory data, store topology, and normal inventory characteristics from store data into an artificial intelligence model to determine whether there are cross-store data entry anomalies in the real-time inventory data, including: Load the pre-trained gradient boosting tree classification model. The model is trained based on historical normal inventory data samples and historical cross-store abnormal data samples. The model contains hierarchical decision rules of multiple decision trees to distinguish between normal inventory data and cross-store abnormal data. Adjust the model's decision threshold and decision tree iteration depth according to the feature distribution of the current store topology and the update status of the normal inventory feature benchmark. In the gradient boosting tree classification model, the first-layer decision tree verifies whether the deviation between the feature distribution of the current store topology and the normal inventory feature benchmark exceeds a reasonable range, filtering out suspected abnormal data whose features deviate from the benchmark; the second-layer decision tree verifies whether the features of the suspected abnormal data match the normal inventory feature benchmark of stores in the same cluster of the currently bound store; through the remaining decision trees, the operation time, device ID, and document information of the real-time data are cross-validated to ensure consistency with the regular business characteristics of the bound store; the model assigns corresponding weights to the judgment results of each decision tree and integrates the judgment conclusions of all decision trees; based on the integrated judgment conclusions, the model determines whether there is cross-store recording anomaly in the real-time inventory data.

5. The method for matching a store standard library based on big data according to claim 1, characterized in that, When cross-store data entry anomalies occur, the process of constructing inventory cross-account anomaly data includes: When cross-store data recording anomalies occur, extract real-time inventory data and the identifier of the currently incorrectly bound store topology node; label the anomaly type, the amount of abnormal data, the degree of deviation between real-time inventory characteristics and the normal inventory characteristics benchmark of the incorrectly bound store, and the characteristics that trigger the anomaly judgment.

6. The method for matching a store standard library based on big data according to claim 1, characterized in that, The method for tracing the source of abnormal inventory cross-store data, combining store topology and store feature values, yields the following tracing and location results: Based on the incorrectly bound store topology nodes, all topology nodes with associated edges to the incorrectly bound store topology nodes are selected in the store topology structure, and nodes in the same cluster are selected as traceability candidate nodes. The features in the abnormal inventory cross-store data are compared with the store feature values ​​corresponding to all traceability candidate nodes one by one. Specifically: when the candidate node store has no latitude and longitude, only the similarity between the features in the abnormal data and the text feature similarity between the traceability candidate node store is compared; when the candidate node store has latitude and longitude, the spatial distance correlation and text feature similarity between the features in the abnormal data and the candidate node store are compared. Based on the feature value similarity comparison results, all source tracing candidate nodes are sorted, and the candidate node with the highest feature value similarity is selected as the source tracing and localization result.

7. A store standard library matching system based on big data, using the store standard library matching method based on big data according to any one of claims 1-6, characterized in that, include: Store Feature Value Calculation Module: Includes: Store Data Acquisition Unit collects store data from all stores in the store standard library; Store Feature Value Calculation Unit calculates store feature values ​​based on the store data; The normal inventory characteristic benchmark establishment module includes: a topology construction unit that uses a single store as an independent topology node and store characteristic values ​​as the basis for association to construct a store topology structure; and an inventory normal characteristic benchmark establishment unit that performs statistical analysis on the historical inventory data of the stores corresponding to each topology node to establish the normal inventory characteristic benchmark for the stores. Cross-store recording anomaly detection module: including: Cross-store recording anomaly detection unit inputs real-time inventory data, store topology and normal inventory characteristics into artificial intelligence model to determine whether there is a cross-store recording anomaly in the real-time inventory data; The data correction module includes: an abnormal data construction unit that constructs abnormal inventory data when cross-store cross-entry anomalies exist; a source tracing and positioning unit that traces the source of abnormal inventory data by combining the store topology and store feature values ​​to obtain the source tracing and positioning results; and a data correction unit that corrects the store standard library and inventory data based on the source tracing and positioning results.

Images