Visualization method for analysis of ex-warehouse goods of e-commerce warehouse
By designing visualization methods, multi-dimensional analysis and interactive display of goods out of e-commerce warehouses is achieved, which solves the non-interactive and non-intuitive problems of data analysis in the existing technology, and improves warehouse operation efficiency.
Patent Information
- Application Number
- CN202411193252.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-28
- Publication Date
- 2024-11-26
AI Technical Summary
In the prior art, the warehousing data analysis of e-commerce warehouses lacks interactivity and intuitiveness, which makes it difficult for warehouse administrators to summarize from the analysis results and cannot formulate accurate optimization measures, which leads to low efficiency in e-commerce warehouse operations.
Design a visualization method for outbound goods for e-commerce warehouses, including product feature projection view, associated product analysis view and cargo location comparison analysis view. Through data processing and dimensionality reduction technology, multi-dimensional analysis of outbound goods characteristics and Interactive display, supporting optimization decisions for cargo space management.
This method can deeply analyze the individual and group characteristics of goods, explore the similarity and differences of goods, check the distribution rules of goods, provide intuitive and understandable analysis results, provide effective decision-making support for the optimization of goods management in e-commerce warehouses, and improve the Overall operational efficiency.
Smart Images

Figure CN119027029A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of information visualization and visual analysis, and in particular to a visualization method for analyzing outbound goods from e-commerce warehouses. Background Art
[0002] With the booming development of the Internet e-commerce economy, the operating efficiency of traditional warehousing can no longer meet the development needs, and digital warehousing has emerged. Digital warehousing uses the Internet of Things and big data as core technologies to automatically collect data on the warehousing and outbound links of goods. These data contain rich information such as product characteristics and warehouse operation performance. Exploring the implicit patterns can assist in supervising the warehousing operation status and guide the allocation of incoming cargo spaces. However, due to the large scale, high dimensionality and time series characteristics of digital warehousing data, the valuable information contained in it is more difficult to mine or understand.
[0003] E-commerce warehouses are the core of modern e-commerce logistics systems. They carry multiple functions such as goods storage, sorting, packaging and distribution, and are the key link to ensure the smooth operation of the supply chain. At present, some studies on warehouse data analysis use automated analysis methods such as machine learning and data mining. However, the analysis process of the above methods lacks interactivity, warehouse managers cannot actively participate in the analysis, and the analysis results lack intuitiveness and comprehensibility. As a result, it is difficult for warehouse managers to summarize and generalize from the general analysis results and formulate accurate optimization measures. Although some studies have simulated and visualized warehouse scenes, few have conducted comprehensive analysis of the characteristics of outbound goods. Therefore, the existing visualization methods are still limited in their ability to provide decision support for optimizing goods management, resulting in low overall operational efficiency of e-commerce warehouses. Summary of the invention
[0004] In order to solve the above-mentioned technical problems existing in the prior art, the purpose of the present invention is to provide a visualization method for analyzing goods shipped out of e-commerce warehouses, and to design and implement interactive goods feature projection views, related goods analysis views, and cargo location comparison analysis views based on various goods-related data in the operation process of e-commerce warehouses. The present invention can assist e-commerce warehouse managers in exploring individual characteristics and group characteristics of goods, deeply analyze the similarities and differences of goods, and support the exploration of related goods combinations, check cargo location distribution patterns, and provide decision support for cargo location management optimization. The specific technical scheme is as follows:
[0005] A visualization method for analyzing outbound goods from an e-commerce warehouse includes the following steps:
[0006] Step S1: Obtain outbound order data and outbound product data from the warehouse management system of the warehousing service provider, and store them in the database after data processing;
[0007] Among them, the outbound order data includes the picking batch code, order code, product code and picking time; the outbound product data includes the order code, product code, the picking quantity of the product and the location code used to mark the picking location of the product;
[0008] Step S2: Perform visual mapping based on the outbound order data and outbound product data;
[0009] Step S3: Perform visual layout based on the result of visual mapping.
[0010] Preferably, the data processing includes:
[0011] (1) Delete the outbound order data or outbound product data with empty picking batch code in batch picking;
[0012] (2) Taking the month as the picking granularity, calculate the product characteristic data, including: outbound frequency, outbound quantity, outbound frequency ratio, outbound quantity ratio, number of storage locations, and product type;
[0013] (3) Extract the product code and the cargo location code, calculate the product location data based on the cargo location placement method of ground square picking;
[0014] (4) The t-SNE (t-Distributed Stochastic Neighbor Embedding) algorithm is used to reduce the dimension of the product feature data. Specifically:
[0015] Use one-hot encoding to create binary features for each categorical feature, so that when a product is in the corresponding category, the value of the binary feature is 1, otherwise it is 0;
[0016] Linear normalization is used to scale all product feature data, and the t-SNE algorithm is applied to set hyperparameters to achieve data dimensionality reduction;
[0017] By using normalization, the absolute coordinates obtained by data dimension reduction are converted into the first relative coordinates, and stored in the picking product feature data in the form of coordinate pairs.
[0018] (5) The dimension of the product location data is reduced by using the MDS (Multidimensional Scaling) algorithm, specifically:
[0019] Store the picking goods location data in the form of a dictionary, where the key is the goods code and the value is the corresponding area, row and column;
[0020] Calculate the distance between each pair of goods using Manhattan distance;
[0021] Reduce the dimension of cargo location data through MDS algorithm;
[0022] The minimum-maximum standardization method is used to convert the absolute coordinates obtained through dimensionality reduction by the MDS algorithm into the second relative coordinates, and the coordinate pairs are stored in the goods location data.
[0023] (6) The Eclat algorithm is used to associate the outbound order data, specifically:
[0024] 1) Data format conversion
[0025] Select the order code, picking batch code and goods included in the order from the outbound order data to form a new data set D;
[0026] Create an empty mapping table M, where the key is the picking batch code and the value is an empty product set;
[0027] Generate a "batch-item" data set based on the mapping table M, where each record represents a picking batch, and the items contained in each picking batch are an item set;
[0028] 2) Use the Eclat algorithm to find frequent product combinations that meet the minimum support (i.e. the proportion of product combination outbound frequency)
[0029] 2.1) Create an empty frequent item set list F and an empty candidate frequent item set list C; traverse the "batch-goods" data set and add each appearing good to the candidate frequent item set list C;
[0030] 2.2) Traverse and calculate the support of each item set in the candidate frequent item set list C in the "batch-goods" data set;
[0031] 2.3) Add the itemsets whose support meets the minimum support requirement to the frequent itemset list F;
[0032] 2.4) In the frequent item set list F, for any two item sets, if the first k-1 elements they share are the same, they are integrated into a new item set and added to the list C;
[0033] Repeat steps 2.2) to 2.4) until the candidate frequent item set list C is empty or no new candidate frequent item set can be synthesized, and output the frequent item set list F;
[0034] Among them, support refers to the proportion of the outbound frequency of the associated product combination;
[0035] 3) Integrate the frequent item set list to obtain the associated product data, including the product combination list, the outbound frequency of the product combination, the outbound frequency ratio of the product combination, the outbound frequency list of each product in the product combination, and the outbound quantity list.
[0036] Preferably, the visual mapping includes:
[0037] (1) Constructing product feature projection views at monthly granularity
[0038] 1) Construct a projection scatter plot:
[0039] All the shipped goods in a single month are represented as dots, where one dot represents one product.
[0040] Encoding the first relative coordinate by the coordinate of the dot;
[0041] Clustering results of classification features are color-coded and cluster centers are represented by preset flat graphics;
[0042] 2) Construct product glyphs:
[0043] The outer donut chart consists of two semicircles of equal size on the left and right. The left semicircle encodes the quantity of goods shipped out of the warehouse, and the right semicircle encodes the frequency of goods shipped out of the warehouse.
[0044] Taking 12 o'clock on the clock as the starting position, the specific value of the outbound quantity is mapped through the length of the ring in the left semicircle, and the specific value of the outbound frequency is mapped through the length of the ring in the left semicircle; different colors are used to map the outbound quantity, outbound frequency, the lower quartile and upper quartile of the outbound frequency, and the lower quartile and upper quartile of the outbound quantity;
[0045] The inner candy chart includes two sectors, upper and lower sectors, and a circle in the middle. The proportion of the central angle of the upper sector to the sector segment represents the proportion of the outbound quantity to the total outbound quantity of the month, and the proportion of the central angle of the lower sector to the sector segment represents the proportion of the outbound frequency to the total outbound frequency of the month. The circle uses text labels to represent the product classification.
[0046] (2) Constructing related product analysis views based on monthly granularity
[0047] 1) Build a node link graph:
[0048] The connection between the dots indicates the relationship between the goods;
[0049] Use colored dots to mark the goods selectedGoods and their related goods otherGoods selected by lasso interaction in the goods feature projection view; the related goods otherGoods are calculated by the data mining algorithm;
[0050] Encode the outbound quantity or outbound frequency through the size of the dots;
[0051] The width of the line encodes the support of the associated product combination;
[0052] Based on the width of the connection line, the dots corresponding to the associated product combinations with high support are set in adjacent positions;
[0053] 2) Build a list of related product combinations:
[0054] Use different rows to map different related product combinations;
[0055] Different columns are used to map the composition of product combinations, the frequency of product combination shipments, the support of product combinations, and optional operations. Optional operations include highlighting, deleting, and marking.
[0056] (3) Constructing a comparative analysis view of cargo locations based on monthly granularity
[0057] 1) Construct a bubble chart:
[0058] Use product glyphs to represent products, and one product glyph represents one product; use bubbles to surround a group of product glyphs to represent a related product combination;
[0059] The second relative coordinate is encoded by the position of the product glyph;
[0060] Different associated product combinations are color-coded, and bubbles of the same color contain all products of the corresponding associated product combination;
[0061] The distribution distance of the cargo locations of the associated product combinations is encoded by the bubble area;
[0062] 2) Build a product portfolio management list:
[0063] Use different rows to represent different related product combinations;
[0064] Different columns are used to represent the goods combination composition GoodsCombination, the goods combination delivery frequency CombinationFreq, the goods combination support Support and the optional operation Operation;
[0065] Among them, the product combination outbound frequency CombinationFreq and the product combination support Support have sorting functions.
[0066] Preferably, the visual layout includes:
[0067] (1) Visual layout of product feature projection view
[0068] The upper right corner of the product feature projection view is set to the lasso control button, cluster counter and time selector, the upper left corner is set to the product glyph chart, and the middle is set to the projection scatter plot;
[0069] 1) Visual layout of projection scatter plot:
[0070] The product feature data after dimension reduction by the t-SNE algorithm is divided into groups through K-Means clustering, specifically:
[0071] 1.1) Select k points as the initial cluster centers;
[0072] 1.2) Calculate the distance from each dot to the center of each initial cluster and divide the clusters according to the distance;
[0073] 1.3) For each cluster, calculate the average value of the coordinates of all origins, and set the coordinates corresponding to the average value as the new cluster center;
[0074] 1.4) Repeat steps 1.2) to 1.3) until the distance between the cluster centers before and after the update is no greater than 0.00001;
[0075] 1.5) Convert the coordinates of all points in each cluster and the center of the cluster into relative coordinates and store them; calculate the actual position of the dot in the projection view of the product feature,
[0076] clusterCenter xi =ScatterSvgW×clusterCenterRelative xi ×scaleFactor;
[0077] clusterCenter yi =ScatterSvgH×clusterCenterRelative yi ×scaleFactor;
[0078] In the formula, (clusterCenter xi ,clusterCenter yi ) is the real coordinate of the center of the i-th cluster, ScatterSvgW and ScatterSvgH are the width and height of the product feature projection view, clusterCenterRelative xi 、clusterCenterRelative yi is the relative coordinate of the center of the ith cluster, and scaleFactor is the scaling factor of the projection view of the product features;
[0079] 2) Visual layout of product glyphs:
[0080] The dotted line in the left semicircle represents the mean of the number of shipments in the current month, and the dotted line in the right semicircle represents the mean of the frequency of shipments in the current month;
[0081] (2) Visual layout of related product analysis view
[0082] In the related product analysis view, there is a node link diagram in the middle and a control panel on the top; the control panel includes a support selection slider, a related product combination number selection slider, a clock button, and a statistics button;
[0083] For a node-link graph,
[0084] 1) Associated item conversion:
[0085] Convert the associated items corresponding to the associated product combination into the data format containing nodes and links required by D3.js;
[0086] If the length of the association item is equal to 2, a link including the source node and the target node is directly generated from the two elements in the association item; if the length of the association item is greater than 2, a closed-loop link including source and target is generated from the elements in the association item;
[0087] 2) Node coordinate mapping:
[0088] Use d3.forceSimulation to build a force-directed layout and determine the coordinates of nodes and links through physical mechanics simulation;
[0089] Specifically, repulsion, attraction and central force are set, and node coordinates are iterated;
[0090] Using the svg element <circle>Create a node using <line>Draw links; define drag behavior functions and interact with elements <circle>Binding to update node coordinates in real time;
[0091] (3) Visual layout of the cargo location comparison analysis view
[0092] In the cargo location comparison analysis view, there is a bubble chart in the middle and a function button in the upper right corner;
[0093] For bubble charts,
[0094] 1) Draw bubble outlines with the help of the bubble sets visualization library;
[0095] 2) Collision avoidance detection:
[0096] 2.1) Initialize the node position to the relative coordinates after dimensionality reduction (x i ,y i ), the node radius is r i ;
[0097] 2.2) Calculate the distance between nodes,
[0098]
[0099] In the formula, (x j ,y j )、(x k ,y k ) are the coordinates of node j and node k respectively;
[0100] 2.3) For each pair of nodes, if the distance between the nodes is less than the sum of the radii, a collision is determined;
[0101] 2.4) For each pair of colliding nodes, move at least one of the nodes along the connecting line until the distance between the nodes is equal to the sum of the radii;
[0102] 2.5) Iterate 2.1) to 2.5) until the displacement of the nodes before and after the update does not exceed 0.00001;
[0103] 3) Sorting of related product combinations:
[0104] Define the sum of the distances between all items in the associated product combination as the shelf distance of the associated product combination;
[0105] Among them, product X 1 and product X 2 The distance d between them is:
[0106] d=|x 1 -x 2 |+|y 1 -y 2 |;
[0107] In the formula, (x 1 ,y 1 )、(x 2 ,y 2 ) are respectively product X 1 、Product X 2 The coordinates of
[0108] Use maximum and minimum normalization to map all product feature data into the interval.
[0109]
[0110] In the formula, z min 、z max are the minimum and maximum eigenvalues, respectively; z is the original eigenvalue, and z' is the eigenvalue normalized by the maximum and minimum values;
[0111] Perform weighted summation on the eigenvalues normalized by maximum and minimum.
[0112]
[0113] In the formula, Score is the ranking score of the associated product combination, S a , S b , S c , S d are the maximum and minimum normalized outbound frequency, cargo location distance, outbound frequency mean and outbound quantity mean of the associated product combination, respectively. a , W b , W c , W d They are the weights corresponding to the maximum and minimum normalized outbound frequency, cargo location distance, mean outbound frequency and mean outbound quantity of the associated product combination.
[0114] Furthermore, it also includes: realizing human-computer interaction based on the result of the visualization layout;
[0115] (1) Interaction of product feature projection views
[0116] Drag: Drag the mouse to pan the view to view local information of the projected scatter plot;
[0117] Zoom: In the projection scatter plot, use the mouse wheel to zoom and switch the observation granularity;
[0118] Click: In the projection scatter plot, click a dot to view the corresponding product glyph;
[0119] Hover: In the product glyph, hover the mouse to view the specific properties of the product;
[0120] Lasso: Analyze the relationship between goods by selecting dots within a specific range;
[0121] (2) Interaction of related product analysis views
[0122] Drag: Click and drag the associated product analysis view with the mouse; after dragging a node, the coordinates of the remaining nodes will be automatically updated and rearranged;
[0123] Zoom: Use the mouse wheel to zoom in and out of the analysis view of related products;
[0124] Click: Switch the encoding information of dot size by clicking the button;
[0125] Floating: By hovering the mouse, you can view the corresponding product code and outbound quantity, or product code and outbound frequency of the dot in the form of a prompt box; by hovering the mouse, you can view the corresponding support of the connection in the form of a prompt box;
[0126] Filter: Dynamically adjust the parameter values of the mining algorithm by dragging the support selection slider and the number of associated product combinations selection slider;
[0127] (3) Interaction of the cargo location comparison analysis view
[0128] Drag: Drag the mouse to pan the view to view the comparison of the product glyphs and product locations of a specific associated product combination;
[0129] Zoom: Use the mouse wheel to zoom in and out the comparison and analysis view of the cargo location;
[0130] Click: Click on the bubble to add the corresponding associated product combination to the management list; click the function button of the view to view the detailed information of the management list;
[0131] Hover: Hover the mouse to view the product information of the product corresponding to the product glyph in the form of a prompt box, including product code, product location, outbound frequency, outbound quantity, outbound frequency ratio and outbound quantity ratio; when hovering, the bubble at the corresponding position is highlighted, remove the hover, the highlight is removed and the prompt box disappears.
[0132] The technical solution provided by the present invention can conduct a comprehensive analysis of the characteristics of outbound goods in terms of individual and group dimensions, and conduct a joint analysis of the correlation between groups of goods and the spatial distribution of goods. It has high interactivity, can meet the warehouse manager's proactive analysis, and the displayed results are highly intuitive and understandable. Furthermore, it can provide good data support for optimizing goods management decisions and ensure that e-commerce warehouses have high overall operational efficiency. BRIEF DESCRIPTION OF THE DRAWINGS
[0133] Figure 1 It is a schematic diagram of the overall process framework of the method for visualizing analysis of outbound goods in the present invention.
[0134] Figure 2 It is a schematic diagram of the overall layout of the product feature projection view in the present invention.
[0135] Figure 3 It is a schematic diagram of the visual mapping of the product glyph diagram in the present invention.
[0136] Figure 4 It is a schematic diagram of the overall layout and visual mapping of the related product analysis view in the present invention.
[0137] Figure 5 It is a schematic diagram of the overall layout of the cargo position comparison and analysis view in the present invention.
[0138] Figure 6 (a) is a schematic diagram of the interactive design of the product glyph in the present invention, Figure 6 (b) Schematic diagram of the interactive design of the projected scatter plot lasso.
[0139] Figure 7 It is a schematic diagram of the interactive design of the node link graph in the present invention.
[0140] Figure 8 It is a schematic diagram of the design of the associated product combination list in the present invention.
[0141] Fig. 9 It is a schematic diagram of the interactive design of the bubble chart in the present invention.
[0142] Fig.10 It is a schematic diagram of the design of the product combination management list in the present invention. DETAILED DESCRIPTION
[0143] The technical solution provided by the present invention will be further explained in detail below with reference to the accompanying drawings.
[0144] The present invention realizes multi-level analysis of e-commerce warehouse product data through information visualization and visual analysis methods, combined with innovative visualization views and flexible interaction methods, helping warehouse managers to deeply explore the multi-dimensional characteristics of products and the spatial distribution patterns of products in the warehouse, and discover potential storage problems. The technical solution includes: data acquisition and processing, visualization mapping, visualization layout and implementation, and interactive design.
[0145] The overall process framework diagram of the method for analyzing and visualizing outbound goods for e-commerce warehouses of the present invention is as follows: Figure 1 As shown, the specific steps are as follows:
[0146] Step 1: Data acquisition and processing
[0147] Obtain product-related data from the warehouse management system of the warehousing service provider, perform data cleaning, data preprocessing, data dimension reduction, and related product mining, and then store it in the database.
[0148] 1. Data acquisition
[0149] Obtain outbound order data and outbound product data from the warehouse management system of the warehousing service provider. The important attributes of outbound order data include: picking batch code, order code, product code, picking time, and the important attributes of outbound product data include: order code, product code, product picking quantity, and cargo location code. The cargo location code marks the picking location of the product, and the coding format is, for example, B001, C056, etc.
[0150] 2. Data cleaning and conversion
[0151] This method processes erroneous records and missing fields in warehouse operation data through data cleaning, thereby optimizing data quality and improving the accuracy of data analysis.
[0152] (1) Delete missing data
[0153] There are two picking methods in warehouse picking operations: batch picking and individual picking. When the warehouse has a large number of orders, picking batches will be created to pick orders in batches to improve the efficiency of picking operations. This method focuses on the operation process supervision under order batch picking, so the fields with empty "picking batch code" attribute values are deleted.
[0154] (2) Extracting the characteristic data of outbound goods
[0155] In order to analyze the flow pattern of goods in picking operations, the frequency of goods leaving the warehouse, the quantity of goods leaving the warehouse, the proportion of the frequency of goods leaving the warehouse, the proportion of the quantity of goods leaving the warehouse, the number of storage locations, and the type of goods are calculated with the month as the time granularity to obtain the characteristic data of the picked goods.
[0156] (3) Extracting product location data
[0157] Separate the "goods code" and "location code" fields, extract the area of the goods location according to the location placement method of the ground square picking, such as "area B", calculate the rows and columns of the goods location, and obtain the picking goods location data.
[0158] 3. Data Dimensionality Reduction
[0159] In order to facilitate users to explore the structure and pattern of product data, the t-SNE algorithm is used to reduce the dimension of product feature data, so as to intuitively reveal the local similarities of goods; the MDS algorithm is used to reduce the dimension of product location data, which can deeply analyze the global distribution of locations.
[0160] The specific implementation of t-SNE algorithm dimensionality reduction is as follows:
[0161] (1) For categorical features (such as product type), convert them into numerical form. Use one-hot encoding to create a new binary feature for each categorical feature. When the category of a sample is that category, the value of the feature is 1, otherwise it is 0.
[0162] (2) Perform linear normalization on the data and scale all features to a uniform range to ensure that all features have the same importance in dimensionality reduction.
[0163] (3) Use the t-SNE algorithm to reduce the dimension of the processed data.
[0164] The result of dimensionality reduction by the t-SNE algorithm is the absolute coordinates of the feature data in the new low-dimensional space. In order to draw the expected scatter plot on the system page, the absolute coordinates need to be converted into relative coordinates. The normalization method is used to convert the absolute coordinates after dimensionality reduction into relative coordinates and store them in the form of coordinate pairs in the feature data of the picked goods.
[0165] After using the t-SNE algorithm and setting hyperparameters to achieve data dimensionality reduction, goods with different characteristics are mapped into specific scattered points on a two-dimensional plane. Through the distribution density of the scattered points, the group characteristics of the goods can be intuitively observed, making it easier to identify the similarities and differences between different types of goods.
[0166] The specific implementation of MDS algorithm dimensionality reduction is as follows:
[0167] (1) The picking item location data is stored in dictionary form, where the key is the item code and the value is the corresponding area, row, and column information.
[0168] (2) Use Manhattan Distance to calculate the distance between each pair of items.
[0169] (3) Apply the MDS algorithm to reduce the dimension of cargo location data.
[0170] (4) The minimum-maximum normalization method is used to convert the absolute coordinates after dimensionality reduction by the MDS algorithm into relative coordinates, and the coordinate pairs are stored in the picking goods location data.
[0171] After dimensionality reduction using the MDS algorithm, the cargo location codes are mapped to different positions on the two-dimensional plane. Through the scattered spatial relationship, the cargo location distribution of the goods can be analyzed intuitively, making it easier to understand the spatial relationship and distribution patterns of the goods.
[0172] 4. Related product mining
[0173] Analyzing the correlation of goods in an order and further exploring the spatial distribution patterns of related product combinations can help managers make decisions to optimize product storage.
[0174] The present invention uses the Eclat algorithm in association rule mining to perform association analysis on product data. First, the data format is converted. The specific steps are as follows:
[0175] (1) Select the order code, picking batch code, and picking items included in the order from the outbound order data to form a new data set D.
[0176] (2) Create an empty mapping table M, whose key is the picking batch code and whose value is an empty set of goods.
[0177] (3) Generate a "batch-item" data set D based on the mapping table M. Each record in this data set represents a picking batch, and all the items it contains are regarded as an item set.
[0178] After obtaining the "batch-product" data set, use the Eclat algorithm to find all frequent product combinations that meet the minimum support. The specific steps are as follows:
[0179] (1) Create an empty frequent item set list F and an empty candidate frequent item set list C. Traverse the "batch-item" data set and add each item that appears to the candidate frequent item set list C.
[0180] (2) Traverse each item set in the candidate frequent item set list C and calculate its support in the "batch-goods" data set.
[0181] (3) Traverse each item set in the candidate frequent item set list C and add the item sets whose support meets the minimum support requirement to the frequent item set list F.
[0182] (4) In the frequent itemset list F, for any two itemsets, if their first k-1 common elements are consistent, they are integrated into a new itemset and added to the list C.
[0183] (5) Repeat steps (2) to (4) until the candidate frequent item set list C is empty or no new candidate frequent item set can be generated, and output the frequent item set list F.
[0184] After executing the Eclat algorithm, the frequent item set list is integrated to obtain the associated product data, including the product combination list, the delivery frequency of the product combination, the delivery frequency ratio of the product combination, the delivery frequency list of each product in the product combination, and the delivery quantity list.
[0185] 5. Data storage
[0186] After completing the above operations, the picked goods feature data, picked goods location data and related goods data are stored in the MySQL database for subsequent visualization.
[0187] Step 2: Visual Mapping
[0188] After data acquisition and processing, Figure 2 , Figure 4 , Figure 5 As shown, a visual mapping scheme is designed for the product feature projection view, the associated product analysis view and the cargo location comparison analysis view in the present invention.
[0189] 1. Product feature projection view
[0190] (1) Projection scatter plot
[0191] Shape: A dot is used to represent a product, and all the dots in the view represent all the shipped products in a certain month.
[0192] Position: The position of dots is encoded using the t-SNE dimensionality reduction algorithm. The goods represented by dots with similar positions have similar product characteristics, while the goods represented by dots with farther positions have greater differences.
[0193] Color: The color of the dots encodes the clustering results of the product characteristics. Contrasting colors are selected to map different groups of products. The dots in the same group are the smallest distance from each other, and the dots between different groups are the largest distance from each other. The cluster center of each group is represented by a gray rectangle.
[0194] (2) Product glyphs
[0195] Shape and position: The outer donut chart consists of two equal-sized circles on the left and right. The left circle encodes the number of goods shipped out of the warehouse, and the right circle encodes the frequency of goods shipped out of the warehouse. The inner candy chart consists of two upper and lower sectors and a circle in the middle. The upper sector indicates the proportion of the number of goods shipped out of the warehouse relative to the total number of goods shipped out of the warehouse in the month, and the lower sector indicates the proportion of the frequency of goods shipped out of the warehouse relative to the total frequency of goods shipped out of the warehouse in the month. The middle circle uses text labels: such as A, B, C, to indicate the classification of goods.
[0196] Color: Blue is used to map the quantity of goods shipped out, pink is used to map the frequency of goods shipped out. Gray rings are used to map the lower and upper quartiles of the frequency or quantity of goods shipped out.
[0197] Length: Take 12 o'clock on the clock as the starting position, and use the length of the ring to map the specific value of the quantity or frequency of goods leaving the warehouse.
[0198] Area: Use the sector area to map the proportion of the number of goods shipped out or the frequency of goods shipped out relative to the number of goods shipped out or the frequency of goods shipped out in the current month.
[0199] 2. Related product analysis view
[0200] (1) Node-link graph
[0201] Shape: Each dot represents a product, and the line between two products represents the relationship between them.
[0202] Color: The goods selectedGoods interactively selected by lasso in the item feature projection view are marked with yellow dots, and the goods otherGoods associated with selectedGoods calculated by the data mining algorithm are marked with blue dots.
[0203] Size: The size of the dot encodes the quantity or frequency of goods shipped out of the warehouse.
[0204] Width: The width of the line between the dots encodes the support of the product combination calculated by the association rule mining algorithm. The greater the support, the thicker the line.
[0205] Position: To visually distinguish related product combinations, related products are placed in adjacent positions through connecting lines, while products that are not directly related are placed farther apart.
[0206] (2) List of related product combinations
[0207] Position: The vertical direction of the list maps different product combinations, and the horizontal direction maps the three attributes and optional operations of "product combination composition, product combination delivery frequency, and product combination support" from left to right. The optional operations include the three function buttons of "Highlight", "Delete" and "Mark".
[0208] 3. Comparative analysis view of cargo location
[0209] (1) Bubble chart
[0210] Shape: Use the product glyphs in the product feature projection view to represent the product. One glyph represents one product. Use bubbles to surround a group of product glyphs to represent a product combination.
[0211] Position: The position encoding of the glyphs is the result of the dimensionality reduction calculation of the MDS algorithm. The actual physical distance between the goods represented by glyphs with similar positions in the warehouse is smaller, and the actual physical distance between the goods represented by glyphs with distant positions in the warehouse is larger.
[0212] Color: Use bubbles of different colors to represent different related product combinations. Bubbles of the same color cover all the products in the combination.
[0213] Area: Use the bubble area to encode the distribution distance of the associated product combination. The smaller the area, the closer the distribution distance of the product combination, and the larger the area, the farther the distribution distance of the product combination.
[0214] (2) Product Portfolio Management List
[0215] Location: Each row in the table represents an associated product combination, showing the detailed information of the combination. The four columns from left to right are: GoodsCombination, the frequency of goods combination delivery CombinationFreq, the support of goods combination Support, and the optional operation Operation. Among them, the CombinationFreq column and the Support column both implement the sorting function, and you can choose to sort in ascending or descending order as needed. In the Operation column, set three function buttons ViewCombination, DeleteCombination, and TagCombination.
[0216] Step 3: Visualization layout and implementation
[0217] Read the e-commerce warehouse data of the specified year and month, complete the layout and implementation of the scatter plot, product glyph map and control panel in the product feature projection view; complete the layout and implementation of the node link diagram, control panel and related product combination list in the related product analysis view; complete the layout and implementation of the bubble chart and product combination management list in the location comparison analysis view, as follows:
[0218] 1. Visual layout and implementation of product feature projection view
[0219] The overall layout of the product feature projection view is as follows Figure 2 In the upper right corner are the lasso control buttons, cluster counter, and time selector. In the upper left corner is the product glyph chart for checking the individual characteristics of the products. In the middle is a scatter plot for an overview of the characteristics of the product group.
[0220] (1) Scatter plot visualization layout and implementation
[0221] The product characteristics are reduced using the t-SNE algorithm, and the K-Means clustering technology is used to realize the group division of the product data. This method divides the n observations in the data set into k clusters, and each observation belongs to the cluster corresponding to the nearest mean (i.e., cluster center). The specific steps are as follows:
[0222] a. Select k data points as initial cluster centers.
[0223] b. For each point in the data set, calculate its distance to the center of each cluster and then divide it into the nearest cluster.
[0224] c. For each cluster, calculate the average coordinates of all points in it, and then set the average as the new cluster center.
[0225] d. Repeat steps b to c until the location of the cluster center no longer changes significantly or the preset maximum number of iterations is reached.
[0226] e. Convert the coordinates of all points in each cluster and the cluster center coordinates into relative coordinates for storage.
[0227] When drawing scattered points and cluster centers, it is necessary to calculate the real coordinates in the view based on their relative coordinates. Taking the cluster center as an example, its real coordinates (clusterCenter xi ,clusterCenter yi ) is calculated as follows:
[0228]
[0229] Where ScatterSvgW and ScatterSvgH are the width and height of the view canvas; clusterCenterRelative xi 、clusterCenterRelative yi is the relative coordinate of the cluster center; scaleFactor is the scaling factor of the image canvas.
[0230] (2) Visualization layout and implementation of product glyphs
[0231] like Figure 3 As shown, the product glyph chart consists of an outer donut chart and an inner candy chart.
[0232] The outer donut chart consists of two equal-sized circles on the left and right. The left circle encodes the number of goods shipped out of the warehouse, and the right circle encodes the frequency of goods shipped out of the warehouse. Taking 12 o'clock as the starting position, the length of the blue ring filled counterclockwise outboundSum represents the specific value of the number of goods shipped out of the warehouse, and the length of the pink ring filled clockwise outboundFreq represents the specific value of the frequency of goods shipped out of the warehouse. The gray ring on the left circle represents the lower quartile outboundSum of the number of goods shipped out of the warehouse in that month Q1 and upper quartile outboundSum Q2 The black dotted line represents the mean number of goods shipped out of the warehouse in that month; the gray ring on the right represents the lower quartile outboundFreq of the goods shipped out of the warehouse in that month Q1 and the upper quartile outboundFreq Q2 , black dashed line outboundSum mean The black dotted line outboundFreq represents the average number of goods shipped out of the warehouse in that month. mean Indicates the average frequency of goods leaving the warehouse in the current month.
[0233] The inner candy chart consists of two sectors and a circle in the middle. The upper sector represents the proportion of the number of goods shipped out of the warehouse relative to the total number of goods shipped out of the warehouse in the month. The area of the blue texture is outboundSum ratio Indicates the proportion of outbound quantity; the lower sector indicates the proportion of the outbound frequency of goods relative to the total outbound frequency of goods in the month, and the area of the pink texture is outboundFreq ratio Indicates the proportion of outbound delivery frequency. The color coding of the upper and lower sectors is consistent with the outer circular chart. The middle circle uses text labels: A, B, C to indicate the classification of goods.
[0234] 2. Visual layout and implementation of related product analysis view
[0235] The overall layout of the related product analysis view is as follows: Figure 4 As shown in the figure, the middle is the node link diagram and the top is the control panel. The control panel includes a support selection slider, a slider for selecting the number of associated product combinations, a clock button, and a statistics button.
[0236] The following focuses on the visual layout and implementation of the node link graph:
[0237] (1) Conversion of related items
[0238] To draw a node-link graph, the association items are converted into the data format containing nodes and links required by D3.js. When the length of the association item is equal to 2, a link containing the source node (source) and the target node (target) can be directly generated from the two elements in the association item; when the length of the association item is greater than 2, n closed-loop links containing source and target need to be generated from n elements in the association item.
[0239] (2) Node coordinate mapping
[0240] Use d3.forceSimulation to build a force-directed layout, using physical mechanics simulation to determine the position of nodes and links. The various parameters of the force-directed simulation are adjusted to affect the presentation of the view: set repulsion to disperse the nodes and increase the readability of the view; set gravity to bring the connected nodes closer to each other to reflect the structure of the view; set the center force to keep the nodes and links in the predetermined position in the view. The interaction of various forces causes the node positions to move continuously. After multiple rounds of iterations, when the force-directed layout reaches a stable state, the position information of all nodes will be returned in the form of an array. Then, use the svg element <circle>Create a node, the size of the node is determined by the product attributes; use <line>Draw a link. The link width is determined by the support. Define the drag behavior function and bind it to <circle>element to achieve real-time update of node position.
[0241] 3. Visual layout and implementation of the comparison analysis view of the storage location
[0242] The overall layout of the cargo location comparison analysis view is as follows Figure 5 As shown, there is a bubble chart in the middle and a function button in the upper right corner. The specific steps for visual layout and implementation of the bubble chart are as follows:
[0243] (1) Bubble chart drawing
[0244] The bubble outline is drawn with the help of the bubblesets visualization library. The process depends on two important parameters: related nodes (Points) and irrelevant nodes (otherPoints). Among them, the Points parameter represents the node position information that the bubble outline needs to include, and the otherPoints parameter represents the node position information that the bubble needs to avoid. In the process of drawing the outline, the goal is to minimize the distance between the outlines of related nodes while ensuring that the outline does not intersect with irrelevant nodes. If an intersection occurs, the outline needs to be adjusted appropriately to avoid irrelevant nodes.
[0245] (2) Collision avoidance detection
[0246] When multiple goods are close to each other, the goods glyphs drawn based on the MDS algorithm may cause mutual occlusion. Therefore, the present invention implements a collision detection and avoidance algorithm for readjusting the glyph positions, preventing element overlap and reducing visual confusion.
[0247] The specific description of the algorithm is as follows:
[0248] a. Initialize the node position. The initialization position of the i-th node is the relative coordinate after dimensionality reduction (x i ,y i ), the node radius is r i .
[0249] b. Calculate the distance between nodes. Calculate the distance between each pair of nodes. When the number of nodes is , the total number of nodes to be calculated is For distance. The distance calculation formula is:
[0250]
[0251] Where: (x i ,y i )、(x j ,y j ) are the coordinates of the two nodes respectively.
[0252] c. Detect collision. For each pair of nodes, check whether the distance between the nodes is less than the sum of the radii. If so, a collision has occurred.
[0253] d. Handle collision. If two nodes collide, the node positions need to be moved until the distance between the nodes is equal to the sum of the radii. The total distance between the collision node and the node being moved is, and the direction of movement is along the direction of the line connecting the two nodes.
[0254] e. Repeat the above steps for 500 times until the node position no longer changes significantly.
[0255] (3) Product combination sorting
[0256] When calculating the order of product combinations, the distance between each combination is defined as the sum of the distances between all pairs of products. This distance is calculated using the Manhattan formula. 1 ,x 2 ) and (y 1 ,y 2 ) is calculated as follows:
[0257] d=|x 1 -x 2 |+|y 1 -y 2 |
[0258] Since there are dimensional differences in the combined outbound frequency, combined cargo location distance, combined outbound frequency mean of each product, and outbound quantity mean, this paper adopts maximum-minimum normalization to map all data into the interval. The calculation formula is as follows:
[0259]
[0260] Where: x min , x max represents the minimum and maximum eigenvalues; x represents the original eigenvalue; x' represents the normalized eigenvalue.
[0261] Finally, the normalized data are weighted and summed using their respective weight factors. The calculation formula is as follows:
[0262]
[0263] Where: S a , S b , S c , S d They represent the normalized values of the combined outbound frequency, the combined cargo location distance, the average outbound frequency of each item in the combination, and the average outbound quantity; W a , W b , W c , W d They respectively represent the weights corresponding to the combined outbound frequency, the combined cargo location distance, the average outbound frequency of each item in the combination, and the average outbound quantity.
[0264] Step 4: Interaction Design
[0265] In the product feature projection view, after clicking any dot, the upper left corner will display the product glyph represented by the dot. Circle the dots in a specific range to update the product combination analysis view. In the associated product analysis view, click any product and a table will pop up on the view to display all the associated combination information of the product in detail. At the same time, the location comparison analysis view will be updated. In the location comparison analysis view, click on the product combination with abnormal location distribution to add it to the management list. Click the button in the upper right corner of the view to display the product combination management list. In addition, when you hover the mouse over the elements in the three views, there will be an information float to display detailed information. You can zoom or pan the view by scrolling the mouse wheel and dragging.
[0266] 1. Interactive design of product feature projection view:
[0267] The product feature projection view supports five interactive operations: dragging, zooming, clicking, hovering, and lassoing.
[0268] Drag and zoom: The projection scatter plot provides zoom and pan interactions. Users can use the mouse wheel to zoom in and out to quickly switch the observation granularity; they can also drag the mouse to pan the view to view the local information of the scatter plot in detail.
[0269] Click: In the projection scatter plot, all goods are represented as dots of the same size. Users can click on any dot with the mouse to view the glyph of the product represented by the dot.
[0270] Hover: In the product glyph, users can hover the mouse to view the specific properties of the product, such as Figure 6 As shown in (a), the selected gray band will be highlighted, and a prompt box will pop up to display the lower quartile and upper quartile of the shipment quantity of the goods represented by the band.
[0271] Lasso: You can further analyze the relationship between products by selecting dots within a specific range, and the related product analysis view will be updated accordingly. Figure 6 As shown in (b), users can set any range of lasso selection as needed. The starting position of the lasso and the lasso selection range will be drawn in the view, and the dots not selected by the lasso will be darkened for distinction.
[0272] 2. Interactive design of related product analysis view
[0273] The related product analysis view supports five interactive operations: click, drag, zoom, hover, and filter.
[0274] Click: Click the button group with the mouse to switch the coding information of the dot size. Click the "Clock" button on the left, and the dot size reflects the frequency of goods leaving the warehouse; click the "Statistics" button on the right, and the dot size reflects the number of goods leaving the warehouse; click any selectedGoods, and the connection lines of all nodes associated with the node are highlighted, and the transparency of other nodes and connections is reduced, such as Figure 7 At the same time, a table pops up in the view to display all the associated combination information of the product, such as Figure 8 As shown, click the Close button in the upper right to remove the table from view.
[0275] Drag and zoom: Use the mouse wheel to zoom in or out the view, making it easier to view the overall structure and local details of the associated product combination; click the mouse to drag the view to move the node of interest to the center of the canvas for observation, and the remaining nodes will automatically update their positions and rearrange themselves.
[0276] Suspension: Figure 8 As shown in the figure, hover the mouse over any dot, and a prompt box will appear to display the product code and the number or frequency of shipments; hover over any line to display the support information of the product combination. When the mouse leaves, the prompt box is removed.
[0277] Filtering: By dragging the "Support" selection slider and the "Number of combinations" selection slider, you can dynamically adjust the parameter values of the mining algorithm to update the algorithm calculation results and then update the drawing data of the node-link graph.
[0278] 3. Interactive design of cargo position comparison analysis view
[0279] The shelf location comparison analysis view supports four interactive operations: floating, dragging, zooming, and clicking.
[0280] Hover: If the user wants to know the specific product information represented by a certain glyph, the user can hover the mouse over the glyph, and a prompt box will pop up above the view to display the detailed information of the product, including: product code, product location, outbound frequency, outbound quantity, outbound frequency ratio and outbound quantity ratio, such as Fig. 9 As shown. When the mouse is moved out, the prompt box disappears automatically. When there are many bubbles in the view, you can hover the mouse over a specific bubble to highlight it, and other bubbles will become dim to highlight the contrast.
[0281] Drag and zoom: You can zoom in or out the view by using the mouse wheel, and you can drag the view by clicking the mouse. By zooming out, you can get an overview of the overall location distribution of all related product combinations; by panning and zooming in, you can view the product glyph information and location comparison of a specific combination in detail.
[0282] Click: Click on a bubble to add the product combination it represents to the management list. Click the function button in the upper right corner of the view to view the detailed information of the management list, such as Fig.10 shown.
[0283] The technical solution provided by the present invention can conduct a comprehensive analysis of the characteristics of outbound goods in terms of individual and group dimensions, and conduct a joint analysis of the correlation between groups of goods and the spatial distribution of goods. It has high interactivity, can meet the warehouse manager's proactive analysis, and the displayed results are highly intuitive and understandable. Furthermore, it can provide good data support for optimizing goods management decisions and ensure that e-commerce warehouses have high overall operational efficiency.
[0284] Specifically, the present invention obtains the product-related data of the e-commerce warehouse, performs data cleaning, data preprocessing, data dimension reduction, and related product mining, and then stores the effective information in the database; designs the product feature projection view, related product analysis view, and cargo location comparison analysis view, and uses visual channels such as position, shape, and color for visual mapping. Combined with the interaction within the view and the linkage between views, it can assist warehouse managers in viewing the multi-dimensional attributes of goods, checking the cargo location distribution of single goods and related product combinations, analyzing abnormal product combinations, and providing support for warehouse managers in making production decisions;
[0285] Furthermore, on the basis of achieving the above-mentioned beneficial effects, each preferred scheme also achieves the following beneficial effects: deleting outbound order data or outbound product data with empty picking batch codes can improve the accuracy of data analysis; calculating product location data can improve the convenience of visualization; data dimensionality reduction is conducive to the mapping of projection scatter plots and bubble charts.< / circle> < / line> < / circle> < / circle> < / line> < / circle>
Claims
1. A visualization method for analyzing outbound goods from e-commerce warehouses, characterized by: The following steps are involved: Step S1: Obtain outbound order data and outbound product data from the warehouse management system of the warehousing service provider, and store them in the database after data processing; The outbound order data includes the picking batch code, order code, product code, and picking time; the outbound product data includes the order code, product code, product picking quantity, and the location code used to mark the picking location of the product; Step S2: Perform visual mapping based on the outbound order data and outbound product data; Step S3: Perform visual layout based on the visual mapping result.
2. A visualization method for analyzing outbound goods from e-commerce warehouses according to claim 1, characterized in that: The data processing includes: (1) Delete the outbound order data or outbound product data with empty picking batch code in batch picking; (2) Taking the month as the picking granularity, calculate the product characteristic data, including: outbound frequency, outbound quantity, outbound frequency ratio, outbound quantity ratio, number of storage locations, and product type; (3) Extract the product code and cargo location code, calculate the product location data based on the cargo location placement method of ground square picking; (4) Dimensionality reduction of product feature data is performed using the t-SNE algorithm, specifically: Use one-hot encoding to create a binary feature for each categorical feature, so that when a product belongs to the corresponding category, the value of the binary feature is 1, otherwise it is 0; Use linear normalization to scale the existing product feature data, apply the t-SNE algorithm, set hyperparameters, and achieve data dimensionality reduction; Normalization is used to convert the absolute coordinates obtained through data dimensionality reduction into first relative coordinates, and the coordinate pairs are stored in the feature data of the picked goods. (5) Reduce the dimension of the goods location data using the MDS algorithm, specifically: Store the picking location data of the goods in the form of a dictionary; the key is the goods code, and the value is the corresponding area, row and column; Calculate the distance between each pair of items using Manhattan distance; Reduce the dimension of cargo location data through MDS algorithm; The minimum-maximum normalization method is used to convert the absolute coordinates obtained by dimensionality reduction of the MDS algorithm into the second relative coordinates, and then store them in the goods location data in the form of coordinate pairs. (6) Use the Eclat algorithm to correlate outbound order data, specifically: 1) Data format conversion From the outbound order data, we select the order code, picking batch code, and items included in the order to form a new data set D; Create an empty mapping table M, where the key is the picking batch code and the value is an empty product set; Generate a "batch-item" data set based on the mapping table M. Each record represents a picking batch, and the items included in each picking batch are an itemset. 2) Use the Eclat algorithm to find frequent product combinations that meet the minimum support 2.1) Create an empty frequent item set list F and an empty candidate frequent item set list C; traverse the "batch-item" data set and add each item that appears to the candidate frequent item set list C; 2.2) Traverse and calculate the support of each item set in the candidate frequent item set list C in the "batch-product" dataset; 2.3) Add the itemsets whose support meets the minimum support requirement to the frequent itemset list F; 2.4) In the frequent itemset list F, for any two itemsets, if the first k-1 elements they share are the same, they are combined into a new itemset and added to the list C; Repeat steps 2.2) to 2.4) until the candidate frequent item set list C is empty or no new candidate frequent item set can be synthesized, and then output the frequent item set list F; Among them, support refers to the proportion of the delivery frequency of the associated product combination; 3) Integrate the frequent item set list to obtain the associated product data, including the product combination list, the delivery frequency of the product combination, the delivery frequency ratio of the product combination, the delivery frequency list and the delivery quantity list of each product in the product combination.
3. The visualization method for analyzing outbound goods from e-commerce warehouses according to claim 2, characterized in that: The visual mapping includes: (1) Constructing product feature projection views at monthly granularity 1) Construct a projection scatter plot: All shipped goods in a single month are represented as dots, where one dot represents one item. Encoding the first relative coordinate by the coordinate of the dot; Clustering results of classification features are color-coded and cluster centers are represented by preset plane graphics; 2) Build product glyphs: The outer donut chart consists of two semicircles of equal size. The left semicircle encodes the quantity of goods shipped out, and the right semicircle encodes the frequency of goods shipped out. Starting at 12 o'clock, the length of the band in the left semicircle maps the specific value of the outbound quantity, and the length of the band in the left semicircle maps the specific value of the outbound frequency. Different colors are used to map the outbound quantity, outbound frequency, the lower and upper quartiles of the outbound frequency, and the lower and upper quartiles of the outbound quantity. The inner candy chart consists of two sectors, upper and lower, and a circle in the middle. The ratio of the central angle of the upper sector to the sector segment represents the proportion of the shipment quantity to the total shipment quantity for the month, while the ratio of the central angle of the lower sector segment to the sector segment represents the proportion of the shipment frequency to the total shipment frequency for the month. The circle uses text labels to represent product categories. (2) Constructing related product analysis views based on monthly granularity 1) Build a node-link graph: The connection between the items is represented by lines between the dots; In the item feature projection view, colored dots are used to indicate the item selectedGoods and its related items otherGoods selected through the lasso interaction. The related items otherGoods are calculated by the data mining algorithm. Encode the quantity or frequency of shipments by the size of the dots; The width of the line encodes the support of the associated product combination; Based on the width of the connection line, the dots corresponding to the related product combinations with high support are placed in adjacent positions; 2) Build a list of related product combinations: Use different rows to map different related product combinations; Use different columns to map the product combination composition, product combination delivery frequency, product combination support, and optional operations. Optional operations include highlighting, deleting, and marking. (3) Construct a comparative analysis view of cargo locations based on monthly granularity 1) Construct a bubble chart: Use product glyphs to represent products, and one product glyph represents one product; use bubbles to surround a group of product glyphs to represent a related product combination; The second relative coordinate is encoded by the position of the product glyph; Different related product combinations are color-coded, and bubbles of the same color contain all products in the corresponding related product combination; The distribution distance of the cargo locations of the associated product combinations is coded by the bubble area; 2) Build a product portfolio management list: Use different rows to represent different related product combinations; Different columns are used to represent the composition of the goods combination GoodsCombination, the frequency of goods combination delivery CombinationFreq, the support of the goods combination Support and the optional operation Operation; Among them, the product combination delivery frequency CombinationFreq and product combination support Support have sorting functions.
4. A visualization method for analyzing outbound goods from e-commerce warehouses according to claim 3, characterized in that: The visual layout includes: (1) Visual layout of product feature projection view The upper right corner of the product feature projection view is set to the lasso control button, cluster counter and time selector, the upper left corner is set to the product glyph chart, and the middle is set to the projection scatter plot; 1) Visual layout of projection scatter plot: The product feature data after t-SNE dimension reduction is divided into groups through K-Means clustering, specifically: 1.1) Select k points as the initial cluster centers; 1.2) Calculate the distance from each point to the center of each initial cluster and divide the clusters according to the distance; 1.3) For each cluster, calculate the average of the coordinates of all origins and set the coordinate corresponding to the average as the new cluster center; 1.4) Repeat steps 1.2) to 1.3) until the distance between the cluster centers before and after the update is no greater than 0.00001; 1.5) Convert the coordinates of all points in each cluster and the cluster center coordinates into relative coordinates and store them; calculate the actual position of the dot in the product feature projection view, clusterCenter xi =ScatterSvgW×clusterCenterRelative xi ×scaleFactor; clusterCenter yi =ScatterSvgH×clusterCenterRelative yi ×scaleFactor; In the formula, (clusterCenter xi ,clusterCenter yi ) is the real coordinate of the center of the i-th cluster, ScatterSvgW and ScatterSvgH are the width and height of the product feature projection view, clusterCenterRelative xi 、clusterCenterRelative yi is the relative coordinate of the center of the i-th cluster, and scaleFactor is the scaling factor of the product feature projection view; 2) Visual layout of product glyphs: The dotted line in the left semicircle represents the mean of the number of shipments in that month, and the dotted line in the right semicircle represents the mean of the frequency of shipments in that month. (2) Visual layout of related product analysis view In the Related Product Analysis view, there is a node-link diagram in the middle and a control panel at the top. The control panel includes a support selection slider, a slider for selecting the number of related product combinations, a clock button, and a statistics button. For a node-link graph, 1) Associated item conversion: Convert the associated items corresponding to the associated product combination into the data format containing nodes and links required by D3.js; Among them, if the length of the association item is equal to 2, a link containing the source node and the target node is directly generated by the two elements in the association item; if the length of the association item is greater than 2, a closed-loop link containing the source and target is generated by the elements in the association item; 2) Node coordinate mapping: Use d3.forceSimulation to build a force-directed layout, and determine the coordinates of nodes and links through physical mechanics simulation; Specifically, repulsion, attraction, and central force are set, and node coordinates are iterated; Using the svg element <circle>Create a node, use <line>Draw links; define drag behavior functions and connect them to elements <circle> Binding to update node coordinates in real time;< / circle> < / line> < / circle> (3) Visual layout of cargo location comparison analysis view In the cargo location comparison analysis view, there is a bubble chart in the middle and a function button in the upper right corner; For bubble charts, 1) Draw bubble outlines with the help of the bubblesets visualization library; 2) Collision avoidance detection: 2.1) Initialize the node position to the relative coordinate after dimensionality reduction (x i ,y i ), the node radius is r i ; 2.2) Calculate the distance between nodes, In the formula, (x j ,y j )、(x k ,y k ) are the coordinates of node j and node k respectively; 2.3) For each pair of nodes, if the distance between the nodes is less than the sum of the radii, a collision is determined; 2.4) For each pair of colliding nodes, move at least one of the nodes along the connecting line until the distance between the nodes equals the sum of the radii; 2.5) Repeat 2.1) to 2.5) until the displacement of the nodes before and after the update does not exceed 0.00001; 3) Sorting of related product combinations: Define the sum of the distances between all items in the associated product combination as the shelf distance of the associated product combination; The distance d between item X1 and item X2 is: d=|x1-x2|+|y1-y2|; Where (x1, y1) and (x2, y2) are the coordinates of item X1 and item X2 respectively; Use maximum and minimum normalization to map all product feature data into the interval. Where z min 、z max are the minimum and maximum eigenvalues, respectively; z is the original eigenvalue, and z' is the eigenvalue normalized by the maximum and minimum values; Perform weighted summation on the eigenvalues normalized by maximum and minimum, In the formula, Score is the ranking score of the associated product combination, S a 、S b 、S c 、S d are the maximum and minimum normalized outbound frequency, cargo location distance, outbound frequency mean, and outbound quantity mean of the associated product combination, respectively. a 、W b 、W c 、W d They are the weights corresponding to the maximum and minimum normalized outbound frequency, cargo location distance, mean outbound frequency and mean outbound quantity of the associated product combination.
5. The visualization method for analyzing outbound goods from e-commerce warehouses according to claim 4, characterized in that: After step S3, the method further includes: realizing human-computer interaction based on the result of the visual layout.
6. A visualization method for analyzing outbound goods from e-commerce warehouses according to claim 5, characterized in that: The human-computer interaction includes: (1) Interaction of product feature projection views Drag: Drag the mouse to pan the view to view local information of the projected scatter plot; Zoom: In the projection scatter plot, use the mouse wheel to zoom in and out to switch the observation granularity; Click: In the projection scatter plot, click a dot to view the corresponding product glyph chart; Hover: In the product glyph image, hover the mouse to view the specific properties of the product; Lasso: Analyze the relationship between products by circling points within a specific range; (2) Interaction of related product analysis views Drag and drop: Click and drag the associated product analysis view with the mouse; after dragging a node, the coordinates of the remaining nodes will be automatically updated and rearranged; Zoom: Use the mouse wheel to zoom in and out of the related product analysis view; Click: switch the dot size encoding information by clicking the button; Hover: Hover the mouse to view the corresponding product code and shipment quantity, or product code and shipment frequency in a tooltip. Hover the mouse to view the corresponding support level of the connection in a tooltip. Filtering: Dynamically adjust the parameter values of the mining algorithm by dragging the support selection slider and the number of associated product combinations selection slider; (3) Interaction of the cargo location comparison analysis view Drag: Drag the mouse to pan the view and view the comparison between the product glyphs and product locations of a specific related product combination; Zoom: Use the mouse wheel to zoom in and out of the cargo location comparison analysis view; Click: Click on the bubble to add the corresponding associated product combination to the management list; click the function button to view the detailed information of the management list; Hover: Hover the mouse to view the product information corresponding to the product glyph image in the form of a prompt box, including product code, product location, outbound frequency, outbound quantity, outbound frequency ratio and outbound quantity ratio; when hovering, the bubble at the corresponding position is highlighted. Remove the hovering button to remove the highlight and the prompt box disappears.
Citation Information
Cited By
Warehouse goods allocation layout optimization method and system based on big data visual analysis
CN120146768A
Warehouse Location Layout Optimization Method and System Based on Big Data Visual Analysis
CN120146768B
