A business data-oriented big data multidimensional analysis visualization method and system

Abstract

The application discloses a kind of big data multidimensional analysis visualization method and system for business data, by obtaining business data set, business semantic configuration, current visual viewport parameter and interactive context parameter, and multidimensional structured processing is carried out to business data set, so that dimension member, index value, hierarchical relationship and query cost are associatedly expressed in unified analysis space, to provide computable basis for subsequent judgment data expression pressure, determine the dimension to be folded and the dimension to be reserved by visual entropy density, so that visualization processing granularity can dynamically change with pixel carrying capacity, index fluctuation and abnormal contribution, and by reversible dimension folding, residual distribution information, abnormal anchor information and local accurate compensation, the traceability and abnormal explainability of compressed data are maintained, so that multidimensional analysis process can reduce the load of full query and redundant display while maintaining the accurate expression ability of key business anomaly, user focus area and high residual area.

Description

Technical Field

[0001] This invention relates to the field of electronic digital data processing technology, and in particular to a method and system for multidimensional analysis and visualization of big data for business data. Background Technology

[0002] With the continuous development of enterprise digital systems, internet business systems, supply chain management systems, financial transaction systems, marketing management systems, and production operation systems, the scale, number of dimensions, hierarchical complexity, and update frequency of business data are constantly increasing. Business data typically includes multiple dimensions such as time, region, organization, product, customer, channel, order, inventory, contract, transaction, risk, and service, as well as various indicators such as sales revenue, order volume, profit margin, conversion rate, inventory turnover rate, fulfillment time, anomaly rate, and churn rate. This type of data is not only large in scale, but also has complex hierarchical relationships, co-occurrence relationships, business rule relationships, and anomaly propagation relationships among its dimensions.

[0003] Existing business data visualization systems typically employ methods such as data warehousing, online analytical processing (OLAP), multidimensional datasets, pre-aggregated queries, caching acceleration, chart rendering, and interactive drill-down to achieve analysis and display. This approach can meet basic needs in scenarios involving fixed reports, dashboards, and fixed metrics analysis. For example, by pre-establishing dimensional models such as time, region, product, and customer, the system can generate line charts, bar charts, heatmaps, maps, or pivot tables based on user-selected dimensions; by setting up data caching and aggregation tables, redundant queries are reduced; and by setting up drill-down and filtering operations, detailed local data can be viewed.

[0004] However, as business data enters a stage characterized by high dimensionality, high cardinality, strong real-time requirements, and strong interactivity, existing technologies have gradually revealed structural deficiencies. First, existing systems typically use data records or pre-aggregated tables as computational objects. When a user is in a specific viewport, the system still tends to query data volumes far exceeding the current display's capacity, resulting in a large number of query results being visually indistinguishable. Second, existing systems typically generate charts with fixed granularity, neglecting the relationship between the current screen pixel capacity, chart representation, and the user's current area of ​​focus. This easily leads to a large number of dimension members being crammed into the same visible area. Third, to improve speed, existing systems often employ methods such as sampling, truncation, fixed aggregation, or cache hits to reduce computational load. However, these methods can easily mask extreme points, abrupt changes, outliers, and business rule anomalies, causing business personnel to overlook critical risks within seemingly smooth charts. Fourth, existing systems typically interpret user interactions as front-end display behaviors, without converting hovering, selection, drill-down, filtering, and rollback operations into control parameters that affect subsequent query plans, aggregation granularity, and supplementary calculation range. This results in the system being unable to dynamically adjust computing resources based on the user's actual concerns.

[0005] Therefore, the core contradiction of existing technologies is not merely insufficient query speed or a lack of chart types, but rather a deep-seated conflict between the high dimensionality and complexity of the data space, the limited computing resources, the limited display space, and the need to maintain data integrity in the event of business anomalies. Simply adding computing nodes, increasing cache capacity, or adding chart templates can only alleviate local performance issues to a certain extent. It cannot solve the problems of averaged anomaly information when high-cardinality dimensions are compressed and displayed, the mismatch between user-focused areas and computing resource allocation, and the difficulty in balancing query accuracy and response speed. Summary of the Invention

[0006] This invention overcomes the shortcomings of existing technologies and provides a method and system for multidimensional analysis and visualization of big data for business data.

[0007] To achieve the above objectives, the technical solution adopted by this invention is: a big data multidimensional analysis and visualization method for business data, comprising: S1. Obtain the business dataset, business semantic configuration, current visualization viewport parameters, and interaction context parameters; perform multidimensional structured processing on the business dataset to generate candidate analysis data containing dimension members, indicator values, hierarchical relationships, and query cost identifiers; S2. Based on the business semantic configuration, the current visualization viewport parameters, and the candidate analysis data, determine the number of dimension states, indicator fluctuation, anomaly contribution, query cost, and pixel capacity corresponding to the candidate visual unit, calculate the visual entropy density of the candidate visual unit, and determine the dimensions to be folded and the dimensions to be retained based on the comparison result of the visual entropy density and the preset critical conditions. S3. Perform reversible dimension folding on the dimension to be folded to generate a semantic cell set, and establish the original dimension member reverse mapping, residual distribution information and anomaly anchor information for the semantic cell set to obtain the folding analysis results; S4. Based on the folding analysis results, the retained dimensions, and the interaction context parameters, generate multidimensional analysis visualization data, and perform precise compensation calculation on candidate visual units that meet the local compensation calculation conditions according to the residual distribution information and the abnormal anchor point information to obtain corrected visualization data; S5. Output the corrected visualization data, and update the interaction context parameters according to the user's interaction behavior with the corrected visualization data, so as to adjust the folding granularity, calculation range and output priority in the next round of multidimensional analysis visualization process.

[0008] In a preferred embodiment of the present invention, the step of acquiring the business dataset, business semantic configuration, current visualization viewport parameters, and interaction context parameters, and performing multi-dimensional structured processing on the business dataset to generate candidate analysis data containing dimension members, indicator values, hierarchical relationships, and query cost identifiers, includes: Retrieve the business dataset containing business object identifier, time field, space field, organization field, product field, customer field, transaction field, and indicator field; Retrieve the business semantic configuration, which includes dimension definitions, dimension levels, indicator definitions, business rules, and dimension relationships; Get the current visualization viewport parameters, including display range, zoom level, chart type, number of screen pixels, visual unit division method, and current filter conditions; Retrieve interaction context parameters that include historical hover records, selected records, drill-down records, filtered records, and dwell time records; According to the business semantic configuration, the business dataset is mapped by fields, normalized by definitions, bound by hierarchy, and aggregated by indicators to generate candidate analysis data.

[0009] In a preferred embodiment of the present invention, determining the number of dimensional states, metric fluctuation, anomaly contribution, query cost, and pixel capacity corresponding to a candidate visual unit based on the business semantic configuration, the current visualization viewport parameters, and the candidate analysis data includes: Based on the current visualization viewport parameters, the current visualization area is divided into multiple candidate visual units; The number of dimension members covered by the candidate dimension within each candidate visual unit is counted to obtain the number of dimension states. The fluctuation of the indicator is determined based on the dispersion, variation range, and local trend differences of the indicator values ​​within each candidate visual unit. The abnormal contribution amount is determined based on the degree of contribution of extreme points, mutation points, outliers, and business rule anomalies within each candidate visual unit relative to its respective visual area. The query cost is determined based on the number of data partitions associated with the candidate visual unit, the number of detailed records, the complexity of indicator aggregation, and the historical query time. The pixel capacity is determined based on the number of pixels of each candidate visual unit in the current viewport and the graphical representation method.

[0010] In a preferred embodiment of the present invention, the step of calculating the visual entropy density of the candidate visual unit and determining the dimension to be folded and the dimension to be retained based on the comparison result of the visual entropy density and a preset critical condition includes: The visual entropy density of the j-th candidate visual unit for the i-th candidate dimension is calculated using the following formula: ;in, Represents visual entropy density, Indicates the number of states in a dimension. Indicates the fluctuation of the indicator. This indicates an abnormal contribution. This indicates the cost of the query. Indicates pixel capacity. Indicates the minimum pixel capacity. and These represent the weight parameters, and ; When the visual entropy density is greater than or equal to a preset entropy density threshold, and the ratio between the number of dimensional states and the pixel carrying capacity is greater than or equal to a preset state carrying capacity threshold, the corresponding candidate dimension is determined as the dimension to be folded. When the visual entropy density is less than the preset entropy density threshold, or when the candidate dimension is marked as a strongly interpreted dimension, the corresponding candidate dimension is determined as a retained dimension.

[0011] In a preferred embodiment of the present invention, performing reversible dimensional folding on the dimension to be folded to generate a semantic cell set includes: Based on the business semantic configuration, calculate the semantic distance between any two dimension members in the dimension to be collapsed; Based on the candidate analysis data, calculate the index distribution distance between any two dimension members; Based on the semantic distance, the indicator distribution distance, and the query cost distance, determine the dimension member merging cost; According to the dimensional member merging cost, dimensional members that meet the merging conditions are folded into the same semantic cell; For each semantic cell, generate a cell identifier, member list, aggregation metric, dominant dimension label, and display granularity label.

[0012] In a preferred embodiment of the present invention, the dimension member merging cost is determined according to the following formula: ;in, This represents the cost of merging dimension members between dimension member a and dimension member b. Indicates semantic distance. Indicates the distance of indicator distribution. Indicates the distance of the query cost. This indicates the amount of reward for hierarchical consistency. and These represent the weight parameters; When the merging cost of the dimension members is less than or equal to the preset merging cost threshold, and dimension members a and b do not contain any abnormal anchors that are forcibly retained, dimension members a and b are assigned to the same semantic cell.

[0013] In a preferred embodiment of the present invention, establishing the original dimensional member reverse mapping, residual distribution information, and anomaly anchor information for the semantic cell set includes: Record the correspondence between each semantic cell and the original dimensional member that has been folded, forming a reverse mapping of the original dimensional member; Based on the difference between the real index value and the aggregate index value of each original dimension member in each semantic cell, residual distribution information is generated. Identify the original dimension members that meet the anomaly preservation conditions, and write the original dimension members as anomaly anchor points into the anomaly anchor point information; The anomaly preservation conditions include at least one of the following: extreme value preservation conditions, mutation preservation conditions, outlier preservation conditions, and business rule anomaly preservation conditions.

[0014] In a preferred embodiment of the present invention, the step of performing precise compensation calculation on candidate visual units that meet the local compensation calculation conditions based on the residual distribution information and the abnormal anchor point information includes: Based on the residual distribution information, calculate the residual risk value for each candidate visual unit; Based on the abnormal anchor point information, calculate the abnormal retention risk value for each candidate visual unit; Based on the interaction context parameters, calculate the interaction attention value for each candidate visual unit; When at least one of the residual risk value, the anomaly retention risk value, and the interactive attention value meets the corresponding threshold condition, the corresponding candidate visual unit is determined as a visual unit to be recalculated. The original dimension members and index values ​​associated with the visual unit to be supplemented are precisely aggregated to generate a partial supplementation result, and the partial supplementation result is used to correct the multidimensional analysis visualization data.

[0015] In a preferred embodiment of the present invention, updating the interaction context parameters based on the user's interaction behavior with the corrected visualization data to adjust the folding granularity, calculation range, and output priority in the next round of multidimensional analysis visualization includes: Capture at least one of the following user interactions performed on the corrected visualization data: hover, click, box selection, zoom, filter, drill-down, and back; An interactive heatfield is generated based on the visual unit position, duration, operation frequency, and dimensional path corresponding to the interactive behavior. Based on the interactive thermal field, the priority of supplementary calculation for the attention visual unit is increased, the display granularity of the unattended visual unit is reduced, and the folding granularity associated with the attention dimension path is updated. In the next round of multidimensional analysis visualization, based on the updated interaction context parameters, the visual units of interest, outlier anchors, and visual units with high residuals will be output first.

[0016] To achieve the above objectives, the second technical solution adopted by this invention is: a big data multidimensional analysis and visualization system for business data, comprising: The data processing module is used to acquire business datasets, business semantic configurations, current visualization viewport parameters, and interaction context parameters, and to perform multi-dimensional structured processing on the business datasets to generate candidate analysis data containing dimension members, indicator values, hierarchical relationships, and query cost identifiers. The entropy density determination module is used to determine the number of dimensional states, index fluctuation, anomaly contribution, query cost, and pixel capacity corresponding to the candidate visual unit based on the business semantic configuration, the current visualization viewport parameters, and the candidate analysis data; calculate the visual entropy density of the candidate visual unit; and determine the dimensions to be collapsed and the dimensions to be retained based on the comparison result of the visual entropy density with the preset critical conditions. The folding processing module is used to perform reversible dimension folding on the dimension to be folded, generate a semantic cell set, and establish the original dimension member reverse mapping, residual distribution information and anomaly anchor information for the semantic cell set to obtain folding analysis results. The supplementary calculation processing module is used to generate multidimensional analysis visualization data based on the folding analysis results, the retained dimensions and the interaction context parameters, and to perform precise supplementary calculation on candidate visual units that meet the local supplementary calculation conditions according to the residual distribution information and the abnormal anchor point information, so as to obtain corrected visualization data. The interactive feedback module is used to output the corrected visualization data and update the interactive context parameters based on the user's interactive behavior on the corrected visualization data, so as to adjust the folding granularity, calculation range and output priority in the next round of multidimensional analysis visualization process.

[0017] This invention addresses the shortcomings of the prior art and has the following beneficial effects: This invention provides a big data multidimensional analysis and visualization method for business data. By acquiring business datasets, business semantic configurations, current visualization viewport parameters, and interaction context parameters, and performing multidimensional structuring on the business datasets, dimension members, indicator values, hierarchical relationships, and query costs can be expressed in a unified analysis space. This provides a computable basis for subsequently assessing data expression pressure. The visual entropy density determines the dimensions to be folded and the dimensions to be retained, allowing the visualization processing granularity to dynamically change with pixel capacity, indicator fluctuation, and anomaly contribution. Through reversible dimension folding, residual distribution information, anomaly anchor information, and local precise compensation, the traceability and anomaly interpretability of the compressed data are maintained. This enables the multidimensional analysis process to reduce the load of full queries and redundant displays while maintaining the accurate expression of key business anomalies, user-focused areas, and high residual areas.

[0018] In this invention, by performing dimension member merging based on business semantic relationships, indicator distribution differences, and query cost differences on the dimensions to be folded, high cardinality dimensions can be transformed into semantic cells with business consistency and indicator similarity. This reduces the computational pressure and display congestion caused by the direct participation of original dimension members in the visualization output. By synchronously establishing the original dimension member reverse mapping, residual distribution information, and anomaly anchor point information for the semantic cells, the folded aggregated expression still retains the original data backtracking path, error assessment basis, and anomaly retention basis. This ensures that dimension compression no longer comes at the cost of irrecoverable information loss, but rather maintains drill-down calculation, anomaly location, and result interpretation capabilities while reducing the display granularity.

[0019] In this invention, by capturing user interactions such as hovering, clicking, box selection, filtering, drilling, and dwelling on the corrected visualization data, and generating an interactive heatfield based on the location of the visual unit, operation frequency, duration, and dimensional path, the user's analytical intent can be converted into control parameters that can participate in subsequent calculation scheduling. By using the interactive heatfield to adjust the folding granularity, calculation range, and output priority in the next round of multidimensional analysis visualization, the query plan can prioritize covering the user's areas of interest, high residual areas, and abnormal anchor point areas, while reducing the processing accuracy requirements of non-interested areas. This transforms the visualization system from a static, responsive query to an interactive, adaptive analysis process, improving the accuracy of resource allocation and the efficiency of key result output in continuous drill-down scenarios. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Figure 1 A flowchart illustrating a big data multidimensional analysis and visualization method for business data provided in Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of a process for triggering reversible dimensional folding based on visual entropy density, provided in Embodiment 2 of the present invention. Figure 3 This is a schematic diagram illustrating the relationship between semantic cells, residual distribution information, and anomaly anchor information provided in Embodiment 3 of the present invention. Figure 4 This is a schematic diagram of the structure of a big data multidimensional analysis and visualization system for business data provided in Embodiment 4 of the present invention; The diagram shows: 410, Data Processing Module; 420, Entropy Density Determination Module; 430, Folding Processing Module; 440, Compensation Processing Module; 450, Interactive Feedback Module. Detailed Implementation

[0021] 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.

[0022] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention is also practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0023] The business data described in this invention refers to the full lifecycle operation and maintenance business data of robot equipment, as well as the electronic component supply chain business data associated with the full lifecycle operation and maintenance of the robot equipment. The full lifecycle operation and maintenance business data of robot equipment includes data generated during the sales and delivery, installation and commissioning, operation and maintenance, fault diagnosis, repair, component replacement, after-sales follow-up, and maintenance evaluation processes of industrial robots, service robots, and intelligent robots. Specifically, it may include robot equipment identification, equipment model, manufacturing batch, installation area, installation environment, customer type, runtime, operating status, fault codes, alarm records, repair work orders, repair actions, repair personnel, replacement parts, repair time, recurring faults, maintenance cycle, and service evaluation data. The electronic component supply chain business data includes power electronic components, model specifications, suppliers, batches, inventory status, purchase records, substitute material relationships, quality inspection records, inbound and outbound records, and fault association records related to robot equipment repair and component replacement. The aforementioned business data exhibits high-dimensional correlations among equipment models, fault codes, installation environments, maintenance actions, replacement parts, and supply batches. Furthermore, critical faults, batch anomalies, and recurring maintenance events typically exhibit low-frequency but high-impact characteristics, and ordinary aggregation or sampling visualization can easily mask these anomalies. Therefore, this invention determines the dimension to be folded through visual entropy density, preserves the reverse mapping between equipment, faults, work orders, and parts through reversible dimensional folding, and maintains the interpretability of critical fault events through anomaly anchors and local supplementary calculations.

[0024] For robot equipment lifecycle maintenance data, existing multidimensional analysis and visualization methods typically employ fixed-dimensional aggregation, sampling display, or full drill-down when processing high-dimensional data such as equipment models, fault codes, maintenance work orders, replacement parts, supply batches, and installation environments. Fixed-dimensional aggregation tends to average out low-frequency but high-impact fault codes, batch quality anomalies, and recurring maintenance events; sampling display may miss key fault objects; and full drill-down can lead to excessive query load and visualization congestion under high cardinality of equipment and component dimensions. Therefore, the technical problem this invention aims to solve is how to reduce the load of full query and redundant display while maintaining the traceable relationship between equipment faults, maintenance work orders, replacement parts, and supply batches, and ensuring that key anomaly events are preserved and accurately calculated during the collapse display process, under the conditions of high-dimensionality, high cardinality, and low-frequency distribution of robot equipment maintenance data.

[0025] Application Overview: Existing big data multidimensional analysis and visualization technologies suffer from a long-standing contradiction at the fundamental level: the dimensional complexity of business data continues to increase, while display space, interaction response time, and user cognitive capacity all have limits. Traditional technical approaches mainly fall into three categories. The first category involves fixed reports or dashboards, offering stability and a clear query path, but its fixed dimension combinations make it unsuitable for adapting to temporary changes in analysis paths by business users. The second category is multidimensional drill-down based on online analytical processing (OLAP), supporting slicing, dicing, roll-up, and drill-down, but prone to query explosion and rendering congestion in high-cardinality dimensional combinations. The third category involves sampling, caching, or pre-aggregation, reducing query latency, but sampling may miss anomalies, fixed pre-aggregation may mask local mutations, and cache hits rely on historical access patterns, making it difficult to respond to new interaction paths.

[0026] None of the three aforementioned technical approaches redefine the computational object based on the pixel capacity of the display viewport. In other words, existing systems typically determine query and aggregation results from the data side first, and then send the results into charts for display. However, in high-dimensional, high-cardinality scenarios, many query results, even if fully returned, cannot be effectively distinguished within the current viewport. When the number of dimensional members significantly exceeds the number of states that a visual unit can represent, continuing to execute a full query not only increases computational costs but also fails to provide a proportional information gain. More importantly, if simple sampling or aggregation is used to improve speed, extreme values, mutations, and business rule anomalies are easily flattened, resulting in a smooth-looking visualization but concealing business risks.

[0027] The technological breakthrough of this invention lies in its unified modeling of the number of dimensional states, index fluctuations, anomaly contributions, query costs, and pixel capacity through visual entropy density calculation, triggering reversible dimensional folding when critical conditions are exceeded. This reversible dimensional folding is not ordinary clustering or static grouping, but rather retains the original dimensional member reverse mapping, residual distribution information, and anomaly anchor point information while generating semantic cells. This allows the system to quickly generate visualization results within the current viewport and perform localized, precise supplementary calculations in user-focused areas, areas with excessively high residuals, and areas with excessively high anomaly contributions. Thus, this application transforms global high-load queries into folded display and localized supplementary calculations under viewport constraints, and converts fixed aggregation displays into a reversible, interpretable, and feedback-enabled dynamic analysis process.

[0028] Example 1: Figure 1 This is a flowchart illustrating a big data multidimensional analysis and visualization method for business data, provided in Embodiment 1 of the present invention. This embodiment is applicable to multidimensional analysis and visualization scenarios of business data, such as enterprise operation analysis, supply chain analysis, financial risk analysis, marketing conversion analysis, production operation analysis, customer behavior analysis, and comprehensive management dashboards. Figure 1 As shown, the method includes: S1. Obtain the business dataset, business semantic configuration, current visualization viewport parameters, and interaction context parameters; perform multidimensional structured processing on the business dataset to generate candidate analysis data containing dimension members, indicator values, hierarchical relationships, and query cost identifiers; S2. Based on business semantic configuration, current visualization viewport parameters and candidate analysis data, determine the number of dimensional states, indicator fluctuation, anomaly contribution, query cost and pixel capacity corresponding to the candidate visual unit, calculate the visual entropy density of the candidate visual unit, and determine the dimensions to be collapsed and the dimensions to be retained based on the comparison results of the visual entropy density and the preset critical conditions. S3. Perform reversible dimension folding on the dimension to be folded to generate a semantic cell set, and establish the original dimension member reverse mapping, residual distribution information and anomaly anchor information for the semantic cell set to obtain the folding analysis results; S4. Based on the folding analysis results, retained dimensions, and interaction context parameters, generate multidimensional analysis visualization data, and perform precise compensation calculation on candidate visual units that meet the local compensation calculation conditions according to residual distribution information and anomaly anchor point information to obtain corrected visualization data. S5. Output corrected visualization data and update the interaction context parameters based on the user's interaction with the corrected visualization data to adjust the folding granularity, calculation range and output priority in the next round of multidimensional analysis visualization process.

[0029] When implementing the above method, the key technical challenges are: the system cannot simply compress high cardinality dimensions into a small number of categories, because ordinary compression will cause a loss of business meaning; the system cannot rely solely on sampling, because sampling may miss low-frequency but high-value outliers; and the system cannot perform precise calculations on all dimension members, because under high-concurrency interaction conditions, full precise calculation will cause response delays. Therefore, this embodiment needs to establish a dynamic balance between limited display capacity, limited computation time, and the need to preserve anomalies. This balance is not achieved through fixed threshold truncation, but rather by identifying the dimensions that are truly causing visual congestion and query load in the current viewport through visual entropy density, then preserving backtracking capabilities through reversible dimension folding, and preserving supplementary calculation capabilities through residual distribution and anomaly anchor points.

[0030] Specifically, in step S1, the business dataset is a collection of data from business systems, transaction systems, customer relationship systems, supply chain systems, financial systems, production systems, risk control systems, or log systems. The business dataset is not limited to a single storage format; it can be relational data, columnar data, detailed transaction data, wide table data, log data, or a dataset that has undergone preliminary cleaning. Business semantic configuration describes the meaning, dimension affiliation, indicator definitions, hierarchical relationships, and business rules of each field in the business data. For example, in a sales analysis scenario, time, region, store, product, customer, and channel are defined as dimensions, while sales revenue, gross profit margin, order volume, and return rate are defined as indicators. In a risk analysis scenario, customer level, account type, transaction channel, transaction region, and risk label are defined as dimensions, while risk score, number of abnormal transactions, and loss amount are defined as indicators. Current visualization viewport parameters describe the current visualization output range and display status, including but not limited to display area size, number of pixels, chart type, zoom level, filter conditions, and the currently displayed combination of dimensions. Interaction context parameters are used to describe the user's actions during the current or historical analysis process, such as a user staying in a certain area for a long time, drilling down a certain product category multiple times, or clicking or selecting a certain anomaly.

[0031] In step S1, acquisition is not simply reading data, but rather obtaining input objects that can participate in subsequent calculations according to the analysis task. Multidimensional structuring of the business dataset refers to transforming the raw business data into a data structure that can be analyzed according to dimensions and indicators. This process includes field identification, field standardization, dimension binding, indicator normalization, hierarchical mapping, missing value handling, initial screening of anomaly labels, and query cost estimation. Generating candidate analysis data refers to forming data objects that can be divided into candidate visual units, have their visual entropy density calculated, and be folded in subsequent steps. Dimension members in candidate analysis data are specific values ​​under a certain dimension, such as a region, a store, a product, a customer group, or a time period; indicator values ​​are the numerical results under the corresponding dimension combinations; hierarchical relationships describe the superior-subordinate relationships between dimension members; and query cost identifiers characterize the relative cost required to read, aggregate, or supplement relevant data in subsequent analysis.

[0032] In one specific implementation, the multidimensional structuring of the business dataset includes the following steps: First, based on the business semantic configuration, the field types in the business dataset are identified, and the fields are divided into dimension fields, indicator fields, time fields, identifier fields, and rule fields. Then, the dimension fields are standardized, for example, mapping region names, organization codes, product codes, or customer classifications from different systems to a unified coding system. Next, indicator fields undergo standardization, for example, converting indicator values ​​recorded from different sources with different currencies, time granularities, or statistical standards into a unified standard. Subsequently, the dimension fields are bound to the dimension hierarchy in the business semantic configuration, for example, binding stores to cities, cities to provinces, products to categories, and categories to business units. Finally, a query cost identifier is generated for each piece or group of business data. The query cost identifier is determined based on the number of partitions containing the data, the number of detailed records, the complexity of indicator aggregation, historical query time, and cache hit status.

[0033] In step S2, a candidate visual unit is a region within the current visualization viewport capable of carrying one or a group of visualization results. A candidate visual unit corresponds to a bar in a histogram, a point or line segment in a line chart, a grid point in a heatmap, a spatial grid in a map, a local area in a scatter plot, or a cell in a pivot table. The number of dimensional states indicates the number of different dimensional members present within the candidate visual unit. The metric volatility indicates the degree of drastic change in metric values ​​within the candidate visual unit. The anomaly contribution indicates the importance of extreme values, mutations, outliers, or business rule anomalies within the candidate visual unit to the final visualization interpretation. The pixel capacity indicates the information capacity that the candidate visual unit can distinguish under the current display conditions.

[0034] Calculating visual entropy density is one of the core steps in this embodiment. Visual entropy density is not a direct replacement of traditional information entropy, but rather a composite control quantity constructed specifically for business data visualization scenarios. This control quantity is used to determine whether the current candidate visual unit has entered a state of visual congestion and whether it is necessary to change the granularity of data representation through folding. In ordinary visualization systems, the system typically displays data according to a preset dimensional granularity without determining whether the current pixel space can accommodate that dimensional granularity. This embodiment incorporates display capabilities into the data processing flow through visual entropy density, enabling the system to pre-identify information overload that the display side cannot effectively represent during the data calculation stage.

[0035] In one specific implementation, visual entropy density is calculated based on the number of dimensional states, metric fluctuation, anomaly contribution, query cost, and pixel capacity. A higher number of dimensional states indicates more dimensional members mixed within the same candidate visual unit, increasing the likelihood of visual overlap and interpretation confusion. Higher metric fluctuation indicates greater differences in metrics between different dimensional members within the candidate visual unit; simple aggregation makes it easier to lose internal differences. Higher anomaly contribution indicates the presence of anomalous objects within the candidate visual unit that significantly impact business decisions. Higher query cost indicates that performing a full query or full recalculation would consume more computational resources. Higher pixel capacity indicates that the current visual unit can carry more information. Therefore, visual entropy density is essentially the business information pressure per unit of display capacity.

[0036] When the visual entropy density exceeds a preset critical condition, the system determines the corresponding candidate dimension as the dimension to be collapsed. The critical condition here is not a simple data volume threshold, but is jointly determined by the current viewport, business metrics, and anomaly contributions. For example, for visual units containing 1000 customers, if these customer metric values ​​are similar and there are no obvious anomalies, they are collapsed into a coarser-grained customer group; if a few high-risk anomaly customers exist among these customers, they cannot be simply collapsed as a whole, but rather the anomaly anchor points need to be retained and independently displayed or recalculated. Therefore, the dimension to be collapsed determination process in this embodiment avoids the anomaly omission problem caused by traditional quantity-based truncation.

[0037] In step S3, reversible dimension folding refers to mapping multiple original dimension members in the dimension to be folded to one or more semantic cells, while preserving the reverse mapping relationship from the semantic cell back to the original dimension members. A semantic cell is not an arbitrary aggregation unit, but rather a visual representation unit formed based on business semantic similarity, indicator distribution similarity, and query cost similarity. For example, in a sales analysis scenario, if multiple low-selling long-tail products belong to the same category, have similar indicator trends, and no outliers, they are folded into the same semantic cell; if one of the products has low sales but a sudden increase in return rate, that product should be retained as an anomaly anchor point and should not be directly merged into a regular long-tail cell.

[0038] Establishing a reverse mapping for original dimension members ensures that the collapsed data still retains drill-down and backtracking capabilities. Traditional aggregation methods typically only retain aggregated values; users need to reconstruct the query to view the source of these values. In this embodiment, the correspondence between cell identifiers and original dimension members is recorded during semantic cell generation, allowing users to quickly locate the relevant original data range during subsequent drill-down, filtering, or recalculation. Residual distribution information records the difference distribution between the actual index values ​​of original dimension members and the aggregated index values ​​of semantic cells. Anomaly anchor information records abnormal objects that should be forcibly retained. By simultaneously establishing these three types of information, the system can achieve synergy between rapid display, reversible querying, and anomaly preservation.

[0039] In step S4, generating multidimensional analysis visualization data based on the folding analysis results, retained dimensions, and interaction context parameters means using the folded semantic cells, the folded retained dimensions, and the user's attention status together to generate the output data. The multidimensional analysis visualization data includes chart data, layer data, pivot table data, prompt information data, drill-down path data, and anomaly annotation data. For candidate visual units that meet the local supplementation conditions, the system no longer relies on the folding results but instead re-aggregates them precisely according to the original dimension members and the actual indicator values ​​to generate the local supplementation results. The corrected visualization data is the visualization data corrected using the local supplementation results.

[0040] The local supplementary calculation conditions are jointly determined by the residual risk value, the anomaly retention risk value, and the interaction attention value. The residual risk value reflects the impact of differences within semantic cells on the accuracy of the visualization results. The anomaly retention risk value reflects the risk that anomaly anchors will be collapsed or occluded. The interaction attention value reflects the user's level of attention to a certain visual unit or a certain dimensional path. For example, if a user hovers over a certain area multiple times and drills down into that area, the interaction attention value of the corresponding visual unit in that area increases, and the system reduces the folding granularity of that area or prioritizes supplementary calculation in the next round of analysis. Thus, this embodiment achieves dynamic matching between computing resources and user-interested areas.

[0041] In step S5, outputting the corrected visualization data means outputting the folded and corrected data as a result that can be displayed, transmitted, stored, or further analyzed. User interactions with the corrected visualization data include hovering, clicking, selecting, zooming, filtering, drilling down, undoing, and marking. Updating the interaction context parameters means converting these interactions into control parameters for the next round of analysis, rather than simply saving them as front-end operation logs. In this way, this embodiment forms a closed loop between data processing, visualization output, user interaction, and the next round of data processing.

[0042] This technical solution identifies the data representation pressure under the current viewport by using visual entropy density, reduces the computational and rendering burden caused by high cardinality by reversible dimensionality folding, mitigates the risk of anomalies being masked by aggregation by using residual distribution information and anomaly anchor point information, and feeds user concerns back to subsequent calculation processes through interactive context parameters. This solution is not simply a patchwork of existing data warehouses, visualization charts, and caching mechanisms, but rather enables viewport capacity, business semantic relationships, anomaly preservation mechanisms, and interactive feedback mechanisms to interact within the same process. This achieves a synergistic improvement in response speed, display interpretability, and anomaly preservation capabilities in high-dimensional business data visualization scenarios.

[0043] Example 2: This example, based on Example 1, further refines the calculation of visual entropy density, determination of the dimension to be folded, reversible dimension folding, residual distribution generation, preservation of abnormal anchor points, and local precise compensation. Compared with Example 1, this example provides more specific calculation methods, parameter meanings, threshold conditions, and logical judgment processes to enable those skilled in the art to reproduce the present invention.

[0044] In this embodiment, the business dataset is first subjected to multidimensional structuring processing. Let the business dataset be: ;in, Let represent the r-th business data record, and n represent the number of business data records. Each business data record is represented as: ;in, to This represents the u-dimensional members corresponding to the r-th business data record. to This represents the v indicator values ​​corresponding to the r-th business data record. This represents the business time field. Dimensions include, but are not limited to, time, space, organization, product, customer, channel, and risk dimensions. Metrics include, but are not limited to, quantity metrics, monetary metrics, percentage metrics, duration metrics, and scoring metrics.

[0045] Business semantic configuration is represented as: Where D represents the dimension set, M represents the indicator set, L represents the dimension hierarchy set, B represents the business rule set, and R represents the dimension relationship set. The dimension hierarchy set L describes the hierarchical relationships between different dimension members, such as store to city, city to province, product to category, and category to business unit. The business rule set B describes the rules for judging business anomalies, such as return rate exceeding a preset percentage, risk score exceeding a preset score, inventory turnover days exceeding a preset number of days, and gross profit margin falling below a preset percentage. The dimension relationship set R describes the strength of the relationship between different dimensions, such as the sales relationship between product and channel, the relationship between customer level and risk level, and the relationship between region and logistics time.

[0046] The current visualization viewport parameters are represented as follows: Where w represents the width of the current visualization area in pixels, h represents the height of the current visualization area in pixels, z represents the zoom level, and g represents the chart type. This indicates the rules for dividing the visual units, where F represents the current filtering condition. The interaction context parameters are represented as follows: Where H represents the set of hovering behaviors, C represents the set of clicking behaviors, K represents the set of box selection behaviors, S represents the set of filtering and drill-down behaviors, and T represents the set of dwell time.

[0047] After the current visualization viewport parameters W are determined, the current visualization area is divided into q candidate visual units: Each candidate visual unit Each visual unit corresponds to a display area and a data coverage area. For bar charts, candidate visual units are bar regions; for line charts, candidate visual units are points or line segment neighborhoods; for heatmaps, candidate visual units are grids; for maps, candidate visual units are spatial tiles or geographic grids; for pivot tables, candidate visual units are row and column intersection cells.

[0048] For candidate visual units and candidate dimensions Determine the number of dimensional states. The number of dimensional states that fall into the candidate visual unit is determined statistically. The number of members in different dimensions of the data is obtained, that is: ; in, This indicates the number of elements in the set. The larger the number of dimensional states, the more dimensional members need to be represented within the candidate visual unit.

[0049] Furthermore, determine the indicator volatility. In one implementation, the metric volatility is determined based on the relative relationship between the standard deviation and the mean of the target metric within the candidate visual unit: ; in, Indicates falling into the candidate visual unit The r-th index value, Indicates candidate visual unit Internal index mean Indicates candidate visual unit The number of records involved in the calculation. This represents a smoothing parameter to prevent the denominator from being zero. Take any value between 0.000001 and 0.001. The greater the fluctuation of the indicator, the greater the internal difference of the candidate visual unit. If it is folded in a coarse-grained manner, it is more likely to produce expression bias.

[0050] Furthermore, determine the abnormal contribution amount. The anomaly contribution is used to represent the importance of an anomaly object in a candidate visual unit to the business interpretation. In one implementation, the anomaly strength of each data object is first calculated: ; in, Indicates the intensity of extreme value anomalies. Indicates the intensity of the mutation anomaly. Indicates the intensity of outlier anomalies. Indicates the abnormal strength of the business rule. and These represent the weight parameters, and The intensity of extreme value anomalies is determined based on the quantile of the indicator value within a local region; the intensity of mutation anomalies is determined based on the rate of change between the current time window and the previous time window; the intensity of outlier anomalies is determined based on the distance between the data object and the local distribution center; and the intensity of business rule anomalies is determined based on the rule hit rate in the business rule set B.

[0051] Candidate visual unit Abnormal contributions are represented as follows: Therefore, when an abnormal object has both high abnormality intensity and a large contribution to local index differences, the abnormal contribution of the candidate visual unit increases.

[0052] Furthermore, determine the query cost. The query cost is determined based on the number of data partitions associated with the candidate visual unit, the number of detailed records, the complexity of metric aggregation, and the historical query time. ; in, This indicates the number of data partitions associated with the candidate visual unit. Indicates the maximum number of reference partitions. This indicates the number of detailed records associated with the candidate visual unit. Indicates the maximum number of detailed records to reference. Indicates the time taken for historical queries. This indicates the maximum reference query time. This indicates a cache hit flag or cache hit probability. and These represent the weight parameters, and .

[0053] Further, determine the pixel capacity. The pixel capacity is related to the number of pixels occupied by candidate viewable units within the current viewport and the chart representation method. For two-dimensional charts, the pixel area of ​​the candidate viewable units is used as the basic capacity; for line or scatter plot representations, it is further modified by combining the point radius, line width, and symbol size. The pixel capacity is expressed as: ;in, Indicates the number of pixels in width of the candidate view unit. Indicates the number of pixels at the height of the candidate visible unit. This indicates the load correction factor corresponding to the chart type. For table-type charts, Take values ​​from 0.8 to 1.2; for heatmaps, Use values ​​between 1.0 and 1.5; for scatter plots, Use values ​​from 0.5 to 1.0; for overlapping map layers, The range is 0.3 to 0.9. This range reflects the difference in the ability of the same pixel area to distinguish different data states under different chart types.

[0054] After obtaining the number of dimensional states, metric fluctuation, anomaly contribution, query cost, and pixel capacity, the visual entropy density is calculated using the following formula: ;in, Represents visual entropy density, This represents the minimum pixel capacity, preventing the calculation results from being unstable due to an excessively low denominator when the candidate visible units are too small. and Configure settings according to the business scenario. In a general business analysis scenario, configure... In risk analysis scenarios, increase the weight of abnormal contribution amounts, for example, by setting... The above parameters do not limit the scope of protection of this invention. Those skilled in the art can adjust them according to the business scenario, provided that the sum of the weights is 1.

[0055] In this embodiment, the preset critical conditions include an entropy density threshold and a state carrying capacity threshold. Specifically, when the following formula is satisfied, the candidate dimension is determined as the dimension to be folded: ;in, This represents the preset entropy density threshold. This indicates the preset state carrying threshold. Determined based on samples of visual congestion in historical analysis tasks, for example, taking the 25th percentile value of the visual entropy density of historical congestion samples; The minimum identifiable area of ​​the visualized symbol is determined. By simultaneously setting entropy density threshold and state carrying capacity threshold, we avoid incorrectly folding low-state dimensions that must be retained in the business simply because of high query costs, and we also avoid ignoring metric fluctuations and abnormal contributions simply because of a large number of dimensions.

[0056] When a candidate dimension is marked as a strongly interpretable dimension by the business semantic configuration, it is determined to be a retained dimension even if it meets some of the folding conditions. A strongly interpretable dimension is a dimension that has direct interpretive significance for the business conclusion in the current business scenario. For example, in regional sales anomaly analysis, the region dimension is set as a strongly interpretable dimension; in customer risk analysis, the customer risk level dimension is set as a strongly interpretable dimension; in production quality analysis, the batch number or production line number is set as a strongly interpretable dimension. By setting strongly interpretable dimensions, the system is prevented from folding critical interpretation paths in pursuit of low load.

[0057] After determining the dimension to be collapsed, reversible dimension collapse is performed on it. The core of reversible dimension collapse is that the collapse process not only generates the aggregation result, but also simultaneously generates the reverse mapping, residual distribution, and outlier anchors, so that subsequent drill-down and recalculation can return to the original dimension members.

[0058] For any two dimension members a and b in the dimension to be collapsed, first calculate the semantic distance. Semantic distance is determined based on dimensional hierarchical distance, business tag similarity, and dimensional association: ; in, Indicates the distance between dimensional levels. Indicates the similarity of business tags. Indicates the similarity of dimensions. and These represent the weight parameters, and Dimensional hierarchical distance is determined by the path length from two dimension members to their nearest common parent node in the dimension hierarchy tree. Business label similarity is determined by the intersection-union ratio of the label sets of two dimension members. Dimensional association similarity is determined by the co-occurrence degree of two dimension members in historical queries or business transactions.

[0059] Next, calculate the distribution distance of the indicators. The indicator distribution distance is used to measure the difference in indicator performance between two dimension members. For business data with time-series indicators, the distance is calculated after normalizing the indicator series corresponding to each dimension member: ; Where K represents the number of time windows or the number of indicator segments. This represents the normalized index value of dimension member a in the k-th window. This represents the normalized index value of dimension member b in the k-th window. The smaller the index distribution distance, the closer the index trends or distributions of the two dimension members are.

[0060] Furthermore, calculate the query cost distance. Query cost distance is used to represent the difference in query path, data partition, or aggregation cost between two dimension members. For example, if two dimension members are located in the same data partition, the same index path, or the same cache area, the query cost distance is low; if two dimension members are distributed in different partitions and require different aggregation paths, the query cost distance is high.

[0061] The cost of merging dimension members is determined based on semantic distance, indicator distribution distance, and query cost distance. ; in, This represents the cost of merging dimension members between dimension member a and dimension member b. This represents the reward amount for hierarchical consistency. If dimension member a and dimension member b belong to the same parent level node and do not violate the constraints in business rule set B, then... A value greater than 0 is used to reduce the cost of merging. If the two items do not belong to the same business level or have mutual exclusion rules, then... Take 0.

[0062] The cost of merging dimension members Less than or equal to the preset merging cost threshold If neither dimension member a nor dimension member b contains any abnormal anchors that are forcibly retained, then dimension member a and dimension member b will be grouped into the same semantic cell. A preset merging cost threshold is set. The threshold is determined based on historical aggregation errors in business data. For example, multiple dimension members are selected from historical data, and the average residual and anomaly recall rate are calculated under different thresholds. The threshold that results in an average residual lower than a preset residual threshold and an anomaly recall rate higher than a preset recall threshold is then determined. .

[0063] Each semantic cell is represented as: ; in, The cell identifier representing the p-th semantic cell. This represents the set of original dimensional members that were collapsed into this semantic cell. This represents the aggregation index of the semantic cell. Indicates the dominant dimension label, This indicates the granularity label to display. Aggregated metrics are determined based on the business metric definition. For example, for sales revenue metrics, summation is used; for proportional metrics such as gross profit margin and conversion rate, a weighted average is used; and for risk score metrics, the maximum value, quantile, or weighted average is used.

[0064] After generating semantic cells, establish a reverse mapping of the original dimensional members: ; in, This indicates that the semantic cell identifier allows traversal back to the original set of dimensional members. This reverse mapping is invoked when the user performs drill-down, selection, or supplementary calculations, making the folding result reversible.

[0065] Furthermore, residual distribution information is generated. For semantic cells... Let the original dimension member 'a' in the original dimension be denoted as 'true index value'. The aggregation index value of semantic cells is The residual is: ; Semantic cell The residual distribution information includes the residual mean, residual variance, largest positive residual, largest negative residual, and residual quantiles. The residual risk value is expressed as: ; in, Represents the standard deviation of the residuals. Represents the maximum absolute residual. This represents the 95th percentile of the absolute value of the residual. and These represent the weight parameters, and When the residual risk value is high, it indicates that there are large differences within the semantic cell. Folding the display may cause interpretation bias, and the priority of supplementary calculation should be increased in subsequent steps.

[0066] Furthermore, abnormal anchor information is generated. An abnormal anchor refers to an original dimension member that should not be completely covered by ordinary aggregation during the collapse process. Abnormal anchors are identified through extreme value preservation conditions, mutation preservation conditions, outlier preservation conditions, and business rule abnormal preservation conditions.

[0067] The condition for preserving extreme values ​​is expressed as: or ; in, The upper quantile threshold represents the distribution of local indicators. This represents the lower quantile threshold of the local index distribution. For example, Take the 95th percentile. Take the 5th percentile.

[0068] The mutation preservation condition is expressed as: ; in, This represents the metric value of dimension member 'a' within the current time window. This represents the metric value of dimension member 'a' in the previous time window. This represents the mutation threshold. The settings are based on the business scenario. For example, the value can be set to 0.3 to 0.5 for daily changes in sales revenue, and 0.1 to 0.3 for changes in risk score.

[0069] The condition for maintaining the group's presence or absence is expressed as: ; in, This represents the median of the index values ​​within a semantic cell. Indicates the absolute deviation of the median. This represents the outlier threshold. Use values ​​between 2.5 and 5.0. The absolute deviation of the median is more stable for extreme values ​​than the variance, making it suitable for scenarios with outlier business points.

[0070] The conditions for maintaining business rule exceptions are determined by the business rule set B. For example, in inventory analysis, if the inventory turnover days exceed the preset number of days and the sales forecast declines, a business rule exception is triggered; in customer analysis, if a high-value customer has not made a repeat purchase for several consecutive periods, a business rule exception is triggered; in financial risk analysis, if the transaction amount, transaction region, and transaction frequency simultaneously meet the risk rules, a business rule exception is triggered. Original dimension members that meet any exception maintenance condition are written into the exception anchor information and retained in the visualization output as independent markers, prompts, or high-priority recalculation objects.

[0071] After generating the folding analysis results, multidimensional analysis visualization data is generated based on the folding analysis results, retained dimensions, and interaction context parameters. This step includes the following processes: First, the retained dimensions are used as the main display path, and semantic cells are used as the display objects of the folded dimensions. Then, basic display values ​​are generated based on the aggregated metrics of each semantic cell. Next, anomaly anchors are overlaid onto the corresponding candidate visual units, and anomaly type, anomaly intensity, and traceability identifier are written into the display data. Subsequently, the error indication level of each candidate visual unit is determined based on the residual distribution information. Finally, the output priority of user-focused areas is increased based on the interaction context parameters.

[0072] For candidate visual units that meet the local compensation criteria, precise compensation is performed. The local compensation criteria are determined based on residual risk values, anomaly retention risk values, and interactive concern values. Let the candidate visual units be... The comprehensive supplementary calculation score is: ; in, This represents the residual risk value. This indicates that the abnormal value is maintained at a risk level. Indicates the interaction attention value. and These represent the weight parameters, and .when At that time, candidate visual units The visual unit is identified as needing supplementary calculation. During supplementary calculation, the set of original dimension members associated with the visual unit is obtained by reverse mapping of the original dimension members, and the corresponding indicators are recalculated according to the business indicator criteria. Therefore, the system only performs accurate calculations on high-risk or high-concern areas, rather than performing redundant calculations on the global data.

[0073] Interactive attention value Calculated based on user interaction behavior: ; in, Indicates candidate visual unit Number of hovering times Indicates the number of clicks. Indicates the number of times the selection box is used. Indicates the number of times the filter or drill-down is performed. Indicates the duration of stay. and These represent the maximum values ​​within the corresponding statistics window. to These represent the weighting parameters. For scenarios that need to highlight proactive analysis behavior, increase the weight of bounding box selection and drill-down; for scenarios that need to capture potential areas of interest, increase the weight of hovering and dwell time.

[0074] In the next round of multidimensional analysis and visualization, the system adjusts the folding granularity, supplementary calculation range, and output priority based on the interaction attention value. For candidate visual units with high interaction attention values, the system lowers the entropy density threshold or the merging cost threshold, making them less likely to be folded with coarse granularity; it also increases the supplementary calculation priority, allowing the user's area of ​​interest to output accurate results earlier. For candidate visual units that have not been noticed for a long time and have low residual risk values, the system increases the folding granularity to reduce computational resource consumption. Through this mechanism, the system no longer treats all visual regions equally, but reallocates resources according to the user's actual analysis path.

[0075] The effect of this embodiment is that when the relationship between the number of dimension members and the pixel capacity exceeds a critical condition, the system does not linearly increase query resources, but changes the computation object, switching from full computation of the original dimension members to a combination of semantic cell computation and local supplementary computation. This change allows the front-end visualization output and interactive response to no longer be completely controlled by the original number of records as the data scale continues to increase, but are mainly affected by the number of semantic cells, the number of abnormal anchor points, and the number of regions to be supplemented within the current viewport. Therefore, this effect differs from the linear optimization brought by conventional caching, indexing, or parallel computing; rather, it involves a switch in processing methods under specific critical conditions.

[0076] Furthermore, this embodiment alleviates the traditional conflict between speed and accuracy. Ordinary sampling improves speed but reduces anomaly recall; ordinary full queries ensure accuracy but reduce response speed. This embodiment outputs low-load results first through reversible dimension folding, then precisely corrects key areas using anomaly anchors and residual compensation, ensuring speed improvement without irrecoverable information loss. A collaborative mechanism is formed between visual entropy density, reversible dimension folding, residual distribution information, anomaly anchor information, and interaction context parameters, jointly determining which data should be folded, which data must be retained, which areas need compensation, and which results should be prioritized for output.

[0077] Example 3: Figure 4 This is a schematic diagram of the structure of a big data multidimensional analysis and visualization system for business data provided in Embodiment 3 of the present invention. Figure 4 As shown, the system includes: a data processing module 410, an entropy density determination module 420, a folding processing module 430, a supplementary calculation processing module 440, and an interactive feedback module 450.

[0078] The data processing module 410 is used to acquire business datasets, business semantic configurations, current visualization viewport parameters, and interaction context parameters, and to perform multi-dimensional structured processing on the business datasets to generate candidate analysis data containing dimension members, indicator values, hierarchical relationships, and query cost identifiers.

[0079] The entropy density determination module 420 is used to determine the number of dimensional states, index fluctuation, anomaly contribution, query cost, and pixel capacity corresponding to the candidate visual unit based on business semantic configuration, current visualization viewport parameters, and candidate analysis data. It calculates the visual entropy density of the candidate visual unit and determines the dimensions to be collapsed and the dimensions to be retained based on the comparison results of the visual entropy density and the preset critical conditions.

[0080] The folding processing module 430 is used to perform reversible dimension folding on the dimension to be folded, generate a semantic cell set, and establish the original dimension member reverse mapping, residual distribution information and abnormal anchor point information for the semantic cell set to obtain the folding analysis results.

[0081] The supplementary calculation processing module 440 is used to generate multidimensional analysis visualization data based on the folding analysis results, retained dimensions and interaction context parameters, and to perform precise supplementary calculation on candidate visual units that meet the local supplementary calculation conditions according to residual distribution information and abnormal anchor point information, so as to obtain corrected visualization data.

[0082] The interactive feedback module 450 is used to output corrected visualization data and update the interactive context parameters based on the user's interactive behavior on the corrected visualization data, so as to adjust the folding granularity, calculation range and output priority in the next round of multidimensional analysis visualization process.

[0083] Specifically, the data processing module 410 includes a data access submodule, a semantic parsing submodule, a viewport parameter processing submodule, and a context processing submodule. The data access submodule receives data records from different business systems, including at least one of a sales management system, customer relationship management system, supply chain management system, inventory management system, financial management system, production management system, risk control management system, and log collection system. Data records include detailed data, summary data, event data, log data, and indicator snapshot data. The semantic parsing submodule reads the business semantic configuration and determines the correspondence between fields and dimensions, fields and indicators, fields and levels, and fields and business rules based on the business semantic configuration. The viewport parameter processing submodule parses the current visualization viewport parameters, enabling the system to obtain the number of pixels in the current display area, chart expression method, filtering conditions, and visual unit division rules. The context processing submodule parses user interaction context parameters, enabling the system to obtain user behavior information such as hovering, clicking, box selection, filtering, drill-down, back navigation, and dwell time in the current analysis path.

[0084] Furthermore, the multidimensional structured processing in the data processing module 410 does not require changing the original storage format of the business data. Instead, it achieves this by constructing intermediate indexes, mapping tables, semantic tags, or temporary analytical views. In other words, the business data remains stored in the original data warehouse, columnar database, object storage, or log system, while the data processing module 410 only generates candidate analytical data for subsequent calculations. This candidate analytical data includes data object identifiers, dimension member identifiers, indicator values, hierarchical codes, business rule tags, data partition identifiers, cache hit identifiers, and historical query cost identifiers. In this way, the system can transform scattered, heterogeneous, and inconsistent business data into unified analytical objects that can participate in visual entropy density calculations without forcibly changing the original business system.

[0085] The entropy density determination module 420 includes a visual unit partitioning submodule, a state statistics submodule, a fluctuation determination submodule, an anomaly determination submodule, a load determination submodule, and a critical judgment submodule. The visual unit partitioning submodule is used to divide the current display area into multiple candidate visual units based on the current visualization viewport parameters. For bar charts, candidate visual units are each bar or a group of bars; for line charts, candidate visual units are the neighborhood of sampling points or the neighborhood of line segments; for heatmaps, candidate visual units are heat grids; for map-type charts, candidate visual units are spatial grids or region blocks; and for pivot tables, candidate visual units are row and column intersection units.

[0086] The status statistics submodule is used to count the number of members in different dimensions within each candidate visual unit, obtaining the dimension status count. The dimension status count is not simply the number of data records, but rather represents the number of business states that need to be distinguished within the current candidate visual unit. For example, a map area may contain 10,000 order records, but these orders may only correspond to 30 cities, 200 stores, or 3,000 customers. The system counts the dimension status count for different candidate dimensions to determine which specific dimension is causing information congestion in the current visual area.

[0087] The volatility determination submodule is used to determine the volatility of indicators based on the dispersion, magnitude of change, and local trend differences of indicator values ​​within candidate visual units. If the indicator values ​​of members in each dimension within a candidate visual unit are relatively similar, the information loss caused by folding may be low, even if the number of dimension states is large. Conversely, if the internal indicator values ​​differ significantly, folding can easily lead to deviations in business interpretation. Therefore, the indicator volatility is used to constrain the system from folding solely based on the number of dimensions.

[0088] The anomaly identification submodule is used to identify extreme points, mutation points, outliers, and business rule anomalies within candidate visual units, and to determine the contribution of these anomalies. Extreme points are identified using local location thresholds, mutation points are identified using the rate of change between adjacent time windows or adjacent business cycles, outliers are identified using local statistical distributions, and business rule anomalies are identified based on the business rules configured in the business semantics. A higher anomaly contribution indicates that the anomaly object in the candidate visual unit is less suitable for complete coverage by ordinary aggregation.

[0089] The pixel capacity determination submodule determines the pixel capacity based on the number of pixels of candidate visual units in the current viewport and the chart representation method. Pixel capacity reflects how much distinguishable information state the current display area can carry. With the same data scale, a larger display area can carry more states, while a smaller display area can only carry fewer states; with the same number of pixels, tables, bar charts, heatmaps, scatter plots, and map overlay layers have different distinguishability of information states. Therefore, the capacity determination submodule introduces a chart representation method correction coefficient, enabling the system to incorporate the display's expressive capabilities into data processing decisions.

[0090] The critical judgment submodule determines the dimensions to be collapsed and the dimensions to be retained based on the comparison between the visual entropy density and preset critical conditions. When the visual entropy density exceeds the preset critical condition, it indicates that the dimension has caused information pressure exceeding the display capacity in the current visual unit. At this time, the system no longer displays all members of the original dimension, but instead determines the corresponding dimension as the dimension to be collapsed. If a candidate dimension is marked as a strongly interpretable dimension by the business semantic configuration, or if the abnormal contribution of the dimension reaches the mandatory retention condition, the critical judgment submodule determines it as a retained dimension, or retains the abnormal anchor points within it during collapse.

[0091] The folding processing module 430 includes a distance determination submodule, a merge cost determination submodule, a semantic cell generation submodule, a reverse mapping establishment submodule, a residual record submodule, and an anomaly anchor point record submodule. The distance determination submodule determines the semantic distance, metric distribution distance, and query cost distance between members of each dimension in the dimension to be folded. Semantic distance characterizes the proximity of dimension members at the business semantic level, metric distribution distance characterizes the proximity of dimension members at the metric performance level, and query cost distance characterizes the proximity of dimension members in the data reading and aggregation paths. The merge cost determination submodule calculates the folding cost of dimension members based on the above distances and the hierarchical consistency reward. The semantic cell generation submodule folds dimension members that meet the folding conditions into the same semantic cell and generates a cell identifier, member list, aggregation metric, dominant dimension label, and display granularity label.

[0092] The reverse mapping establishment submodule is used to establish the reverse mapping relationship between semantic cells and original dimension members. Since this reverse mapping relationship is generated synchronously during collapse, subsequent user drill-down, selection, or precise calculations can directly obtain the set of original dimension members based on the semantic cell identifier, without needing to re-traverse all business data. The residual recording submodule is used to record the difference between the actual indicator value and the aggregate indicator value of each original dimension member within each semantic cell, and generates the residual mean, residual variance, maximum residual, residual quantile, and residual risk value. The anomaly anchor recording submodule is used to write original dimension members that meet the anomaly preservation conditions into anomaly anchor information, and records the anomaly type, anomaly strength, the semantic cell to which it belongs, and the location of the visible unit.

[0093] The supplementary calculation processing module 440 includes a basic visualization data generation submodule, a supplementary calculation score determination submodule, a supplementary calculation object determination submodule, a precise aggregation submodule, and a data correction submodule. The basic visualization data generation submodule generates initial multidimensional analysis visualization data based on semantic cells, retained dimensions, and aggregation indicators. The supplementary calculation score determination submodule determines the comprehensive supplementary calculation score for each candidate visual unit based on residual risk values, anomaly retention risk values, and interaction attention values. The supplementary calculation object determination submodule identifies candidate visual units whose comprehensive supplementary calculation scores reach the supplementary calculation threshold as visual units to be supplemented. The precise aggregation submodule obtains the set of original dimension members corresponding to the visual units to be supplemented based on the reverse mapping of original dimension members, and re-executes precise aggregation according to business indicator standards. The data correction submodule uses the precise aggregation results to correct the basic visualization data, obtaining corrected visualization data.

[0094] The interactive feedback module 450 includes an interaction capture submodule, an interaction heatmap generation submodule, a context update submodule, and a priority adjustment submodule. The interaction capture submodule captures user actions such as hovering, clicking, selecting, zooming, filtering, drilling down, and going back when interacting with the visualization data. The interaction heatmap generation submodule generates an interaction heatmap based on the visual unit position, duration, operation frequency, and dimensional path corresponding to the interaction action. The context update submodule writes the interaction heatmap into the interaction context parameters. The priority adjustment submodule adjusts the folding granularity, calculation range, and output priority based on the interaction context parameters during the next round of multidimensional analysis visualization.

[0095] The modules in this embodiment are implemented through software, hardware, or a combination of both. Modules may be configured in the same electronic device or distributed across multiple electronic devices. The division between modules is merely for the purpose of describing the technical solution of this invention and should not be construed as a limitation on the actual deployment method. In specific implementations, the data processing module 410, entropy density determination module 420, folding processing module 430, supplementary calculation processing module 440, and interactive feedback module 450 are deployed as different software functional units, or they may be combined into one or more processing programs for execution. Any method that can achieve the big data multidimensional analysis and visualization method for business data provided in this embodiment of the invention should fall within the protection scope of this invention.

[0096] The system uses a data processing module 410 to convert scattered business data into candidate analysis data. An entropy density determination module 420 establishes a critical balance between display capacity and business information pressure. A folding processing module 430 generates semantic cells with reverse mapping capabilities. A supplementary calculation module 440 performs localized, precise supplementary calculations on high-risk areas. An interactive feedback module 450 transforms user behavior into subsequent calculation control parameters. These modules are not simply connected in series, but rather form a cyclical feedback relationship. The calculation result of visual entropy density determines the folding range; the folding result carries residual and anomaly information; the residual and anomaly information determine the supplementary calculation range; and user interaction, in turn, adjusts the folding granularity and supplementary calculation priority for the next round. Therefore, the system can reduce full-scale computation in high-dimensional, high-cardinality business data scenarios while maintaining the analytical accuracy of anomalies and user-focused areas.

[0097] The above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, combinations, splits, or variations made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0098] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments are implemented by computer program instructions to related hardware. The computer program is stored in a computer-readable storage medium. When the computer program is executed by a processor, it implements the processes of the above method embodiments. The computer-readable storage medium may be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, solid-state drive, or other media suitable for storing computer programs.

[0099] It should be understood that although this specification describes the invention according to embodiments, not every embodiment contains only one independent technical solution. This narrative style is for clarity only, and those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can be appropriately combined to form other implementation methods understood by those skilled in the art.

[0100] It should also be understood that the terms "including," "contains," or any other variations thereof used in this specification and claims are intended to cover non-exclusive inclusion. A process, method, system, product, or apparatus that includes a list of elements includes not only those expressly listed but also other elements not expressly listed, or elements inherent to such process, method, system, product, or apparatus. Unless otherwise specified, the inclusion of a defined element by a statement does not exclude the presence of other identical elements in the process, method, system, product, or apparatus that includes that element.

[0101] Furthermore, in this document, relational terms such as "first," "second," etc., are used only to distinguish one entity or operation from another, and do not necessarily require or imply any actual order or sequence between these entities or operations. "Multiple" refers to two or more entities or operations, unless otherwise explicitly specified.

[0102] Those skilled in the art should understand that the flowcharts and structural diagrams in the accompanying drawings only illustrate some steps or partial structures in possible implementations of the systems, methods, and computer program products according to various embodiments of the present invention. Each block in the drawings represents a functional module, a program segment, or a processing step. The blocks are executed in the order shown in the drawings, but may also be executed in different orders or in parallel as needed. Any action that achieves the technical effects of the present invention should not be construed as departing from the scope of protection of the present invention.

[0103] Based on the preferred embodiments of the present invention described above, and through the above description, those skilled in the art can make various changes and modifications without departing from the technical concept of the present invention. The technical scope of the present invention is not limited to the contents of the specification, but must be determined according to the scope of the claims.

Claims

1. A method for multi-dimensional analysis and visualization of big data oriented business data, characterized in that, include: S1. Obtain the business dataset, business semantic configuration, current visualization viewport parameters, and interaction context parameters; perform multidimensional structured processing on the business dataset to generate candidate analysis data containing dimension members, indicator values, hierarchical relationships, and query cost identifiers; S2. Based on the business semantic configuration, the current visualization viewport parameters, and the candidate analysis data, determine the number of dimension states, indicator fluctuation, anomaly contribution, query cost, and pixel capacity corresponding to the candidate visual unit, calculate the visual entropy density of the candidate visual unit, and determine the dimensions to be folded and the dimensions to be retained based on the comparison result of the visual entropy density and the preset critical conditions. S3. Perform reversible dimension folding on the dimension to be folded to generate a semantic cell set, and establish the original dimension member reverse mapping, residual distribution information and anomaly anchor information for the semantic cell set to obtain the folding analysis results; S4. Based on the folding analysis results, the retained dimensions, and the interaction context parameters, generate multidimensional analysis visualization data, and perform precise compensation calculation on candidate visual units that meet the local compensation calculation conditions according to the residual distribution information and the abnormal anchor point information to obtain corrected visualization data; S5. Output the corrected visualization data, and update the interaction context parameters according to the user's interaction behavior with the corrected visualization data, so as to adjust the folding granularity, calculation range and output priority in the next round of multidimensional analysis visualization process. 2.The business data oriented big data multidimensional analysis and visualization method according to claim 1, characterized in that, The process involves acquiring the business dataset, business semantic configuration, current visualization viewport parameters, and interaction context parameters; performing multidimensional structured processing on the business dataset to generate candidate analysis data containing dimension members, indicator values, hierarchical relationships, and query cost identifiers, including: Retrieve the business dataset containing business object identifier, time field, space field, organization field, product field, customer field, transaction field, and indicator field; Retrieve the business semantic configuration, which includes dimension definitions, dimension levels, indicator definitions, business rules, and dimension relationships; Get the current visualization viewport parameters, including display range, zoom level, chart type, number of screen pixels, visual unit division method, and current filter conditions; Retrieve interaction context parameters that include historical hover records, selected records, drill-down records, filtered records, and dwell time records; According to the business semantic configuration, the business dataset is mapped by fields, normalized by definitions, bound by hierarchy, and aggregated by indicators to generate candidate analysis data. 3.The business data oriented big data multidimensional analysis and visualization method of claim 1, wherein, The step of determining the number of dimensional states, metric fluctuation, anomaly contribution, query cost, and pixel capacity corresponding to candidate visual units based on the business semantic configuration, the current visualization viewport parameters, and the candidate analysis data includes: Based on the current visualization viewport parameters, the current visualization area is divided into multiple candidate visual units; The number of dimension members covered by the candidate dimension within each candidate visual unit is counted to obtain the number of dimension states. The fluctuation of the indicator is determined based on the dispersion, variation range, and local trend differences of the indicator values ​​within each candidate visual unit. The abnormal contribution amount is determined based on the degree of contribution of extreme points, mutation points, outliers, and business rule anomalies within each candidate visual unit relative to its respective visual area. The query cost is determined based on the number of data partitions associated with the candidate visual unit, the number of detailed records, the complexity of indicator aggregation, and the historical query time. The pixel capacity is determined based on the number of pixels of each candidate visual unit in the current viewport and the graphical representation method.

4. The method of claim 1, wherein, The calculation of the visual entropy density of the candidate visual units, and the determination of the dimension to be folded and the dimension to be retained based on the comparison result of the visual entropy density with the preset critical condition, includes: The visual entropy density of the j-th candidate visual unit for the i-th candidate dimension is calculated using the following formula: ;in, Represents visual entropy density. Indicates the number of states in a dimension. Indicates the fluctuation of the indicator. This indicates an abnormal contribution. This indicates the cost of the query. Indicates pixel capacity. Indicates the minimum pixel capacity. and These represent the weight parameters, and ; When the visual entropy density is greater than or equal to a preset entropy density threshold, and the ratio between the number of dimensional states and the pixel carrying capacity is greater than or equal to a preset state carrying capacity threshold, the corresponding candidate dimension is determined as the dimension to be folded. When the visual entropy density is less than the preset entropy density threshold, or when the candidate dimension is marked as a strongly interpreted dimension, the corresponding candidate dimension is determined as a retained dimension.

5. The business data oriented big data multidimensional analysis visualization method of claim 1, wherein, The step of performing reversible dimension folding on the dimension to be folded to generate a semantic cell set includes: Based on the business semantic configuration, calculate the semantic distance between any two dimension members in the dimension to be collapsed; Based on the candidate analysis data, calculate the index distribution distance between any two dimension members; Based on the semantic distance, the indicator distribution distance, and the query cost distance, determine the dimension member merging cost; According to the dimensional member merging cost, dimensional members that meet the merging conditions are folded into the same semantic cell; For each semantic cell, generate a cell identifier, member list, aggregation metric, dominant dimension label, and display granularity label.

6. The method of claim 5, wherein, The cost of merging the dimension members is determined according to the following formula: ;in, This represents the cost of merging dimension members between dimension member a and dimension member b. Indicates semantic distance. Indicates the distance of indicator distribution. Indicates the distance of the query cost. This indicates the amount of reward for hierarchical consistency. and These represent the weight parameters; When the merging cost of the dimension members is less than or equal to the preset merging cost threshold, and dimension members a and b do not contain any abnormal anchors that are forcibly retained, dimension members a and b are assigned to the same semantic cell.

7. The business data oriented big data multidimensional analysis visualization method of claim 1, wherein, The process of establishing the original dimensional member reverse mapping, residual distribution information, and anomaly anchor information for the semantic cell set includes: Record the correspondence between each semantic cell and the original dimensional member that has been folded, forming a reverse mapping of the original dimensional member; Based on the difference between the real index value and the aggregate index value of each original dimension member in each semantic cell, residual distribution information is generated. Identify the original dimension members that meet the anomaly preservation conditions, and write the original dimension members as anomaly anchor points into the anomaly anchor point information; The anomaly preservation conditions include at least one of the following: extreme value preservation conditions, mutation preservation conditions, outlier preservation conditions, and business rule anomaly preservation conditions.

8. The business data oriented big data multidimensional analysis visualization method of claim 1, wherein, The step of performing precise compensation calculation on candidate visual units that meet the local compensation calculation conditions based on the residual distribution information and the abnormal anchor point information includes: Based on the residual distribution information, calculate the residual risk value for each candidate visual unit; Based on the abnormal anchor point information, calculate the abnormal retention risk value for each candidate visual unit; Based on the interaction context parameters, calculate the interaction attention value for each candidate visual unit; When at least one of the residual risk value, the anomaly retention risk value, and the interactive attention value meets the corresponding threshold condition, the corresponding candidate visual unit is determined as a visual unit to be recalculated. The original dimension members and index values ​​associated with the visual unit to be supplemented are precisely aggregated to generate a partial supplementation result, and the partial supplementation result is used to correct the multidimensional analysis visualization data.

9. The business data oriented big data multidimensional analysis visualization method of claim 1, wherein, The step of updating the interaction context parameters based on the user's interaction with the corrected visualization data to adjust the folding granularity, calculation range, and output priority in the next round of multidimensional analysis visualization includes: Capture at least one of the following user interactions performed on the corrected visualization data: hover, click, box selection, zoom, filter, drill-down, and back; An interactive heatfield is generated based on the visual unit position, duration, operation frequency, and dimensional path corresponding to the interactive behavior. Based on the interactive thermal field, the priority of supplementary calculation for the attention visual unit is increased, the display granularity of the unattended visual unit is reduced, and the folding granularity associated with the attention dimension path is updated. In the next round of multidimensional analysis visualization, based on the updated interaction context parameters, the visual units of interest, outlier anchors, and visual units with high residuals will be output first.

10. A business data oriented big data multidimensional analysis visualization system, characterized by, include: The data processing module is used to acquire business datasets, business semantic configurations, current visualization viewport parameters, and interaction context parameters, and to perform multi-dimensional structured processing on the business datasets to generate candidate analysis data containing dimension members, indicator values, hierarchical relationships, and query cost identifiers. The entropy density determination module is used to determine the number of dimensional states, index fluctuation, anomaly contribution, query cost, and pixel capacity corresponding to the candidate visual unit based on the business semantic configuration, the current visualization viewport parameters, and the candidate analysis data; calculate the visual entropy density of the candidate visual unit; and determine the dimensions to be collapsed and the dimensions to be retained based on the comparison result of the visual entropy density with the preset critical conditions. The folding processing module is used to perform reversible dimension folding on the dimension to be folded, generate a semantic cell set, and establish the original dimension member reverse mapping, residual distribution information and anomaly anchor information for the semantic cell set to obtain folding analysis results. The supplementary calculation processing module is used to generate multidimensional analysis visualization data based on the folding analysis results, the retained dimensions and the interaction context parameters, and to perform precise supplementary calculation on candidate visual units that meet the local supplementary calculation conditions according to the residual distribution information and the abnormal anchor point information, so as to obtain corrected visualization data. The interactive feedback module is used to output the corrected visualization data and update the interactive context parameters based on the user's interactive behavior on the corrected visualization data, so as to adjust the folding granularity, calculation range and output priority in the next round of multidimensional analysis visualization process.

Images