Model management and data analysis method, computer device, medium, program product

Abstract

This application discloses a model management and data analysis method, computer equipment, media, and program product, relating to the field of artificial intelligence technology. The model management method includes constructing a topological association between analytical indicators in the current business scenario as an indicator honeycomb; obtaining business instructions and parsing the sales indicators to be analyzed as target indicators; querying the related indicators of the target indicators selected from the indicator honeycomb through self-analysis; and using the related indicators as input to a business analysis model to construct the business analysis model. The business analysis model is used to analyze the target indicators of the current business scenario based on the related indicators. This solves the technical problems of poor data analysis flexibility and insufficient reliability of analysis results in a commercial business environment, achieving the technical effect of adaptively reconstructing the business analysis model to optimize the flexibility and reliability of data analysis.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a model management method, a data analysis method, a computer device, a computer-readable storage medium, and a computer program product. Background Technology

[0002] In current scenarios such as business analytics, separate analytical models and content generation models are typically trained for different analytical objectives and outputs to adapt to varying analytical goals. Therefore, in practical business analytics scenarios, a large number of analytical models and content generation models need to be deployed in parallel, resulting in bloated model data, poor data analysis flexibility, and difficulty in dynamically adjusting models according to the actual scenario, which can easily lead to unreliable analytical results. Summary of the Invention

[0003] This application provides a model management method, a data analysis method, a computer device, a computer-readable storage medium, and a computer program product to at least solve the problems of poor data analysis flexibility and insufficient reliability of analysis results in the business environment in related technologies.

[0004] This application provides a model management method, which includes: constructing a topology of relationships between analytical indicators in the current business scenario as an indicator honeycomb; obtaining business instructions and parsing the sales indicators to be analyzed as target indicators; querying the related indicators of the target indicators selected by the indicator honeycomb from the self-analysis indicators; and using the related indicators as input to the business analysis model to construct the business analysis model; wherein, the business analysis model is used to analyze the target indicators of the current business scenario based on the related indicators.

[0005] In one embodiment of this application, using related indicators as input to construct an operational analysis model includes: obtaining the current analysis model and decoupling the indicators in the current analysis model; reconstructing the current analysis model by using related indicators as new model inputs, and fusing the related indicators at the model layer based on dimension values ​​to form an operational analysis model; wherein, the dimension values ​​include at least one of time, region, and product category.

[0006] In one embodiment of this application, the associated indicators include candidate indicators for associated stores. The candidate indicators represent analytical indicators within the indicator honeycomb that meet preset association conditions in terms of their correlation with the target indicator. Before the query indicator honeycomb selects the associated indicators for the target indicator from the self-analysis indicators, it further includes: identifying whether the business instruction carries a main keyword; wherein, the main keyword includes the scene store information of the current business scenario; in response to the business instruction carrying the main keyword, the scene store identified by the main keyword is taken as the target store, and other scene stores with a similarity threshold with the target store are selected as associated stores.

[0007] In one embodiment of this application, constructing a correlation topology among analytical indicators in the current business scenario as an indicator honeycomb includes: evaluating at least one correlation factor among analytical indicators; weighting and fusing the correlation factors among the same analytical indicators to obtain the correlation degree between them; constructing a correlation topology with analytical indicators as topology nodes and correlation degree as edge weights, and constructing a multi-level correlation layer of the correlation topology based on the correlation relationship between analytical indicators; wherein, the correlation relationship includes direct correlation relationship and indirect correlation relationship; analytical indicators that have a direct correlation relationship with the central node of the correlation topology are closer to the central node than analytical indicators that have an indirect correlation relationship.

[0008] In one embodiment of this application, the correlation factor includes at least one of the following: number of common data source tables, business domain correlation, business logic dependency, time series correlation, and dimension value overlap; wherein, time series correlation represents the similarity of indicator values ​​between analysis indicators over time.

[0009] In one embodiment of this application, parsing the sales indicator to be analyzed indicated by the business instruction as the target indicator includes: identifying whether the business instruction carries sales indicator information; in response to the business instruction not carrying sales indicator information, extracting the instruction target of the business instruction, and matching the sales indicator associated with the instruction target as the target indicator.

[0010] This application also provides a data analysis method, which includes: obtaining a data analysis instruction; identifying the target indicator to be analyzed in the current business scenario carried by the data analysis instruction; constructing a business analysis model for analyzing the target indicator using the model management method in any of the above embodiments; and performing data analysis on the target indicator using the business analysis model to generate analysis results.

[0011] This application also provides a computer device, which includes: a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program to implement the steps of any of the above-described model management methods; or, to implement the steps of the above-described data analysis methods.

[0012] This application also provides a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, it implements the steps of any of the above-described model management methods; or, implements the steps of the above-described data analysis methods.

[0013] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of any of the above-described model management methods; or, implements the steps of the above-described data analysis methods.

[0014] This application enables the pre-construction of a correlation topology, known as an indicator honeycomb, based on the relationships between analytical indicators. Therefore, when receiving data analysis requirements (business instructions), it allows for adaptive selection of related indicators for the target indicators being analyzed. This adaptively constructs a business analysis model for the current operational scenario, flexibly adapting to different data analysis needs and enabling model reconstruction. Furthermore, the pre-constructed indicator honeycomb allows for the analysis and correlation of analytical indicators, effectively selecting input data during business analysis model reconstruction and improving the reliability of the model. Thus, it solves the technical problems of poor data analysis flexibility and insufficient reliability of analysis results in a commercial business environment, achieving the technical effect of adaptively reconstructing business analysis models to optimize both the flexibility and reliability of data analysis. Attached Figure Description

[0015] To more clearly illustrate the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram illustrating an application scenario of an embodiment of the data analysis method of this application; Figure 2 This is a flowchart illustrating an embodiment of the model management method of this application; Figure 3 This is a flowchart illustrating another embodiment of the model management method of this application; Figure 4 This is a flowchart illustrating an embodiment of the data analysis method of this application; Figure 5 This is a schematic diagram of the structure of an embodiment of the data analysis system of this application. Detailed Implementation

[0017] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the protection scope of this application.

[0018] It should be noted that, in the description of this application, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. The terms "first," "second," etc., in this application are used to distinguish similar objects and are not used to describe a specific order or sequence.

[0019] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0020] The specific application environment architecture or specific hardware architecture on which the execution of the model management method depends is described here.

[0021] Please see Figure 1 , Figure 1 This is a schematic diagram illustrating an application scenario of one embodiment of the data analysis method of this application.

[0022] In one embodiment, the network service serves as the interaction entry point with external users, enabling users to complete operations such as issuing business instructions, configuring indicators, and viewing analysis results. The application service can act as a scheduling layer, connecting to the network service to receive requests, and can also link the two major functional modules to achieve scheduling and collaboration of the entire process tasks. The first is equivalent to a basic data service module, which can cover four sub-modules: metadata management, model management, data scheduling, and data storage. Metadata management provides basic data support for the construction of the indicator honeycomb, data scheduling ensures the flow and synchronization of indicator data, data storage completes the persistence of all data, and model management is responsible for the full lifecycle management of the business analysis model, jointly building the underlying data and model support system for business analysis. The second is equivalent to an artificial intelligence computing power service module, which can include artificial intelligence application services, artificial intelligence services, and graphics processor clusters. The artificial intelligence services rely on algorithms to realize core analysis logic such as indicator association queries, model training, and inference. The artificial intelligence application services encapsulate analysis capabilities to complete task execution, and the graphics processor clusters provide high-performance parallel computing power to support large-scale data computation and model calculation needs.

[0023] Thus, the overall architecture in this embodiment achieves collaborative linkage between data management, computing power support, and business applications through a layered and decoupled design. It can support the business analysis process based on indicator honeycomb, that is, users issue instructions through network services, which are parsed by application services and then call the basic data service module to obtain related indicators. The business analysis model is constructed and reasoned by relying on the artificial intelligence computing power service module, so as to feed the analysis results back to the front end and ensure the scalability, maintainability, and analysis efficiency of the system.

[0024] The embodiments of this application provide a model management method, and the model management method is described in detail in conjunction with the execution flow of the model management method.

[0025] Please see Figure 2 , Figure 2 This is a flowchart illustrating an embodiment of the model management method of this application.

[0026] S101: Construct the topology of relationships between analytical indicators in the current business scenario as an indicator honeycomb.

[0027] In this embodiment, a relational topology is constructed among various analytical indicators in the current business scenario, and this topology is defined as an indicator honeycomb. It is easy to understand that an indicator honeycomb is used to represent the inherent logical relationships and mutual influence paths between different analytical indicators, such as the relational structure between sales indicators and customer flow indicators, inventory indicators, and promotion indicators. By constructing an indicator honeycomb, a systematic foundation of indicator relationships can be provided for subsequent analysis, reducing the potential bias of single-indicator analysis and improving the comprehensiveness and accuracy of business analysis.

[0028] S102: Obtain business instructions and parse the sales indicators to be analyzed as target indicators.

[0029] In this embodiment, after obtaining the business instruction, its content can be parsed to extract the sales indicator to be analyzed and identified as the target indicator. Business instructions typically originate from business analysts, management systems, or automated analysis tasks, used to clarify the core focus of the current analysis. Determining the target indicator by parsing the business instruction provides a clear direction and focus to the analysis process, ensuring that the analysis revolves around actual business needs.

[0030] S103: Query the related indicators of the target indicator selected by the honeycomb self-analysis indicator.

[0031] In this embodiment, after determining the target indicator, the existing indicator honeycomb can be queried, and indicators that are related to the target indicator can be selected from the analytical indicators contained therein. It is easy to understand that related indicators refer to indicators that have a direct, indirect, or synergistic relationship with the target indicator in terms of business logic. Systematic querying through the indicator honeycomb can quickly and accurately identify various indicators related to the target indicator, reducing potential omissions or biases that may occur in manual analysis.

[0032] S104: Use related indicators as input to construct the business analysis model. The business analysis model is used to analyze target indicators in the current business scenario based on related indicators.

[0033] In this embodiment, the selected related indicators can be used as input to the business analysis model to construct a business analysis model for analyzing target indicators in the current business scenario. The business analysis model can employ algorithms such as regression analysis, causal inference, and machine learning to interpret the current status of target indicators, identify potential influencing factors, or predict future trends based on historical and real-time data of the related indicators. By combining the indicator relationships revealed by the indicator honeycomb with model construction, the entire process from indicator relationship identification to quantitative analysis can be automated, improving the efficiency and depth of business analysis and providing interpretable data support for business decisions.

[0034] Thus, by constructing an indicator honeycomb to structurally model the relationships between analytical indicators, after obtaining business instructions and determining target indicators, the indicator honeycomb automatically selects related indicators and builds an operational analysis model based on these related indicators. This ensures that the analysis process closely follows actual business needs while systematically incorporating various influencing factors related to the target indicators, avoiding limitations and subjectivity in the analytical perspective, thereby improving the comprehensiveness, accuracy, and automation level of operational analysis and enhancing the reliability of business decisions.

[0035] In other words, because a correlation topology, known as an indicator honeycomb, can be pre-constructed based on the relationships between analytical indicators, when data analysis needs (i.e., business instructions) are received, adaptive correlation indicators can be selected for the target indicators to be analyzed. This allows for the adaptive construction of the business analysis model for the current business scenario, enabling flexible reconstruction of the business analysis model to adapt to different data analysis needs. Furthermore, the pre-constructed indicator honeycomb allows for the analysis and correlation of the relationships between analytical indicators, thus enabling the effective selection of analytical input data during business analysis model reconstruction. This improves the reliability of the business analysis model, allowing for adaptive reconstruction of the data analysis business analysis model to optimize both the flexibility and reliability of data analysis.

[0036] Please see Figure 3 , Figure 3 This is a flowchart illustrating another embodiment of the model management method of this application.

[0037] S201: Obtain business instructions.

[0038] S202: Parse the instruction keywords carried by the business instruction.

[0039] In this embodiment, the keyword type of the instruction keyword can include a subject type and a target type. Therefore, the subject keyword and / or operational keyword can be extracted from the instruction keywords carried by the business instruction.

[0040] It's easy to understand that the main keywords can be considered as indicating the business entity being analyzed, such as stores, regions, and channels; while the operational keywords indicate the specific operational indicators or analytical objectives to be analyzed, such as sales revenue, gross profit margin, and customer traffic. By categorizing and parsing the instruction keywords, the accuracy of the identification and analysis can be improved.

[0041] S203: Evaluate the construct factors that match the indicator keywords. The construct factors represent the analytical objectives of the current analysis model.

[0042] In this embodiment, based on the parsed instruction keywords, matching construction factors are evaluated. Construction factors can be used to characterize the analytical objectives that the current analysis model needs to achieve, such as diagnosing the reasons for fluctuations in a sales indicator, predicting the future trend of a business entity, or comprehensively evaluating multiple operational dimensions. By mapping instruction keywords to construction factors, business instructions can be transformed into executable analytical tasks, thereby helping to ensure the consistency between the model construction objectives and business needs.

[0043] S204: Select a building factor as the current factor to construct the current analysis model. In response to the current analysis model completing its analysis output on the current factor, deconstruct the current analysis model and select another building factor as the new current factor to reconstruct the current analysis model, allowing the new current analysis model to analyze and output on the new current factor. This process continues until all building factors have been traversed.

[0044] In this embodiment, a construction factor can be selected as the current factor, and the current analysis model is constructed based on the current factor. In response to the current analysis model completing the analysis of the current factor and outputting the analysis results, the current analysis model is deconstructed, releasing model resources or decoupling the current factor from the model structure. Another unanalyzed construction factor is selected as the new current factor, and the analysis model is reconstructed, allowing the new current analysis model to analyze and output the new current factor. This process is repeated until all construction factors have been traversed and analyzed. This approach reduces model concurrency complexity while facilitating sequential coverage of multi-objective and multi-dimensional analysis tasks, reducing model interference caused by multi-objective coupling, and improving the flexibility and controllability of the analysis process.

[0045] Specifically, in response to the instruction keywords, which include subject keywords and operational keywords, subject indicators are constructed using subject keywords, and the sales indicators to be analyzed, indicated by the operational keywords, are used as target indicators. Subject indicators are used to define the business boundaries of the analysis, such as specific stores or regions; target indicators are used to clarify the core objects of the analysis, such as sales revenue or average order value. By combining subject indicators and target indicators, a complete analysis task definition can be formed, providing clear input constraints for subsequent model construction.

[0046] The current analysis model can include the business analysis model. When constructing the business analysis model as the current analysis model, the related indicators of the target indicator can be selected by querying the inter-indicator honeycomb self-analysis indicator, and the related indicators can be used as inputs to construct the business analysis model.

[0047] Specifically, the current analysis model is acquired, and its indicators are decoupled. Indicator decoupling refers to breaking down the originally coupled input indicators in the model into independent indicator units, allowing for flexible replacement and recombination for different related indicators. The related indicators are then used as new model inputs to reconstruct the current analysis model, and these related indicators are fused at the model layer based on dimension values ​​to form the business analysis model. Dimension values ​​include at least one of time, region, and product category. By fusing related indicators and dimension values ​​at the model layer, the distribution characteristics and interaction relationships of indicators across different dimensions can be preserved, giving the business analysis model stronger interpretability and business adaptability.

[0048] Therefore, this embodiment can parse business instructions to extract instruction keywords, evaluate and match construction factors, and complete the analysis task of multiple construction factors sequentially through construction, deconstruction, and reconstruction. Simultaneously, when constructing the business analysis model, the decoupling of indicators, reconstruction based on related indicators, and model layer fusion based on dimensional values ​​facilitate flexible adaptation of the model structure and dynamic expansion of analytical capabilities. This helps meet the needs of multi-objective, multi-dimensional business analysis and reduces the risk of increased model complexity and decreased analytical accuracy caused by multi-task coupling, thereby improving the flexibility, scalability, and reliability of the analysis results of the business analysis system.

[0049] The following example illustrates the detailed working principle of the combined indicator honeycomb.

[0050] When constructing a topology of relationships between analytical indicators in the current business scenario, known as an indicator honeycomb, at least one relationship factor between the analytical indicators can be evaluated. This relationship factor quantifies the degree of correlation between different indicators in terms of data source, business logic, statistical characteristics, etc. By comprehensively considering relationship factors across multiple dimensions, the bias inherent in relying solely on a single dimension to judge indicator relationships can be reduced, thereby improving the data reliability of the indicator relationship topology.

[0051] The correlation degree between identical analytical indicators is obtained by weighted fusion of correlation factors. The weights of multiple correlation factors can be summed to 1. Different correlation factors may represent different dimensions of the actual correlation between analytical indicators. Assigning appropriate weights to different correlation factors can improve the alignment between the correlation degree and the actual business logic and data relationships. The correlation degree obtained after weighted fusion can serve as a quantitative basis for subsequently constructing the indicator topology.

[0052] The correlation factors include at least one of the following: the number of tables sharing a common data source, the degree of relevance to the business domain, the degree of dependence on business logic, time series correlation, and the degree of overlap in dimension values. Time series correlation refers to the similarity of the changes in indicator values ​​between analytical metrics over time.

[0053] Among these factors, the number of common data source tables reflects the degree of homogeneity between two indicators at the underlying data level; business domain relevance measures the similarity of the business scope to which the indicators belong; business logic dependency characterizes whether there is a causal or sequential relationship between the indicators in the business process; time series correlation indicates the similarity of indicator values ​​changing over time, for example, by measuring the degree of synchronous fluctuation of two indicators in the time dimension through the correlation coefficient; and dimensional value overlap measures the overlap of the value ranges of two indicators in dimensions such as time, region, and product category. By introducing multi-dimensional correlation factors, different types and levels of correlation between indicators can be comprehensively captured, improving the accuracy and robustness of correlation assessment.

[0054] The analytical indicators are used as topological nodes, and the correlation degree is used as the edge weight to construct a correlation topology. A multi-level correlation layer is then built based on the correlation relationships between the analytical indicators. These correlation relationships include direct and indirect relationships. Analytical indicators with direct correlations to the central node of the correlation topology are located closer to the central node than those with indirect correlations.

[0055] In other words, each analytical indicator is treated as a node in a topology, and the degree of correlation is used as the weight of the edges between nodes to construct an indicator correlation topology. Furthermore, a multi-level correlation layer is constructed based on the correlations between the analytical indicators. These correlations include direct and indirect relationships. Analytical indicators with a direct correlation to the central node of the correlation topology are closer to the central node than those with an indirect correlation. The central node is typically the core indicator of interest in the current analytical task. By constructing a multi-level correlation layer, the hierarchical correlation structure between indicators can be visually presented, and the distance between an analytical indicator and the central node is directly proportional to the closeness of their relationship in business logic or data.

[0056] The related indicators include candidate indicators for related stores. Candidate indicators represent analytical indicators whose correlation with the target indicator within the indicator hive meets the preset correlation conditions.

[0057] Therefore, before selecting related indicators for the target indicator in the query metrics analysis by Hive, it can also identify whether the business instruction carries a main keyword. The main keyword includes the scenario store information of the current business scenario. By identifying the main keyword, it can be determined whether the current analysis task needs to limit the scope of the subject, serving as the basis for whether to filter related stores.

[0058] In response to business instructions carrying key keywords, the system identifies the target store within the specified scenario and then filters other stores in the same scenario that meet a similarity threshold to the target store as associated stores. It's easy to understand that store similarity can be comprehensively evaluated based on one or more dimensions, such as store location, store size, customer characteristics, and product category. By introducing the concept of associated stores, data from stores with similar operating characteristics to the target store can be included in the analysis, providing more comparable data support for the subsequent selection of association indicators and enhancing the reference value and generalization ability of the analysis results.

[0059] The following section further elaborates on the correlation factors and the degree of correlation.

[0060] As explained above, correlation factors include at least one of the following: the number of tables sharing a common data source, the degree of relevance to the business domain, the degree of dependence on business logic, the correlation of time series data, and the degree of overlap of dimension values. Correlation factors may also include the co-occurrence rate of behavioral paths.

[0061] The number of common data source tables indicates the number of common data source tables or the proportion of shared intermediate calculation tables among statistical analysis indicators in the process, based on data warehouse metadata.

[0062] Business domain relevance analysis can identify whether indicators likely belong to the same business domain (such as finance or marketing) or participate in the same business process, based on business knowledge. This can be achieved through metadata tags or classification systems.

[0063] Business logic dependencies can be determined based on a business rule base by extracting mathematical relationships (such as addition, subtraction, multiplication, division, ratios, and conditional judgments) or by mining association rules. For example, analysis indicator A (average order value) = analysis indicator B (total sales revenue) / analysis indicator C (order volume).

[0064] Time series correlation can be considered as the correlation coefficient of indicator values ​​over time through historical data analysis (such as the Pearson correlation coefficient). If the trends of two analytical indicators are consistent, their time series correlation is relatively higher.

[0065] Dimensional overlap indicates that if two analytical indicators frequently have data on the same dimension values ​​(such as the same time period, region, or product category), they are considered to be more related. This can be measured by calculating the Jaccard similarity of the dimension values.

[0066] User behavior path co-occurrence refers to the proportion of co-occurrence between statistical analysis indicators in a user behavior sequence during traffic analysis. For example, analyzing the co-occurrence ratio of the indicators "add to cart" and "payment successful" in a user behavior sequence.

[0067] Furthermore, after calculating each correlation factor, the correlation factors can be normalized to form a standardized score correlation factor. A specific calculation formula can be illustrated as follows: Standardized Score = (max(all correlation factor values) - min(all correlation factor values)) / (current correlation factor - min(all correlation factor values)). Where, "all correlation factor values" represents the values ​​of all correlation factors; "max" represents the maximum value among all correlation factor values; "min" represents the minimum value among all correlation factor values; and "current correlation factor" represents the correlation factor value currently being standardized.

[0068] The following provides specific numerical examples of the weighted weights of the relevant factors involved in the sales system:

[0069] When constructing an indicator honeycomb, analytical indicators can be used as nodes, and the overall similarity can be used as edge weights to build a relational topology such as an undirected weighted graph. In this way, given a target indicator, one or more related indicators with the highest correlation can be found through the shortest path.

[0070] The core layer consists of the central node representing the current analytical metric (e.g., "Sales Revenue"). The first-level related layer displays directly related metrics (e.g., "Average Order Value" and "Order Quantity"), sorted by correlation degree, and can use different colors / thickness to indicate intensity. The second-level related layer can be expanded by clicking on the first-level node to display indirectly related metrics (e.g., "Advertising Volume" indirectly affects "Sales Revenue" through "Traffic").

[0071] Metrics Hive provides an API (Application Programming Interface) query interface. By inputting a metric as the core layer, it can return a list of related metrics sorted by their degree of relevance through hierarchical association.

[0072] Thus, in the data preparation phase, metadata, query logs, and time series data can be extracted from the indicator platform and dimension platform. In the parameter calculation phase, the correlation degree of each parameter can be calculated for each pair of indicators. In the model training phase, the parameter correlation degrees can be combined and weighted to train a correlation scoring model. In the correlation topology construction phase, indicators and correlation scores can be constructed into a graph structure and stored in a graph database and memory. In the iterative optimization phase, correlation parameters can be updated periodically, incorporating user feedback to adjust weights.

[0073] Based on the target metrics specified in the business instructions, other metrics with the highest correlation can be queried through the API interface provided by Metrics Hive. The data models containing these metrics can then be split and reorganized to generate lightweight new data models that focus on data analysis.

[0074] Business instructions serve as the atomic indicator layer, and combined with the indicator honeycomb as the original visible layer, the target indicators are extracted to form one or more single-scenario model layers. Combining and reorganizing the single-scenario model layers can form complex scenario model layers.

[0075] The honeycomb association graph primarily provides highly influential auxiliary analysis indicators through the API interface of the indicator honeycomb. Indicator extraction automatically identifies high-frequency analysis dimension combinations based on the analysis history of the corresponding indicators, decoupling the indicators in the data model based on columnar representation. Model reorganization merges cross-model analysis indicators across multiple scenario model layers through dimension values, aiming to generate the optimal execution plan. The data model after fission and reorganization has a small storage size and a short lifespan, existing only in the analysis and tracing stages. While solving the problems of model redundancy and large data storage ratios, it can provide data support for result tracing and improve the credibility of the results.

[0076] In layman's terms, this embodiment will be illustrated with specific examples in particular scenarios. Based on analysis instructions, an intelligent agent uses a model derived from fission and recombination to interpret and analyze data, providing feasible execution strategies.

[0077] The reasoning direction of an intelligent agent is generally generated by business instructions, and the reasoning logic varies across different scenarios. Taking a marketing campaign as an example, when the business issues the instruction "Recommend a marketing campaign that can be implemented by store X," the agent will first read the store X's level, location, and environment, and extract a combination of stores with a high degree of similarity to store X from stores nationwide. Based on the marketing campaign as the central indicator, it will analyze the stores in the combination that have the marketing campaign. Then, using sales amount, sales completion rate, etc., as new central indicators, it will generate a new analytical data model based on the indicator honeycomb through fission and recombination to analyze the data, obtain information such as the content of the marketing campaign, the individual products that achieved the completion rate, the unpopular individual products, and the stock shortage situation, and generate an execution strategy report from the marketing plan, individual product inventory, target audience, to the ranking of the best-selling individual products.

[0078] Over a period of time, the intermediate results of the execution strategy are evaluated to determine whether they meet expectations. If expectations are met, the strategy is finalized and can be extended to similar scenarios; if expectations are not met, a similar reasoning analysis is performed on the results at Store X to analyze the unknown influencing factors of the strategy's anomalies and make corrections.

[0079] Specifically, let's take the migration of marketing activities between Store X and Store Y as an example.

[0080] The metrics platform includes indicators such as sales amount, sales volume, inventory, inventory transfer, and inventory transfer ratio, and is associated with dimensions such as stores, product categories, and brands.

[0081] The business input command is store marketing promotion, which is (sales segment ∩ store level ∩ user age group) ∪ (industry sales ∩ weather data). Here, ∩ represents the intersection, and ∪ represents the union.

[0082] The current analytical model's first fission and recombination analysis selects store X, which has seen significant recent marketing activities. A logical chain of "(region-sales domain, region-city-sales domain, region-city-store-sales domain, store-category-sales domain, store-brand-sales domain...)" is achieved through methods such as difference-in-differences.

[0083] The current analytical model's second fission and recombination analysis, based on preliminary data from selected stores' marketing activities, initially defines a list of comparable stores by sales level, sales brand, and sales capability. Through methods such as synthetic control, a logical chain of "(store-store type-store format-sales domain, store-store level-sales domain...)" is formed.

[0084] The importance of features is analyzed. External data can be obtained to perform feature analysis on store X and the set of stores, including but not limited to weather, regional holidays, competitor dynamics, etc., and a suitable comparison store Y can be selected.

[0085] The current analytical model's third fission and recombination analysis primarily analyzes the advantages and disadvantages of store X in the marketing process. This includes aspects such as the marketing group, target sales group, groups with product demand, best-selling products, restocking of best-selling products, and less popular products. The analysis generates reports on collaborative filtering recommendations, causal transfer learning, and grouping strategy optimization, facilitating rapid marketing validation by comparing with store Y.

[0086] The current analysis model's fourth fission and recombination analysis mainly requires updating model parameters in real time based on the analysis results through online learning optimization algorithms, adapting to the dynamic changes of Y stores, integrating time series prediction functions, and monitoring and comparing the expected and actual gaps after strategy migration.

[0087] In one embodiment, it can also be identified whether the business instruction carries sales target information. In response to the business instruction not carrying sales target information, the instruction target of the business instruction is extracted, and the sales target associated with the instruction target is matched as the target indicator.

[0088] In other words, during the parsing of business instructions, it is necessary to identify whether the instructions carry sales target information. Sales target information directly indicates the specific sales metrics to be analyzed, such as sales revenue, sales volume, and gross profit margin. By identifying whether sales target information is carried, it is possible to determine the clarity of the business instruction's connection to the analysis objective, and thus decide how to determine the target metrics.

[0089] If the business instruction does not contain explicit sales target information, the instruction objective contained within the business instruction is further extracted. The instruction objective represents the business intent of the analysis task, such as evaluating the effectiveness of promotional activities, analyzing the reasons for sales decline, or predicting next month's sales volume. Based on the instruction objective, associated sales indicators are matched as target indicators. For example, when the instruction objective is to evaluate the effectiveness of a promotional activity, sales revenue, promotion conversion rate, and average order value can be matched as target indicators. Through the mapping of instruction objectives to sales indicators, the system can automatically identify the focus of the analysis task even when the business instruction does not explicitly specify sales indicators, improving the system's intelligence and user-friendliness.

[0090] The embodiments of this application provide a data analysis method, and the data analysis method is described in detail in conjunction with the execution flow of the data analysis method.

[0091] Please see Figure 4 , Figure 4 This is a flowchart illustrating an embodiment of the data analysis method of this application.

[0092] S301: Obtain data analysis instructions.

[0093] In this embodiment, acquiring data analysis instructions serves as the initiation signal for the business analysis process. Data analysis instructions can originate from manual triggering by business analysts, automatic pushes from upper-level business systems, or scheduled execution of pre-set analysis tasks. Clearly defining the acquisition stage of data analysis instructions helps provide a clear input source and execution basis for subsequent analysis tasks, thereby ensuring that the analysis process has clear initiation boundaries and traceability.

[0094] S302: Identify the target indicators to be analyzed in the current business scenario carried by the data analysis instruction.

[0095] In this embodiment, the acquired data analysis instructions are parsed to identify the target indicators to be analyzed in the current business scenario. These target indicators are the core focus of this analysis task, such as sales revenue, customer traffic, gross profit margin, and inventory turnover rate. It is easy to understand that data analysis instructions can carry target indicator information in various ways, including directly specifying the indicator name, implicitly indicating it through instruction keywords, or matching the target through instruction mapping. Accurately identifying the target indicators helps improve the accuracy of the analysis direction.

[0096] S303: Utilize model management methods to construct business analysis models for analyzing target indicators.

[0097] In this embodiment, the model management method can be as illustrated in any of the previous embodiments, and will not be repeated here.

[0098] S304: Use business analysis models to perform data analysis on target indicators and generate analysis results.

[0099] In this embodiment, a constructed business analysis model is used to perform data analysis on target indicators, and analysis results are generated based on the analysis process and model output. Data analysis may include interpretive analysis of the current state of the target indicators, importance assessment of influencing factors, root cause diagnosis of abnormal fluctuations, or predictive analysis of future trends. Analysis results can be presented in the form of structured reports, visual charts, key conclusion summaries, or data interface outputs for use by business analysts or upper-level decision-making systems. It is easy to understand that systematic data analysis through a business analysis model forms a reproducible, interpretable, and quantifiable automated analysis process, effectively improving analysis efficiency and the objectivity of the results.

[0100] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method.

[0101] Embodiments of this application also provide a data analysis system. Please refer to... Figure 5 , Figure 5 This is a schematic diagram of the structure of an embodiment of the data analysis system of this application.

[0102] In one embodiment, the data analysis system includes an application platform, a tool platform, and an intelligent platform.

[0103] As the business interaction layer, the application platform constructs indicator honeycomb data through standard indicator management and standard dimension management. It also provides business instruction management and strategy pilot functions, enabling users to complete operations such as indicator configuration, analysis instruction issuance, and strategy verification.

[0104] As the basic support layer, the tool platform includes modules for metadata management, model management, task scheduling, and data storage. It receives the metrics data from the application platform and provides basic capabilities such as metadata control, model lifecycle management, automated task scheduling, and persistent data storage for the entire process.

[0105] As the analysis and computing layer, the intelligent platform collects multi-source data through internal and external data acquisition modules. It then combines this data with a honeycomb-like indicator system to perform trend analysis and anomaly attribution for target indicators. Finally, the strategy report generation module outputs the analysis results back to the application platform. This achieves a complete closed loop for business analysis. The layered architecture decouples business interaction, basic support, and intelligent analysis, ensuring system scalability and maintainability, and improving the efficiency of supporting intelligent business analysis needs based on the indicator honeycomb system.

[0106] For a description of the features in the embodiments corresponding to the data analysis system, please refer to the relevant descriptions in the embodiments corresponding to the data analysis method; they will not be repeated here.

[0107] Embodiments of this application also provide a computer device, including a memory and a processor, wherein the memory stores a computer program and the processor is configured to run the computer program to perform the steps in any of the above-described data analysis method embodiments.

[0108] Embodiments of this application also provide a computer-readable storage medium storing a computer program, wherein the computer program is configured to execute the steps in any of the above-described data analysis method embodiments when it is run.

[0109] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard disk, magnetic disk, or optical disk.

[0110] Embodiments of this application also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the above-described model management method embodiments; or, implements the steps in the above-described data analysis method embodiments.

[0111] Embodiments of this application also provide another computer program product, including a non-volatile computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps in any of the above-described model management method embodiments; or, implements the steps in the above-described data analysis method embodiments.

[0112] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0113] The foregoing has provided a detailed description of the model management method, data analysis method, computer equipment, computer-readable storage medium, and computer program product provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and core ideas of this application. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A model management method, characterized in that, The model management method includes: Construct an indicator honeycomb by establishing a topology of relationships between analytical indicators in the current business scenario; Obtain business instructions and parse the sales indicators to be analyzed as target indicators. The query function selects related indicators of the target indicator from the analysis indicators. The associated indicators are used as inputs to construct an operational analysis model; wherein the operational analysis model is used to analyze the target indicators of the current operational scenario based on the associated indicators.

2. The model management method according to claim 1, characterized in that, The step of using the aforementioned related indicators as input to construct the business analysis model includes: Obtain the current analysis model and decouple the indicators of the current analysis model; The correlation indicators are used as new model inputs to reconstruct the current analysis model, and the correlation indicators are fused at the model layer based on the dimension values ​​to form the business analysis model; wherein, the dimension values ​​include at least one of time, region, and product category.

3. The model management method according to claim 1, characterized in that, The correlation indicators include candidate indicators for related stores. The candidate indicators represent analytical indicators within the indicator hive that meet preset correlation conditions with the target indicator. The query of the indicator honeycomb before selecting the related indicator of the target indicator from the analysis indicator also includes: Identify whether the business instruction carries a main keyword; wherein, the main keyword includes the scene store information of the current business scenario; In response to the business instruction carrying the main keyword, the scene store identified by the main keyword is selected as the target store, and other scene stores with a similarity threshold with the target store are selected as associated stores.

4. The model management method according to claim 1 or 3, characterized in that, The topology of relationships between analytical indicators in the current business scenario, referred to as an indicator honeycomb, includes: Evaluate at least one correlation factor among the analytical indicators; The correlation factors between the same analytical indicators are weighted and fused to obtain the degree of correlation between them. The analytical indicators are used as topological nodes, and the correlation degree is used as the edge weight to construct the correlation topology. A multi-level correlation layer of the correlation topology is constructed based on the correlation relationship between the analytical indicators. The correlation relationship includes direct correlation relationship and indirect correlation relationship. Analytical indicators that have a direct correlation relationship with the central node of the correlation topology are closer to the central node than analytical indicators that have an indirect correlation relationship.

5. The model management method according to claim 4, characterized in that, The correlation factors include at least one of the following: number of common data source tables, business domain relevance, business logic dependency, time series correlation, and dimension value overlap; wherein, the time series correlation represents the similarity of indicator values ​​among the analytical indicators over time.

6. The model management method according to claim 1, characterized in that, The process of parsing the sales indicators to be analyzed as target indicators by the business instructions includes: Identify whether the business instruction carries sales target information; In response to the fact that the business instruction does not carry sales target information, the instruction target of the business instruction is extracted, and the sales target associated with the instruction target is matched as the target indicator.

7. A data analysis method, characterized in that, The data analysis methods include: Get data analysis instructions; Identify the target indicators to be analyzed in the current business scenario carried by the data analysis command; An operational analysis model for analyzing the target indicators is constructed using the model management method as described in any one of claims 1 to 6; The business analysis model is used to perform data analysis on the target indicators and generate analysis results.

8. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the model management method according to any one of claims 1 to 6; or, implements the steps of the data analysis method according to claim 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the model management method according to any one of claims 1 to 6; or, implements the steps of the data analysis method according to claim 7.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the model management method according to any one of claims 1 to 6; or, implements the steps of the data analysis method according to claim 7.

Images