An engineering predictive analysis method

Abstract

An engineering-based predictive analytics method, based on multi-knowledge fusion of financial or business data, includes: generating a predictive model based on financial or business information obtained through an application interface; the predictive model is based on a high-order directed graph structure, expressing logical judgments and / or calculation rules through multiple nodes, the attributes of the multiple nodes, and the pairwise relationships between the multiple nodes, and expressing prediction paths through directed edges between nodes; configuring the common attributes of the predictive model, the attributes of the prediction path, and the personalized attributes of the nodes according to instantiation configuration information; wherein, the common attributes of the predictive model are used to set the connection relationships with the data source, input table, and output table; the attributes of the prediction path are used to set the correspondence between the logical judgments and / or calculation rules and the data; and the personalized attributes of the nodes are used to set the prediction object and prediction dimension expressed by each node.

Description

Technical Field

[0001] The embodiments of this application relate to the field of financial forecasting, and more particularly to an engineered forecasting analysis method. Background Technology

[0002] Financial forecasting is based on data from a company's past business and financial activities, external information such as the macro market and competitors, and relevant rules along the company's value chain. It utilizes systematic quantitative analysis techniques to predict the company's future financial condition and operational performance. The purpose of financial forecasting is to enhance the proactive nature of financial management, anticipate and quantify the potential impact of risks, reduce uncertainty in corporate management, and ensure that the expected goals of financial plans are aligned with changing external environments and economic conditions. It also allows for timely quantification of the effectiveness of financial plan implementation. Financial forecasting is a crucial basis for corporate managers to conduct lean management and make scientific decisions.

[0003] For a complex enterprise with multiple product lines and subsidiaries, the scope of business involved in the forecast, the time span of the forecast, and the frequency of analysis vary. A complete forecasting process often involves multiple disconnected systems. This situation leads to time-consuming and labor-intensive collaboration among various roles, resulting in low efficiency. Furthermore, the forecasting models and expert experience accumulated during the forecasting process cannot be systematically stored, making it difficult to form a systematic management of corporate financial forecasting knowledge and to effectively integrate it with corporate business data to support knowledge accumulation and reuse.

[0004] On the other hand, traditional enterprises typically have long business processes, complex logical relationships and financial rules. Forecasts carried by traditional financial systems emphasize the system's executability, making it difficult to trace the interpretability of the forecast data. This makes it difficult for non-IT personnel to understand, and experts find it difficult to adjust information or logic in the system to incorporate their own experience.

[0005] Enterprises commonly face the following challenges when building financial forecasting and analysis systems: integrating multiple data sources of business and financial data; complex and periodically updated business logic and financial rules that are difficult to continuously build and maintain; limited data volume in certain scenarios making it difficult to integrate expert judgments or achieve high-precision forecasts; difficulty in cross-validating with multi-instance forecasts; difficulty in testing and simulating different forecast assumptions; difficulty in interpreting the forecasting model; and difficulty in conducting cross-business, multi-role collaborative forecasting. Solving these challenges is a problem that needs to be addressed. Summary of the Invention

[0006] To address the aforementioned problems, embodiments of this application provide an engineering-based predictive analysis method. This method, based on multi-knowledge fusion of financial or business data, includes: generating a predictive model using financial or business information obtained through an application interface; the predictive model is based on a high-order directed graph structure, expressing logical judgments and / or calculation rules through multiple nodes, the attributes of the multiple nodes, and the pairwise relationships between the multiple nodes, and expressing predictive paths through directed edges between nodes; configuring the common attributes of the predictive model, the attributes of the predictive path, and the personalized attributes of the nodes according to instantiation configuration information; wherein the common attributes of the predictive model are used to set the connection relationships with the data source, input table, and output table; the attributes of the predictive path are used to set the logical... The algorithm establishes a correspondence between logical judgments and / or calculation rules and data; the personalized attributes of each node are used to set the prediction object and prediction dimension expressed by each node; the semantic parsing engine transforms the prediction model and the instantiation configuration information into computer-executable first parsing code; the first parsing code is used to engineer the prediction model with the data source, input table, and output table according to the common attributes of the prediction model; on the input table of the data source, the logical judgments and calculation rules expressed by the prediction model are executed according to the prediction path; the prediction data of the prediction object expressed by each node on the prediction dimension is determined, and the prediction data is output to the output table or displayed on the application interface. In this way, an algorithmic model that effectively captures patterns can transform business logic and financial rules into computer-executable parsing code, apply it to integrated data, and, with the assistance of expert experience and operational personnel's analysis and judgment, ultimately perform data prediction analysis.

[0007] In one possible implementation, the prediction path of the prediction model includes a node relationship path, which is a high-order directed graph in which the plurality of nodes are progressively connected from bottom to top, wherein the same node is associated with multiple nodes at different levels. Thus, this prediction path can express the progressively connected relationships between nodes from bottom to top under a certain dimension of the rules for financial prediction.

[0008] In one possible implementation, the prediction path of the prediction model includes a dimensional convergence path, which is a multi-branch tree structure connected by the plurality of nodes based on the prediction dimensional convergence principle; thus, the prediction path can express bottom-up convergence relationships based on dimensions such as company business classification, customer management, or organizational structure.

[0009] In one possible implementation, the prediction path of the prediction model includes a baseline transformation path, which comprises multiple baselines, each of which is a multi-branch tree structure consisting of the multiple nodes. Thus, the prediction path can express the transformation relationship of integrated baseline information between indicators.

[0010] In one possible implementation, generating a predictive model based on financial or business information obtained through a user application interface includes: invoking one or more of the predictive models based on information input through the user application interface. This enables collaborative operation among multiple predictive models in complex predictive workflows.

[0011] In one possible implementation, generating a predictive model based on financial or business information obtained through a user application interface includes: retrieving a template from a financial or business template library using information from the user application interface; the template's structure being the high-order directed graph; determining logical judgments and calculation rules based on the template by defining multiple nodes, configuring the attributes of the multiple nodes, and defining the pairwise relationships between the multiple nodes; determining the predictive path based on the connection relationships of directed edges between nodes; and generating a predictive model. In this way, users can save the high-order directed graph as a template for generalization in future predictive scenarios, and can also apply existing templates to newly generated business data, forming a virtuous cycle of iterative learning for corporate financial prediction knowledge.

[0012] In one possible implementation, generating a predictive model based on financial or business information obtained through a user application interface includes: selecting multiple nodes from an indicator library and / or baseline library based on the information obtained through the user application interface; determining the logical judgment and calculation rules by defining the multiple nodes, configuring the attributes of the multiple nodes, and the pairwise relationships between the multiple nodes based on the financial or business information obtained through the user application interface; determining the prediction path by connecting the directed edges of the multiple nodes; and generating a predictive model. In this way, users can create personalized templates and can also save high-order directed graphs as templates for generalization in future prediction scenarios.

[0013] In one possible implementation, determining the logical judgment and calculation rules by defining the plurality of nodes, configuring the attributes of the plurality of nodes, and the pairwise relationships between the plurality of nodes includes: defining the prediction object, step size / frequency, dimension, and analysis granularity of the nodes; configuring the adjustment attributes, calculation order attributes, prediction object type, and prediction constraints of the nodes; and setting the arrow direction of the directed edges between the nodes to determine the logical judgment and calculation rules. In this way, users can effectively capture the algorithmic model of patterns and express business logic and financial rules through the prediction model.

[0014] In one possible implementation, the personalized attributes of the node are also used to set the node's adjustment attributes, computation order attributes, prediction object type, and prediction constraints. This allows for personalized configuration of the prediction model.

[0015] In one possible implementation, the adjustment attributes of the nodes include adjustable or non-adjustable modes; the calculation order attributes of the nodes include the order in which the nodes perform logical judgments or calculations; the prediction object types of the nodes include calculation-type nodes, condition judgment-type nodes, information-type nodes, and algorithm-type nodes; and the prediction constraints of the nodes include the range of parameter values. Thus, the prediction model can be optimized and adjusted by combining expert experience and the analysis and judgment of operators.

[0016] In one possible implementation, configuring the common attributes of the prediction model based on the obtained instantiation configuration information includes: setting the connection relationships between the prediction model and local data sources, input tables, and output tables based on the obtained instantiation configuration information; or setting the connection relationships between the prediction model and remotely invoked data sources, input tables, and output tables based on the obtained instantiation configuration information. This enables connections between different prediction rules, instance data, prediction models, and prediction objects, improving the operability of prediction model generation, the interpretability of the prediction process, and the rapid generalization ability of the prediction model.

[0017] In one possible implementation, the step of converting the prediction model and the instantiation configuration information into computer-executable parsing code by the semantic parsing engine includes: using a JSON or XML semantic parsing engine to convert the prediction model and the instantiation configuration information into computer-executable parsing code. This allows for the engineered connection of the prediction model with the input and output tables of the data source based on the common attributes of the prediction model.

[0018] In one possible implementation, the data source includes one or a combination of financial databases, business databases, baseline databases, dimensional databases, and external databases. This allows enterprises to access a large amount of data from multiple systems when building a financial forecasting system, integrating relevant data on business activities and financial status across systems.

[0019] In one possible implementation, the method further includes: adjusting the common attributes of the prediction model, the attributes of the prediction path, and the personalized attributes of the nodes based on data information obtained through the application interface, to determine a prediction model that meets the prediction requirements. This allows for optimization and adjustment of the prediction model, supplemented by expert experience and the analytical judgment of operational personnel.

[0020] In one possible implementation, determining the predictive model that meets the prediction requirements includes: displaying the rate of change of the predicted data for each node affected by the adjustment through the application interface, and determining the predictive model that meets the prediction requirements based on the predicted data of the predicted object expressed by each node in the prediction dimension. In this way, the logical judgment and calculation process of the predictive model linked to the instance data can be intuitively displayed, the prediction results can be shown, and a predictive model that meets the prediction requirements can be obtained based on the process and the prediction results.

[0021] In one possible implementation, the method further includes: generating a first prediction model based on financial or business information obtained through the application interface; configuring the public attributes of the first prediction model, the attributes of the first prediction path, and the personalized attributes of the first node based on the obtained first instantiation configuration information, and determining a first calculation rule;

[0022] A second prediction model is generated based on financial or business information obtained through the application interface; the public attributes of the second prediction model, the attributes of the second prediction path, and the personalized attributes of the second node are configured based on the obtained second instantiation configuration information, and a second calculation rule is determined; the linkage relationship between the first calculation rule and the second calculation rule is configured; the semantic parsing engine converts the first and second prediction models, the first and second instantiation configuration information, and the linkage relationship between the first and second calculation rules into computer-executable second parsing code; the second parsing code is used to connect the first and second prediction models in an orderly manner according to the linkage relationship between the first and second calculation rules, generating a prediction job flow. In this way, multiple prediction models can be integrated to construct a prediction job flow, enabling collaborative operations between multiple nodes in a complex prediction job flow.

[0023] In one possible implementation, the first prediction model is configured with a second and / or a third calculation rule, and the second prediction model is configured with a first and / or a fourth calculation rule. This allows for the configuration of one or more prediction rules based on the complexity of the prediction model for each node.

[0024] In one possible implementation, generating a predictive model based on financial or business information obtained through the application interface includes connecting the predictive model to the output interface of a trained algorithm model, invoking the algorithm model in real time, and combining it with the financial or business information obtained through the application interface to generate the predictive model. This allows for more efficient learning of knowledge from predictive data, thereby improving the efficiency of algorithm learning and enhancing the quality of algorithmic predictions using deterministic knowledge.

[0025] In one possible implementation, the method further includes configuring permissions for the prediction model, the permissions including permissions for baseline nodes in the prediction model, permissions for each prediction path in the prediction model, and permissions for the prediction results corresponding to the prediction paths. This can meet the specific requirements of timeliness in financial forecasting, flexibly implement time-segmented operation permission control, and provide a permission configuration unit that can be defined on demand for complex permission configurations.

[0026] For a complex enterprise with multiple product lines and subsidiaries, the embodiments of this application can quickly and efficiently reuse the knowledge and data accumulated in the forecasting process, and can flexibly integrate expert experience into the forecasting and analysis business, and verify the expert experience to realize the engineering of financial forecasting. Attached Figure Description

[0027] To more clearly illustrate the technical solutions of the various embodiments disclosed in this specification, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only a few embodiments disclosed in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] The accompanying drawings used in the description of the embodiments or prior art are briefly introduced below.

[0029] Figure 1 A system architecture diagram of an engineering predictive analysis method provided for embodiments of this application;

[0030] Figure 2 A schematic diagram of a layered decoupled architecture for an engineering predictive analysis method provided for an embodiment of this application;

[0031] Figure 3 A flowchart of an engineering predictive analysis method provided for embodiments of this application;

[0032] Figure 4a A schematic diagram of three prediction paths for the engineering predictive analysis method provided for embodiments of this application;

[0033] Figure 4b The engineering predictive analysis method provided for the embodiments of this application includes a combination graph of three types of models;

[0034] Figure 4c for Figure 4a The inverse model diagram of type ② in the middle;

[0035] Figure 5a A schematic diagram illustrating the construction of a prediction model through a user application interface for an engineering predictive analysis method provided in an embodiment of this application;

[0036] Figure 5b A schematic diagram illustrating the configuration of a prediction model instantiation via a user application interface for an engineering predictive analysis method provided in an embodiment of this application.

[0037] Figure 5c A schematic diagram illustrating the adjustment of prediction process parameters and the linkage of prediction results in the engineering predictive analysis method provided for embodiments of this application;

[0038] Figure 6a A schematic diagram illustrating the workflow of profit forecasting at a fixed point in time for the engineering predictive analysis method provided in the embodiments of this application;

[0039] Figure 6b A schematic diagram illustrating the scheduling of relevant forecast instances on a profit forecasting workflow using an engineering forecasting analysis method provided in an embodiment of this application;

[0040] Figure 7 A functional block diagram of the engineering predictive analysis method provided for embodiments of this application. Detailed Implementation

[0041] Financial indicators are quantitative metrics used by companies to summarize and evaluate their financial condition and operating results. They involve financial activities within a company's business operations and are commonly applied to various aspects of a company's financial, operational, and risk management. As a fundamental component of predictive models, financial indicators possess the formal characteristics of finance and can also serve as relational nodes connecting various aspects of a company's business activities, revealing the linkage between business activities and financial results, and reflecting the essence of a company's various economic activities. Commonly used financial indicators for companies are categorized into profit and loss indicators, operating assets, and cash flow indicators, which respectively reflect a company's profitability, solvency, and operational efficiency.

[0042] The first approach proposes a method for predicting corporate financial indicators and cash flow. This method acquires transaction data and auxiliary data from a cash center, extracts features from these data, and inputs them into multiple pre-trained individual prediction models to obtain corresponding predicted values. A non-dominated sorting genetic algorithm with an elitist strategy is then used to process the predicted values ​​from each individual prediction model to obtain the cash inventory usage prediction result. The individual prediction models can be ridge regression, categorical regression tree, extreme random regression tree, LightGBM regression, or convolutional neural network models. This approach builds models based on data and seeks patterns to make predictions; however, it lacks a solution for integrating financial and business knowledge involved in the prediction, resulting in poor interpretability of the data-driven approach.

[0043] The second approach proposes a method for generating a corporate financial indicator forecasting model. Based on an integrated approach, it assigns weights to the accuracy of historical forecasts from individual institutions or individuals, combining them to form the final result. This approach, for a predetermined time period, acquires historical forecast data generated by multiple institutions and individuals for corporate financial indicators corresponding to one or more financial products, as well as actual data for these indicators. These institutions and individuals are associated with multiple institutions. The historical and actual data are then filtered. Based on the filtered historical and actual data, the multiple institutions and individuals are scored. Based on the scoring results, the weights corresponding to the forecasts generated by the multiple institutions and individuals for each of the one or more financial products are determined when using the combined forecasting model to predict corporate financial indicators. This method comprehensively scores historical knowledge of individuals and dynamically adjusts the weights of the combined forecasting model by integrating the forecasts published by different institutions and individuals to more accurately predict corporate financial indicators. However, this approach only synthesizes the forecast results and does not involve the integration of different knowledge and experiences.

[0044] When building a financial forecasting system, enterprises need to integrate large amounts of data from multiple systems, design and develop algorithmic models that can effectively capture patterns, transform business logic and financial rules into executable code, apply it to the integrated data, and supplement it with expert experience and operational staff analysis and judgment to ultimately conduct data forecasting and analysis. This process involves multiple professional knowledge and IT skills across various fields of enterprise business, financial management, and IT, and needs to evolve and iterate along with the development of the enterprise's own business and changes in the external environment. The enterprise's financial forecasting chain is complex and ever-changing. If each link cannot accumulate knowledge and experience, and if the linkage between links cannot be timely applied and analyzed, it will be unable to meet the increasingly sophisticated management needs of enterprises.

[0045] The following describes the engineering predictive analysis method provided by the embodiments of this application.

[0046] Financial forecasting, as a key tool in enterprise management, permeates every important link in the enterprise's value chain. Within an enterprise, forecasting can be conducted in several ways: ① Identifying the leading indicator and then forecasting based on the complex logical relationship between the leading indicator and business or financial activities; ② Forecasting separately for each sub-department / product and then aggregating the results; ③ Using historical data and big data technologies such as statistical analysis and machine learning to create algorithmic models for forecasting. In practical applications of enterprise forecasting, these methods can be used to cross-verify the results. Based on these forecasting methods, a forecast requires cross-system integration of relevant data on enterprise business activities and financial status, clarifying business logic and financial rules that conform to the actual business operations, with IT personnel implementing the corresponding parsing code or building algorithmic models based on the data for engineered forecasting analysis. In practice, the final forecast results are further refined through expert judgment or adjustments before reaching a definitive conclusion.

[0047] The engineering predictive analysis method provided in the embodiments of this application adopts a layered decoupling model, decoupling the prediction rules from the prediction objects of the instances, allowing the prediction rules to be reused. Prediction rules refer to the logical judgments used for data analysis and / or the computational rules used for data operations. Optionally, the prediction rules can also be algorithms implemented based on historical data through big data technologies such as statistical analysis and machine learning.

[0048] Decoupling is a state of IT environment where two or more models or components work or connect in some way without direct interaction. For example, there is no direct relationship between prediction rules and the objects being predicted; the same prediction rule can be applied to different instances of the objects being predicted.

[0049] In some application scenarios, the forecasting rules for complex enterprises may correspond to different instance data depending on the scope of the forecast object, but the forecasting rules used in the forecasting process are the same. For example, taking a certain revenue item of the enterprise as the forecast object, the forecasting rule is: Revenue = Product Price × Quantity, Quantity = Product Demand Orders. Here, the instance data corresponding to the forecast object "Revenue" are "Product Price," "Quantity," and "Product Demand Orders." Regardless of the product, the forecasting rule for the forecast object "Revenue" is consistent. However, when applied to different product lines, different instance data are corresponding to them. For example, different products correspond to different price data and demand order data.

[0050] The engineering predictive analysis method provided in the embodiments of this application templates the prediction rules, abstracts various types of prediction templates based on multiple prediction paths, and combines or edits the templates to build prediction models for specific business problems.

[0051] The engineering predictive analysis method provided in the embodiments of this application configures instances of various types of predictive models and generates executable code corresponding to the predictive models through configurable interactive design.

[0052] The engineering predictive analysis method provided in the embodiments of this application can support the construction of complex predictive models across types based on the above-mentioned engineering capabilities, perform engineering predictive analysis of data and knowledge with different logical judgments and / or calculation rules, and the prediction results can be mutually verified on different instances.

[0053] To flexibly adapt to different business and / or financial forecasting rules, the engineering forecasting analysis method provided in the embodiments of this application constructs a forecasting model through an interactive application interface, connects forecasting rules, data, and algorithms, and establishes and publishes various types of forecasting models. Business experts can adjust the forecasting model at any node in the forecasting process based on their business experience. The semantic parsing engine in the background generates computer-executable parsing code based on the forecasting model adjusted by the business experts, and performs relevant logical judgments and / or calculations to conduct data forecasting analysis.

[0054] In one possible implementation, business personnel first load a template expressing business and / or financial forecasting rules through an interactive application interface, constructing a forecasting model in the form of a high-order directed graph, and determining the logical judgments and calculation rules involved in the model. Then, IT personnel configure and integrate relevant instance data and forecasting dimensions at each node level based on the granular requirements of business and / or financial analysis. Next, the backend parsing engine integrates the constructed forecasting model with the configuration information input by IT personnel and converts it into computer-executable parsing code. Through the API interface, it performs corresponding calculations on the relevant database according to the logical judgments and calculation rules to perform data forecasting analysis.

[0055] For example, a template is an expression of prediction rules abstracted from the company's accumulated business and / or financial knowledge, stored in the template library, indicator library, baseline library, or algorithm library of the company's business and / or financial data prediction model management platform. The template's expression can include a high-order directed graph or a multi-layered tree diagram. Analysis granularity refers to the level of detail in the scope of business and / or financial analysis for a specific prediction scenario. For example, analysis granularity can range from coarse to fine (from large to small) or can be divided into levels such as group-subsidiary, group-subsidiary-production workshop, group-subsidiary-production workshop-product line, production workshop-product line-product model, etc.

[0056] Forecasting dimensions refer to the level of detail in the data hierarchy for business and / or financial analysis in a specific forecasting scenario. For example, forecasting dimensions are divided according to the principle of data from low to high levels, and forecasting dimensions are set and adjusted for forecasting products, forecasting customers, forecasting subjects, nodes, forecasting regions, forecasting types, and currencies.

[0057] The prediction model is based on a high-order directed graph structure. It expresses logical judgments and / or calculation rules through multiple nodes, node attributes, and pairwise relationships between nodes, and represents prediction paths through directed edges between nodes. The prediction model reflects the prediction process of making logical judgments and calculations based on instance data and can intuitively display the prediction results. The prediction model can include multiple prediction paths to achieve different logical judgments and calculations for different instance data.

[0058] The prediction path refers to the process from integrating input information to generating prediction results for a specific prediction target. For complex enterprises, this process includes defining the prediction target, integrating relevant data, building the model, and setting the prediction dimensions. Defining the prediction target requires clearly defining the scope of the enterprise's business involved, the time span of the prediction, and the frequency of the analysis.

[0059] The core of building a forecasting model is to determine the forecasting path, including forecasting paths based on the progressive relationship between indicators, forecasting paths based on convergence of enterprise architecture dimensions, and forecasting paths based on baseline transformation. In the actual application scenarios of enterprise forecasting, the modeling of forecasting models often involves the combined modeling of the above three forecasting paths.

[0060] Based on their experience, business experts can adjust and test the key nodes of the prediction model. After verification, the results are stored as knowledge for specific business and / or financial prediction scenarios. This knowledge is saved in the backend instance graph database in the form of a high-order directed graph to support the verification, search, and deduction of prediction paths.

[0061] The engineering predictive analysis method provided in the embodiments of this application offers calling and configurable integration between different predictive models through multi-knowledge representation technology.

[0062] The technical solutions in the embodiments of this application will be described below with reference to the accompanying drawings.

[0063] Figure 1 A system architecture diagram of the engineering predictive analysis method provided for embodiments of this application. (See diagram below.) Figure 1 As shown, the system architecture of the engineering predictive analysis method provided in the embodiments of this application includes an application layer 11, a data adaptation layer 12, and a knowledge management layer 13.

[0064] The application layer 11 generates a prediction model based on financial or business information obtained through the application interface. The prediction model is based on a high-order directed graph structure, expressing logical judgments and / or calculation rules through multiple nodes, the attributes of multiple nodes, and the pairwise relationships between multiple nodes. The prediction path is expressed through the directed edges between nodes. The prediction results of the prediction object are displayed.

[0065] For example, different prediction objects, prediction step size / frequency, prediction dimensions, and analysis granularity can be defined for each node.

[0066] Data adaptation layer 12 is used to adapt prediction rules and data for application scenarios. It employs semantic parsing engine technology to perform formatted configuration transformation on the data of prediction objects, relationships between prediction objects, and corresponding logical judgments and calculation rules. This formatted configuration transformation can be performed online or offline.

[0067] For example, semantic parsing engine technology can be used to perform formatting configuration and transformation online for remotely invoked prediction models and / or prediction instances. It can also be used offline for locally invoked prediction models and / or prediction instances.

[0068] Optionally, the semantic parsing engine can be JSON (JavaScript Object Notation). JSON is a lightweight data exchange language that uses a text format that is completely independent of programming languages. It also uses C-like language conventions, including C, C++, C#, Java, JavaScript, Perl, Python, etc., which makes it easier for the processor to format and configure the data of nodes, the relationships between nodes, and the corresponding logical judgments and calculation rules.

[0069] Optionally, the semantic parsing engine can be XML (Extensible Markup Language). XML is a cross-platform, content-dependent technology in the Internet environment and a powerful tool for processing structured document information. XML uses a series of simple tags to format and configure data, including nodes, relationships between nodes, and corresponding logical judgments and calculation rules.

[0070] It should be noted that business or financial templates formatted using IT technologies such as JSON and XML are easy for machines to parse and generate, and support cross-platform calls in the Internet environment.

[0071] The knowledge management layer 13 includes a graph library 131, a configuration database 132, and a business database 133, which are used to store and manage graphs, configuration data, and business data.

[0072] Specifically, the graph library 131 is used to store the prediction model configured and transformed by the data adaptation layer 12, including a high-order directed graph and multiple formatted nodes on the graph, the attributes of the multiple nodes and the pairwise relationships between the multiple nodes, and to perform knowledge management, providing template support for relation mining of prediction objects, prediction model construction, knowledge reasoning, conflict detection and correlation calculation.

[0073] Database 132 is configured to store and manage instantiated configuration data, transformed by the data adaptation layer 12, in relational (Oracle) format. This includes dimension data, metadata, and data related to defining prediction models and prediction instances. It contains the types of prediction nodes, common attributes of the prediction model, attributes of the prediction path, and personalized attributes of the nodes. The common attributes of the prediction model are used to set the connection relationships between the prediction model and the data source, input tables, and output tables. The attributes of the prediction path are used to set the procedural relationships between logical judgments and / or calculation rules and the data. The personalized attributes of the nodes are used to set the prediction object and prediction dimension expressed by each node.

[0074] For example, configuration data such as prediction objects, node dimensions, version management, node configuration information, path configuration information, and path integration relationships can be defined in the application layer 11. After configuration transformation in the configuration data adaptation layer 12, the configuration data is stored in the configuration database 132. The background parsing engine converts the configuration data into computer-executable parsing code and connects to the business database 133 via an API interface.

[0075] Business database 133 is used to store and manage enterprise financial or business data sources in the form of relational data (Oracle). This data provides relevant data for instantiation calculations such as cash forecasting, group forecasting, corporate forecasting, tax forecasting, and BG forecasting. It includes input tables consisting of data from the data sources and output tables consisting of forecast result data.

[0076] By executing the above parsing code on the relevant data source through the API interface, the prediction model can be engineered to connect the prediction model with the input and output tables of the data source according to the common attributes of the prediction model; on the input table of the data source, the logical judgment and calculation rules expressed by the prediction model are executed according to the prediction path order; the prediction data of the prediction object expressed by each node in the prediction dimension are determined, and the prediction result data is output to the output table or displayed on the application interface of application layer 11, and data services for optimizing and adjusting the prediction model are provided.

[0077] For example, data sources include BG databases, group databases, tax databases, cash databases, legal entity databases, etc.

[0078] For example, the forecast objects of different forecast dimensions include BG forecasts, group forecasts, tax forecasts, cash flow forecasts, and corporate forecasts.

[0079] For example, the output tables include BG forecast output tables, group forecast output tables, tax forecast output tables, cash flow forecast output tables, legal entity forecast output tables, etc.

[0080] The embodiments of this application provide a layered decoupling design pattern, which decouples the template (application layer) from the prediction rules (graph library 131), the execution code of the prediction rules (graph library 131) (configuration database 132) from the database (business database 133), and the automated generation of prediction paths (data adaptation layer 12) from adjustments made by operators. This decoupling design allows the template to be reused during the construction of the prediction model, and model builders can adjust it based on existing models, greatly improving work efficiency. The decoupling design also supports the parsing code of the prediction model being called by different prediction applications and executed in the corresponding databases.

[0081] Figure 2 The engineering predictive analysis method provided for embodiments of this application is based on a layered decoupled architecture using multi-knowledge representation technology. (See diagram for example.) Figure 2 As shown, this method uses multi-knowledge representation technology to logically configure the prediction rules and paths expressed by the prediction model, integrates instance data to generate model instances, and allows business experts to review the prediction paths and results. Based on the experts' experience, the parameters of the model instances are adjusted, and the backend parsing engine converts this into computer-executable parsing code. The adjusted prediction model is then executed on the data source, and the prediction results, the adjusted instance model, and the adjusted logical configuration are displayed and saved. This method achieves connectivity between different prediction rules, instance data, prediction models, and prediction objects, improving the operability of prediction model generation, the interpretability of the prediction process, and the rapid generalization ability of the prediction model.

[0082] Figure 3 A flowchart illustrating the method based on engineering predictive analysis provided for embodiments of this application. Figure 3 As shown, the specific steps of the engineering predictive analysis method provided in this application embodiment include:

[0083] S301 obtains a template through the application interface, edits multiple nodes based on the template, and defines the prediction dimension of each node, the relationship between nodes, and the corresponding logical judgment and calculation rules.

[0084] In one possible implementation, S301 is specifically implemented by performing the following steps S3011-S3012.

[0085] S3011 obtains the template loaded by the user through the user application interface via the application interface.

[0086] For example, the processor loads templates from the prediction template library, indicator library, baseline library, and algorithm library based on user input in the application interface. For example, financial indicator templates and / or business indicator templates are loaded from the indicator library; business baseline templates, business conversion templates, and / or financial baseline templates are loaded from the baseline library; and statistical learning templates, machine learning templates, and / or deep learning templates are loaded from the algorithm library.

[0087] The template library is set in the graph library 131, and the index library, baseline library and algorithm library are set in the configuration database 132.

[0088] S3012 obtains the configuration information of each node on the template based on the user's input data, determines the business scope, prediction object, prediction step size / frequency and prediction dimension of each node based on the configuration information, and configures the relationship between nodes and nodes in multiple nodes as well as the corresponding logical judgment and calculation rules.

[0089] For example, the processor obtains configuration information such as nodes, relationships between nodes, and corresponding logical judgments and calculation rules defined by the user through the application interface, and constructs a prediction model based on the configuration information. The defined nodes include the business scope, prediction object, prediction step size / frequency, and prediction dimension of each node.

[0090] S302 initializes parameters based on template configuration, configures the attributes of multiple nodes, configures data source link information, determines the prediction path, and generates a prediction model; the semantic parsing engine translates the prediction path into computer-executable parsing code and links it to the data source.

[0091] For example, the public attributes of the prediction model, the attributes of the prediction path, and the personalized attributes of the nodes can be configured according to the instantiation configuration information.

[0092] The semantic parsing engine transforms the prediction model and instantiation configuration information into first-order parsing code that can be executed by a computer.

[0093] In one possible implementation, S302 is specifically achieved by performing the following steps S3021-S3023.

[0094] S3021, Configure initialization parameters based on user input data. For example, initialization parameters include initial values ​​for information such as prediction step size / frequency, prediction dimension, and analysis granularity.

[0095] S3022, Based on user input data, configure personalized attributes for multiple nodes. For example, the personalized attributes of a node include at least setting the prediction object expressed by each node, the prediction dimension, the node's adjustment enabling attributes, and information constraint attributes.

[0096] S3023, Configure data source link information, determine the prediction path, and generate a prediction model. For example, the data source link information is a public attribute of the prediction model, including setting the connection relationships between the model and the data source, input table, and output table; determining the prediction path includes determining the correspondence between logical judgments and / or calculation rules and the data.

[0097] S3024, the parsing engine parses the prediction model into computer-executable parsing code, and establishes a link between the computer-executable parsing code and the data source through an API interface. For example, the computer-executable parsing code establishes links to multi-dimensional data sources such as financial databases, business databases, baseline databases, dimensional databases, and external databases.

[0098] For example, the parsing code is used to engineer the connection between the prediction model and the data source, input table, and output table based on the common properties of the prediction model.

[0099] S303 runs computer-executable parsing code on the data source to perform checksum calculations on the predicted path and conduct data prediction analysis.

[0100] In one possible implementation, the processor can verify and parse the configured prediction model through the processor's parsing engine to generate executable parsing code; connect the computer-executable parsing code to the corresponding data source through an API; and execute the parsing code on the data source for the prediction object to perform data prediction analysis.

[0101] For example, when the processor runs computer-executable parsing code on the input table of the data source, it executes the logical judgments and calculation rules expressed by the prediction model according to the prediction path; it determines the prediction data of the prediction object expressed by each node in the prediction dimension, and outputs the prediction data to the output table or displays it on the application interface. S304, the prediction results are reviewed, node attributes and prediction paths are adjusted, and the prediction model is optimized.

[0102] In one possible implementation, S304 is specifically achieved by performing the following steps:

[0103] S3041: The user reviews the prediction results displayed on the application interface using their professional knowledge. If the review result indicates that the prediction results are unreasonable, proceed to S3042. If the review result indicates that the prediction results are reasonable, the process ends.

[0104] S3042, adjust the configuration information of the prediction model, generate computer-executable parsing code from the parsing engine, input the computer-executable parsing code into the data source through the API interface; execute the parsing code on the data source, and perform data prediction analysis for prediction objects of different prediction dimensions.

[0105] For example, adjusting the configuration information of the prediction model includes adjusting one or more of the information values ​​such as prediction step size / frequency, prediction dimension and / or analysis granularity.

[0106] For example, adjusting the configuration information of the prediction model includes adjusting the node prediction objects, the node enable adjustment attributes, and / or configuration information restriction attributes.

[0107] For example, adjusting the configuration information of the prediction model includes adjusting the prediction path, resetting the connection relationship between the data source, input table and output table; and reconfiguring the correspondence between logical judgment and / or calculation rules and data.

[0108] S3043: The user reviews and verifies the prediction results displayed on the application interface using their professional knowledge. If the review result indicates that the prediction result is unreasonable, proceed to S301. If the review result indicates that the prediction result is reasonable, the process ends.

[0109] S305, Save the prediction model.

[0110] S306 enables service-oriented invocation of prediction models, allowing for the combination of multiple prediction models or the combination of prediction models with other applications.

[0111] In one possible implementation, the predicted path includes a first type of path, which is a node relationship class. This path expresses the association between nodes under a certain dimension configuration. It is generally a bottom-up, progressively layered directed graph structure composed of multiple nodes, where the same node can be associated with multiple other nodes at different levels. For example... Figure 4a As shown in type ①, this directed graph structure is based on financial forecasting rules. Its lower-level nodes are indicators related to business activities, such as positive and negative revenue; the upper-level nodes are indicators related to financial management, such as equipment revenue data, service revenue data, and other revenue data. The positive and negative revenue data are used to obtain equipment revenue data. Further, the net sales revenue forecast is obtained by summing the equipment revenue data, service revenue data, and other revenue data.

[0112] In one possible implementation, the prediction path includes a second type of path, which is a dimensional convergence type. This path represents a bottom-up convergence of relationships between nodes based on dimensions such as company business classification, customer management, or organizational structure. It is typically a multi-branch tree structure with multiple levels of nodes, which can be parsed from dimensional data related to the company's products, customers, and organization. For example... Figure 4a As shown in type ②, the company's equipment revenue forecast is based on product lines. The revenue forecast data of each product line is converged into the company's equipment revenue forecast. For example, the equipment revenue forecast data of product line A, product line B, and product line C are summed to obtain the company's equipment revenue forecast result.

[0113] In one possible implementation, the prediction path includes a third type of path, which is a baseline transformation class. This path represents the transformation relationship between indicators that integrates baseline information, and is often a binary or multi-branch tree structure. For example, a multi-branch tree structure is a multi-level data structure with hierarchical transformations, and can be formed by multiple transformations of a single-level tree structure. Figure 4a As shown in type ③, this type is characterized by equipment cost being converted into equipment revenue through the gross margin baseline.

[0114] In one possible implementation, the predictive model can be based on a combination and overlay of multiple types of predictive paths. For example, the generation of a predictive model for corporate-level financial management needs involves the overlay and combination of multiple different types of predictive paths. Further, as... Figure 4b As shown, the net sales revenue forecasting model integrates three different types of forecasting paths: node relationship type, dimensional convergence type, and baseline transformation type. Type ① shows the forecasting path based on indicator relationships. In the figure, the net sales revenue indicator is directly related to financial indicators such as equipment revenue, service revenue, and other revenue, and indirectly related to various lower-level business indicators. Type ② shows the bottom-up forecasting path for the service revenue indicator, where the forecasting dimensions converge from low to high. This path, where the forecasting dimensions converge from fine to coarse, is reflected in the convergence of service revenue forecasts from multiple product lines to the company's service revenue. Type ③ shows the superposition of multiple baseline transformation models. In the figure, equipment material costs are transformed into equipment costs through the material cost conversion rate baseline, and equipment costs are transformed into equipment revenue through the gross profit margin baseline.

[0115] In one possible implementation, such as Figure 4cAs shown, the prediction model can be a reverse tree structure for multidimensional analysis. Based on the configuration in the existing prediction path ②, it can be adjusted and derived accordingly to solve the decision problem of resource allocation in the reverse of the same type ②. Furthermore, by adjusting the configuration of the intermediate prediction path and the key information, it can effectively provide a deductive model for this type of decision problem and provide quantitative basis for different decision schemes.

[0116] In one possible implementation, the different prediction paths mentioned above are configured accordingly, relevant business and / or financial configuration information is matched according to the different prediction paths, a prediction model is constructed, and the prediction paths and business configuration information are parsed into development code for calculation.

[0117] In one possible implementation, templates for data configuration and rule configuration are provided for the abstracted prediction paths, translating each type of prediction path into computer code that can be executed in the corresponding database.

[0118] One of the common management decision-making problems in the daily business activities of enterprises is the resource allocation problem. This problem generally begins with a node of the total amount of limited resources, and then, based on the business scenario of the company for that resource at a certain management level, relevant constraints and allocation principles are defined to allocate the limited resources to multiple nodes at the next management level.

[0119] For example, based on the definitions of the three financial statements, a general prediction path is stored as a template in the graph library 131 of this application in the form of a high-order directed graph. The prediction object represented by the upper-level nodes of the high-order directed graph can be set as an icon in the prediction node type selection box of the application interface. In different prediction scenarios, the user can drag and drop this icon to call the basic template.

[0120] Figures 5a-5c The diagram illustrates the application interface of the application layer of the engineering predictive analysis method.

[0121] in, Figure 5a A schematic diagram of the application interface for building a prediction model; such as Figure 5a As shown, users can configure and build templates. In one possible implementation, business personnel can click the indicator or baseline button in the prediction node area 51 to search the system-integrated business and financial indicator library and baseline library. They can then drag and drop indicator or baseline nodes to the intermediate model editing area 52 and connect different nodes using the mouse. The prediction path, expressed by directed arrows, shows the node relationship between parent and child nodes, with the arrow pointing to the parent node, thus generating a template. Business personnel can then configure the node level and template level based on the built template.

[0122] The pairwise relationships between multiple nodes simultaneously express logical judgments and / or calculation rules. For example, in a prediction path, if there is a calculation relationship between nodes, the starting node of a directed edge is on the right side of the equation, and the starting node serves as the input for the calculation rule; the ending node pointed to by the directed edge is on the left side of the equation, and the ending node serves as the output for the calculation rule.

[0123] For example, the bottom node in a template represented by a high-order directed graph is called a leaf node. The template also includes intermediate nodes and a parent node (leading indicator).

[0124] For example, users can configure the node attributes and public attributes of the template in the attribute configuration area 53.

[0125] For example, in the prediction model area 52, the user double-clicks a node to configure its attributes, including defining the node, configuring node attributes, and setting algorithms between nodes, such as logical judgment and numerical calculation algorithms. Defining a node includes defining the prediction object, step size / frequency, dimension, and analysis granularity.

[0126] Node attributes define the prediction objects for that node. For example, configuring node attributes includes configuration adjustment attributes, computation order attributes, node type, prediction constraints, etc.

[0127] For example, configuring adjustment attributes includes setting the node's business scope, prediction object, prediction step size / frequency, and prediction dimension to an adjustable or non-adjustable mode.

[0128] The configuration calculation order attribute includes configuring the order of logical judgments and the order of calculations.

[0129] The types of configuration nodes include configuring computational nodes, conditional judgment nodes, information nodes, or algorithm nodes.

[0130] For example, the relationship between nodes can be many-to-many, meaning that a node can be the input of multiple upper-level nodes at the same time.

[0131] For example, a template can be a high-order directed graph consisting of 2-3 layers of nodes and directed edges between nodes; or it can be a mesh structure with directed connections between nodes, with a multi-branch tree structure being a relatively simple template.

[0132] For example, the configuration of public attributes includes the template's application attributes and template ID, etc. Application attributes include financial statement attributes, business scope, etc., and the ID is an encoding that represents the target's location and identity on the computer. The financial statement includes the balance sheet, income statement, or cash flow statement.

[0133] Users can save high-order directed graphs as templates for generalization in future forecasting scenarios, and can also apply existing templates to newly generated business data, forming a virtuous cycle of iterative knowledge for corporate financial forecasting.

[0134] In one possible implementation, the template can be a system-hosted template. Business personnel can enter keywords in the dialog box in the prediction node area 51 to search for system-hosted templates, import existing templates into the prediction model area 52 with one click, and then modify them based on the application scenario of business or financial operations, performing addition, deletion, modification, and query at the node level.

[0135] Users can also edit the template, configuring the node attributes and common attributes of the template in the attribute configuration area 53.

[0136] exist Figure 5a The application interface has a button that allows you to enter... Figure 5b The interface shown and Figure 5c The interface shown. Clicking the "Instantiate" button on the application interface will take you to... Figure 5b Configure the instantiation of the prediction model. Figure 5b A schematic diagram of the user application interface configured for instantiating the prediction model; such as Figure 5b As shown, users configure template instantiation to establish a connection between the template and instance data. IT engineers can input the relevant information required for instantiation through the configuration box in the attribute configuration area 53 to configure instance at the model level and node level. The semantic parsing engine in the background then converts this information into executable parsing code, such as SQL statements, which can be executed by a computer.

[0137] For example, the attribute configuration area 53 displays the instance configuration information of the template's model level and the configuration information of the node instances.

[0138] For example, users can configure model-level instances in the attribute configuration area 53, including configuration of public attribute instantiation, such as the data source ID, input table ID, output table ID, visualization configuration, prediction object ID, and dimension measure.

[0139] For example, a user can double-click a node in the prediction model area 52 and enter the information configuration and data retrieval conditions for each node in the attribute configuration area 53.

[0140] For example, a user double-clicks a node in the prediction model area 52 and enters the node instance's attributes in the attribute configuration area 53, including basic configuration information, calculation logic, SQL automatic generation vs. SQL manual editing buttons, attribute adjustment, and associated attributes.

[0141] For example, configure basic information about the node instance, such as the node instance name, ID, and encoding.

[0142] For example, the calculation logic can be configured such as Revenue = Positive Revenue + DOS Deferred Amount, etc.

[0143] For example, a button can be set up to select whether the SQL is generated automatically or manually.

[0144] For example, configure node adjustment attributes, such as whether the node is adjustable, whether there are adjustment restrictions, and restriction conditions.

[0145] For example, configure node association attributes, such as whether to associate them with other input tables or prediction models.

[0146] The instantiation and configuration process of the prediction model firstly enables the reuse of the same template on different databases and converges to different prediction dimensions for different prediction objectives; secondly, it enables the editing of the prediction model, supporting the rapid generalization of the prediction model in similar prediction scenarios and avoiding the repeated development of reusable prediction models.

[0147] For example, IT engineers can input common attributes and personalized node-level attributes involved in instantiation through the configuration box in attribute configuration area 53. Double-clicking a specific node in prediction model area 52 will open a configuration box in prediction object attribute area 53, where the node instantiation configuration can be performed based on the data entered in the configuration box, and the backend will combine it with... Figure 5a The templates defined in the code are converted by the parsing engine into executable parsing code. For nodes that require review or adjustment by IT personnel, a double-click button can be used to verify the prediction path and prediction dimension information for that node, and the prediction results will be output to the output table.

[0148] In one possible implementation, the prediction model is instantiated and configured by the user following the steps S601-S603:

[0149] S601, select a leaf node in the prediction model area 52, and configure the link information of the data source corresponding to the leaf node in the configuration box of the prediction object attribute area 53.

[0150] S602, Select an intermediate child node in the prediction model area 52, configure the relevant attributes of the level in the configuration box of the prediction object attribute area 53, including public attributes and personalized attributes of the node level, configure the prediction dimension of the node; generate the prediction model.

[0151] S603, Examine the prediction model and verify the prediction results and prediction dimensions at the nodes.

[0152] In one possible implementation, users can configure multidimensional data sources during the instantiation of the predictive model to connect to databases from different enterprises.

[0153] After configuring the database connection and prediction dimensions, the prediction model and its instantiation configuration information are parsed into executable code by the backend parsing engine.

[0154] The embodiments of this application decouple the construction and instantiation configuration processes of the prediction model, allowing different personnel to perform these processes. Figure 5a and Figure 5b The interface allows for the construction and instantiation configuration of the prediction model, while the semantic parsing engine implements the engineering connection between the prediction model and the data source.

[0155] The method provided in the embodiments of this application provides matching configuration information based on the deployment environment of enterprise data to achieve data source connection between different nodes.

[0156] Users can Figure 5b In the prediction model section (52), examine the prediction model to ensure that the logical judgments and calculation rules expressed in the prediction model are consistent with the computer-executable parsing code. Furthermore, for nodes in the prediction model that require special handling, IT engineers can double-click the node and adjust it in the edit box. Adjustments include adding new nodes, deleting nodes, modifying parameters, and checking configurations. Figure 5b A button is provided on the user application interface to access... Figure 5c and Figure 5a The interface shown is accessible by clicking the "Visualization" button on the application interface. Figure 5c The predictive model is reviewed and adjusted based on the user application interface.

[0157] Figure 5c The user interface diagram was adjusted to reflect the predicted results; Figure 5c The user application interface includes a "Submit" button; clicking this button displays the prediction and analysis results. For example, clicking... Figure 5c After the "Submit" button is set on the application interface, the relevant data sources are connected based on the prediction model, and data prediction analysis is performed at the configured prediction dimension level.

[0158] like Figure 5cAs shown, the prediction model area 52 displays the input data table information for leaf nodes, the calculation process information for intermediate nodes, and the prediction result information for parent nodes. It also displays the impact information of node adjustments. Taking the positive revenue prediction model in the diagram as an example, adjusting the average daily sales from 1199 to 1299, an increase of 8%, makes the impact path of this adjustment immediately visible in the prediction model area 52. Simultaneously, the prediction model area 52 displays the rate of change of relevant nodes affected by the average daily sales, with upward arrows indicating the increase rate and downward arrows indicating the decrease rate.

[0159] For example, in the attribute configuration area 53, the instance name can be displayed, such as **positive income; the instance result dimension filtering information can be displayed, such as the dimension values ​​common to all nodes in the model, including product, date, version, region, etc.; the node level dimension filtering information can also be displayed, such as the product, date, version, region, etc. of nodes that need to be configured with personalized dimension values.

[0160] For example, users can adjust the logical judgment and calculation rules in the attribute configuration area 53, and connect to data calculation in real time to display the impact model of rule adjustment and the change rate of affected nodes.

[0161] For example, when a user adjusts the common dimension values ​​of all nodes in the model in the attribute configuration area 53, and connects to the data for calculation in real time, the model affected by the adjustment of the common dimension values ​​and the rate of change of the affected nodes will be displayed in the prediction model area 52.

[0162] For example, after selecting a node that requires personalized configuration of dimension values, the dimension values ​​are filtered based on the situation of that node, and the data is connected for calculation in real time. The model affected by the personalized configuration dimension value adjustment and the rate of change of the affected nodes will be displayed in the prediction model area 52.

[0163] The engineering predictive analysis method provided in the embodiments of this application constructs predictive models by type, adapts to interactive configuration design and semantic parsing engine, supports the construction of complex predictive models across types, and supports multi-instance prediction verification.

[0164] In one possible implementation, the visualization of the predicted path and data is achieved by executing S701-S703, with the specific process as follows:

[0165] S701, Adjust the analysis granularity of the nodes based on the prediction results. For example, the analysis granularity can be a common attribute of all nodes in the model, or it can be a node that needs to be customized.

[0166] S702, Select nodes with adjustment attributes to adjust dimension values, and perform hypothesis testing based on the link between the prediction model and the data source.

[0167] S703 sets the indicator of the impact of node adjustment, and at the same time obtains the rate of change of the predicted data after the adjustment of each node on the model.

[0168] S704, adjust the prediction model based on the rate of change of the predicted data, perform extrapolation and verification on the uncertain prediction model, and obtain the ideal prediction model.

[0169] For example, if the prediction result increases positively after node adjustment, the impact indicator can be set to display as an up arrow; if the prediction result decreases negatively after node adjustment, the impact indicator can be set to display as a down arrow.

[0170] This scheme enables forecasting operators to adjust the forecast path in real time based on the forecast results and connect to the data source for trial calculations, thereby obtaining the impact process of the forecast path adjustment and the rate of change of information of the affected nodes.

[0171] In one possible implementation, forecasting personnel build a forecasting model for uncertain forecasting scenarios, perform hypothesis testing based on the link between the forecasting model and the data source, adjust the forecasting model according to the forecasting results of the test, and deduce and verify uncertain forecasting paths to obtain an ideal forecasting model.

[0172] This solution links relevant data sources based on instance configuration information. Using graph query technology, it provides real-time feedback on all subgraphs affected by the prediction model adjustment, quantifies the rate of information change for each affected node, and quantitatively analyzes the impact of the prediction model adjustment. The impact of the prediction model is obtained by quantifying changes in local prediction paths.

[0173] The decoupling between prediction rules and data enables the division of labor among different professions, while the process-oriented prediction path and data linkage enable efficient collaborative work among different professionals.

[0174] The embodiments of this application utilize a configuration application interface based on a high-order directed graph and a process visualization-based exploratory analysis of the prediction path. After the prediction model is initialized and configured, business personnel can independently complete the entire process of discovery, exploration, and reasoning without the assistance of IT professionals. Graph data retrieval is faster, capable of calculating the attribute distribution of massive prediction objects, and achieving results returned in seconds. This supports real-time decision-making by prediction business personnel, enabling agile responses to business changes or sudden market events. Financial prediction knowledge extracted from massive amounts of data, combined with the accumulated experience and judgment of business experts, forms an application verification, solidifying it into new knowledge stored in the graph library.

[0175] The engineering predictive analysis method provided in the embodiments of this application can realize collaborative operation among multiple nodes in a complex predictive workflow by orderly connecting multiple predictive paths on the predictive model and configuring related attributes.

[0176] In one feasible approach, constructing a prediction job flow based on the prediction model includes: a prediction model A and at least one prediction model B; configuring a first prediction rule and first configuration information for the initiation and generation of the first prediction rule on prediction model A; the first configuration information includes a first initiation condition, a first database, and a linkage relationship with a second prediction rule; configuring a second prediction rule and second configuration information for the initiation and generation of the second prediction rule on prediction model B; the second configuration information includes a second initiation condition, a second database, and a linkage relationship with the first prediction rule; the first prediction model and the second prediction model are connected in an orderly manner according to the linkage relationship between the first prediction rule and the second prediction rule to form a prediction job flow.

[0177] Taking profit forecasting, a common practice in business management, as an example, the routine profit forecasting workflow at fixed points in time is as follows: Figure 6a As shown, the first forecasting line, which consists of revenue forecasting nodes, cost forecasting nodes, gross sales forecasting nodes, and other expense forecasting nodes, is connected to... Figure 6b The marketing expense forecasting workflow operates in parallel, and the two are ultimately aggregated to form the profit forecast. This forecasting workflow can express revenue forecasting nodes through a revenue forecasting model, cost forecasting nodes through a cost forecasting model, gross sales forecasting nodes through a gross sales forecasting model, and other expense forecasting nodes through other expense forecasting models. Each forecasting node / model is configured with relevant forecasting rules and related configuration information for the initiation and generation of these rules, such as initiation conditions, relevant databases, and the linkage relationships between other related forecasting rules.

[0178] Based on the complexity of the prediction model representing each node, one or more prediction rules can be configured. For example, a "revenue prediction" node typically has two prediction rules: positive revenue prediction and negative revenue prediction. Based on the complexity of the prediction model representing each node, one or more prediction instances can be configured. For example, a "revenue prediction" node typically has two prediction instances: positive revenue prediction and negative revenue prediction. The negative revenue prediction can be further divided into multiple sub-instances based on the complexity of the enterprise's negative revenue prediction. Besides the mutual invocation of prediction rules between nodes, multiple prediction rules on the same node can also be configured to collaborate. For example, the positive revenue amount in positive revenue prediction can be used as input information for negative revenue prediction. The dependency between these two prediction rules can be implemented through the configuration of the prediction rule hierarchy.

[0179] The decoupled design effectively avoids the interference between multiple versions of the prediction job. As the initial automatically generated version, it is independent of the subsequent adjustments of multiple jobs. Incremental scheduling is performed by identifying relevant nodes on the prediction path, and prediction results are quickly generated based on the adjustment job.

[0180] Financial forecasting often uses algorithms such as statistical learning or machine learning to capture patterns from historical data and then uses these patterns to predict future changes.

[0181] In the engineering predictive analysis method provided in the embodiments of this application, the predictive model can be connected to the results of algorithm prediction, that is, the results of algorithm prediction are used as data input to the predictive model as leaf nodes; or the predictive model can be directly connected to the output interface of a trained algorithm model, calling the algorithm in real time for prediction, and combined with the latest business or financial data input from the system integration to finally perform data predictive analysis. In the process of applying the algorithm to make predictions for a certain prediction object, the predictive model composed of prediction rules provides effective assistance to the algorithm, enabling it to learn knowledge from the prediction data more efficiently; the predictive model stored in the form of graph data can integrate business logic, financial rules, and expert experience into the steps of building the algorithm, such as abnormal data identification, feature engineering, information adjustment, and segmentation of prediction scenarios, thereby improving the efficiency of algorithm learning and improving the quality of algorithm prediction by utilizing deterministic knowledge.

[0182] For example, using the sales forecast of a product as the underlying input node for the revenue forecast of that product, and using algorithms to predict future sales based on historical data, the sales forecast result is used as an input node for revenue forecast. By performing relationship and instance queries based on the total profit and loss calculation rule graph, the calculation rules related to sales forecast and the important business rules carried on the instances in the context of enterprise profit forecast are identified. This background knowledge is then passed back to the algorithm construction process, which can greatly improve the quality of sales forecast.

[0183] After the predictive model automatically generates a version of the forecast based on instances, business experts review the generated forecasts and, based on their professional knowledge and experience, make judgments and adjustments to the forecasts. The completion of the forecasting task involves experts from different fields successively reviewing and adjusting the forecasts within a certain forecasting time window until the final version of the forecast is submitted by the deadline.

[0184] The engineering predictive analysis method provided in the embodiments of this application can realize the predictive review and adjustment function, thereby systematically storing the adjustment information made by business experts based on a specific predictive object and work process. The adjustment information may also include the adjustment information of a predictive rule and the time of adjustment of the predictive rule.

[0185] In an implementable manner, the adjustment information can be used as the behavioral data of business experts to create a digital predictive behavior profile for them. On the one hand, the predictive behavior profile can be fed back to business experts to support their review and retrospection; on the other hand, when business experts face emergencies or enterprise policy adjustments and need to make judgments, it can provide the input of the prior model, which is integrated into the construction process of the predictive model in the form of probability distribution or automatically integrated into the prediction path of the predictive model for similar events.

[0186] In addition, the enterprise has split the prediction permissions among its various financial management modules.

[0187] The method of engineering prediction analysis provided by the embodiments of the present application, based on the special requirements of financial prediction timeliness, the prediction workflow can be flexibly configured to implement the control of job permissions in different time periods, and provides a permission configuration unit that can be defined according to needs for complex permission configurations. This permission configuration unit can be a certain baseline node in a prediction model, or a certain branch in a prediction model, or one or more prediction rules, or the entire prediction workflow.

[0188] In the embodiments of the present application, for the interactive design of connection configuration, transformation and parsing, adjustment feedback, and physical execution between different knowledge forms, a physical connection between data, knowledge, data and knowledge, and knowledge and operators is established. At the same time, it provides systematic linkage feedback during the integration process of each knowledge, and provides systematic retrospective management. On this basis, the precipitation and review of knowledge, as well as reasoning and verification, can be efficiently realized.

[0189] The embodiments of the present application provide a method of engineering prediction analysis. Based on a unified knowledge representation and multi-type prediction knowledge bases, a prediction knowledge network is formed for the consumption, integration, and accumulation of knowledge. The system architecture of the method of engineering prediction analysis provided by the embodiments of the present application builds a prediction model management platform for enterprise financial prediction, enabling the precipitation of business and financial knowledge related to prediction, reducing the dependence on human inheritance experience for enterprise financial prediction, and supporting the integrated collaborative prediction analysis operations of multiple data, multiple knowledge, and multiple roles.

[0190] [[ID=​​​A functional block diagram of the system for engineering predictive analysis provided for embodiments of this application. (See diagram below.) Figure 7 As shown, the data prediction engineering analysis apparatus of this application comprises a definition module 70, a construction module 71, an instantiation module 72, a review and adjustment module 73, and a service invocation module 74. Specifically, the definition module 70 obtains the business scope, prediction object, prediction step size / frequency, and prediction dimension defined by business experts and solution analysts based on nodes through the application interface.

[0192] The construction module 71 is used to import prediction rules in batches based on templates, or to build templates in the form of high-order directed graphs based on prediction objects, and to edit and modify the templates.

[0193] The instantiation module 72 is used to link to the data source related to the prediction object of the prediction model. According to the user's input parameters, the prediction dimension of the prediction model is configured, personalized nodes are configured, node adjustment attributes are configured, and parameter information limit thresholds are configured. The prediction model is parsed into computer-executable parsing code through a JSON and XML parsing engine and input into the linked data source through the API interface. The parsing code is executed on the data source to perform data prediction analysis for prediction objects of different prediction dimensions.

[0194] The review and adjustment module 73 is used to review the configuration information and prediction results of the nodes of the prediction model, adjust the nodes, node dimensions, node configuration information, adjust the version of the prediction model, adjust the configuration of prediction paths and rule integration relationships.

[0195] The service-oriented invocation module 74 is used to store prediction models, invoke prediction models in a service-oriented manner, and integrate multiple prediction models / models with other applications.

[0196] In a specific business scenario, business experts and solutions analysts define nodes through module 70, including the predicted business scope, metrics, step size, frequency, and dimensions. First, a prediction model is defined in module 71. Then, IT engineers configure the connection between the data source and the prediction model in module 72. The parsing engine parses the prediction model into computer-executable code, which is then run on the data source to obtain the prediction results. Next, business experts and solutions analysts review and adjust the prediction results in module 73. Finally, the service invocation module 74 saves the prediction model and makes service invocations, enabling the integration of multiple prediction models or the integration of prediction models with other applications.

[0197] The above-mentioned operable functional modules based on application engineering are supported by five components related to the IT technology of the prediction model management platform: multi-knowledge fusion component 701, algorithm component 702, technology component 703, service component 704, and management component 705.

[0198] The knowledge fusion component 701 is used for knowledge representation, knowledge extraction, knowledge matching, rule verification, knowledge reasoning, and knowledge storage based on multi-knowledge representation technology. With the support of the knowledge fusion component 701, the definition module 70 can obtain the business scope, prediction object, prediction step size / frequency, and prediction dimension prediction based on node definitions, thus realizing knowledge representation. With the support of the knowledge fusion component 701, the construction module 71 can edit and modify templates based on nodes, batch import prediction paths, and build prediction models based on high-order directed graphs, thus realizing knowledge extraction.

[0199] Algorithm component 702 is used for statistical modeling, machine learning, deep learning, transfer learning, and semantic parsing to learn from data. With the support of algorithm component 702, the review and adjustment module 73 can review the configuration information and prediction results of the nodes of the prediction model, adjust the nodes, node dimensions, node configuration information, adjust the version of the prediction model, adjust the configuration of prediction paths and rule integration relationships, and realize rule verification and relationship reasoning.

[0200] Technical component 703 is used to link database interfaces such as ADI, Jalors, MySQL, DB2, Oracle, GES, and Neo4j. With the support of technical component 703, instantiation module 72 can connect to data sources, configure prediction dimensions, configure personalized nodes, configure node adjustment attributes, configure information limit thresholds, parse the prediction model into computer-executable parsing code through the parsing engine, and calculate and obtain prediction results. Instantiation module 72 can also be configured and connected with knowledge management layer 13 to achieve knowledge matching.

[0201] Service component 704 is used for integrated scheduling, log analysis, remote configuration, expert behavior analysis, and performance comparison. With the support of service component 704, service-oriented invocation module 74 can perform service-oriented invocations on the predictive model, including integrated scheduling, log analysis, remote configuration, expert behavior analysis, and performance comparison.

[0202] The management component 705 is used for managing permissions, workflows, models, dimensions, and lifecycles for service-oriented calls to predictive models.

[0203] With the support of management component 705, service-oriented invocation module 74 can make service-oriented invocations of prediction models, and combine multiple prediction models or combine prediction models with other applications.

[0204] The embodiments of this application provide a flexible and simple human-computer interaction method for building prediction models, which can handle the complex and diverse relationships between prediction objects, build an interactive engineering system, and meet the needs of different stages in prediction, including analysis and exploration, generalization implementation, and knowledge management.

[0205] The embodiments of this application propose an engineering predictive analysis system. This system provides a predictive model management platform based on multi-knowledge representation fusion technology, expresses multiple types of predictive knowledge bases based on a unified knowledge representation method, and adopts a high-order directed graph predictive knowledge network for knowledge consumption, integration, and accumulation. This enables the systematic accumulation of business and financial knowledge related to nodes, reduces the reliance of enterprise financial forecasting on human-inherited experience, and supports integrated predictive analysis operations involving multiple data, multiple knowledge, and multiple roles.

[0206] This application provides an engineering-based predictive analysis system that employs multi-knowledge representation fusion technology to establish a structured predictive model for complex knowledge in the financial field. The predictive object, predictive path, business logic, and financial rules are stored in the predictive model using a unified knowledge method at different management and analysis levels, and these predictive models are interconnected and integrated. The advantage of this predictive model construction is that it can represent descriptive expert knowledge in business language, rule information existing in expert experience judgment, and knowledge learned from data through various algorithms in a machine-friendly and easily stored manner. Furthermore, the application interface of the engineering-based predictive analysis system provided in this application allows for interaction with the predictive user, efficiently integrating new knowledge and providing timely feedback on the impact of this knowledge on the existing predictive system, thereby effectively realizing the generation and iteration of predictive models.

[0207] Those skilled in the art will 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, or a combination of computer software and electronic hardware. 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 the embodiments of this application.

[0208] Furthermore, various aspects or features of the embodiments of this application can be implemented as methods, apparatus, or articles of manufacture using standard programming and / or engineering techniques. The term "article of manufacture" as used in this application encompasses a computer program accessible from any computer-readable device, carrier, or medium. For example, computer-readable media may include, but are not limited to: magnetic storage devices (e.g., hard disks, floppy disks, or magnetic tapes), optical discs (e.g., compact discs (CDs), digital versatile discs (DVDs), etc.), smart cards, and flash memory devices (e.g., erasable programmable read-only memory (EPROMs), cards, sticks, or key drives, etc.). Additionally, the various storage media described herein may represent one or more devices and / or other machine-readable media for storing information. The term "machine-readable medium" may include, but is not limited to, wireless channels and various other media capable of storing, containing, and / or carrying instructions and / or data.

[0209] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of this application.

[0210] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0211] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0212] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0213] If the aforementioned function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application embodiment, essentially, or the part that contributes to the prior art, or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a server, or an access network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0214] The above description is merely a specific implementation of the embodiments of this application, but the protection scope of the embodiments of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the embodiments of this application should be covered within the protection scope of the embodiments of this application.

Claims

1. An engineering-based predictive analysis method, which performs predictions based on the fusion of multiple knowledge sources from financial or business data, characterized in that: The method includes: A prediction model is generated based on financial or business information obtained through the application interface; the prediction model is based on a high-order directed graph structure, which expresses logical judgments and / or calculation rules through multiple nodes, the attributes of the multiple nodes, and the pairwise relationships between the multiple nodes, and expresses the prediction path through the directed edges between the nodes; Configure the common attributes of the prediction model, the attributes of the prediction path, and the personalized attributes of the nodes according to the instantiation configuration information; wherein, the common attributes of the prediction model are used to set the connection relationship between the prediction model and the data source, input table, and output table; the attributes of the prediction path are used to set the correspondence between the logical judgment and / or calculation rules and the data; the personalized attributes of the nodes are used to set the prediction object and prediction dimension expressed by each node. The semantic parsing engine converts the prediction model and the instantiation configuration information into first parsing code that can be executed by a computer. The first parsing code is used to engineer the prediction model with the data source, input table and output table according to the common attributes of the prediction model; on the input table of the data source, the logical judgment and calculation rules expressed by the prediction model are executed according to the prediction path; the prediction data of the prediction object expressed by each node on the prediction dimension are determined, and the prediction data is output to the output table or displayed on the application interface.

2. The method according to claim 1, characterized in that, The prediction path of the prediction model includes a node relationship path, which is a high-order directed graph in which the plurality of nodes are progressively connected from bottom to top, wherein the same node is associated with multiple nodes at different levels.

3. The method according to claim 1, characterized in that, The prediction path of the prediction model includes a dimensional convergence path, which is a multi-branch tree structure connected by the plurality of nodes based on the prediction dimensional convergence principle.

4. The method according to claim 1, characterized in that, The prediction path of the prediction model includes a baseline transformation path, which includes multiple baselines, each of which is a multi-branch tree structure composed of the multiple nodes.

5. The method according to claim 1, characterized in that, The step of generating a predictive model based on financial or business information obtained through a user application interface includes: Based on the information input through the user application interface, one or more of the prediction models are invoked.

6. The method according to claim 1, characterized in that, The step of generating a predictive model based on financial or business information obtained through a user application interface includes: Templates are retrieved from a financial or business template library based on information from the user application interface, and the structure of the template is the higher-order directed graph. Based on the template, the logical judgment and calculation rules are determined by defining the multiple nodes, configuring the attributes of the multiple nodes, and the pairwise relationships between the multiple nodes. The prediction path is determined according to the connection relationship of the directed edges between the nodes; and a prediction model is generated.

7. The method according to claim 1, characterized in that, The step of generating a predictive model based on financial or business information obtained through a user application interface includes: Multiple nodes are selected from the indicator library and / or baseline library based on information obtained through the user application interface; Based on the financial or business information obtained from the user application interface, the logical judgment and calculation rules are determined by defining the multiple nodes, configuring the attributes of the multiple nodes, and determining the pairwise relationships between the multiple nodes. The directed edges connecting the multiple nodes are used to determine the prediction path, and a prediction model is generated.

8. The method according to claim 6 or 7, characterized in that, The step of determining the logical judgment and calculation rules by defining the multiple nodes, configuring the attributes of the multiple nodes, and the pairwise relationships between the multiple nodes includes: By defining the prediction object, step size / frequency, dimension, and analysis granularity of the node, configuring the adjustment attributes, calculation order attributes, prediction object type, and prediction constraints of the node, setting the arrow direction of the directed edges between nodes, and determining the logical judgment and calculation rules.

9. The method according to claim 1, characterized in that, The personalized attributes of the node are also used to set the node's adjustment attributes, calculation order attributes, prediction object type, and prediction constraints.

10. The method according to claim 8, characterized in that, The adjustment attributes of the node include adjustable or non-adjustable modes; The computation order attribute of the node includes the order in which the node performs logical judgments or the order in which it performs calculations; The prediction object types of the nodes include calculation nodes, condition judgment nodes, information nodes, and algorithm nodes; The prediction constraints for the nodes include the range of parameter values.

11. The method according to claim 1, characterized in that, The step of configuring the public attributes of the prediction model based on the instantiation configuration information includes: Configure the connection relationships between the prediction model and the local data source, input table, and output table according to the instantiation configuration information; or The connection relationships between the prediction model and the remote call data source, input table, and output table are set according to the instantiation configuration information.

12. The method according to claim 1, characterized in that, The step of converting the prediction model and the instantiation configuration information into computer-executable first parsing code by the semantic parsing engine includes: The prediction model and the instantiation configuration information are converted into computer-executable first parsing code using a JSON or XML semantic parsing engine.

13. The method according to claim 1 or 11, characterized in that, The data sources include one or a combination of financial databases, business databases, baseline databases, dimensional databases, and external databases.

14. The method according to claim 1, characterized in that, The method further includes: adjusting the public attributes of the prediction model, the attributes of the prediction path, and the personalized attributes of the nodes based on the data information obtained through the application interface, to determine a prediction model that meets the prediction requirements.

15. The method according to claim 14, characterized in that, The step of determining a prediction model that meets the prediction requirements includes: displaying the rate of change of the prediction data of each node affected by the adjustment through the application interface, and determining a prediction model that meets the prediction requirements based on the prediction data of the prediction object expressed by each node in the prediction dimension.

16. The method according to claim 1, characterized in that, The method further includes: A first prediction model is generated based on the financial or business information obtained through the application interface. Configure the public attributes of the first prediction model, the attributes of the first prediction path, and the personalized attributes of the first node based on the obtained first instantiation configuration information, and determine the first calculation rule; A second prediction model is generated based on the financial or business information obtained through the application interface. Configure the public attributes of the second prediction model, the attributes of the second prediction path, and the personalized attributes of the second node based on the obtained second instantiation configuration information, and determine the second calculation rule; Configure the linkage between the first calculation rule and the second calculation rule; The semantic parsing engine converts the first and second prediction models, the first and second instantiation configuration information, and the linkage relationship between the first calculation rule and the second calculation rule into computer-executable second parsing code. The second parsing code is used to connect the first prediction model and the second prediction model in an orderly manner according to the linkage relationship between the first calculation rule and the second calculation rule, so as to generate a prediction job flow.

17. The method according to claim 16, characterized in that, The first prediction model is configured with a second and / or a third calculation rule, and the second prediction model is configured with a first and / or a fourth calculation rule.

18. The method according to claim 1, characterized in that, The step of generating a prediction model based on financial or business information obtained through the application interface includes connecting the prediction model to the output interface of a trained algorithm model, calling the algorithm model in real time, and combining it with the financial or business information obtained through the application interface to generate the prediction model.

19. The method according to claim 1, characterized in that, The method further includes: configuring permissions for the prediction model, including permissions for baseline nodes in the prediction model, permissions for each prediction path in the prediction model, and permissions for the prediction results corresponding to the prediction paths.

Images