Cognitive decision-making system fusing BI and AI capabilities

Abstract

The invention relates to the technical field of big data and data visualization, and discloses a cognitive decision-making system fusing BI and AI capabilities, and the system comprises a data access and management module which accesses multi-modal data, carries out the preprocessing of the multi-modal data, and further carries out the full-life-cycle management; the index interaction and analysis module is used for performing index caliber unification processing on the data to obtain standardized index data; a data interaction panel is constructed, and an analysis result is visually displayed; the decision deduction module is used for performing cognitive calculation and dynamic deduction on the standardized index data and generating and outputting decision suggestions; executing a decision behavior to obtain actual effect data, and feeding back the actual effect data to a cognitive calculation process to realize continuous optimization of the module; and the authority workbench module is used for configuring a personalized Web interface according to role authority and integrating multiple functions so as to realize differentiated data access and operation.

Description

TECHNICAL FIELD

[0001] The present specification relates to the technical field of big data and data visualization, in particular to a cognitive decision system integrating BI and AI capabilities. BACKGROUND

[0002] Business intelligence analysis and enterprise decision-making, as the core of enterprise operation and development, the key is to transform massive data into accurate, forward-looking and executable business insights.

[0003] The prior art has multiple deficiencies: first, traditional business intelligence systems focus on statistical, descriptive and visual presentation of historical data, and their analysis logic is based on correlation rather than causality. This model, when faced with key decisions full of uncertainty, the need to predict the future and bear responsibility, can only provide historical regularities as a reference, and cannot provide decision support based on causal logic and simulation deduction, resulting in a decision-making process that relies on experience, is biased and difficult to quantify risks. Second, existing data analysis and decision support tools are often fragmented. Data processing, indicator construction, visualization analysis, model training and business deduction belong to different platforms, resulting in inconsistent data, broken analysis processes, and delayed decision feedback. This architecture makes it difficult to form a decision-making loop, and the system cannot automatically and continuously return real effect data after decision execution to the analysis model and data governance process to achieve self-optimization and evolution, ultimately becoming a static and one-time analysis tool.

[0004] Therefore, there is an urgent need for a cognitive decision system integrating BI and AI capabilities to overcome the limitations of traditional analysis systems, such as weak explanatory power, static decision support, and fragmented tool chains, to achieve closed-loop intelligent decision-making from data to insights, from insights to simulation, and from simulation to optimized action. SUMMARY

[0005] Therefore, the present application aims to provide a cognitive decision system integrating BI and AI capabilities to solve the problem of one-sided business insights and decision-making process relying on experience and unable to quantify pre-play caused by extensive data governance, chaotic indicator caliber, post hoc analysis conclusions and missing decision deduction in existing data analysis systems.

[0006] To achieve the above-mentioned purpose, the present application adopts the following technical solutions: A cognitive decision system integrating BI and AI capabilities, the system comprising: A data access and management module for accessing multi-modal data and performing standardized processing; performing full life cycle management on the multi-modal data after standardized processing; the full life cycle management includes data asset cataloging, quality monitoring, master data maintenance and access permission control based on RBAC model.

[0007] An index interaction and analysis module is configured to uniformly process indexes and indicators of the multi-modal data managed through the whole life cycle to obtain standardized index data to eliminate data ambiguity, and to construct a data interaction panel through a semantic model to visually display multi-dimensional analysis results of the standardized index data.

[0008] A decision deduction module is configured to perform cognitive computing and dynamic deduction on the standardized index data to generate and output decision suggestions, to perform decision behaviors according to the decision suggestions to obtain actual effect data, and to feed back the actual effect data and the decision suggestions to the process of cognitive computing and dynamic deduction to continuously optimize the capability thereof.

[0009] A permission workbench module is configured to configure a personalized Web interface according to user role permissions and to integrate analysis, deduction and report functions to realize differentiated data access and operation.

[0010] The system of the present application has the following beneficial effects: in the prior art, there are problems such as fragmented data sources, inconsistent governance standards, analysis conclusions being post-facto and lacking of forward deduction in the process of enterprise data analysis and decision-making, which restricts the depth of business insight and the timeliness and scientificity of decision-making. The system realizes the leap from passive analysis relying on historical statistics to active intelligent decision-making based on simulation deduction through unified access and whole life cycle management of multi-modal data, standardized definition and interactive visualization of business indexes, and closed-loop decision deduction based on causal inference and dynamic simulation, effectively improving the value of data assets, analysis efficiency, decision-making foresight and system self-evolution capability.

[0011] Further, the data access and management module comprises a data access unit; the data access unit comprises an internal data access unit, an external data access unit and a multi-modal data processing unit.

[0012] The internal data access unit is configured to access data in a business database, a data warehouse, log files and an ERP / CRM system.

[0013] The external data access unit is configured to access crawler data, third-party API data and Internet of Things sensor data.

[0014] The multi-modal data processing unit is configured to support access and preprocessing of text, image and audio unstructured data.

[0015] Beneficial effect: the configuration realizes comprehensive integration and uniform preprocessing of enterprise internal core data, external expanded data and multi-modal unstructured data through construction of multi-level and multi-type data access pipelines, thereby laying a solid and complete data foundation for subsequent high-quality data governance and in-depth analysis.

[0016] Further, the data access and management module further comprises a data management unit; the data management unit comprises a metadata management unit, a data integration development unit, a data quality monitoring unit, a master data management unit and a data security and permission unit.

[0017] The metadata management unit is used to realize cataloging, blood analysis and influence analysis of data assets.

[0018] The data integration development unit is used to build real-time and batch data processing pipelines.

[0019] The data quality monitoring unit is used to define and monitor data quality rules and generate quality reports.

[0020] The master data management unit is used to provide unique and trusted versions of key business entities to each business system by constructing a master data master library with unique identification and maintaining master data, and the master data master library is a special database that only stores data of key business entities.

[0021] The data security and permission unit is used to implement data access control and audit based on the RBAC model.

[0022] Beneficial effects: through metadata management, data traceability and controllable influence are realized, through integrated development, efficient and stable data processing flow is ensured, through quality monitoring, the reliability of analysis basis is ensured, through master data management, core business entity ambiguity is eliminated to support accurate analysis, and through permission control and audit, data security and compliance requirements are met, thereby building a trusted, usable and secure high-quality data asset base, providing a solid guarantee for upper intelligent analysis and decision-making.

[0023] Further, the index interaction and analysis module comprises an index management subunit and an index analysis subunit.

[0024] The index management subunit is used to define the multi-modal data managed through the whole life cycle through a visual interface or SQL, precompute and query it using an OLAP database to form standardized index data, and provide a unified API externally.

[0025] The index analysis subunit is used to map physical tables to business logic models, support users to perform multi-dimensional queries in a drag-and-drop manner, integrate NLP engines to receive voice or text queries and return visual results, and create and display real-time refreshable data interaction panels through semantic models.

[0026] Beneficial effects: By deeply integrating index definition management and interactive analysis, an efficient and friendly channel from data to business insight is constructed. On the one hand, the index management subunit converts raw data into standardized indexes with unified caliber and efficient calculation, and releases them through a unified API, completely eliminating cross-department data ambiguity and repeated calculation problems, and ensuring the consistency of analysis sources. On the other hand, the index analysis subunit greatly reduces the technical threshold of business users for complex multi-dimensional data analysis through semantic models, drag-and-drop queries, natural language interaction and real-time visualization panels, realizes the agility of the analysis process and the intuitiveness of the result presentation, thereby significantly improving the efficiency and popularity of data-driven decision-making.

[0027] Further, the decision deduction module comprises: a cognitive computing engine for full-process machine learning modeling of feature engineering, model training, evaluation and deployment of the standardized index data, and integration of a causal inference library to automatically discover a causal graph and perform causal effect estimation from the standardized index data.

[0028] a dynamic deduction engine for constructing a simulation environment, performing multi-scenario counterfactual deduction based on the causal model output by the cognitive computing engine, and converting the deduction results into decision recommendations; and feeding actual effect data back to the cognitive computing engine to realize continuous optimization of the model.

[0029] Beneficial effects: By combining cognitive computing and dynamic deduction, an intelligent decision-making core with understanding-prediction-decision-optimization closed-loop capability is constructed. The cognitive computing engine not only realizes traditional machine learning modeling, but more importantly introduces causal inference to reveal the essential causal relationship between variables from data, making the model interpretable and stable, and fundamentally improving the scientific nature of decision recommendations. The dynamic deduction engine performs counterfactual analysis on various intervention strategies in a simulation environment based on the causal model, quantitatively evaluates the potential results of different decision schemes, and changes the decision-making from relying on historical experience to forward-looking selection based on simulation experiments. At the same time, by feeding the actual effect of the decision back to the model, a closed loop of continuous learning and optimization is formed, enabling the system to dynamically adapt to business changes and continuously improve decision accuracy and adaptability, realizing the leap from a static analysis tool to a dynamic decision-making agent.

[0030] Further, the cognitive computing engine comprises: a machine learning platform configured to perform feature engineering based on Scikit-learn, TensorFlow or PyTorch framework to construct input features of the causal model, use machine learning algorithms in the framework to train the features, evaluate the trained causal model through preset performance indicators, and deploy the causal model that passes the evaluation to a production environment.

[0031] a causal inference unit configured to integrate a causal inference library to automatically discover a causal structure graph among variables from the standardized indicator data, and to perform causal effect estimation based on the causal structure graph to quantify the impact of an intervention measure on a business indicator.

[0032] Beneficial effects: By organically combining general machine learning processes with special causal inference capabilities, an advanced analysis kernel with both prediction accuracy and causal explanation is constructed. The machine learning platform provides standardized, engineered whole-process support from feature construction to model deployment, ensuring reliable production and efficient iteration of analysis models. The causal inference unit breaks through the limitations of traditional correlation analysis, can automatically identify the causal structure among variables from observed data, and accurately quantify the impact of specific intervention measures on key business indicators. This combination enables the system not only to predict future trends, but also to deeply understand the internal mechanisms of business operations, answering the core decision-making questions of "why" and "how to change", providing a solid and reliable theoretical and technical foundation for subsequent simulation and scientific decision-making based on causal models.

[0033] Further, the dynamic inference engine includes: a simulation environment configured to construct a business scenario simulation framework based on an agent simulation model or a system dynamics model, for performing dynamic process simulation of multivariate interaction and state evolution within the framework.

[0034] a counterfactual inference module configured to receive user input intervention conditions based on the trained causal model output by the cognitive computing engine, and perform counterfactual reasoning in the simulation environment to infer potential business results under different intervention conditions.

[0035] an analysis workbench configured to provide a graphical operation interface, enabling users to set different decision hypothesis conditions, run multiple inference scenarios in parallel, and visually compare and analyze the results of each scenario.

[0036] Beneficial effects: By integrating a simulation environment, counterfactual reasoning, and a visual analysis workbench, a complete dynamic decision inference system is constructed. The simulation environment provides a high-fidelity dynamic mapping for the business system; the counterfactual inference module quantitatively evaluates the potential effects of different intervention strategies based on causal models; and the analysis workbench supports multi-scenario parallel design and comparative analysis through a graphical interface. The three work together to achieve controllable experiments, visual comparisons, and quantitative verification of the decision-making process, significantly improving the foresight, systematicness, and scientificity of decision-making. BRIEF DESCRIPTION OF DRAWINGS

[0037] The specification will be further illustrated in the way of example embodiments, which will be described in detail with the aid of the accompanying drawings. These embodiments are not restrictive, and in these embodiments, the same numbers represent the same structures, in which: Figure 1 is an example structure diagram of a cognitive decision system fusing BI and AI capabilities. DETAILED DESCRIPTION

[0038] In order to more clearly illustrate the technical solutions of the embodiments of the specification, the drawings needed to be used in the embodiment description will be briefly introduced below. Obviously, the drawings in the following description are only some examples or embodiments of the specification, and for those skilled in the art, the specification can be applied to other similar scenarios without creative labor on the basis of these drawings. Unless the context clearly indicates otherwise or otherwise stated, the same reference numbers in the drawings represent the same structure or operation.

[0039] As shown in the specification and claims, unless the context clearly indicates otherwise or otherwise stated, "one", "a", "an", and / or "the" do not refer to the singular, but can also include the plural. Generally speaking, the terms "comprise" and "include" only indicate the inclusion of the steps and elements explicitly identified, and these steps and elements do not constitute an exclusive list, and the method or device can also include other steps or elements.

[0040] Flowcharts are used in the specification to illustrate the operations performed by the system according to the embodiments of the specification. It should be understood that the preceding or subsequent operations are not necessarily performed in sequence. On the contrary, each step can be processed in reverse order or simultaneously. At the same time, other operations can also be added to these processes, or one or more steps of operation can be removed from these processes.

[0041] Further details will be described in the specific embodiments as follows: Figure 1 is an example structure diagram of a cognitive decision system fusing BI and AI capabilities, as shown in Figure 1 A cognitive decision system fusing BI and AI capabilities, the system includes a data access and management module, an index interaction and analysis module, a decision deduction module, and a permission workbench module.

[0042] The data access and management module is used to access multi-modal data and perform standardized processing thereon; to perform full life cycle management on the multi-modal data after standardized processing; the full life cycle management includes data asset cataloging, quality monitoring, master data maintenance, and access permission control based on the RBAC model.

[0043] In this embodiment, the multi-modal data includes internal data and external data.

[0044] Internal data is the core business data generated in the process of daily business operation of enterprises. For example, transaction order data, user registration information data stored in business databases; historical sales summary data deposited in data warehouses; system operation logs, user behavior trajectory logs recorded in log files; and supply chain management data, customer relationship maintenance data retained in ERP / CRM systems.

[0045] External data is external associated data involved in the extension of enterprise business scenarios. For example, industry competitor dynamic data and market trend analysis data obtained through web crawlers; social media public opinion data, third-party payment data, and industry policy release data accessed through third-party API interfaces; and environmental monitoring data and equipment operating status data collected by Internet of Things sensors.

[0046] Further, the data access and management module includes a data access unit; the data access unit includes an internal data access unit, an external data access unit, and a multi-modal data processing unit.

[0047] The internal data access unit is configured to access data in business databases, data warehouses, log files, and ERP / CRM systems.

[0048] In this embodiment, the internal data access unit accesses data from different channels through multiple different interfaces.

[0049] For example, for business databases, the internal data access unit establishes a stable connection through JDBC / ODBC standard interfaces, supports batch synchronization at a preset period or real-time incremental synchronization based on binlog, and ensures the real-time and completeness of core business data such as transaction orders and user registration information. The preset period can be minute-level or hour-level.

[0050] For data warehouses, Hadoop ecological components or data warehouse native import and export tools are used to efficiently pull batch data such as historical sales summaries and business indicators. Hadoop ecological components can include Sqoop, DataX, etc.

[0051] For log files, a log collection component is used to listen to log storage directories or log streams in real time, support data parsing in JSON, CSV, or custom delimiter log formats, and transmit data to a data processing channel through TCP / UDP protocols. Log collection components can include Flume, FileBeat, etc.

[0052] For ERP / CRM systems, access data through the system's open RESTful API or WebService interface, complete identity authentication based on OAuth2.0 authorization mechanism, support active data push or timed pull triggered by business scenarios, and ensure accurate docking of supply chain management, customer relationship maintenance and other data.

[0053] The external data access unit is used for accessing crawler data, third-party API data and Internet of Things sensor data.

[0054] In this embodiment, the external data access unit accesses data from different channels through multiple different interfaces.

[0055] For crawler data, configure crawling rules through a distributed crawler framework, and use XPath, CSS selector or regular expression to parse and extract target field data from target sites.

[0056] For third-party API data, initiate a data request through a third-party open interface, and after identity authentication, extract target data from the returned JSON / XML format response.

[0057] For Internet of Things sensor data, receive sensor transmission data through Internet of Things special protocols, and after format conversion, extract core data such as device operating status and environmental monitoring.

[0058] The multi-modal data processing unit is used to support the access and preprocessing of unstructured text, image and audio data.

[0059] In this embodiment, the multi-modal data processing unit uses different methods for standardization processing of text data, image data and audio data.

[0060] For text data, the system performs encoding unification, format regularization, word segmentation, part-of-speech tagging and entity recognition to convert unstructured text into structured feature vectors. The encoding can be unified as UTF-8; format regularization includes removing redundant spaces / line breaks, etc.; word segmentation includes Chinese and English word segmentation, stop word filtering, etc.

[0061] For image data, the system performs size normalization, format conversion, grayscale / color space standardization, noise removal and image feature extraction, and outputs standardized image data and structured features.

[0062] For audio data, the system performs sampling rate / bit depth unification, format conversion, noise reduction processing, speech segmentation and feature extraction to convert audio data into structured feature sequences.

[0063] In the embodiment, the setting realizes comprehensive fusion and unified preprocessing of enterprise internal core data, external expansion data and multi-modal unstructured data by constructing multi-level and multi-type data access pipelines, thereby laying a solid and complete data foundation for subsequent high-quality data governance and deep analysis.

[0064] Further, the data access and management module further comprises a data management unit; the data management unit comprises a metadata management unit, a data integration development unit, a data quality monitoring unit, a master data management unit and a data security and permission unit.

[0065] The metadata management unit is used to realize cataloging, blood relationship analysis and influence analysis of data assets.

[0066] Metadata refers to structured information describing data properties, sources, formats, relationships and management rules.

[0067] In the embodiment, metadata of each data source is automatically collected, structured and classified according to a preset classification system (such as business domain and data type), a standardized data asset directory is generated, and keyword search and multi-dimensional filtering are supported.

[0068] In the embodiment, the whole-link flow track of recorded data from access, processing, conversion to application is recorded, a data blood relationship map is constructed by analyzing ETL scripts, SQL statements and interface call logs, and the upstream and downstream dependency relationships of data such as field level and table level association are determined.

[0069] In the embodiment, based on the data blood relationship map, when a data source or a data field is changed (such as field addition / deletion, format adjustment), the affected downstream data assets, indicators and application scenarios are automatically identified, and an influence range report is output.

[0070] The data integration development unit is used to construct real-time and batch data processing pipelines.

[0071] The data quality monitoring unit is used to define and monitor data quality rules and generate quality reports.

[0072] In the embodiment, the data quality monitoring unit defines data quality rules by preset standardized verification dimensions and supports individualized extension configuration, so as to realize comprehensive and accurate verification of data quality.

[0073] Specifically, for data integrity, the integrity verification standards at the data record and field levels are defined, the core field null rate threshold and batch data missing rate threshold are determined, and the core business fields such as “order number” and “user ID” in order data are specified as mandatory items. If there is null value in such mandatory field, it is determined that the data quality is unqualified.

[0074] For data format accuracy, data type matching specifications are developed, and requirements such as "amount" field being numeric and "mobile phone" field being 11 digits are specified. At the same time, format regular check standards are set, such as "email" field conforming to the preset xxx@xxx.xxx format regular expression, "date" field uniformly adopting YYYY-MM-DD format, and text data uniformly using UTF-8 encoding.

[0075] For data logical consistency, field logical association rules are defined, such as "order total amount" being equal to "product unit price x quantity + freight - discount amount". At the same time, cross-table association consistency check requirements are set to ensure that "user ID" in the order table can be matched to the corresponding record in the user table, and the value range of "order status" and other fields is limited to ensure the logical consistency of data in the same dimension.

[0076] In combination with specific business scenarios, business validity check rules are defined, including setting reasonable intervals for "single order amount", "user registration time" not being later than "first order time", and other business logic rationality requirements, as well as real-time data access check time effectiveness, offline data update time, and other time effectiveness rules.

[0077] In addition, a visual configuration interface and a SQL script editing entry are provided to support users to customize personalized data quality check rules according to specific business needs such as retail and finance, such as "product gross profit rate ≥ 5%" and "customer risk level matches credit limit", to realize flexible expansion and precise adaptation of data quality rules.

[0078] On the basis of the above, the statistical analysis results are presented in a structured manner according to the preset standardized report template. The preset standardized report template is set based on human experience.

[0079] The master data management unit is configured to build a master data master library with a unique identifier and maintain the master data, and provide a unique trusted version of key business entities to each business system. The master data master library is a special database that only stores data of key business entities.

[0080] In this embodiment, the master data management unit sorts out the scope of key business entities such as customers and products, defines a unified data model, extracts corresponding entity data from each business system for cleaning, deduplication, and integration, assigns a unique identifier to each entity, and loads it into a special database to build a master data master library.

[0081] The data security and permission unit is configured to implement data access control and audit based on the RBAC model.

[0082] In the embodiment, the "user-role-permission" mapping relationship is established by presetting the roles of administrator, analyst, etc. and binding the corresponding operation permissions and data access ranges; when a user initiates an access request, the system checks the role permission of the user, and only allows operations with matching permissions; meanwhile, full logs of user access behaviors are recorded in real time, multi-dimensional query analysis is supported, and an audit report is generated, so that security risks such as unauthorized access can be identified in a timely manner.

[0083] In the embodiment, the data traceability and controllable influence are realized through metadata management, the efficient and stable data processing flow is ensured through integrated development, the reliability of analysis basis is ensured through quality monitoring, the core business entity ambiguity is eliminated through master data management to support accurate analysis, and the data security and compliance requirements are met through permission control and audit, so that a reliable, usable and secure high-quality data asset foundation is built, and a solid guarantee is provided for upper-layer intelligent analysis and decision-making.

[0084] The index interaction and analysis module is configured to perform unified processing on the standardized index data obtained by performing unified processing on the multi-modal data managed through the whole life cycle to eliminate data ambiguity, and to construct a data interaction panel through a semantic model to visually display multi-dimensional analysis results of the standardized index data.

[0085] The standardized index data is a structured business measurement value that can be directly used for analysis, comparison and decision support after being standardized based on unified business caliber, calculation logic and data sources.

[0086] Further, the index interaction and analysis module includes an index management subunit and an index analysis subunit.

[0087] The index management subunit is configured to define the multi-modal data managed through the whole life cycle through a visual interface or SQL, to perform pre-computation and query on the multi-modal data through an OLAP database to form the standardized index data, and to provide a unified API externally.

[0088] The index analysis subunit is configured to map a physical table to a business logic model, to support a user to perform multi-dimensional query in a drag-and-drop manner, to integrate an NLP engine to receive voice or text query and return a visual result, and to create and display a data interaction panel that can be refreshed in real time through a semantic model.

[0089] In the embodiment, the index analysis subunit defines field association rules and index calculation caliber based on business scenario requirements, maps original fields in a physical table to dimensions and indexes with business meanings, and constructs a business logic model.

[0090] In the embodiment, the index analysis subunit is built-in visual drag-and-drop interactive interface and NLP engine, through the preset physical table field and the association mapping relationship of business dimension and index, supports the user to drag and select the target dimension and index and configure the filtering and aggregation conditions to execute multi-dimensional query; at the same time, the voice or text query statement input by the user is analyzed and converted into a standardized query instruction by the NLP engine, the corresponding data is called and automatically matched with the column chart, line chart and other visual charts for output; based on the above mapping relationship and query logic, a semantic model is constructed, the user's common analysis scene is solidified into a data interaction panel, the panel data and chart are automatically refreshed by real-time listening to the update state of the underlying data, and efficient cooperation of data query and visual analysis is realized.

[0091] In the embodiment, by deeply integrating index definition management and interactive analysis, an efficient and friendly channel from data to business insight is constructed. On the one hand, the index management subunit converts raw data into standardized indexes with unified caliber and efficient calculation, and releases them through a unified API, completely eliminating cross-department data ambiguity and repeated calculation problems, and ensuring the consistency of analysis source. On the other hand, the index analysis subunit greatly reduces the technical threshold of business users for complex multi-dimensional data analysis through semantic model, drag-and-drop query, natural language interaction and real-time visualization panel, realizes the agility of analysis process and the intuitiveness of result presentation, and thus significantly improves the efficiency and popularity of data-driven decision-making.

[0092] The decision deduction module is used for cognitive calculation and dynamic deduction of the standardized index data, generates and outputs decision suggestions; according to the decision suggestions, the actual effect data of the decision behavior is obtained, and the actual effect data and the decision suggestions are fed back to the process of cognitive calculation and dynamic deduction, so as to realize the continuous optimization of the ability.

[0093] The decision suggestion refers to a policy scheme or optimization guide with clear business orientation and executability.

[0094] In the embodiment, the decision suggestion is generated by the decision deduction module, and its content covers business adjustment direction, resource allocation scheme, risk avoidance measure, target achievement path, etc.

[0095] The actual effect data refers to the quantitative and qualitative data reflecting the effectiveness of the decision landing after the specific decision behavior is executed according to the decision suggestion.

[0096] In the embodiment, the actual effect data includes the change value of the core business index after the decision execution, the target achievement rate, the resource input-output ratio, the risk occurrence rate, etc. This data will be fed back to the cognitive calculation and dynamic deduction process of the decision deduction module together with the corresponding decision suggestion, and through the comparison between the expected and actual effectiveness, the deduction algorithm model and the cognitive calculation rule are optimized, and the continuous iterative upgrade of the decision deduction ability is realized.

[0097] Further, the decision deduction module comprises: a cognitive computing engine for full-process machine learning modeling of feature engineering, model training, evaluation and deployment of standardized indicator data, and integration of a causal inference library to automatically discover a causal graph and perform causal effect estimation from the standardized indicator data.

[0098] a dynamic deduction engine for constructing a simulation environment, performing multi-scenario counterfactual deduction based on the causal model output by the cognitive computing engine, and converting the deduction result into a decision suggestion; and feeding actual effect data back to the cognitive computing engine to realize continuous optimization of the model.

[0099] In the embodiment, by combining cognitive computing with dynamic deduction, an intelligent decision core with understanding-prediction-decision-optimization closed-loop capability is constructed. The cognitive computing engine not only realizes traditional machine learning modeling, but more importantly introduces causal inference to reveal the essential causal relationship between variables from data, making the model interpretable and stable, and fundamentally improving the scientificity of decision suggestions. The dynamic deduction engine performs counterfactual analysis on various intervention strategies in a simulation environment based on the causal model, quantitatively evaluates the potential results of different decision schemes, and changes the decision from relying on historical experience to forward-looking selection based on simulation experiments. At the same time, by feeding the actual effect of the decision back to the model, a closed loop of continuous learning and optimization is formed, enabling the system to dynamically adapt to business changes and continuously improve decision accuracy and adaptability, realizing the leap from a static analysis tool to a dynamic decision agent.

[0100] Further, the cognitive computing engine comprises: a machine learning platform configured to perform feature engineering based on Scikit-learn, TensorFlow or PyTorch framework to construct causal model input features, use machine learning algorithms in the framework to train the features, evaluate the trained causal model through preset performance indicators, and deploy the evaluated causal model to a production environment.

[0101] The machine learning platform is an intelligent algorithm support system that provides standardized management of feature engineering, model training, evaluation and deployment for the needs of causal model construction. The platform has a multi-framework adaptation layer that encapsulates the core interfaces of mainstream machine learning frameworks such as Scikit-learn, TensorFlow and PyTorch, forms a unified framework calling standard, and supports users to select the adaptation framework as needed without additional interface adaptation development.

[0102] In this embodiment, for the feature engineering link, the platform has built-in feature extraction, feature conversion, feature selection and other functional modules. It can automatically generate a candidate feature set based on the input standardized index data, and provide feature correlation analysis and feature importance sorting tools to assist in screening effective features with strong association with causal models, and generate a feature matrix that meets the model training requirements.

[0103] In the model training phase, the system supports flexible configuration of multi-dimensional training parameters according to business needs, including iteration times, learning rate, batch size, regularization coefficient and other core parameters. It also provides distributed training capabilities, which can dynamically allocate training tasks to multi-node computing clusters according to computing resources, and improve model training efficiency through data parallelism or model parallelism.

[0104] Taking the gradient boosting decision tree causal model training based on the TensorFlow framework as an example, users can set specific parameters such as tree depth, learning rate step, and sub-sample ratio for business scenarios. The system will automatically split the training data set and distribute it to each computing node. Each node independently completes the training and gradient calculation of the base learner, and then synchronously integrates the gradient information and model parameters through the parameter server. During the training process, the system monitors key indicators such as training loss value, validation loss value, causal effect estimation accuracy, and feature importance distribution in real time, generates visual training process curves at preset time intervals, clearly shows the downward trend of loss value with iteration times, the fluctuation of validation set indicators, and the model convergence state, which helps users judge whether the model is overfitting or underfitting in time, and dynamically adjust the training parameters to optimize the model performance.

[0105] In the model evaluation link, the platform presets a multi-dimensional performance indicator system, including accuracy, recall rate, F1 value, AUC value and other classification indicators, as well as mean square error, mean absolute error and other regression indicators, and supports user-defined evaluation indicator weights. The platform automatically calls the trained model to validate the test data set, generates a quantitative evaluation report, and determines whether the model is qualified according to the preset evaluation threshold.

[0106] For the model deployment link, the platform provides model packaging functions to convert the qualified causal model into a standardized deployment file, and supports containerized deployment and API interface deployment modes. Containerized deployment encapsulates the model running environment through Docker to ensure consistency across environments. API interface deployment encapsulates the model as a RESTful API for decision-making and reasoning modules to call on demand, achieving fast online and efficient reuse of the model.

[0107] The causal inference unit is configured to integrate a causal inference library to automatically discover causal structure diagrams between variables from standardized index data; and perform causal effect estimation based on the causal structure diagrams to quantify the impact of intervention measures on business indicators.

[0108] In this embodiment, in the causal structure diagram automatic discovery stage, the causal inference unit pre-processes the input standardized indicator data, screens out a variable set related to business decision, and guarantees data quality through data type conversion, missing value filling, and outlier removal. The variable set related to business decision includes intervention variables, result variables, and potential confounding variables.

[0109] Based on the pre-processed data, a structure learning algorithm in the causal inference library is called to automatically mine the causal correlation between variables by analyzing the conditional independence, mutual information value, and conditional probability distribution characteristics between variables, and to remove false variable pairs. Finally, a causal structure diagram is outputted, in which nodes represent variables and directed edges represent causal directions, and the confidence of each causal path is labeled to clearly present the direct and indirect influence paths of intervention variables to business indicators.

[0110] Among them, the causal inference library can be an open source library for general causal inference scenarios, such as the DoWhy library, the EconML library, etc. Such causal inference libraries encapsulate structure learning algorithms such as PC algorithm, FGES algorithm, and causal effect estimation methods such as propensity score matching, double machine learning, and causal forest, with perfect algorithm logic and reusable interfaces.

[0111] In the causal effect estimation stage, the causal inference unit identifies and blocks potential confounding paths based on the automatically discovered causal structure diagram to avoid interference of confounding variables on effect estimation results. Then, according to the complexity of the causal structure diagram and the data characteristics, a matching estimation method is selected from the integrated causal inference library. If it is a linear causal relationship, double machine learning method is used, and if there is a nonlinear or heterogeneous effect, causal forest method based on tree model is used. Through the selected method, the standardized indicator data is calculated to quantitatively obtain the average treatment effect, conditional average treatment effect, and other core parameters of the intervention measures on business indicators, to clearly define the specific impact of different intervention intensities and different business scenarios on business indicators, and to form an effect estimation report.

[0112] In this embodiment, through the cooperation of the machine learning platform and the causal inference unit, the training, evaluation, and deployment of the causal model are completed relying on mainstream framework specifications to ensure the reliability and scene adaptability of the model. At the same time, the causal inference library is integrated to mine the causal correlation between variables rather than correlation, to quantify the impact of intervention measures on business indicators, to avoid decision bias, to combine the iterative optimization mechanism, to promote the decision from experience-driven to data and model-driven, and to improve the scientificity and accuracy of decision deduction.

[0113] Further, the dynamic deduction engine comprises: The simulation environment is configured to construct a business scenario simulation framework based on an agent simulation model or a system dynamics model, for performing dynamic process simulation of multivariate interaction and state evolution within the framework.

[0114] In this embodiment, a model adaptation engine is built into the simulation environment, which selects an agent simulation model or a system dynamics model according to the characteristics of the business scenario, the former being suitable for multi-agent game scenarios and the latter being suitable for system-level dynamic evolution scenarios, while providing a parameter customization interface to adapt to differentiated needs.

[0115] Based on standardized index data, the core elements of the scenario are extracted, and the corresponding simulation framework is constructed: for the agent simulation model, the agent attributes, behavior rules and interaction modes are defined; for the system dynamics model, the variable feedback relationship is sorted out through the causal loop diagram, and the stock-flow calculation model is built.

[0116] Relying on the time step or event trigger mechanism to drive the simulation, the variable state is calculated in real time and the interaction relationship is updated in each round of reasoning, and the subject behavior or system state data is recorded synchronously; dynamic adjustment of simulation parameters is supported to carry out multiple rounds of simulation, and the simulation results are fed back to the decision-making reasoning module to provide a basis for decision-making effect prediction.

[0117] The counterfactual reasoning module is configured to receive user input intervention conditions based on the trained causal model output by the cognitive computing engine, and perform counterfactual reasoning in the simulation environment to deduce potential business results under different intervention conditions.

[0118] In this embodiment, the counterfactual reasoning module takes the trained causal model output by the cognitive computing engine as the core basis, first analyzes the user input intervention conditions, extracts key parameters such as intervention variables, intervention intensity and constraint boundaries, and converts them into a standardized input format recognizable by the causal model, while based on the causal association rules between variables in the causal model, the rationality and feasibility of the intervention conditions are verified, and invalid intervention conditions that conflict with business logic or causal relationships are eliminated.

[0119] Subsequently, the module calls the simulation environment, injects the verified intervention conditions into the pre-set business scenario simulation framework, drives the causal model to perform counterfactual reasoning in the simulation environment: fixes the state of other variables except the intervention variables, simulates the causal transmission process between variables under the intervention conditions, calculates the potential business indicator change values corresponding to different intervention conditions, generates a multi-dimensional reasoning result report, and clearly presents the correlation between intervention measures and potential business results, providing a quantitative reference for decision-making.

[0120] The analysis workstation is configured to provide a graphical operation interface, allowing users to set different decision-making hypothesis conditions, run multiple reasoning scenarios in parallel, and visually compare and analyze the reasoning results of each scenario.

[0121] In the embodiment, a complete dynamic decision deduction system is constructed through the integrated simulation environment, counterfactual reasoning and visual analysis workbench. The simulation environment provides a high-fidelity dynamic mapping for the business system; the counterfactual reasoning module quantitatively evaluates the potential effects of different intervention strategies based on the causal model; and the analysis workbench supports multi-scenario parallel design and comparative analysis through a graphical interface. The three work together to realize controllable experiments, visual comparison and quantitative verification of the decision-making process, significantly improving the forward-looking, systematic and scientific nature of the decision-making.

[0122] The permission workbench module is configured to configure a personalized Web interface according to user role permissions and integrate analysis, deduction and report functions to realize differentiated data access and operation.

[0123] The permission workbench module takes the RBAC permission model as the core, constructs a three-layer mapping relationship of user-role-permission, and configures differentiated permissions sets for different business roles based on a preset role permission matrix. The permission scope covers data access permissions, function operation permissions and interface visual permissions. Different business roles can include administrators, analysts, decision makers and the like.

[0124] The module has a built-in interface rendering engine that automatically filters and loads the analysis, deduction and report function components that the currently logged-in user has access to according to the role permissions of the user, and shields the function entry without permission. At the same time, the standardized index data and business models within the permission scope are called to generate a personalized Web operation interface, realizing that different role users can only view authorized data and use authorized functions, which not only guarantees data security and operation compliance, but also improves the business operation efficiency of each role user.

[0125] In the embodiment, the system realizes the leap from passive analysis relying on historical statistics to active intelligent decision-making based on simulation deduction through unified access and full-life-cycle management of multi-modal data, standardized definition and interactive visualization of business indicators, and closed-loop decision deduction based on causal inference and dynamic simulation, effectively improving the value of data assets, analysis efficiency, decision-making foresight and system self-evolution ability.

[0126] The foregoing detailed description has been set forth to illustrate the basic concepts of the present disclosure. Obviously, the above detailed description is only used as an example for those skilled in the art, and does not constitute a limitation on the present disclosure. Although the present disclosure does not explicitly state it, those skilled in the art can make various modifications, improvements and corrections to the present disclosure. Such modifications, improvements and corrections are suggested in the present disclosure, so such modifications, improvements and corrections still belong to the spirit and scope of the exemplary embodiments of the present disclosure.

[0127] Furthermore, the order of the processing elements and sequences described in this specification are not intended to be construed as a limitation, unless specifically stated, but are included to provide a complete description of one or more embodiments of the present specification. Regardless of the particular sequence of processing elements and sequences, however, the description herein of a process should be understood to include any and all combinations of one or more elements of a process independently selected from each sequence. For example, although the system components described above can be implemented by hardware devices, they can also be implemented by software solutions, such as installing the described system on an existing server or mobile device.

[0128] Similarly, it is to be noticed that the term "comprising", used in the description, is not intended to exclude other elements or steps. It is to be understood that the description and the examples are intended to be illustrative, but not limiting, of the scope of the present specification. Thus, the scope of the present specification should be given by the appended claims, along with their full scope of equivalents, and not by an restricting interpretation of the description or the examples.

[0129] Some embodiments use numerical designations to describe components, quantities of attributes. It is to be understood that such numerical designations used in the description of embodiments are, in some examples, modified by the adjectives "about", "approximately", or "generally". Unless otherwise stated, "about", "approximately", or "generally" indicates that the stated numerical value is allowed ±20% variation. Accordingly, numerical values used in the description and claims are approximations that can vary depending on the desired properties of the individual embodiments. In some embodiments, numerical values should be considered in the context of the number of significant figures used in the description and claims. Although the numerical ranges and parameters setting forth the broadest scope of the embodiments herein are approximations, the numerical values set forth in the specific examples are reported as precisely as practicable. The numerical values set forth in the specific examples are provided to give a general understanding of the embodiments.

[0130] Each patent, patent application, patent publication, and other material cited in this specification is hereby incorporated by reference in its entirety for the purpose of describing and disclosing the materials described in the documents in connection with the embodiments of the present specification. Citation of a document is not an admission that it is prior art with respect to the present specification. Citation of a document herewith provides open admission that the document is incorporated by reference. In the event that any conflict exists between the description, definitions, and / or terms used in this specification and those of the documents incorporated by reference, the description, definitions, and / or terms in this specification take precedence. Thus, the description and examples set forth herein are not intended to be exhaustive or to be construed as limiting the scope of the embodiments.

[0131] Finally, it should be understood that the embodiments described herein are only given by way of example and that other modifications can occur to persons skilled in the art. Therefore, the scope of the present description is not intended to be limited to the embodiments described herein but is only limited by the claims that follow.

Claims

1. A cognitive decision-making system integrating BI and AI capabilities, characterized in that, The system includes: The data access and management module is used to access multimodal data and perform standardized processing on it; it performs full lifecycle management on the standardized multimodal data; the full lifecycle management includes data asset cataloging, quality monitoring, master data maintenance, and access control based on the RBAC model; The indicator interaction and analysis module is used to unify the indicator definitions of the multimodal data after full lifecycle management to obtain standardized indicator data to eliminate data ambiguity; and to build a data interaction panel through a semantic model to visually display the multidimensional analysis results of the standardized indicator data. The decision-making and deduction module is used to perform cognitive calculations and dynamic deductions on the standardized indicator data, generate and output decision suggestions; execute decision-making actions based on the decision suggestions to obtain actual effect data, and feed the actual effect data and the decision suggestions back into the cognitive calculation and dynamic deduction process to achieve continuous optimization of its capabilities; The Permissions Workbench module is used to configure a personalized web interface based on user role permissions and integrate analysis, inference and reporting functions to achieve differentiated data access and operation.

2. The system according to claim 1, characterized in that, The data access and management module includes a data access unit; the data access unit includes an internal data access unit, an external data access unit, and a multimodal data processing unit. The internal data access unit is used to access data from business databases, data warehouses, log files, and ERP / CRM systems. The external data access unit is used to access crawler data, third-party API data, and IoT sensor data; The multimodal data processing unit is used to support the access and preprocessing of unstructured data such as text, images, and audio.

3. The system according to claim 2, characterized in that, The data access and management module further includes a data management unit; the data management unit includes a metadata management unit, a data integration and development unit, a data quality monitoring unit, a master data management unit, and a data security and access control unit; The metadata management unit is used to catalog, lineage analysis, and impact analysis of data assets; The data integration development unit is used to build real-time and batch data processing pipelines; The data quality monitoring unit is used to define and monitor data quality rules and generate quality reports; The master data management unit is used to provide each business system with a unique and trusted version of the key business entity by constructing a master data master database with a unique identifier and maintaining the master data. The master data master database is a dedicated database that stores only the data of the key business entities. The data security and access control unit is used to implement data access control and auditing based on the RBAC model.

4. The system according to claim 3, characterized in that, The indicator interaction and analysis module includes an indicator management subunit and an indicator analysis subunit; The indicator management subunit is used to define the multimodal data after full lifecycle management through a visual interface or SQL, pre-calculate and query it using an OLAP database to form standardized indicator data, and provide a unified API to the outside world; The indicator analysis subunit is used to map physical tables to business logic models, support users to execute multi-dimensional queries by dragging and dropping, integrate an NLP engine to receive voice or text queries and return visual results, and create and display a data interaction panel that can be refreshed in real time through a semantic model.

5. The system according to claim 4, characterized in that, The decision deduction module includes: The cognitive computing engine is used for the entire process of machine learning modeling, including feature engineering, model training, evaluation and deployment of the standardized indicator data, and integrates a causal inference library to automatically discover causal graphs and estimate causal effects from the standardized indicator data. The dynamic inference engine is used to construct a simulation environment based on the causal model output by the cognitive computing engine, perform counterfactual inferences in multiple scenarios, and transform the inference results into decision suggestions; it also feeds back actual effect data to the cognitive computing engine to achieve continuous model optimization.

6. The system according to claim 5, characterized in that, The cognitive computing engine includes: The machine learning platform is configured to perform feature engineering based on the Scikit-learn, TensorFlow, or PyTorch framework to construct input features for a causal model, train the model using machine learning algorithms in the framework, evaluate the trained causal model using preset performance metrics, and deploy the qualified causal model to the production environment. The causal inference unit is configured to integrate a causal inference library to automatically discover causal structure graphs between variables from the standardized indicator data; and to estimate causal effects based on the causal structure graphs to quantify the impact of intervention measures on business indicators.

7. The system according to claim 5, characterized in that, The dynamic simulation engine includes: The simulation environment is configured to build a business scenario simulation framework based on an intelligent agent simulation model or a system dynamics model, and to perform dynamic process simulation of multivariable interaction and state evolution within the framework. The counterfactual reasoning module is configured to receive user-inputted intervention conditions based on the trained causal model output by the cognitive computing engine, and perform counterfactual reasoning in the simulation environment to deduce potential business outcomes under different intervention conditions. The analysis workbench is configured to provide a graphical user interface, enabling users to set different decision assumptions, run multiple simulation scenarios in parallel, and perform visual comparative analysis of the simulation results of each scenario.

Images