Business decision model construction method and device and computer program product

Abstract

The invention discloses a business decision model construction method and device and a computer program product. The method relates to the field of artificial intelligence, and comprises the following steps: receiving an input business rule file, and carrying out text preprocessing on the business rule file to obtain multi-modal data; extracting rule features from the multi-modal data, and converting the rule features into semantic vectors through a pre-trained large language model; determining a service type of the service rule file, determining a knowledge graph corresponding to the service type, and performing semantic enhancement processing on the semantic vector based on a related entity associated with the rule feature in the knowledge graph to obtain a target semantic vector; and extracting a target rule template matched with the target semantic vector from a rule template library, and constructing a business decision model based on the target rule template and the rule features. Through the method and the device, the problem of low efficiency of manually constructing the business decision model in related technologies is solved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence, and more specifically, to a method, apparatus, and computer program product for constructing a business decision model. Background Technology

[0002] In business operations, automating the processing of laws, regulations, management systems, and procedures has become a core requirement for ensuring business compliance. Currently, the parsing of regulatory documents relies heavily on manual labor, which not only consumes significant time and manpower but also frequently introduces errors due to differences in individual understanding. This inefficient approach struggles to meet the high-frequency demands of regulatory changes, especially when businesses operate in an international environment where the speed and complexity of regulatory updates increase dramatically, making the limitations of manual processing increasingly apparent. During the rule design phase, businesses need to transform manually extracted rule entries into computer-processable rule models.

[0003] However, rule designers in related technologies (such as decision trees and conditional logic tables) have not yet effectively handled the correlation issues of multi-source data such as text, images, and tables. For example, optical character recognition technology may make mistakes when recognizing text in images, and if these errors are not corrected in time, they will directly affect the accuracy and completeness of the rule model. Furthermore, the lack of collaborative analysis capabilities for text and table content often leads to the model omitting key details when formulating rules involving specific numerical or structured data, reducing its applicability and reliability. Rule engines in related technologies are all based on static rule bases and cannot respond to regulatory updates in real time. Rule maintenance and updates require periodic manual intervention and tedious manual updates. This increases maintenance costs, easily leads to rule base version chaos, and delayed rule updates, thereby affecting the compliance and stability of business systems.

[0004] There is currently no effective solution to the problem of low efficiency in manually building business decision-making models in related technologies. Summary of the Invention

[0005] The main objective of this application is to provide a method, apparatus, and computer program product for constructing a business decision model, so as to solve the problem of low efficiency in constructing business decision models manually in related technologies.

[0006] To achieve the above objectives, according to one aspect of this application, a method for constructing a business decision model is provided. The method includes: receiving an input business rule file; performing text preprocessing on the business rule file to obtain multimodal data; extracting rule features from the multimodal data; converting the rule features into semantic vectors using a pre-trained large language model, wherein the rule features include at least one of the following: rule conditions, rule actions, and rule constraints; determining the business type of the business rule file and determining the knowledge graph corresponding to the business type; performing semantic enhancement processing on the semantic vectors based on relevant entities associated with the rule features in the knowledge graph to obtain a target semantic vector, wherein the knowledge graph stores various entities related to the business type and the relationships between entities; extracting a target rule template matching the target semantic vector from a rule template library; and constructing a business decision model based on the target rule template and the rule features.

[0007] Optionally, extracting a target rule template matching the target semantic vector from the rule template library includes: extracting multiple rule templates from the rule template library, determining the template vector of each rule template, wherein the rule template library stores multiple types of rule templates, and the types of rule templates include at least one of the following: business approval, permission setting, and early warning; calculating the similarity between the target semantic vector and each template vector to obtain the similarity between each rule template and the target semantic vector; if there is only one rule template with a similarity greater than or equal to a first similarity threshold, the rule template corresponding to the similarity is determined as the target rule template; if the similarity of all rule templates is less than the first similarity threshold, the rule template corresponding to the maximum similarity is determined as the target rule template; if there are at least two rule templates with a similarity greater than or equal to the first similarity threshold, the rule templates with a similarity greater than or equal to the first similarity threshold are determined as candidate templates, and the candidate template corresponding to the maximum execution effect evaluation value is determined as the target rule template, wherein the execution effect evaluation value is determined by at least one of the rule template's historical execution efficiency, historical execution accuracy, and historical feedback results.

[0008] Optionally, preprocessing the business rule file to obtain multimodal data includes: denoising the business rule file to obtain a denoised business rule file; processing the denoised business rule file to obtain a processed business rule file, wherein the data processing includes at least one of the following: image enhancement processing of image data, segmentation and sentence segmentation processing of text data, and structured data transformation processing of table data; extracting text information and graphic elements from the processed business rule file using a text recognition model to obtain initial data; and processing the initial data using a natural language processing model to obtain multimodal data, wherein the multimodal data includes at least one of the following: text, tables, and images.

[0009] Optionally, processing the initial data through a natural language processing model to obtain multimodal data includes: segmenting the initial data into words using the word segmentation module in the natural language processing model to obtain segmented data; analyzing the initial data using the dependency parsing module in the natural language processing model to obtain the grammatical structure of the initial data; extracting multiple entities from the initial data using the entity recognition module in the natural language processing model; and standardizing the segmented data, grammatical structure, and multiple entities to obtain multimodal data.

[0010] Optionally, constructing a business decision model based on a target rule template and rule features includes: extracting rule variables from the target rule template, wherein the rule variables include at least one of the following: condition variables, action variables, and constraint variables; matching rule features with rule variables; and filling the rule features into the positions of the rule variables in the target rule template that match the rule features, thereby obtaining the business decision model.

[0011] Optionally, after constructing the business decision model based on the target rule template and rule features, the method further includes: running the business decision model in a test environment to obtain test results, wherein the test results include at least one of the following: rule triggering conditions, decision execution results, and boundary conditions; issuing a first prompt message when the similarity between the test results and the expected results is less than a second similarity threshold, wherein the first prompt message is used to indicate that the business decision model does not match the business rule file; formally verifying the business decision model through a logic reasoning engine when the similarity between the test results and the expected results is greater than or equal to the second similarity threshold, and obtaining verification results; issuing a second prompt message when the verification results indicate that the business decision model has logical vulnerabilities or conflicts, wherein the second prompt message is used to prompt the operation and maintenance personnel to fix the logical vulnerabilities or conflicts; and determining that the business decision model has passed the test when the verification results indicate that the business decision model does not have logical vulnerabilities or conflicts.

[0012] Optionally, after confirming that the business decision model has passed the test, the method further includes: determining the target business system to which the business rule file will be applied, and deploying the business decision model through the rule service interface of the target business system; if the deployment of the business decision model fails, controlling the rule service interface to roll back the business rules of the target business system to the previous version of the business decision model; if the deployment of the business decision model is successful, checking every preset time interval whether the rule service interface has received a new version of the business decision model; if the rule service interface is detected to have received a new version of the business decision model, deploying the new version of the business decision model through the rule service interface.

[0013] To achieve the above objectives, according to another aspect of this application, a business decision model construction apparatus is provided. The apparatus includes: a receiving unit for receiving an input business rule file, performing text preprocessing on the business rule file to obtain multimodal data; an extraction unit for extracting rule features from the multimodal data, converting the rule features into semantic vectors using a pre-trained large language model, wherein the rule features include at least one of the following: rule conditions, rule actions, and rule constraints; a first determining unit for determining the business type of the business rule file and determining the knowledge graph corresponding to the business type, performing semantic enhancement processing on the semantic vectors based on relevant entities associated with the rule features in the knowledge graph to obtain a target semantic vector, wherein the knowledge graph stores various entities related to the business type and the relationships between entities; and a construction unit for extracting a target rule template matching the target semantic vector from a rule template library, and constructing a business decision model based on the target rule template and rule features.

[0014] To achieve the above objectives, according to another aspect of this application, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps of the business decision model construction method described in various embodiments of this application.

[0015] This application employs the following steps: receiving an input business rule file; preprocessing the business rule file to obtain multimodal data; extracting rule features from the multimodal data; converting the rule features into semantic vectors using a pre-trained large language model, wherein the rule features include at least one of the following: rule conditions, rule actions, and rule constraints; determining the business type of the business rule file and the corresponding knowledge graph; performing semantic enhancement processing on the semantic vectors based on the relevant entities associated with the rule features in the knowledge graph to obtain a target semantic vector, wherein the knowledge graph stores various entities related to the business type and the relationships between entities; extracting a target rule template matching the target semantic vector from a rule template library; and constructing a business decision model based on the target rule template and rule features, thus solving the problem of low efficiency in manually constructing business decision models in related technologies. By employing multimodal data processing technology for business rule documents, the entire process from inputting regulatory documents to outputting standardized business decision-making models is automated. This overcomes technical bottlenecks such as reliance on manual intervention, limitations of rule designers, and static rule engines. Through the deep semantic understanding capabilities of large language models and entity association information from knowledge graphs, more accurate business decision-making models can be generated. Through template matching and decision-making model construction processes, it is ensured that even when faced with complex rule documents, applicable models can be generated quickly, thereby reducing business response time, improving overall operational efficiency, and ultimately enhancing the efficiency of business decision-making model construction. Attached Figure Description

[0016] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0017] Figure 1 This is a flowchart of a method for constructing a business decision model according to an embodiment of this application;

[0018] Figure 2 This is a schematic diagram of a business rule generation apparatus provided according to an embodiment of this application;

[0019] Figure 3 This is a schematic diagram of a business rule generation method provided according to an embodiment of this application;

[0020] Figure 4 This is a schematic diagram of a business decision model construction apparatus provided in the embodiments of this application;

[0021] Figure 5 This is a schematic diagram of an electronic device provided according to an embodiment of this application. Detailed Implementation

[0022] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0023] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0024] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate for the embodiments of this application described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0025] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for display, data used for analysis, etc.) involved in this disclosure are all information and data authorized by the user or fully authorized by all parties.

[0026] It should be noted that the information collected is information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data all comply with the relevant laws, regulations and standards of the relevant regions, necessary confidentiality measures have been taken, and they do not violate public order and good morals. Corresponding operation portals are provided for users to choose to authorize or refuse.

[0027] The present invention will now be described in conjunction with preferred implementation steps. Figure 1 This is a flowchart of a method for constructing a business decision model according to an embodiment of this application, such as... Figure 1 As shown, the method includes the following steps:

[0028] Step S101: Receive the input business rule file, perform text preprocessing on the business rule file, and obtain multimodal data.

[0029] In step S101, the business rule file can be corporate policies, contract terms, etc., and the file can include text files (PDF, Word documents), spreadsheets (such as Excel files), and image files (such as JPG, PNG). The business rule file undergoes text preprocessing, which may include data cleaning (noise removal), format standardization, character recognition, word segmentation, and image enhancement to ensure the accuracy of subsequent processing. By preprocessing the received business rule file, multimodal data containing text, tables, and images is obtained.

[0030] Step S102: Extract rule features from multimodal data and convert the rule features into semantic vectors using a pre-trained large language model. The rule features include at least one of the following: rule conditions, rule actions, and rule constraints.

[0031] In step S102, natural language processing techniques are used to extract rule features from the preprocessed multimodal data, including rule conditions (e.g., "If condition A is met, then perform operation B"), rule actions (e.g., "Generate a warning", "Trigger an approval process"), and rule constraints (e.g., "Applicable only to a specific department", "Valid within a specific time frame"). For mixed multimodal data, such as tabular data in a PDF file, specialized table parsing techniques are used to extract numerical and categorical information to ensure the comprehensiveness of the rule features. Semantic embedding techniques are used to transform the rule features into high-dimensional semantic vectors using a pre-trained large language model. For example, the semantic vector transformation formula is as follows:

[0032] V = SemanticEmbedding(S);

[0033] Where S is the input rule feature, and V is the transformed high-dimensional semantic vector.

[0034] Step S103: Determine the business type of the business rule file and the knowledge graph corresponding to the business type. Based on the relevant entities associated with the rule features in the knowledge graph, perform semantic enhancement processing on the semantic vector to obtain the target semantic vector. The knowledge graph stores various entities related to the business type and the relationships between entities.

[0035] In step S103, semantic completion and contextual analysis are performed using a knowledge graph corresponding to the business type in the business rule document to resolve semantic ambiguity and vagueness in the text. A multimodal pre-trained large model is used as the core, combined with a dynamic knowledge graph to achieve semantic enhancement. After inputting enterprise policy documents or contract terms, the model generates vector representations through semantic embedding technology and matches them with entity relationships in the knowledge graph to extract conditions, actions, and constraints from the rules. It supports unified parsing of heterogeneous data from multiple sources such as text, tables, and images, and effectively resolves semantic ambiguity by replacing traditional keyword matching with semantic vector matching (such as cosine similarity calculation). For example, for clauses like "contract signing requires legal approval," "legal approval" can be automatically identified as an action node and associated with the approval process entity in the knowledge graph to generate a semantically enhanced target semantic vector.

[0036] Step S104: Extract the target rule template that matches the target semantic vector from the rule template library, and build a business decision model based on the target rule template and rule features.

[0037] In step S104, the template selection strategy is optimized using a reinforcement learning algorithm. Logical elements in the text are extracted using regular expressions and dependency parsing. The template matching engine dynamically selects the target rule template based on historical execution results, eliminating the need for a manually predefined template library. The rule template library provides a visual interface for users to customize templates and evaluates template applicability in real time, reducing development costs and improving flexibility. For example, for rules like "purchases exceeding 500,000 yuan require dual approval," the "amount threshold + number of approvers" template is automatically matched, dynamically generating a rule structure that matches the business scenario. After determining the target rule template, a business decision model is automatically constructed based on the rule characteristics and template structure. The business decision model includes the conditions, actions, and constraints of specific business rules, as well as the decision process logic, ensuring that the business rules can be understood and executed by the business system.

[0038] After generating the business decision model, a chain operation technique using a large language model is employed for multi-model fusion optimization. Rule descriptions are generated through pre-trained large models and semantic calibration is performed. The rule engine then performs logical verification to ensure the accuracy and executability of the business decision model.

[0039] The method for constructing a business decision model provided in this application embodiment receives an input business rule file, performs text preprocessing on the business rule file to obtain multimodal data; extracts rule features from the multimodal data, and converts the rule features into semantic vectors using a pre-trained large language model, wherein the rule features include at least one of the following: rule conditions, rule actions, and rule constraints; determines the business type of the business rule file and the corresponding knowledge graph of the business type, performs semantic enhancement processing on the semantic vectors based on the relevant entities associated with the rule features in the knowledge graph, and obtains a target semantic vector, wherein the knowledge graph stores various entities related to the business type and the relationships between entities; extracts a target rule template matching the target semantic vector from a rule template library, and constructs a business decision model based on the target rule template and rule features, thus solving the problem of low efficiency in constructing business decision models manually in related technologies. By employing multimodal data processing technology for business rule documents, the entire process from inputting regulatory documents to outputting standardized business decision-making models is automated. This overcomes technical bottlenecks such as reliance on manual intervention, limitations of rule designers, and static rule engines. Through the deep semantic understanding capabilities of large language models and entity association information from knowledge graphs, more accurate business decision-making models can be generated. Through template matching and decision-making model construction processes, it is ensured that even when faced with complex rule documents, applicable models can be generated quickly, thereby reducing business response time, improving overall operational efficiency, and ultimately enhancing the efficiency of business decision-making model construction.

[0040] Template matching can be based on the similarity between the target semantic vector and the template. Optionally, in the method for constructing the business decision model provided in this application embodiment, extracting the target rule template that matches the target semantic vector from the rule template library includes: extracting multiple rule templates from the rule template library and determining the template vector of each rule template. The rule template library stores multiple types of rule templates, and the types of rule templates include at least one of the following: business approval, permission setting, and early warning; calculating the similarity between the target semantic vector and each template vector to obtain the similarity between each rule template and the target semantic vector; when there is only one rule template with a similarity greater than or equal to a first similarity threshold, the rule template corresponding to the similarity is determined as the target rule template; when the similarity of all rule templates is less than the first similarity threshold, the rule template corresponding to the maximum similarity is determined as the target rule template; when there are at least two rule templates with a similarity greater than or equal to the first similarity threshold, the rule template with a similarity greater than or equal to the first similarity threshold is determined as a candidate template, and the candidate template corresponding to the maximum execution effect evaluation value is determined as the target rule template. The execution effect evaluation value is determined by at least one of the historical execution efficiency, historical execution accuracy, and historical feedback results of the rule template.

[0041] In some embodiments, the most matching template is selected from a predefined rule template library based on the similarity between the rule template and the target semantic vector. The rule template library contains various template types, such as approval rules, permission rules, and alert rules, and supports flexible rule customization. For example, the calculation formula used to filter candidate templates is as follows:

[0042]

[0043] Where T is the rule template library, F is the target semantic vector of the input rule features, and t is the template vector of each rule template. * For the selected candidate templates, similarity is the similarity approximation function. When there are multiple selected candidate templates, the execution effect evaluation value can be calculated comprehensively based on the historical execution efficiency, historical execution accuracy, and historical feedback results of the rule template, and the candidate template with the highest execution effect evaluation value can be selected as the most matching target rule template.

[0044] This embodiment uses a template matching algorithm based on similarity calculation to automatically identify and call the most matching target rule template, replacing the fixed pattern of manually predefined template libraries. This highly automates the construction and adaptation process of the rule model, significantly reducing development complexity. Based on the historical execution performance evaluation of the templates, the template with the highest applicability is selected, ensuring that the generated business decision model can be executed efficiently and accurately in practical applications.

[0045] To facilitate the automated construction of business decision models from business rule files, it is necessary to preprocess the business rule files to generate multimodal data. Optionally, in the method for constructing a business decision model provided in this application embodiment, preprocessing the business rule files to obtain multimodal data includes: denoising the business rule files to obtain denoised business rule files; processing the denoised business rule files to obtain processed business rule files, wherein the data processing includes at least one of the following: image enhancement processing of image data, segmentation and sentence segmentation processing of text data, and structured data transformation processing of table data; extracting text information and graphic elements from the processed business rule files using a text recognition model to obtain initial data; and processing the initial data using a natural language processing model to obtain multimodal data, wherein the multimodal data includes at least one of the following: text, tables, and images.

[0046] In some embodiments, text preprocessing may include noise removal, format conversion, and quality checks to ensure data consistency and integrity. For image data, image enhancement techniques (such as contrast adjustment and deblurring) are used to improve the accuracy of character recognition; while for text data, noise (such as special symbols and stop words) is removed, and segmentation and sentence segmentation are performed. Next, a deep learning-based character recognition model extracts text information from the images and identifies graphical elements such as tables and charts. In the natural language processing stage, a base segmenter is used to segment the text, and a pre-trained large model is combined for syntactic analysis and named entity recognition to extract key entities and semantic features. Finally, the processed text, image, and table data are standardized and uniformly stored as structured data to obtain multimodal data.

[0047] This embodiment utilizes chained operation techniques of large language models, combining multiple large models and employing chained calls to further extract and enhance key information from multimodal data, improving the accuracy and efficiency of data parsing. The preprocessing workflow covers comprehensive data processing from text and tables to images, ensuring the integrity and accuracy of multimodal information. Using deep learning models for text recognition and natural language processing enables more accurate extraction and understanding of semantic information within business rules. Interference information is removed, and techniques such as image enhancement and data transformation are used to improve the clarity and structure of key information. Text, table, and image information are effectively integrated to form a unified multimodal dataset, providing rich and comprehensive data support for subsequent rule generation.

[0048] Natural language processing can include word segmentation, entity recognition, and syntactic analysis. Optionally, in the method for constructing a business decision model provided in this application embodiment, processing initial data through a natural language processing model to obtain multimodal data includes: segmenting the initial data using the word segmentation module in the natural language processing model to obtain segmented data; analyzing the initial data using the dependency parsing module in the natural language processing model to obtain the syntactic structure of the initial data; extracting multiple entities from the initial data using the entity recognition module in the natural language processing model; and standardizing the segmented data, syntactic structure, and multiple entities to obtain multimodal data.

[0049] In some embodiments, the word segmentation module in the natural language processing model can use rule-based word segmentation algorithms (such as forward maximum matching and backward maximum matching) or statistical word segmentation algorithms to segment the initial data. The dependency resolution module can capture long-distance dependency relationships based on a deep learning model. The dependency resolution module generates a dependency tree, identifying the dependency relationships and roles of each word, thereby understanding the structure of complex sentences. The entity recognition module can identify named entities in text by training a natural language processing model, which helps to automatically identify key roles and objects from regulatory documents. The identified entities are linked to entities in a knowledge graph to enhance the semantic information of the entities. All processed data is ensured to be in a consistent format for easy reading and processing by subsequent modules.

[0050] For example, the dependency parsing algorithm of a syntactic parser parses the grammatical structure of text and identifies the subject-verb-object relationship. The expression for the identification process is as follows:

[0051] G = SyntaxParser(S);

[0052] Where G represents the syntax graph, S is the initial input data, the entity recognizer uses the named entity recognition algorithm to identify key entities in the text, and combines the knowledge graph for semantic completion, as shown in the following expression:

[0053] E'=SemanticEn richment(E,KG);

[0054] Where E represents the identified entity, KG represents the knowledge graph, and E' represents the entity information after completion and deambiguation.

[0055] This embodiment utilizes natural language processing (NLP) technology to ensure that the rule features extracted from the initial data are both accurate and comprehensive, avoiding potential biases and omissions that may occur during manual parsing. Through standardization of the segmented data, grammatical structure, and multiple entities, the system can integrate text, image, and tabular information into a unified data structure, facilitating subsequent processing by deep learning models. The automated NLP process increases data processing speed, reduces the need for manual intervention, and thus reduces error rates and costs.

[0056] After selecting a matching target rule template, a business decision model is constructed by filling in rule features. Optionally, in the method for constructing a business decision model provided in this application embodiment, constructing a business decision model based on a target rule template and rule features includes: extracting rule variables from the target rule template, wherein the rule variables include at least one of the following: condition variables, action variables, and constraint variables; matching rule features with rule variables; and filling the rule features into the positions of the rule variables in the target rule template that match the rule features, thereby obtaining the business decision model.

[0057] In some embodiments, rule variables are extracted from a selected target rule template. These variables can be condition variables (such as triggering conditions like "if..." or "when..."), action variables (such as operation instructions like "execute..." or "generate..."), or constraint variables (such as limiting conditions like "only applicable to..." or "must be completed within..."). The rule variables are categorized to ensure that each variable matches a specific function or role in the business rule. For example, condition variables may be related to logical judgments or state detection scenarios, while action variables may involve system operations or data processing. During the multimodal data processing stage, the system has generated rule features, including text rule descriptions, numerical conditions in tables, examples in images, or flowcharts. These rule features are matched one-to-one with the rule variables extracted from the template. Ideally, each feature should correspond to a variable in the template. For example, the feature "if the amount exceeds 1000 yuan" would correspond to the "amount threshold" position in the condition variables.

[0058] The matched rule features are then populated into the corresponding variable positions in the rule template to generate a preliminary framework for the business decision model. For complex templates containing multiple rule variables, logical integration is required to connect the populated features and form coherent business logic. For example, multiple condition variables can be connected using logical operators (such as AND and OR) to form complex conditional statements.

[0059] For example, combining the chaining technique of large language models, assuming the rule generation model is G, the expression for generating the business decision model is:

[0060] R = G(F,t) * );

[0061] Where R is the generated business decision model.

[0062] This embodiment utilizes a pre-trained large language model and deep learning algorithms to automatically identify and match rule templates and rule features, reducing errors and inconsistencies caused by manual intervention. The entire process, from rule template selection to feature matching, and then to model building and optimization, is highly automated, greatly improving efficiency and reducing labor costs. Automating the generation of business decision models from complex business rule documents significantly improves the efficiency and accuracy of compliance management and business decision-making.

[0063] After constructing the business decision model, it needs to be tested. Optionally, in the business decision model construction method provided in this application embodiment, after constructing the business decision model based on the target rule template and rule features, the method further includes: running the business decision model in a test environment to obtain test results, wherein the test results include at least one of the following: rule triggering conditions, decision execution results, and boundary conditions; if the similarity between the test results and the expected results is less than a second similarity threshold, issuing a first prompt message, wherein the first prompt message is used to indicate that the business decision model does not match the business rule file; if the similarity between the test results and the expected results is greater than or equal to the second similarity threshold, performing formal verification of the business decision model through a logic reasoning engine to obtain verification results; if the verification results indicate that the business decision model has logical vulnerabilities or conflicts, issuing a second prompt message, wherein the second prompt message is used to prompt maintenance personnel to fix the logical vulnerabilities or conflicts; if the verification results indicate that the business decision model does not have logical vulnerabilities or conflicts, determining that the business decision model test has passed.

[0064] In some embodiments, the effectiveness of the generated business decision model is verified in a simulated testing environment. Testing includes verifying the triggering conditions and execution effects of the business decision model under various business scenarios, checking the boundary conditions of the rules, and performing logical reasoning and conflict detection using formal verification tools to ensure the correctness and consistency of the rules. For example, multiple business scenarios (such as approval processes, financial accounting, and access control) are constructed to test the triggering conditions and execution effects of the rules under different scenarios; the behavior of the rules under extreme conditions (such as time boundaries and monetary limits) is verified; and the rules are formally verified using a logical reasoning engine to ensure their correctness and conflict-free nature.

[0065] This embodiment comprehensively verifies the effectiveness of the rules through scenario simulation and boundary condition testing. Formal verification tools are used to ensure the correctness and consistency of the rules. An automated testing framework and logical reasoning engine enable independent model evaluation and verification, reducing manual intervention and improving efficiency. Timely feedback from the first and second error messages allows the team to quickly identify problems and make targeted corrections, accelerating the model optimization process.

[0066] If the business decision model passes the test, it is deployed to the target business system. Optionally, in the method for constructing the business decision model provided in this application embodiment, after determining that the business decision model has passed the test, the method further includes: determining the target business system to which the business rule file will be applied, and deploying the business decision model through the rule service interface of the target business system; if the deployment of the business decision model fails, controlling the rule service interface to roll back the business rules of the target business system to the previous version of the business decision model; if the deployment of the business decision model is successful, checking whether the rule service interface has received a new version of the business decision model every preset time interval; if the rule service interface is detected to have received a new version of the business decision model, deploying the new version of the business decision model through the rule service interface.

[0067] In some embodiments, standardized rule output, integration and adaptation, and version management and updates ensure the efficient application and management of the generated business decision-making model. First, the generated business decision-making model is output in a standardized format (such as various code statements) to ensure compatibility with existing business systems. Next, the business decision-making model is integrated into existing business systems through standardized interfaces, supporting real-time invocation and updates. Simultaneously, application interface documentation and integration guidelines are provided to facilitate interfacing with other systems (such as approval systems and financial systems). Furthermore, version management of the business decision-making model is supported, recording historical information for each update and providing a rollback mechanism to ensure rapid recovery to a stable version in case of update failure. These measures collectively guarantee the stability and flexibility of rule generation, application, and maintenance, enabling the entire system to adapt to changes in business needs and ensuring efficient and reliable operation. Through a canary release mechanism and impact assessment algorithm, new rule versions are deployed in batches, avoiding the business interruption risk caused by full updates. A visual template management interface is also provided to enable real-time adjustment and rapid iteration of the rule model, reducing manual intervention and improving the adaptability and business agility of the device.

[0068] This embodiment features an automatic rollback strategy that triggers when model deployment fails, enabling rapid system recovery to a stable state and avoiding business interruption risks. By automatically detecting and deploying new model versions through the rules service interface, continuous optimization of business decisions and regulatory compliance are achieved, reducing manual intervention. The version management function tracks and manages the model's update history, facilitating problem localization and version rollback, ensuring the traceability and controllability of model updates.

[0069] According to another embodiment of this application, a business rule generation apparatus is also provided. Figure 2 This is a schematic diagram of a business rule generation apparatus provided according to an embodiment of this application. For example... Figure 2 As shown, the device includes: an input terminal, a rule-generating agent, and an output terminal.

[0070] Specifically, the input terminal supports input from various data sources, including text files, spreadsheets, and image data. The multimodal data parsing module, based on a deep learning-based text recognition engine, can efficiently extract text and graphic information from images. Combined with a pre-trained model, it performs text segmentation, syntactic analysis, named entity recognition, and semantic understanding. Data mining techniques are used to extract structured information from tabular data, analyze rule parameters, and construct a rule feature library to ensure comprehensive data parsing and processing.

[0071] The dynamic rule modeling framework constructs flexible and adaptable rule models through syntactic analysis, entity recognition, and semantic embedding techniques. Dependency parsing is used to extract the syntactic structure of the text, identify subject-verb-object relationships, and identify key entities in the text, such as time, location, amount, and legal terms. Semantic embedding techniques are used to transform textual information into high-dimensional semantic vectors to support semantic matching and rule modeling, ensuring the flexibility and adaptability of rule generation.

[0072] The rule template matching engine, based on a predefined rule template library and leveraging the multi-model fusion capabilities of chained operations in large language models, achieves high-precision rule matching and generation. It selects the optimal rule template through a template matching algorithm based on edit distance and semantic similarity. Multi-model fusion optimizes the accuracy and semantic completeness of rule generation by integrating pre-trained models. Based on a dynamic rule model, it performs syntactic and semantic optimization on the generated rules, ensuring their executability and logical consistency.

[0073] The rule self-checking and simulation testing modules conduct comprehensive testing through a simulated testing environment. Scenario simulations construct various business scenarios, such as approval processes, financial accounting, and access control, testing the triggering conditions and execution effects of rules under different scenarios. Boundary condition testing verifies the behavior of rules under extreme conditions (such as time limits and monetary limits). A logical reasoning engine performs formal verification of the rules to ensure their correctness and conflict-free nature.

[0074] The device architecture comprises a data interface layer (for unstructured data acquisition and preprocessing), a core processing layer (for multimodal parsing modules, dynamic rule frameworks, and template matching engines), a storage layer (for rule models, knowledge graphs, and historical version data), and a user interaction layer (for a visual interface that allows for template customization, execution monitoring, and parameter adjustment). Each module operates collaboratively through interface calls and message queues to achieve fully automated processing and end-to-end closed-loop optimization.

[0075] The business rule generation device achieves automatic rule generation through the following steps: Figure 3 This is a schematic diagram of a business rule generation method provided according to an embodiment of this application. For example... Figure 3 As shown, the method includes: receiving multimodal data, performing syntactic analysis and named entity recognition, and extracting key entities and semantic features; extracting rule features from standardized data, and constructing a dynamic rule model by combining semantic understanding and association analysis; selecting matching templates from a rule template library, and generating specific business rule models by combining the dynamic rule model. The effectiveness of the rule model is verified in a simulated environment, testing triggering conditions, execution effects, and boundary conditions. Based on the rule model, rule services are automatically published, and versions are created or updated.

[0076] The business rule generation device in this embodiment achieves automated generation and dynamic adaptation of enterprise business rules through the collaborative design of multimodal data parsing, a dynamic rule framework, a template matching engine, and a dynamic rule service update mechanism. This device uses a multimodal pre-trained large model as the core parsing component, combined with dynamic knowledge graphs and reinforcement learning algorithms, to overcome the technical bottlenecks of traditional rule generation, such as reliance on manually predefined templates, weak semantic ambiguity handling capabilities, and high version update risks. The overall architecture includes a data interface layer, a processing core layer, a storage layer, and a user interaction layer. Module collaboration is achieved through interface calls and message queues, completing the fully automated processing from unstructured data input to rule service output.

[0077] The business rule generation device introduces intelligent and automated tools to replace manual processes. It utilizes optical character recognition, table parsing, and cross-modal alignment technologies to achieve end-to-end automated processing of regulatory documents, improving rule extraction efficiency and reducing human error. Deep learning algorithms optimize the correlation analysis of text, images, and tables, ensuring the integrity and accuracy of the rule model. A rule engine that dynamically responds to regulatory changes is designed, enabling automatic updates and version management of the rule model, reducing manual maintenance costs and improving the device's adaptability. Through the collaborative design of intelligent agents and a dynamic rule framework, the device significantly improves the automation level and flexibility of business rule generation, solving problems such as low efficiency, rigid rule templates, and high risks associated with version updates in traditional manual processes, providing efficient and scalable technical support for enterprise compliance management.

[0078] Leveraging the reasoning capabilities of large-scale pre-trained models, an intelligent agent with autonomous learning and dynamic adaptation capabilities is constructed to achieve deep integration and optimization of the entire process of enterprise business rule generation. Through the semantic understanding capabilities of the large model, the agent can accurately parse complex regulatory texts and, combined with multimodal data fusion technology, significantly improve the processing accuracy of unstructured data such as tables and images, addressing the shortcomings of traditional rule designers in cross-modal data correlation analysis. Simultaneously, utilizing the logical reasoning capabilities of the large model, the agent can analyze the contextual semantics of regulatory changes in real time, automatically match preset rule templates, generate adapted rule models, and dynamically update rule services through rule service interfaces, thereby achieving version management of the rule base and synchronous adaptation with business systems. Furthermore, through the unified representation learning of text, images, and tables by the large model, the agent solves the semantic fragmentation problem between multi-source data, ensuring the reliability of the rule model in terms of logical consistency and completeness.

[0079] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0080] This application also provides an apparatus for constructing a business decision model. It should be noted that this apparatus can be used to execute the method for constructing a business decision model provided in this application. The following describes the apparatus for constructing a business decision model provided in this application.

[0081] Figure 4 This is a schematic diagram of a business decision model construction apparatus provided according to an embodiment of this application. Figure 4 As shown, the device includes:

[0082] The receiving unit 401 is used to receive the input business rule file, perform text preprocessing on the business rule file, and obtain multimodal data;

[0083] Extraction unit 402 is used to extract rule features from multimodal data and convert the rule features into semantic vectors through a pre-trained large language model. The rule features include at least one of the following: rule conditions, rule actions, and rule constraints.

[0084] The first determining unit 403 is used to determine the business type of the business rule file and the knowledge graph corresponding to the business type. Based on the relevant entities associated with the rule features in the knowledge graph, the semantic vector is semantically enhanced to obtain the target semantic vector. The knowledge graph stores various entities related to the business type and the relationships between entities.

[0085] Construction unit 404 is used to extract target rule templates that match the target semantic vector from the rule template library, and to build a business decision model based on the target rule templates and rule features.

[0086] The business decision model construction apparatus provided in this application embodiment receives an input business rule file through a receiving unit 401, performs text preprocessing on the business rule file to obtain multimodal data; an extraction unit 402 extracts rule features from the multimodal data, and converts the rule features into semantic vectors through a pre-trained large language model, wherein the rule features include at least one of the following: rule conditions, rule actions, and rule constraints; a first determining unit 403 determines the business type of the business rule file and determines the knowledge graph corresponding to the business type, performs semantic enhancement processing on the semantic vectors based on the relevant entities associated with the rule features in the knowledge graph, and obtains the target semantic vector, wherein the knowledge graph stores various entities related to the business type and the relationships between entities; and a construction unit 404 extracts the relevant entities from the rule template library. The target rule template for semantic vector matching is used to construct a business decision model based on the target rule template and rule features. This solves the problem of low efficiency in constructing business decision models manually in related technologies. By using multimodal data processing technology for business rule files, the entire process from input of regulatory documents to output of standardized business decision models is automated. This overcomes technical bottlenecks such as reliance on manual labor, limitations of rule designers, and static rule engines. Through the deep semantic understanding capabilities of large language models and entity association information of knowledge graphs, more accurate business decision models can be generated. Through template matching and decision model construction processes, it can ensure that even when faced with complex rule files, applicable models can be generated quickly, thereby reducing business response time, improving overall operational efficiency, and ultimately improving the efficiency of business decision model construction.

[0087] Optionally, in the business decision model construction apparatus provided in this application embodiment, the construction unit 404 includes: a first extraction module, used to extract multiple rule templates from a rule template library and determine the template vector of each rule template, wherein the rule template library stores multiple types of rule templates, and the types of rule templates include at least one of the following: business approval, permission setting, and early warning; a calculation module, used to calculate the similarity between the target semantic vector and each template vector to obtain the similarity between each rule template and the target semantic vector; a first determination module, used to determine the rule template corresponding to the similarity as the target rule template when there is only one rule template with a similarity greater than or equal to a first similarity threshold; a second determination module, used to determine the rule template corresponding to the maximum similarity as the target rule template when the similarity of all rule templates is less than the first similarity threshold; a third determination module, used to determine the rule template with a similarity greater than or equal to the first similarity threshold as a candidate template and determine the candidate template corresponding to the maximum execution effect evaluation value as the target rule template when there are at least two rule templates with a similarity greater than or equal to the first similarity threshold, wherein the execution effect evaluation value is determined by at least one of the historical execution efficiency, historical execution accuracy, and historical feedback results of the rule template.

[0088] Optionally, in the business decision model construction apparatus provided in this application embodiment, the receiving unit 401 includes: a noise reduction module, used to perform noise reduction processing on the business rule file to obtain a noise-reduced business rule file; a data processing module, used to perform data processing on the noise-reduced business rule file to obtain a processed business rule file, wherein the data processing includes at least one of the following: image enhancement processing on image data, segmentation and sentence segmentation processing on text data, and structured data conversion processing on table data; a second extraction module, used to extract text information and graphic elements in the processed business rule file through a text recognition model to obtain initial data; and a processing module, used to process the initial data through a natural language processing model to obtain multimodal data, wherein the multimodal data includes at least one of the following: text, tables, and images.

[0089] Optionally, in the business decision model construction apparatus provided in this application embodiment, the processing module includes: a first processing submodule, used to perform word segmentation processing on the initial data through the word segmentation module in the natural language processing model to obtain the segmented data; an analysis submodule, used to analyze the initial data through the dependency parsing module in the natural language processing model to obtain the grammatical structure of the initial data; an extraction submodule, used to extract multiple entities from the initial data through the entity recognition module in the natural language processing model; and a second processing submodule, used to perform standardization processing on the segmented data, grammatical structure, and multiple entities to obtain multimodal data.

[0090] Optionally, in the business decision model construction apparatus provided in this application embodiment, the construction unit 404 includes: a third extraction module, used to extract rule variables from the target rule template, wherein the rule variables include at least one of the following: condition variables, action variables, and constraint variables; and a matching module, used to match rule features with rule variables, and fill the rule features into the positions of the rule variables that match the rule features in the target rule template to obtain the business decision model.

[0091] Optionally, in the business decision model construction apparatus provided in this application embodiment, the apparatus further includes: a testing unit, configured to run the business decision model in a test environment and obtain test results, wherein the test results include at least one of the following: rule triggering conditions, decision execution results, and boundary conditions; a first prompting unit, configured to issue a first prompting message when the similarity between the test results and the expected results is less than a second similarity threshold, wherein the first prompting message is used to indicate that the business decision model does not match the business rule file; a verification unit, configured to perform formal verification of the business decision model through a logical reasoning engine when the similarity between the test results and the expected results is greater than or equal to the second similarity threshold, and obtain verification results; a second prompting unit, configured to issue a second prompting message when the verification results indicate that the business decision model has logical vulnerabilities or conflicts, wherein the second prompting message is used to prompt maintenance personnel to fix the logical vulnerabilities or conflicts; and a second determining unit, configured to determine that the business decision model test has passed when the verification results indicate that the business decision model does not have logical vulnerabilities or conflicts.

[0092] Optionally, in the business decision model construction apparatus provided in this application embodiment, the apparatus further includes: a third determining unit, configured to determine the target business system to which the business rule file will be applied, and deploy the business decision model through the rule service interface of the target business system; a rollback unit, configured to control the rule service interface to roll back the business rules of the target business system to the previous version of the business decision model if the deployment of the business decision model fails; a detection unit, configured to detect whether the rule service interface has received a new version of the business decision model at preset intervals if the deployment of the business decision model is successful; and a deployment unit, configured to deploy the new version of the business decision model through the rule service interface if the rule service interface is detected to have received a new version of the business decision model.

[0093] The business decision model construction device includes a processor and a memory. The receiving unit 401, the extraction unit 402, the first determining unit 403, and the construction unit 404 are all stored in the memory as program units. The processor executes the program units stored in the memory to realize the corresponding functions.

[0094] The processor contains a kernel, which retrieves the corresponding program units from memory. One or more kernels can be configured, and adjusting kernel parameters can improve the efficiency of building business decision models.

[0095] The memory may include non-permanent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0096] This invention provides a computer-readable storage medium storing a program that, when executed by a processor, implements a method for constructing a business decision model.

[0097] This invention provides a processor for running a program, wherein the program executes a method for constructing a business decision model during runtime.

[0098] Figure 5 This is a schematic diagram of an electronic device provided according to an embodiment of this application. For example... Figure 5 As shown, electronic device 501 includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it performs the following steps: receiving an input business rule file; preprocessing the business rule file to obtain multimodal data; extracting rule features from the multimodal data; converting the rule features into semantic vectors using a pre-trained large language model, wherein the rule features include at least one of the following: rule conditions, rule actions, and rule constraints; determining the business type of the business rule file and the corresponding knowledge graph; performing semantic enhancement processing on the semantic vectors based on relevant entities associated with the rule features in the knowledge graph to obtain a target semantic vector, wherein the knowledge graph stores various entities related to the business type and the relationships between entities; extracting a target rule template matching the target semantic vector from a rule template library; and constructing a business decision model based on the target rule template and rule features. The device in this paper can be a server, PC, PAD, mobile phone, etc.

[0099] This application also provides a computer program product, which, when executed on a data processing device, is suitable for executing an initialization program with the following method steps: receiving an input business rule file; performing text preprocessing on the business rule file to obtain multimodal data; extracting rule features from the multimodal data; converting the rule features into semantic vectors through a pre-trained large language model, wherein the rule features include at least one of the following: rule conditions, rule actions, and rule constraints; determining the business type of the business rule file and determining the knowledge graph corresponding to the business type; performing semantic enhancement processing on the semantic vectors based on the relevant entities associated with the rule features in the knowledge graph to obtain a target semantic vector, wherein the knowledge graph stores various entities related to the business type and the relationships between entities; extracting a target rule template matching the target semantic vector from a rule template library; and constructing a business decision model based on the target rule template and rule features.

[0100] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0101] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0102] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0103] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0104] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0105] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, like read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0106] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0107] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0108] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0109] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for constructing a business decision-making model, characterized in that, include: Receive the input business rule file, perform text preprocessing on the business rule file, and obtain multimodal data; Rule features are extracted from the multimodal data, and the rule features are converted into semantic vectors using a pre-trained large language model. The rule features include at least one of the following: rule conditions, rule actions, and rule constraints. The business type of the business rule file is determined, and the knowledge graph corresponding to the business type is determined. Based on the relevant entities in the knowledge graph that are associated with the rule features, the semantic vector is semantically enhanced to obtain the target semantic vector. The knowledge graph stores various entities related to the business type and the association relationships between the entities. Extract target rule templates that match the target semantic vector from the rule template library, and construct a business decision model based on the target rule templates and the rule features.

2. The method according to claim 1, characterized in that, Extracting target rule templates that match the target semantic vector from the rule template library includes: Multiple rule templates are extracted from the rule template library, and a template vector for each rule template is determined. The rule template library stores multiple types of rule templates, and the types of rule templates include at least one of the following: business approval, permission setting, and early warning. Calculate the similarity between the target semantic vector and each template vector to obtain the similarity between each rule template and the target semantic vector; If there is only one rule template with a similarity greater than or equal to the first similarity threshold, the rule template corresponding to the similarity is determined as the target rule template. If the similarity of all rule templates is less than the first similarity threshold, the rule template with the highest similarity is determined as the target rule template. If there are at least two rule templates with a similarity greater than or equal to the first similarity threshold, the rule template with a similarity greater than or equal to the first similarity threshold is determined as a candidate template, and the candidate template corresponding to the maximum execution effect evaluation value is determined as the target rule template. The execution effect evaluation value is determined by at least one of the rule template's historical execution efficiency, historical execution accuracy, and historical feedback results.

3. The method according to claim 1, characterized in that, The business rule file is preprocessed to obtain multimodal data, including: The business rule file is subjected to noise reduction processing to obtain a noise-reduced business rule file; The denoised business rule file is subjected to data processing to obtain a processed business rule file, wherein the data processing includes at least one of the following: image enhancement processing of image data, segmentation and sentence segmentation processing of text data, and structured data transformation processing of table data; The text information and graphic elements in the processed business rule file are extracted using a text recognition model to obtain initial data. The initial data is processed by a natural language processing model to obtain the multimodal data, wherein the multimodal data includes at least one of the following: text, tables, and images.

4. The method according to claim 3, characterized in that, The initial data is processed by a natural language processing model to obtain the multimodal data, which includes: The initial data is segmented using the word segmentation module in the natural language processing model to obtain segmented data. The initial data is analyzed by the dependency parsing module in the natural language processing model to obtain the syntactic structure of the initial data; Multiple entities are extracted from the initial data using the entity recognition module in the natural language processing model. The segmented data, the grammatical structure, and the multiple entities are standardized to obtain the multimodal data.

5. The method according to claim 1, characterized in that, Constructing a business decision model based on the target rule template and the rule features includes: Rule variables are extracted from the target rule template, wherein the rule variables include at least one of the following: condition variables, action variables, and constraint variables; The rule features are matched with the rule variables, and the rule features are filled into the positions of the rule variables that match the rule features in the target rule template to obtain the business decision model.

6. The method according to claim 1, characterized in that, After constructing the business decision model based on the target rule template and the rule features, the method further includes: Run the business decision model in a test environment to obtain test results, wherein the test results include at least one of the following: rule triggering conditions, decision execution results, and boundary conditions; If the similarity between the test result and the expected result is less than a second similarity threshold, a first prompt message is issued, wherein the first prompt message is used to indicate that the business decision model does not match the business rule file; If the similarity between the test result and the expected result is greater than or equal to the second similarity threshold, the business decision model is formally verified through a logical reasoning engine to obtain the verification result. If the verification result indicates that the business decision model has a logical vulnerability or conflict, a second prompt message is issued, wherein the second prompt message is used to prompt the operation and maintenance personnel to fix the logical vulnerability or conflict; If the verification results indicate that the business decision model does not have the aforementioned logical flaws or conflicts, then the business decision model is deemed to have passed the test.

7. The method according to claim 6, characterized in that, After confirming that the business decision model has passed the test, the method further includes: Determine the target business system to which the business rule file will be applied, and deploy the business decision model through the rule service interface of the target business system; In the event that the business decision model fails to deploy, the rule service interface is controlled to revert the business rules of the target business system to the previous version of the business decision model. If the business decision model is successfully deployed, the rule service interface is checked every preset time interval to see if a new version of the business decision model has been received. Upon detecting that the rule service interface has received the new version of the business decision model, the new version of the business decision model is deployed through the rule service interface.

8. A device for constructing a business decision-making model, characterized in that, include: The receiving unit is used to receive the input business rule file, perform text preprocessing on the business rule file, and obtain multimodal data. An extraction unit is configured to extract rule features from the multimodal data and convert the rule features into semantic vectors using a pre-trained large language model, wherein the rule features include at least one of the following: rule conditions, rule actions, and rule constraints; The first determining unit is used to determine the business type of the business rule file and the knowledge graph corresponding to the business type, and to perform semantic enhancement processing on the semantic vector based on the relevant entities associated with the rule features in the knowledge graph to obtain the target semantic vector. The knowledge graph stores various entities related to the business type and the association relationships between the entities. The construction unit is used to extract target rule templates that match the target semantic vector from the rule template library, and to construct a business decision model based on the target rule templates and the rule features.

9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the method for constructing the business decision model according to any one of claims 1 to 7.

10. An electronic device, characterized in that, It includes one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the method for constructing a business decision model as described in any one of claims 1 to 7.

Images