A method for developing enterprise business application software

Abstract

The application discloses a kind of enterprise business application software development method, it is related to application software development technical field.The method includes the following steps: constructing business semantic knowledge graph, and enterprise business demand is decomposed into reusable business semantic unit;Core business entity, business action and business rule element are extracted by carrying out word segmentation processing, entity identification and intention understanding to business demand description text through semantic analysis engine;The extracted element is matched with knowledge graph, and semantic mapping relationship chain is generated;Code generation rule engine is called based on mapping relationship chain, and program code framework is automatically generated;And semantic integrity check and logic consistency verification are carried out to code framework, and final code is output.The application establishes the automatic conversion mechanism of business demand to code, realizes the quick response of business change, guarantees demand semantic accuracy, and effectively reduces code error rate.

Description

Technical Field

[0001] This invention relates to the field of application software development technology, and in particular to a method for developing enterprise business application software. Background Technology

[0002] In the development of enterprise business application software, frequent changes in business requirements have always been a core challenge in the field of software engineering. How to respond to these changes quickly and accurately at the code level directly determines the efficiency and quality of software development.

[0003] Common business application software development methodologies typically employ modular design, design patterns, or low-code development platforms to address requirement changes by enhancing code structure and component reusability. However, these methods are often limited by the complexity of business logic in practice. When faced with fine-grained adjustments to business rules, developers still need to delve into code details to modify line by line or change the logic, leading to low development efficiency and a high risk of introducing new errors during the modification process. Furthermore, the lack of an effective semantic mapping mechanism between business logic and code implementation makes it difficult to fundamentally guarantee code understandability and maintainability. Accurate response to business changes has become a critical challenge that urgently needs to be addressed in current software development methodologies. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method for developing enterprise business application software based on a semantic-driven engine, which can effectively solve the problems mentioned in the background technology.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: A method for developing enterprise business application software includes the following steps: Build a business semantic model library, break down enterprise business requirements into reusable business semantic units, establish the relationship between each business semantic unit, and form a structured business semantic knowledge graph; Based on the business semantic knowledge graph, the business requirement description text is received, and the business requirement description text is processed by word segmentation, entity recognition and intent understanding through the semantic parsing engine to extract core business entities, business actions and business rule elements. The extracted core business entities, business actions, and business rule elements are semantically matched with the corresponding business semantic units in the business semantic knowledge graph to generate a semantic mapping relationship chain for business requirements. Based on the semantic mapping relationship chain, the code generation rule engine is invoked to automatically generate a program code framework corresponding to the business requirements, according to the code template library and the mapping relationship between business semantics and code. The generated program code framework is subjected to semantic integrity verification and logical consistency verification. Based on the verification results, the program code framework is automatically corrected or manually annotated and adjusted to output the final enterprise business application software code.

[0006] Preferably, the construction of the business semantic model library specifically includes: Extract business description information, business process definition files, and business rule configuration files from the enterprise's existing business application systems; Natural language processing is performed on the extracted business description information to identify and extract business entity nouns, business action verbs, and business conditional adverbs. The extracted business entity nouns are conceptually abstracted and classified into hierarchical levels to construct a hierarchical structure system for business entities.

[0007] Preferably, the extracted business actions are semantically standardized to establish a mapping relationship between business actions and system operations, specifically including: Semantic disambiguation is performed on the identified business action verbs to determine their specific meaning in a particular business scenario based on the context of the business action verbs. Different verb forms expressing the same meaning are processed by synonym unification and standardized action terms. Establish a mapping table between business actions and system operations.

[0008] Preferably, the extracted business condition adverbs are transformed into logical expressions to construct computable business rule condition expressions, specifically including: Identify the condition body, condition operators, and condition thresholds in business condition adverbs, where condition operators include comparison operators, logical operators, and inclusion operators; The parsed conditional adverbs are converted into standardized logical expression formats, which support condition combination, condition nesting, and condition branching. Verify the computability of each conditional expression.

[0009] Preferably, the business requirement description text is received based on the business semantic knowledge graph, and word segmentation, entity recognition, and intent understanding are performed through a semantic parsing engine. Specifically, this includes: inputting the business requirement description text into a pre-trained natural language processing model, whereby the natural language processing model performs Chinese word segmentation and part-of-speech tagging on the business requirement description text. The tag set for part-of-speech tagging includes basic part-of-speech categories such as nouns, verbs, adjectives, numerals, classifiers, pronouns, adverbs, prepositions, conjunctions, auxiliary words, modal particles, and interjections.

[0010] Preferably, based on word segmentation and part-of-speech tagging results, a named entity recognition algorithm is used to identify key information in the business requirement description text, specifically including: A two-layer architecture combining conditional random fields and pre-trained models is used for named entity recognition. The pre-trained model layer generates a dynamic vector representation of each word, while the sequence labeling layer identifies the named entity boundaries and entity types in the text. The identified entity types include business entity name, business amount, business time, business personnel, business location, and business quantity.

[0011] Preferably, the business intent of the business requirement description text is classified using an intent classification model, specifically including: using a multi-level classification model to classify business domains, business functions, and operation types. The first level determines the major business category to which the business requirement belongs; the second level identifies the specific business function type under the determined business domain; and the third level identifies the specific operation type corresponding to the business requirement.

[0012] Preferably, the semantic role labeling model identifies the roles of each semantic component in the business requirement description text. Specifically, it includes: using a semantic role labeling method based on dependency parsing to identify the grammatical dependency relationships between words in the sentence and determine the semantic role of each sentence component; the semantic role types include agent role, patient role, participant role, instrument role, time role, location role, condition role, and result role.

[0013] Preferably, the extracted core business entities are semantically matched with the corresponding business semantic units in the business semantic knowledge graph, specifically including: The semantic similarity calculation involves multi-dimensional feature fusion between the extracted core business entities and the business entity nodes in the business semantic knowledge graph. The dimensions of the calculation include lexical similarity, semantic similarity, contextual similarity, and relational similarity. The extracted business actions are matched with the business action nodes in the business semantic knowledge graph using an action semantic framework.

[0014] Preferably, the generated program code framework is subjected to semantic integrity verification and logical consistency verification, specifically including: The code statements in the program code framework are reverse-mapped and verified to the semantic mapping relationship chain to ensure that each code statement can be traced back to the corresponding business semantic unit; Perform logical coverage analysis on the conditional statements in the program code framework to detect whether there are any unprocessed business rule branches; Perform data integrity checks on data manipulation statements in the program code framework. The check items include required fields, data types, value range, referential integrity, and uniqueness. Perform conflict detection on the business rule logic in the program code framework.

[0015] Compared with the prior art, the advantages and positive effects of the present invention are as follows: By constructing a structured business semantic knowledge graph, a semantic mapping between business requirement descriptions and system code implementation is achieved. This allows business changes to be automatically propagated to the code level for corresponding adjustments through the semantic mapping relationship chain, eliminating the need for developers to modify code line by line and significantly improving the response efficiency of business changes. A semantic parsing engine performs deep semantic understanding of the business requirement description text, extracting core business entities, business actions, and business rule elements, ensuring the completeness and accuracy of the business requirement semantics. A code generation rule engine automatically generates the program code framework based on the semantic mapping relationship, and combined with semantic integrity verification and logical consistency verification, effectively reducing the code error rate and improving the quality and reliability of software development. By establishing a reusable mechanism for business semantic units, the reusability of business components is improved, reducing the cost of repetitive development. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the overall technical solution architecture of the enterprise business application software development method based on a semantic-driven engine according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the core principle framework of the semantic-driven engine in the enterprise business application software development method based on the semantic-driven engine according to an embodiment of the present invention. Figure 3 This is a logical flow framework diagram of the semantic parsing and mapping matching of business requirements in the enterprise business application software development method based on a semantic-driven engine according to an embodiment of the present invention. Figure 4 This is a flowchart illustrating the automatic code generation and verification process in the semantic-driven engine-based enterprise business application software development method of this invention. Detailed Implementation

[0017] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0018] Example 1, referring to Figure 1-4 This is a method for developing enterprise business application software. By constructing a structured business semantic knowledge graph and combining it with natural language processing technology to perform deep semantic understanding of business requirement descriptions, an automated mapping channel from business requirements to code implementation is established, thereby enabling efficient development of enterprise business application software.

[0019] In the specific implementation process, the first step is to build a business semantic model library: specifically, the enterprise's business requirements are broken down into reusable business semantic units, and the relationships between these units are established to form a structured business semantic knowledge graph. This step is the foundational layer of the entire semantic-driven engine, responsible for systematically modeling the knowledge in the enterprise's business domain, providing standardized semantic support for subsequent semantic parsing and code generation.

[0020] The construction of the business semantic model library involves the following specific operations: First, business description information, business process definition files, and business rule configuration files for business functional modules are extracted from the enterprise's existing business application systems. During the extraction process, business description information includes technical documents such as design documents, user manuals, and interface specifications for each functional module; business process definition files include flowchart files, process configuration files, and workflow definition documents; and business rule configuration files encompass business rule base files, validation rule definition files, and calculation logic configuration files. These raw materials are aggregated through a unified data acquisition interface and standardized according to preset format conversion rules to form structured data suitable for subsequent processing.

[0021] Natural language processing (NLP) is performed on the extracted business description information to identify and extract business entity nouns, business action verbs, and business conditional adverbs. This operation employs a deep learning-based NLP model for lexical and syntactic analysis of the business description information. In the lexical analysis stage, Chinese word segmentation algorithms are used to segment the business description text and perform part-of-speech tagging, marking the part-of-speech category of each word in the text. In the syntactic analysis stage, dependency parsing techniques are used to identify the grammatical relationships between words in the sentence, thereby accurately locating business entity nouns, business action verbs, and business conditional adverbs.

[0022] The extracted business entity names are conceptually abstracted and categorized to construct a hierarchical structure. This process begins by conceptually abstracting the identified business entity names, categorizing specific business entity instances into higher-level conceptual categories. Based on this conceptual abstraction, a hierarchical entity classification system is established according to business domain classification standards. The top layer is divided into major categories based on enterprise business domains, the middle layer into subcategories based on business object types, and the bottom layer into categories based on specific business entity instances. This hierarchical structure is organized in a tree structure, with each entity node containing attribute fields such as entity identifier, entity name, entity definition, entity attributes, and entity relationships.

[0023] The extracted business action verbs undergo semantic standardization to establish a correspondence between business actions and system operations. This process begins with semantic disambiguation and synonym unification of the identified business action verbs. Semantic disambiguation determines the specific meaning of each verb in a particular business scenario based on its context; synonym unification standardizes different verb forms expressing the same meaning into standardized action terms. After semantic standardization, a mapping table between business actions and system operations is established. This mapping table defines the correspondence between each standardized business action and the underlying system operation.

[0024] The extracted business condition adverbs are transformed into logical expressions to construct computable business rule condition expressions. This process begins with parsing the business condition adverbs to identify their constituent elements, such as the condition body, condition operators, and condition thresholds. The condition body refers to the object on which the business condition acts; condition operators include comparison operators, logical operators, and inclusion operators; and the condition threshold refers to the specific numerical value or set of values ​​to be evaluated. The parsed condition adverbs are then converted into standardized logical expression formats. These logical expressions are organized using prefix or infix notation, supporting complex logical structures such as condition combinations, nested conditions, and condition branches. Finally, the constructed logical expressions are verified for computability to ensure that each condition expression can be executed programmatically.

[0025] The processed business entities, business actions, and business rule conditions are integrated into standardized business semantic units, and each business semantic unit is assigned a unique semantic identifier. In this operation, each business semantic unit consists of three parts: a business entity definition describing the business objects and their attributes involved in the semantic unit; a business action definition describing the business operations contained in the semantic unit and their execution conditions; and a business rule definition describing the business constraints followed by the semantic unit and their verification logic. Business semantic units are connected through pre-defined association relationships, including composition, inheritance, reference, and mutual exclusion. Each business semantic unit is assigned a unique semantic identifier, which consists of a semantic domain number, a semantic category number, and a semantic sequence number, used to uniquely identify and retrieve the semantic unit within the system.

[0026] After the business semantic model library is built, the system automatically generates a structured business semantic knowledge graph. This knowledge graph is stored using a graph database, with business semantic units as nodes and the relationships between them as edges. Each node in the knowledge graph stores the complete definition information of its corresponding business semantic unit, including semantic identifier, semantic name, semantic type, semantic attribute, and semantic content; each edge stores the type, direction, weight, and constraints of the relationship. The knowledge graph supports multi-dimensional index structures, including classification indexes based on semantic types, feature indexes based on semantic attributes, and relationship indexes based on relationships, to meet the semantic query and reasoning needs of different scenarios.

[0027] In the specific implementation process, step two involves receiving the business requirement description text based on the business semantic knowledge graph. Specifically, the semantic parsing engine performs word segmentation, entity recognition, and intent understanding on the business requirement description text to extract the core business entities, business actions, and business rule elements. This step is the core processing layer of the semantic-driven engine, responsible for transforming the business requirements in natural language form into a structured semantic representation.

[0028] The receipt of business requirement description text is achieved through a dedicated requirement input interface. This interface supports multiple input methods, including direct text input, document upload and import, speech-to-text input, and integration with existing requirement systems. Text input supports rich text formatting, preserving formatting information in the requirement description; document upload supports common document formats; speech-to-text integrates real-time speech recognition, supporting conversational requirement descriptions from business personnel; and requirement system integration uses standardized interfaces to integrate with other requirement management tools. The received business requirement description text first undergoes preprocessing operations, including encoding format standardization, character standardization, space handling, line break standardization, and special character filtering, to ensure the standardization of text data.

[0029] The business requirement description text is input into a pre-trained natural language processing (NLP) model, which performs Chinese word segmentation and part-of-speech tagging. This operation uses a pre-trained language model based on the Transformer architecture as the core processing engine. This model was pre-trained on a large-scale Chinese corpus and fine-tuned for the enterprise business domain. Chinese word segmentation employs a sequence labeling-based method, transforming the segmentation task into predicting a label for each character. Part-of-speech tagging, based on the segmentation results, labels each word with its part-of-speech category. The part-of-speech tag set uses a general Chinese part-of-speech tag set, including basic categories such as nouns, verbs, adjectives, numerals, measure words, pronouns, adverbs, prepositions, conjunctions, auxiliary words, modal particles, and interjections, as well as extended categories such as proper noun tags, business term tags, and business symbol tags specific to the business domain. The results of word segmentation and part-of-speech tagging are organized in the form of word sequences, with each term containing four fields: word text, start position, end position, and part-of-speech tag.

[0030] Based on word segmentation and part-of-speech tagging results, a named entity recognition (NAME) algorithm is used to identify key information such as business entity names, business amounts, business times, and business personnel in business requirement description texts. This operation employs a two-layer architecture combining conditional random fields (CRF) and a pre-trained model for NAME recognition. The first layer is the pre-trained model layer, which uses the contextual representation capabilities of a pre-trained language model to encode the input text, generating dynamic vector representations for each word. The second layer is the sequence labeling layer, which uses a CRF model to perform sequence labeling on the vector representations output by the pre-trained model, identifying named entity boundaries and entity types in the text. Entity types include the following categories: business entity name, business amount, business time, business personnel, business location, and business quantity. The NAME recognition results are organized as an entity list, with each entity item containing five fields: entity text, entity type, entity start position, entity end position, and entity confidence score.

[0031] The intent classification model categorizes the business intent of business requirement description texts to determine the business domain type and business function type to which the business requirement belongs. This operation employs a multi-level classification model for intent analysis. The first level is business domain classification, determining the broad business category to which the business requirement belongs; the second level is business function classification, further identifying specific business function types within the determined business domain; and the third level is operation type classification, identifying the specific operation type corresponding to the business requirement. The intent classification model uses a hierarchical attention mechanism, employing different attention weights for feature aggregation at different levels to capture intent information at different granularities. The results of intent classification include business domain type labels, business function type labels, and operation type labels, each accompanied by a corresponding confidence score.

[0032] This paper identifies the roles of semantic components in a business requirement description text using a semantic role labeling model, distinguishing semantic roles such as action performer, action receiver, action condition, and action result. This operation employs a semantic role labeling method based on dependency parsing. First, dependency parsing is performed on the business requirement description text to identify the grammatical dependencies between words in the sentences. Then, based on the grammatical dependencies and predefined semantic role rules, the semantic role of each sentence component is determined. Semantic role types include: agent, patient, participant, instrument, time, place, condition, and result. The semantic role labeling results are organized in the form of a semantic role graph, clearly expressing the semantic relationships between the components in the business requirement description.

[0033] In the specific implementation process, step three involves semantically matching the extracted core business entities, business actions, and business rule elements with the corresponding business semantic units in the business semantic knowledge graph to generate a semantic mapping relationship chain for business requirements. This step serves as the bridge layer of the semantic-driven engine, responsible for establishing the association mapping between the semantics of business requirements and the semantics of the knowledge graph.

[0034] The extracted core business entities are semantically similar to business entity nodes in the business semantic knowledge graph. Nodes with semantic similarity exceeding a preset threshold are selected as matching results. This semantic similarity calculation employs a multi-dimensional feature fusion method. The first dimension is lexical similarity, calculating the literal match between the business entity name extracted from the business requirements and the entity node name in the knowledge graph. The second dimension is semantic similarity, calculating the vector distance between the business entity and the knowledge graph entity node in the semantic space. The third dimension is contextual similarity, considering the context of the business entity in the requirement description. The fourth dimension is relational similarity, considering the degree of relational match between other entities associated with the business entity and other entities associated with the knowledge graph entity node. The similarity across these four dimensions is comprehensively calculated using a weighted fusion method. When the comprehensive semantic similarity of a knowledge graph entity node exceeds the preset threshold, it is used as the matching result for the business entity. The matching result includes the matched business entity node identifier, similarity score, and matching feature description.

[0035] The extracted business actions are semantically matched with business action nodes in the business semantic knowledge graph to identify the corresponding system operation type and operation parameters. This process begins by parsing the extracted business actions to extract elements such as the action subject, action type, action object, action method, and action constraints. Then, the parsed action elements are matched with business action nodes in the knowledge graph. The matching process uses an action semantic framework matching method, mapping business actions to a predefined action semantic framework. The action semantic framework includes the following slots: Action Name slot stores standardized action terminology; Action Type slot stores the action's category label; Action Parameter slot stores the input parameters and their types required for the action; Action Return Value slot stores the action's output result and its type; Action Precondition slot stores the conditions that must be met before the action is executed; and Action Post-Effect slot stores the state changes that occur after the action is executed. By filling the business action elements into the slots of the action semantic framework, the precise matching position of the business action in the knowledge graph is determined.

[0036] The extracted business rule elements are pattern matched with business rule nodes in the business semantic knowledge graph to identify the condition structure, action structure, and constraint structure of the business rules. This operation first parses the business rule elements, breaking them down into condition, action, and constraint parts. The condition part represents the prerequisites for triggering the business rule, including condition expressions and condition combination logic; the action part is the operation executed after the business rule is triggered, including the operation sequence and operation parameters; the constraint part consists of the restrictions that the business rule must satisfy, including data constraints, permission constraints, and consistency constraints. Then, the parsed rule structure is pattern matched with the business rule nodes in the knowledge graph. Pattern matching employs a dual-track approach combining rule template matching and instance-based learning. The matching result outputs a structured representation of the business rule, including the rule identifier, rule name, rule condition expression, rule action sequence, and rule constraint list.

[0037] A semantic mapping relationship chain is constructed based on the matching results. This chain records the complete mapping path from business requirement description to business semantic units and then to system implementation elements. In this operation, the semantic mapping relationship chain is organized using a directed graph structure. The graph contains three types of nodes: requirement nodes represent semantic elements in the business requirement description, including business entity nodes, business action nodes, and business rule nodes; semantic nodes represent corresponding semantic units in the business semantic knowledge graph, including semantic entity units, semantic action units, and semantic rule units; and implementation nodes represent implementation elements at the system code level, including code modules, function interfaces, and data structures. The graph contains two types of edges: mapping edges connect requirement nodes and semantic nodes, representing the correspondence between business requirements and business semantics; and implementation edges connect semantic nodes and implementation nodes, representing the correspondence between business semantics and code implementation. The semantic mapping relationship chain is serialized and stored in a unified data format, supporting incremental updates and version management.

[0038] In the specific implementation process, step four involves calling the code generation rule engine based on the semantic mapping relationship chain. According to the preset code template library and the mapping relationship between business semantics and code, it automatically generates a program code framework corresponding to the business requirements. This step is the generation layer of the semantic-driven engine, responsible for transforming structured semantic representations into executable program code.

[0039] The semantic mapping relationship chain is parsed to extract the business semantic unit types, semantic unit attributes, and relationships between semantic units. This operation first performs structured parsing of the semantic mapping relationship chain, traversing all nodes and edges in the graph to extract the required semantic information. For each requirement node, its corresponding business semantic type, business semantic name, business semantic attribute, and business semantic relationship are extracted; for each mapping edge, its mapping type, mapping direction, and mapping weight are extracted; for each implementation edge, its implementation method, implementation constraints, and implementation parameters are extracted. The parsing results are organized in the form of semantic feature vectors for easy invocation by subsequent processing modules.

[0040] Based on the extracted business semantic unit type, the corresponding code template unit is selected from the code template library. The code template library stores code templates for different business domains and functions. In this operation, the code template library adopts a hierarchical organizational structure. The first layer is categorized by programming language; the second layer by business domain; the third layer by code hierarchy; and the fourth layer by code function. Each template unit in the code template library contains attribute fields such as template identifier, template name, template type, applicable scenarios, usage constraints, template content, and parameter definitions. The template content is written using syntax supported by the template engine and supports advanced features such as variable placeholders, conditional statements, loop iterations, and macro definitions.

[0041] The semantic unit attributes in the semantic mapping relationship chain are mapped to variable parameters in the code template unit, completing the parameterized population of the code template. This operation employs a combined strategy of attribute mapping and type mapping. Attribute mapping maps the specific attribute values ​​of the semantic unit to the corresponding variable placeholders in the code template. Type mapping maps the attribute types of the semantic unit to the data type declarations in the code template. The parameter mapping process also includes auxiliary operations such as type conversion and format processing to ensure compatibility between semantic attributes and code parameters. The populated code template unit retains the original code structure, only replacing the variable placeholders with the specific attribute values.

[0042] The populated code template units are sequentially combined according to preset code assembly rules to generate a complete program code framework. In this operation, the code assembly rules define the combination methods and assembly order between different types of code template units. The assembly rules include: hierarchical assembly rules defining the hierarchical and calling relationships between templates at different code levels; module assembly rules defining the dependency and referencing relationships between different functional module templates within the same code level; and file assembly rules defining the physical organization and naming conventions of the generated code files. The code assembly process is executed in the following order: first, the data entity layer code is assembled; then, the data access layer code is assembled; next, the business logic layer code is assembled; and finally, the presentation layer code is assembled. The assembled program code framework includes a complete directory structure, source code files, configuration files, and documentation comments.

[0043] In the specific implementation process, step five involves performing semantic integrity and logical consistency verification on the generated program code framework. Based on the verification results, the program code framework is automatically corrected or manually annotated and adjusted to output the final enterprise business application software code. This step is the quality assurance layer of the semantic-driven engine, responsible for ensuring the semantic consistency and logical correctness between the generated code and business requirements.

[0044] The reverse mapping verification of code statements within the program code framework ensures that each code statement can be traced back to its corresponding business semantic unit. This operation combines code traversal and semantic tracing. First, a static analysis of the program code framework is performed, parsing all statements, expressions, variable references, and method calls. Then, a reverse index is established between code elements and semantic elements, mapping each code element to its corresponding business semantic unit. During verification, the system checks for code statements that cannot be traced back to business semantics; if found, they are marked as invalid or redundant code. It also checks for instances where business semantics are not covered by code; if found, they are marked as semantically missing.

[0045] Logical coverage analysis is performed on conditional statements in the program code framework to detect any unprocessed business rule branches. This operation employs path traversal and branch coverage methods. First, all conditional statements in the program code framework are extracted, and a conditional branch graph is constructed. Then, the graph is traversed to analyze the triggering conditions and execution results of each conditional branch. Next, the analysis results are compared with the business rule definitions in the semantic mapping relationship chain to check whether each combination of conditions in the business rules has a corresponding code branch for processing. If a business rule condition combination is detected that is not covered by a code branch, it is marked as a branch missing defect; if a code branch is detected that does not correspond to any business rule condition, it is marked as an invalid branch.

[0046] Perform data integrity checks on data manipulation statements within the program code framework to verify the consistency of business entity attributes during data persistence. This operation includes the following aspects: Required field checks verifying that all required attributes of the business entity are assigned values ​​in the data manipulation statements; Data type checks verifying that the attribute value types of the business entity are consistent with the data definitions; Value range checks verifying that numeric attribute values ​​are within the allowed range; Referential integrity checks verifying the existence of objects referenced by foreign keys; and Uniqueness checks verifying that the values ​​of unique identifier attributes are not duplicated. The results of the data integrity checks are output in the form of an inspection report.

[0047] Conflict detection is performed on the business rule logic within the program code framework to identify contradictory rule definitions. This operation involves pairwise comparisons of multiple business rules within a semantic mapping chain. Conflict types include: mutual exclusion conflicts (detecting whether two business rules require mutually exclusive actions under the same condition); circular dependency conflicts (detecting whether circular dependencies exist between multiple business rules); priority conflicts (detecting whether explicit execution priorities are defined when multiple business rules are simultaneously satisfied); and overlapping condition conflicts (detecting whether the triggering conditions of two business rules overlap and their action definitions are inconsistent). Conflict detection employs a logical reasoning-based approach. Detected conflicts are output as conflict reports.

[0048] Based on the verification results, the system automatically generates code correction suggestions, or outputs the verification results to a manual review interface for developers to adjust and confirm. In this operation, the automatic correction suggestions are generated for verification defects that can be automatically repaired. For semantic missing defects, the system automatically generates supplementary code templates based on the missing semantic units and inserts them into the program code framework; for invalid branch defects, the system marks the detected invalid branch code as comments or deletes it; for data integrity defects, the system automatically generates corrective code based on the defect type. For verification defects that cannot be automatically repaired, the system outputs the defect information to the manual review interface. The manual review interface displays the verification results in a visual manner, including defect location, defect description, impact analysis, and correction suggestions. Developers can confirm ignoring defects, mark them for repair, or provide feedback on modifications on the interface. Defects that are confirmed for repair are manually adjusted by developers, and the adjusted code re-enters the verification process until all defects are addressed. Finally, the program code framework that passes verification is formatted and documented before being output as the final enterprise business application software code.

[0049] Example 2

[0050] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0051] This invention also provides an enterprise business application software development system based on a semantic-driven engine. This system implements the various functional modules of the methods proposed in the embodiments. The hardware architecture of this system includes five core functional modules: a semantic model construction module, a semantic parsing module, a semantic matching module, a code generation module, and a verification and correction module. These modules interact and coordinate processes through an internal message bus.

[0052] The semantic model building module is used to construct a business semantic model library, breaking down enterprise business requirements into reusable business semantic units and establishing relationships between these units to form a structured business semantic knowledge graph. This module includes a data acquisition unit, a natural language processing unit, a concept abstraction unit, an action standardization unit, a rule transformation unit, and a knowledge graph generation unit. The data acquisition unit is responsible for extracting business description information, business process definition files, and business rule configuration files from the enterprise's existing business application systems. This unit interfaces with various business systems through standardized data adapters, supporting both direct extraction of structured data and parsing of unstructured data. The natural language processing unit is responsible for performing lexical and syntactic analysis on the business description information, identifying business entity nouns, business action verbs, and business conditional adverbs. This unit uses a deep learning-based natural language processing model for text analysis. The concept abstraction unit is responsible for hierarchically classifying and unifying the concepts of business entity nouns, constructing a hierarchical structure system of business entities. The action standardization unit is responsible for semantic disambiguation and synonym unification of business action verbs, establishing a mapping relationship between business actions and system operations. The rule conversion unit is responsible for converting business conditional adverbs into logical expressions, constructing computable business rule conditional expressions. The knowledge graph generation unit is responsible for integrating the processed business semantic units into standardized semantic units, assigning unique identifiers to each semantic unit, and generating a structured business semantic knowledge graph. This unit uses a graph database as its storage engine, supporting efficient storage and retrieval of large-scale semantic data.

[0053] The semantic parsing module receives business requirement description text based on a business semantic knowledge graph. It then uses a semantic parsing engine to perform word segmentation, entity recognition, and intent understanding on the text, extracting core business entities, actions, and rules. This module includes a requirement receiving unit, a word segmentation and tagging unit, a named entity recognition unit, an intent classification unit, and a semantic role labeling unit. The requirement receiving unit receives the business requirement description text, supporting various input formats and styles. The word segmentation and tagging unit performs Chinese word segmentation and part-of-speech tagging on the text, using a pre-trained language model-based segmentation method. The named entity recognition unit identifies key information such as business entity names, amounts, times, and personnel, employing a two-layer architecture combining conditional random fields and a pre-trained model. The intent classification unit categorizes the business intent in the requirement description text, determining the business domain and function type, using a multi-level classification model. The semantic role labeling unit identifies the roles of each semantic component in the requirement description text, distinguishing between action executors, action recipients, action conditions, and action results.

[0054] The semantic matching module is used to semantically match the extracted core business entities, business actions, and business rule elements with the corresponding business semantic units in the business semantic knowledge graph, generating a semantic mapping relationship chain for business requirements. This module includes an entity matching unit, an action matching unit, a rule matching unit, and a mapping chain construction unit. The entity matching unit is responsible for calculating the semantic similarity between the extracted core business entities and business entity nodes in the business semantic knowledge graph, selecting business entity nodes with similarity higher than a preset threshold as the matching result, and using a multi-dimensional feature fusion semantic similarity calculation method. The action matching unit is responsible for semantically matching the extracted business actions with business action nodes in the business semantic knowledge graph, identifying the system operation type and operation parameters corresponding to the business actions, and using an action semantic framework matching method. The rule matching unit is responsible for pattern matching between the extracted business rule elements and business rule nodes in the business semantic knowledge graph, identifying the condition structure, action structure, and constraint structure of the business rules. The mapping chain construction unit is responsible for constructing a semantic mapping relationship chain based on the matching results, recording the complete mapping path from business requirement description to business semantic units and then to system implementation elements, and organizing and storing it using a directed graph structure.

[0055] The code generation module invokes the code generation rule engine based on the semantic mapping relationship chain. According to the preset code template library and the mapping relationship between business semantics and code, it automatically generates a program code framework corresponding to business requirements. This module includes a mapping parsing unit, a template selection unit, a parameter filling unit, and a code assembly unit. The mapping parsing unit is responsible for parsing the semantic mapping relationship chain, extracting the business semantic unit types, semantic unit attributes, and relationships between semantic units. The template selection unit is responsible for selecting the corresponding code template unit from the code template library based on the extracted business semantic unit types. The code template library adopts a hierarchical and categorized organizational structure. The parameter filling unit is responsible for mapping the semantic unit attributes in the semantic mapping relationship chain to variable parameters in the code template unit, completing the parameterized filling of the code template. The code assembly unit is responsible for orderly combining the filled code template units according to preset code assembly rules to generate a complete program code framework.

[0056] The verification and correction module performs semantic integrity verification and logical consistency verification on the generated program code framework. Based on the verification results, it automatically corrects or manually annotates and adjusts the program code framework, outputting the final enterprise business application software code. This module includes a reverse mapping verification unit, a logic coverage analysis unit, a data integrity check unit, a rule conflict detection unit, and a correction processing unit. The reverse mapping verification unit is responsible for performing reverse mapping verification between code statements in the program code framework and semantic mapping relationship chains, ensuring that each code statement can be traced back to its corresponding business semantic unit; the logic coverage analysis unit is responsible for performing logic coverage analysis on conditional judgment statements in the program code framework, detecting whether there are any unprocessed business rule branches; the data integrity check unit is responsible for performing data integrity checks on data operation statements in the program code framework, verifying the consistency of business entity attributes during data persistence; the rule conflict detection unit is responsible for detecting conflicts in the business rule logic in the program code framework, identifying whether there are any contradictory rule definitions; the correction processing unit is responsible for automatically generating code correction suggestions based on the verification results or outputting the verification results to a manual review interface for confirmation and adjustment.

[0057] Example 3

[0058] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0059] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the enterprise business application software development method based on a semantic-driven engine proposed in Embodiment 1. This electronic device, as the runtime carrier of the semantic-driven engine, provides computing resources, storage resources, and communication resources for the implementation of the method.

[0060] The electronic device's memory stores computer programs and data, employing a hybrid memory architecture that combines volatile and non-volatile memory. Volatile memory serves as a temporary working area during processor operation, used to load and execute computer programs, store runtime data and intermediate computation results; its capacity is configured based on processor performance and concurrent processing requirements. Non-volatile memory is used for persistent storage of data such as business semantic knowledge graphs, code template libraries, and semantic mapping relationship chains; its capacity is configured based on data scale and management strategies. The memory also stores system resources such as the operating system, runtime environment, dependency library files, and configuration files.

[0061] The processor in this electronic device executes computer programs stored in memory, implementing various processing steps of the semantic-driven engine-based enterprise business application software development methodology. The processor's architecture employs a combination of multi-core parallel processing and pipeline optimization. Semantic parsing, semantic matching, code generation, and verification / correction processes can be distributed across different processor cores for parallel execution, improving system throughput. Each processing step is connected via a data pipeline, with the output of one step directly serving as the input for the next, reducing data transfer overhead. The processor also integrates a dedicated artificial intelligence acceleration unit for efficiently performing computationally intensive tasks such as natural language processing model inference and semantic similarity calculation.

[0062] The electronic device also includes a communication interface for data interaction with other business systems, requirement management systems, and code repository systems within the enterprise. The communication interface supports multiple communication protocols and data formats, enabling seamless integration with heterogeneous systems. Business requirement description text can be automatically imported from the requirement management system via the communication interface; the completed business semantic knowledge graph can be synchronized to the enterprise's knowledge management system via the communication interface; and the generated program code framework can be automatically submitted to the code repository system for version management via the communication interface.

[0063] This invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, this computer program implements the enterprise business application software development method based on a semantic-driven engine disclosed in Embodiment 1. The computer-readable storage medium is implemented using an erasable programmable read-only memory, flash memory, or other suitable storage technologies. The computer program stored thereon includes five main program modules: a semantic model building program, a semantic parsing program, a semantic matching program, a code generation program, and a verification and correction program. Each program module is written according to the processing logic of the aforementioned method and includes corresponding instruction sequences and data structure definitions. When the computer-readable storage medium is connected to an electronic device and the computer program is loaded into the memory, the processor executes the computer program to realize all the functions of the enterprise business application software development method based on a semantic-driven engine.

[0064] This invention also provides a computer program product, including a computer program that, when executed by a processor, implements the enterprise business application software development method based on a semantic-driven engine disclosed in Embodiment 1. The computer program product is distributed as a software package, comprising an installer, configuration files, dependency library files, and help documentation. The computer program product supports multiple deployment methods, allowing enterprises to choose from local deployment, cloud deployment, or hybrid deployment modes according to their actual needs. In cloud deployment mode, the computer program product is provided as a cloud service, and users can access the service through a browser or client application to use the semantic-driven engine for enterprise business application software development.

[0065] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for developing enterprise business application software, characterized in that, Includes the following steps: Build a business semantic model library, break down enterprise business requirements into reusable business semantic units, establish the relationship between each business semantic unit, and form a structured business semantic knowledge graph; Based on the business semantic knowledge graph, the business requirement description text is received, and the business requirement description text is processed by word segmentation, entity recognition and intent understanding through the semantic parsing engine to extract core business entities, business actions and business rule elements. The extracted core business entities, business actions, and business rule elements are semantically matched with the corresponding business semantic units in the business semantic knowledge graph to generate a semantic mapping relationship chain for business requirements. Based on the semantic mapping relationship chain, the code generation rule engine is invoked to automatically generate a program code framework corresponding to the business requirements, according to the code template library and the mapping relationship between business semantics and code. The generated program code framework is subjected to semantic integrity verification and logical consistency verification. Based on the verification results, the program code framework is automatically corrected or manually annotated and adjusted to output the final enterprise business application software code.

2. The enterprise business application software development method according to claim 1, characterized in that, The construction of the business semantic model library specifically includes: Extract business description information, business process definition files, and business rule configuration files from the enterprise's existing business application systems; Natural language processing is performed on the extracted business description information to identify and extract business entity nouns, business action verbs, and business conditional adverbs. The extracted business entity nouns are conceptually abstracted and classified into hierarchical levels to construct a hierarchical structure system for business entities.

3. The enterprise business application software development method according to claim 2, characterized in that, The extracted business actions are semantically standardized to establish a mapping relationship between business actions and system operations, specifically including: Semantic disambiguation is performed on the identified business action verbs to determine their specific meaning in a particular business scenario based on the context of the business action verbs. Different verb forms expressing the same meaning are processed by synonym unification and standardized action terms. Establish a mapping table between business actions and system operations.

4. The enterprise business application software development method according to claim 2, characterized in that, The extracted business condition adverbs are transformed into logical expressions to construct computable business rule condition expressions, specifically including: Identify the condition body, condition operators, and condition thresholds in business condition adverbs, where condition operators include comparison operators, logical operators, and inclusion operators; The parsed conditional adverbs are converted into standardized logical expression formats, which support condition combination, condition nesting, and condition branching. Verify the computability of each conditional expression.

5. The enterprise business application software development method according to claim 1, characterized in that, Based on the business semantic knowledge graph, the text describing business requirements is received, and word segmentation, entity recognition, and intent understanding are performed through a semantic parsing engine. Specifically, this includes: inputting the text describing business requirements into a pre-trained natural language processing model, which performs Chinese word segmentation and part-of-speech tagging on the text describing business requirements. The tag set for part-of-speech tagging includes basic part-of-speech categories such as nouns, verbs, adjectives, numerals, classifiers, pronouns, adverbs, prepositions, conjunctions, auxiliary words, modal particles, and interjections.

6. The enterprise business application software development method according to claim 5, characterized in that, Based on word segmentation and part-of-speech tagging results, key information in the business requirement description text is identified using a named entity recognition algorithm, specifically including: A two-layer architecture combining conditional random fields and pre-trained models is used for named entity recognition. The pre-trained model layer generates a dynamic vector representation of each word, while the sequence labeling layer identifies the named entity boundaries and entity types in the text. The identified entity types include business entity name, business amount, business time, business personnel, business location, and business quantity.

7. The enterprise business application software development method according to claim 1, characterized in that, The business intent of the business requirement description text is classified using an intent classification model. Specifically, this includes: using a multi-level classification model to classify business domains, business functions, and operation types. The first level determines the major business category to which the business requirement belongs; the second level identifies the specific business function type within the determined business domain; and the third level identifies the specific operation type corresponding to the business requirement.

8. The enterprise business application software development method according to claim 1, characterized in that, The semantic role labeling model identifies the roles of each semantic component in the business requirement description text. Specifically, it employs a semantic role labeling method based on dependency parsing to identify the grammatical dependency relationships between words in a sentence and determine the semantic role of each sentence component. The semantic role types include agent role, patient role, participant role, instrument role, time role, location role, condition role, and result role.

9. The enterprise business application software development method according to claim 1, characterized in that, The extracted core business entities are semantically matched with the corresponding business semantic units in the business semantic knowledge graph, specifically including: The semantic similarity calculation involves multi-dimensional feature fusion between the extracted core business entities and the business entity nodes in the business semantic knowledge graph. The dimensions of the calculation include lexical similarity, semantic similarity, contextual similarity, and relational similarity. The extracted business actions are matched with the business action nodes in the business semantic knowledge graph using an action semantic framework.

10. The enterprise business application software development method according to claim 1, characterized in that, The generated program code framework is subjected to semantic integrity verification and logical consistency verification, specifically including: The code statements in the program code framework are reverse-mapped and verified to the semantic mapping relationship chain to ensure that each code statement can be traced back to the corresponding business semantic unit; Perform logical coverage analysis on the conditional statements in the program code framework to detect whether there are any unprocessed business rule branches; Perform data integrity checks on data manipulation statements in the program code framework. The check items include required fields, data types, value range, referential integrity, and uniqueness. Perform conflict detection on the business rule logic in the program code framework.

Images