Mapping generation method from business terms to coding technical terms

Abstract

The invention discloses a method for generating mapping from business terms to coding technology terms, particularly relates to the technical field of computer software, and is used for solving the problems that in a large-scale industry informatization project, business requirements exist in a multi-source natural language text, terms are not uniform, rule boundaries are easy to omit, and the efficiency is low. Therefore, the problem of difficulty in automatically and stably generating coding technical terms matched with a specific engineering environment is solved. On the basis of a unified service term library and service change key points, candidate technical terms are screened by using a double-branch semantic coding model in combination with engineering environment tags and service rules, and finally an AI coding technical term list with version numbers is generated. Therefore, a technical term set capable of directly driving automatic coding can be automatically and stably generated for a specific engineering environment under the business demand conditions of multiple sources, inconsistent expression and boundary omission, and mapping deviation and inconsistency caused by manual understanding and translation and repeated communication are reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer software, in particular to a mapping generation method from business terms to coding technical terms. BACKGROUND

[0002] Currently, the business requirements of large-scale industry informatization projects usually exist in the form of multi-source natural language documents, such as business requirement descriptions, product configuration descriptions, marketing plans, contract terms, and operation manuals, etc. These documents are written by different roles at different stages, and the same business meaning often uses different expressions, and when the requirements change, only the "change points" are written, and the original rule boundaries are no longer restated completely. The existing approach relies on business personnel and developers to manually read these documents, and based on experience, selects or creates function names, interface names, configuration key names, and database field names in the code repository, and manually disassembles the business rules into implementable coding technical terms. With the increase in system size and version number, this manual approach is prone to understanding deviations, inconsistent naming, missing rule boundaries, and other issues, making it difficult to ensure that the mapping relationship between different projects and different versions is traceable and reusable.

[0003] In recent years, there have also been auxiliary coding tools based on natural language processing and large models. One approach is to directly input the complete business requirement text as a prompt, and generate code snippets or interface definitions from the model. Another approach is to maintain a relatively simple business dictionary or rule template, and through keyword matching or fixed rules, map part of the business expression to a pre-set technical term template. Such solutions usually lack systematic management of business terms, do not construct a versioned business term library for specific projects, and rarely extract business change points at the "new and old requirement comparison" granularity and complete the omitted boundary conditions such as amount interval, applicable user group, and time range. In addition, when associating business semantics with technical terms, most existing solutions do not introduce constraints on engineering environment labels and model versions, making it difficult to distinguish between different technical naming for the same business meaning in different languages and frameworks, and lacking version locking and evidence traceability mechanisms from business changes to technical terms to code generation.

[0004] In the above background, when the business requirements come from various sources, the terms are not uniformly expressed, and the rule boundaries are omitted, the existing technology often cannot automatically and stably generate a set of coding technical terms that match a specific engineering environment at the granularity of "specific business change points", and it is also difficult to ensure that the mapping relationship remains consistent and traceable throughout the project lifecycle, thereby affecting the reliability and landing effect of automatic code generation and continuous delivery. SUMMARY

[0005] In view of the deficiencies of the prior art, the application provides a service term to coding technology term mapping generation method to solve the problems in the background art.

[0006] To achieve the above object, the application provides the following technical scheme: a service term to coding technology term mapping generation method, comprising: S1, collecting multi-source service texts, clustering service terms according to domains, modules and logics, and generating service term libraries with version numbers; S2, aligning and differentiating new and old service requirement texts, combining the service term library to complete omitted boundary conditions, generating service change points marked with change objects and change types; S3, constructing service change point and technology term pairing samples, training a double-branch semantic coding model composed of service branches and technology branches, and using contrastive learning to establish the corresponding relationship between service semantic vectors and technology semantic vectors; S4, receiving engineering environment labels, filtering technology term subsets according to languages and frameworks from technology term vector indexes, encoding service change points into service semantic vectors, and searching candidate technology terms with semantic similarity exceeding a threshold in the subsets; S5, sorting and scoring the candidate technology terms according to business logic categories and environment matching degrees, verifying whether the candidate technology terms meet the rules in the service term library according to the business rules, and eliminating the candidate technology terms that do not meet the verification conditions; S6, combining the verified technology terms and corresponding service change points to generate AI coding technology term records, and writing the AI coding technology term list with version numbers and engineering environment labels for code generation components to call.

[0007] Further, S1 comprises: Incrementally grabbing service requirement descriptions, product configuration descriptions, marketing schemes, contract terms and business operation manuals with source system identifiers and version numbers from enterprise document platforms according to project identifiers; Standardizing and dividing the service texts into sentences and words, filtering colloquial words, counting the occurrence times of candidate word groups in the same project, and marking the word groups that reach the set threshold and refer to customer types, product categories, billing periods and discount methods as service terms; Clustering the service terms based on semantic similarity, generating service term records according to domain labels, module labels and logic labels, and assigning version numbers to the service term record sets and writing them into the central term library table at the end of each extraction period.

[0008] Further, S2 comprises: The new and old service requirement texts with the same module identifier are segmented by periods and converted into sentence vectors, and within the observation window determined by the old version number and the new version number, an old sentence is selected for each new sentence according to similarity and the newly added description and the description to be deleted are marked; The aligned sentence pairs are differentiated by words and phrases, and the change fragments containing service terms are obtained by the service term library to obtain term identifiers and service rules; When it is detected that the new requirement text does not contain the amount boundary, user group and time range fields, the boundary conditions are supplemented according to the old requirement version service rule; The change object, change type, pre-change value, post-change value and supplemented boundary conditions are combined to form the service change points and written into the service change point table.

[0009] Further, S3 includes: The function names, interface paths, configuration key names, table names and field names are parsed from the interface design document, configuration description, database design document and code repository meta information; The parsed results are combined with the service name, module identifier, file path, parameter meaning and field meaning to generate technical term records and written into the technical term table; According to the historical release record, the service change points that have been executed in the production environment are selected in the service change point table, a positive pairing relationship is established between each service change point and the actual adopted technical term in the technical term table, and a negative pairing relationship is generated by extracting technical terms unrelated to the service change point from the technical term table in the same batch to form a pairing sample set of service change points and technical terms.

[0010] Further, in the double-branch semantic encoding model, a business branch and a technical branch are set up, the business branch receives the service change point text, the service term identifier, the domain label, the module label and the logic label to generate the business semantic vector; The technical branch receives the technical term text, the service name, the module identifier and the field meaning to generate the technical semantic vector, and the two types of vectors have the same dimension; The business semantic vector and the technical semantic vector in the positive pairing are set as the close vector pair, and the business semantic vector and the technical semantic vector in the negative pairing are set as the far vector pair; The double-branch semantic encoding model representing the corresponding relationship between the service change points and the technical terms is obtained by iteratively adjusting the parameters of the business branch and the technical branch.

[0011] Further, S4 includes: The unique identifier of the service change point and the engineering environment label are received, the structured service description is read from the service change point table through the unique identifier of the service change point, and the business semantic vector is generated through the business branch; Screening technical term records consistent in language label and framework label according to engineering environment label in technical term vector index to form a technical term subset, and calculating semantic similarity between business semantic vector and each technical semantic vector in the technical term subset; Determining technical terms with semantic similarity higher than a preset semantic similarity threshold as candidate technical terms, and writing the candidate technical terms together with business change point unique identification and engineering environment label into a candidate technical term table.

[0012] Further, S5 includes: Reading candidate technical terms from the candidate technical term table according to business change point identification and engineering environment label; Calculating comprehensive scores according to business logic category of the business change point, module label of the candidate technical term and semantic similarity score and sorting; Extracting business rules from the business term library according to the business term library version number cited by the business change point; Checking whether field meaning, parameter meaning and configuration mode of the sorted candidate technical terms item by item cover the amount interval, member level and time range specified by the business rules completely or not; Eliminating candidate technical terms with incomplete coverage, and writing the remaining candidate technical terms together with comprehensive scores and business term library version number into a passed technical term table.

[0013] Further, S6 includes: Reading a technical term set from the passed technical term table according to business change point identification and engineering environment label; Supplementing function name, interface name, parameter name list, configuration key name and database field name for each technical term according to design document and configuration description, and checking consistency according to amount, time and user group in the business change point; Combining technical term text, the above name information, business change point identification, business logic category, engineering environment label and model version number to generate AI coded technical term records, and writing the AI coded technical term records into an AI coded technical term list according to business version number and engineering environment label, for a code generation component to query according to business version number, module label and business change point identification.

[0014] Compared with the prior art, the present application has the following beneficial effects: 1. By using the double-branch semantic coding model combined with engineering environment tags and business rules to filter candidate technical terms based on a unified business term library and business change points, and finally generating an AI-coded technical term list with version number, so as to achieve that under the conditions of multi-source, inconsistent expression and boundary omission of business requirements, a technical term set that can directly drive automatic coding can also be automatically and stably generated for specific engineering environment, reducing the mapping deviation and inconsistency caused by manual understanding, translation and repeated communication.

[0015] 2. By using version locking, evidence chain marking, idempotent and sequence control strategies in each link of the business term library, business change points, double-branch semantic coding model, candidate technical terms and AI-coded technical term list, and decoupling the model training station from the intermediate service close to the production environment, so as to achieve traceable, rollbackable and reusable unified management of the whole process of business to technical mapping under the premise of ensuring data security and compliance, and facilitate stable integration and continuous iterative optimization in the continuous delivery pipeline. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 The flowchart of the mapping generation method of the business term to the coded technical term according to the present application. DETAILED DESCRIPTION

[0017] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0018] Embodiment: A mapping generation method of business term to coded technical term, comprising: S1, collecting multi-source business texts, clustering business terms by domain, module and logic, and generating a business term library with version number, specifically: On a production line for large-scale industry informatization projects, this step is installed between the enterprise document platform and the research and development platform, used to continuously absorb multi-source business texts formed in the whole life cycle of the project and build a business term library with version number. Multi-source business texts refer to business requirement specifications, product configuration specifications, marketing plans, contract terms and business operation manuals formed by business personnel, product personnel and project management personnel. Each business text is a record, carrying project identification, source system identification, business line identification, version number and release time, and unified using agreed character encoding. The document platform provides a list of new business texts to this step by article.

[0019] Business terms refer to word groups with stable meanings in the above business texts, used to refer to business objects or business rules such as customer types, product categories, billing periods, discount methods, etc., preferably two to twenty characters in length, composed of several consecutive words, and distinguished from ordinary function words in natural language. The domain label refers to a classification label used to distinguish the scope of the power, finance, e-commerce and other industries. The module label refers to a classification label used to distinguish the business modules such as billing, settlement, and member management. The logic label refers to a classification label used to distinguish the business logic categories such as discount rules, threshold rules, and segmented billing rules. The three types of labels use a limited set of encodings, which are pre-maintained in system configuration and can be customized by project.

[0020] In this step, the system incrementally crawls new business texts with timestamps from the enterprise document platform on a daily basis. For each text, first perform uniform format specification, remove headers and footers, watermarks, directories, empty lines, and redundant punctuation, then perform sentence segmentation according to period, semicolon, and line feed, perform word segmentation on the sentence according to the word list and segmentation strategy, filter words that are obviously colloquial connectors, and obtain a structured sequence of sentences. In the rolling observation window, the system counts the number of occurrences and distribution positions of each candidate word group within the same project and the same business line. Word groups that appear in multiple business texts and have a number of occurrences that meet a certain threshold are marked as candidate business terms. For high-sensitivity business concepts such as "customer type, product category, billing period, and discount method", preferably include numerical values, time units, or business keywords in the word group to directly add to the candidate set.

[0021] Subsequently, the system clusters the candidate business terms based on semantic similarity, and classifies the candidate business terms with similar semantics and similar business scenarios into the same cluster. The clustering results are classified according to the configured domain label, module label, and logic label. For clusters with high semantic ambiguity, the business personnel familiar with the project can specify the label and standard expression through the manual review interface, so that each business term record contains at least the term text, domain label, module label, logic label, term explanation, and appearance source summary.

[0022] The business term library refers to the complete set of business term records formed within an extraction period. The system assigns a unique business term library version number to this set after each extraction and clustering is complete, records the generation time and the identity of the human reviewer, and writes all business term records into the central term library table. To ensure idempotency, the same business text is repeatedly collected by combining the project identifier and the text hash value for deduplication. For different date tasks collected in parallel, the effective order is determined according to the lexicographic order of the project identifier and the version number. The business term library version number that has been issued is locked and no longer updated by covering the records in it. Only new versions can be extended or corrected.

[0023] The central term library table can be implemented by relational storage or graph storage, and the version number of the business term library is indexed, so that the subsequent steps can accurately retrieve the corresponding business term records according to the project and version when analyzing the business change points and completing the business rules. In actual production lines, this step is applicable to industries such as power billing systems, financial product configuration systems, and e-commerce promotion platforms that require long-term maintenance of a large number of natural language business documents, and its process capability covers cross-system document aggregation, cross-project term identification, and cross-version term evolution records.

[0024] Preferably, in the e-commerce industry scenario, one thousand to five thousand business texts with project identifiers and release times can be collected daily, the latest three versions of the business term library are kept in a usable state within a rolling thirty-day window, and the end-to-end time of a single business term library version construction is controlled within ten minutes, to balance the timeliness of version updates and the computational resource overhead. Another preferred way is to link the business term library version construction with project milestones, and the project manager triggers an extraction and clustering operation through the management interface after the demand review is passed. The system only updates the business term library at these key points, thereby reducing the daily resource occupation in projects with infrequent demand changes.

[0025] S2, aligning and differentiating new and old business requirement texts, completing the omitted boundary conditions in combination with the business term library, generating business change points marked with change objects and change types, and the specific implementation is: In a production line for large-scale industry informatization projects, this step is deployed in the demand change management link, and the new and old business requirement texts for the same business module are automatically compared and the business change points are formed. The new and old business requirement texts refer to two requirement descriptions with the same module identifier and different version numbers, each of which records the submission time, submitter identifier, and associated business term library version number. The module identifier is used to uniquely mark the business modules such as billing, settlement, and member management, and the version number is used to distinguish the demand content of the same module in different iterations.

[0026] The business change point refers to the smallest change unit extracted from the new and old business requirement texts, which is used to describe the rule difference of a certain business object before and after the change. It at least includes the change object, change type, pre-change value, post-change value, and referenced business term library version number, which is used to establish a corresponding relationship with the coding technical terms. After receiving a set of new and old business requirement texts, the system first checks whether the two belong to the same business module and the version order is reasonable according to the module identifier and version number, and then divides the two texts into two sentence sequences according to the period, semicolon, and line feed character, and retains the paragraph number and in-line position of the sentence in the original text.

[0027] To measure the semantic closeness between sentences, the system converts each sentence into a fixed-length vector representation, with the dimension of the sentence vector configured as a fixed value at system deployment, so as to calculate the similarity between sentences in the numerical space. The observation window refers to the range of old demand versions involved in the comparison, which is set as the combination of the latest old demand and the current new demand in the main embodiment, and the specific old version number is passed in by the demand change management system when the comparison is triggered; within the observation window, the system finds the sentence with the highest similarity for each new demand sentence in the sequence of old demand sentences, and marks the new demand sentence with a similarity lower than the similarity threshold as a newly added description, and marks the sentence without any new demand sentence matching it in the sequence of old demand sentences as a description to be deleted.

[0028] The similarity threshold is a value between zero and one, used to distinguish between sentences with "slightly different expressions but similar semantics" and sentences with "larger semantic differences", which is uniformly configured by the business line and recorded in the version record. For successfully aligned old and new sentence pairs, the system further differentiates at the word and phrase level within the sentence, and compares each word and phrase at each position to find out the segments that have been added, replaced or deleted; the business terms appearing in these segments obtain uniform term identifiers, domain tags and logical tags by querying the business term library generated in the previous stage, thereby eliminating different expressions for the same business meaning in different documents.

[0029] The omitted boundary condition refers to the case where only the discount ratio, amount change, trigger condition change, etc. "change point" is written in the new demand text, and the original amount lower limit, amount upper limit, applicable user group, time range, etc. limit factors are not written again. When the system detects that the new sentence only contains part of the key fields and lacks complete boundary description, it looks back to the old demand version binding business rule definition from the business term library according to the term identifier of the related business term, and fills in the original boundary condition into the current change record to form a complete structured business description.

[0030] The structured business description refers to recording the business object name, rule segment before change, rule segment after change and specific value of each field of the boundary condition involved in this change in a fixed field form, so that the subsequent process can understand the true meaning of this change without looking back at the original text. The system registers each structured business description as a business change point, assigns it a unique business change point identifier, and records the module identifier, old and new demand version number, business term library version number and generation time, and writes it into the business change point table, providing a searchable basis for subsequent mapping to coding technical terms.

[0031] To ensure the stability of the order and the de-duplication, the system uses the combination of the new requirement version number, the old requirement version number and the module identifier to check whether the same pair of requirement texts has been compared before recording the business change points. If the combination already exists, the system will not generate the business change points again and only return the existing records. If unexpected situations such as missing documents, coding inconsistencies, empty sentence sequences, etc. are encountered during the comparison process, the system will mark the change batch as pending manual review status, preserve the current alignment results and error reason summary, and not overwrite the already generated business change point records, thereby ensuring traceability and robustness of the overall process.

[0032] This step is suitable for demand change scenarios in industries such as power, finance, e-commerce, etc. As long as the demand explanation is recorded in natural language and has a module identifier and version number, it can be automatically aligned and differentiated within the above process capabilities. Preferably, in the e-commerce billing module, a version change usually generates five to fifty business change points. The similarity threshold can be set to 0.8, the observation window can be set to the last two versions of the demand, and the time from triggering the comparison to writing the business change point table can be set to one to three minutes. For risk control modules that change frequently and are sensitive to accuracy, a wider observation window range can be set. The system first gives the sentence alignment relationship, and then the business personnel generate business change points by checking and confirming each valid sentence through the manual confirmation interface, thereby ensuring automation efficiency while improving the reliability of key modules.

[0033] S3, construct business change point and technical term pairing samples, train double-branch semantic encoding model composed of business branch and technical branch, and establish the corresponding relationship between business semantic vector and technical semantic vector by contrast learning, which is specifically implemented as: On a production line for large-scale industry informatization projects, this step runs as a model training station in the research and development environment, forming a double-branch semantic encoding model for subsequent mapping through offline training, which is not directly exposed to the business line terminal. Technical terms refer to named entities used to identify functions, interfaces, configuration items, database tables or fields in code repositories and design documents, such as configuration key names representing order amount thresholds, field names representing membership levels, and interface paths representing billing rules. These named entities have stable meanings in the code and can be located through static parsing.

[0034] The business change point refers to a record generated in the previous stage, describing the rule difference of a certain business object before and after a change, and at least includes the change object, the change type, the value before the change, the value after the change, and the referenced business terminology library version number, for reflecting the configuration logic actually adopted by the business side. In the model training station, the system first reads the interface design document, configuration description, database design document and corresponding code repository meta information from the historical project, identifies the function name, interface path, configuration key name, table name and field name through rule analysis, and arranges each technical term together with the service name, module identifier, file path, parameter meaning and field meaning, and forms a technical term record written in the technical term table; the parameter meaning and field meaning can be extracted through the description in the document or inline comments, for providing semantic supplement during training.

[0035] Subsequently, the system finds the business change points that have been implemented in the production environment according to the historical release records and change descriptions, retrieves the corresponding technical implementation, establishes a positive pairing relationship between each business change point and one or several technical terms actually adopted, and at the same time, randomly selects several technical terms from the technical term table within the same batch that are not directly related to the business change point to form a negative pairing relationship. These positive and negative pairing relationships together constitute the pairing samples of business change points and technical terms.

[0036] The business branch refers to the part in the double-branch semantic coding model that is specially used to represent the business side information, by jointly coding the text content of the business change point, the associated business terminology identifier, the domain label, the module label and the logic label into a fixed-length business semantic vector; the technical branch refers to the part in the same model that is specially used to represent the technical side information, by jointly coding the technical term text, the service name, the module identifier and the field meaning into a fixed-length technical semantic vector. The dimensions of the business semantic vector and the technical semantic vector are pre-set to be the same in the model configuration, so as to compare the distances in the same vector space.

[0037] In the training process, the system takes the pairing samples as units, takes the business semantic vector and the technical semantic vector in the positive pairing as the vector pair that needs to be close, takes the business semantic vector and the technical semantic vector in the negative pairing as the vector pair that needs to be far away, adjusts the parameters of the business branch and the technical branch through multiple iterations, so that the distance between the technical terms corresponding to the same business change point in the vector space is reduced, and the distance between the irrelevant technical terms is increased, forming a double-branch semantic coding model that can reflect the stable correspondence between the business change point and the technical term.

[0038] In the training process, the system divides the sample set into a training set and a validation set, calculates performance indicators such as matching accuracy and recall rate based on the validation set after each training round, controls the training rounds according to the preset rules, and freezes the parameters when the model performance reaches the target. After each training is completed, the system assigns a unique model version number to the obtained double-branch semantic encoding model, records the paired sample time range, business change key version range, technical terminology table version range, training rounds, main configuration parameters, and performance indicators into the model repository, and leaves traces with the model file path, creation time, and responsible person identity. It is prohibited to modify the parameters of the released version, and only new versions are allowed to replace old versions, ensuring the traceability and version consistency of subsequent mapping results.

[0039] The training station and the business line running environment pass the desensitization copies of the business change key table and the technical terminology table through a controlled channel, without directly accessing the production database and online services, to meet the security and compliance requirements.

[0040] In practical applications, as long as the target industry has structured business change records and analyzable interface documents, configuration instructions, and database design documents, this step can build a double-branch semantic encoding model adapted to the industry in the development environment, providing basic capabilities for subsequent steps to retrieve candidate technical terms based on business change keys in the engineering environment; preferably, in the e-commerce member billing scenario, the system can collect 10,000 to 50,000 paired samples during a single training, the length of the business semantic vector and the technical semantic vector can be set to 128 to 256 dimensions, the training period can be controlled within one hour, and only a single model version is enabled for each business domain; another preferred way is to establish multiple double-branch semantic encoding models for different industries such as power, finance, and e-commerce, and automatically select the corresponding model version based on the domain label referenced by the business change key during runtime to reduce cross-industry semantic interference and improve matching accuracy in their respective industries.

[0041] S4, receive the engineering environment label, filter the technical terminology subset according to language and framework from the technical terminology vector index, encode the business change key as a business semantic vector, and retrieve candidate technical terms with a semantic similarity exceeding a threshold value in the subset, which is implemented as: In a production line facing large-scale industry informatization projects, this step is deployed in the intermediate service close to the production environment, working with the continuous delivery pipeline to retrieve online candidate between business change points and technical terms. The engineering environment tag refers to the combination of language tags, framework tags and module tags used to describe the current engineering running environment, which is used to constrain the scope of technical implementation allowed in this mapping. The tag is preferably generated by parsing the language field, framework field in the project configuration file and the module name field in the build script, and is delivered together with the build product in the continuous delivery pipeline.

[0042] The technical term vector index refers to the index set established in the vector storage system after generating the technical semantic vector for each technical term record sorted out in the previous step according to the double-branch semantic coding model. Each index record contains at least technical term identification, technical semantic vector and engineering environment tag available for the technical term, wherein the technical term identification corresponds to the function name, interface path, configuration key name or database field name in the code repository.

[0043] The business semantic vector refers to the fixed-length numerical sequence generated in the business branch of the double-branch semantic coding model for a certain business change point, which is used to represent the business meaning and rule characteristics of the business change point in the numerical space. The technical semantic vector refers to the fixed-length numerical sequence obtained after encoding the technical term text and its service name, module tag and field meaning by the technical branch, which is used to represent the implementation meaning of the technical term in the same numerical space.

[0044] The candidate technical term refers to the technical term whose semantic similarity to the business semantic vector reaches a certain threshold under the constraint of the current engineering environment tag, which is used for sorting and rule checking in the subsequent steps. After receiving the business change point unique identification and engineering environment tag from the upstream, the intermediate service first reads the structured business description and the associated business term library version number and domain tag from the business change point table through the business change point unique identification, calls the business branch of the online double-branch semantic coding model to generate the business semantic vector, and records the binding of this generation action with the model version number. Subsequently, the intermediate service filters out the index records whose language tags and framework tags are completely consistent with the current engineering environment tag in the technical term vector index according to the engineering environment tag, forming a technical term subset; records outside the technical term subset are not involved in this similarity calculation, thereby avoiding selecting technical terms incompatible with the current language or framework.

[0045] In the technical term subset, the intermediate service calculates the semantic similarity between the business semantic vector and each technical semantic vector, marks the technical terms with a semantic similarity higher than the pre-configured semantic similarity threshold as candidate technical terms, and sorts the corresponding semantic similarity scores and technical term identifications into a candidate technical term list.

[0046] To ensure the latency and stability in the actual production environment, the intermediate service sets the maximum response time and the maximum number of concurrent requests for each candidate retrieval, and the semantic similarity calculation is preferably completed in memory, and the business semantic vector of the recent business change point and the corresponding candidate technical term list are cached.

[0047] In the design of the idempotent strategy, the intermediate service generates a request identifier for each call, and combines the request identifier with the business change point unique identifier and the engineering environment label, which is used to identify repeated calls in a short time, and directly returns the generated candidate technical term list for repeated calls without repeating the semantic vector generation and index query, thereby ensuring the consistency of the candidate set of the same business change point under the same engineering environment label.

[0048] When network fluctuations occur or the node where the technical term vector index is located is temporarily unavailable, the intermediate service re-accesses the index storage according to the configured retry number and interval, and for requests that exceed the retry number and still cannot obtain complete results, records the current state as a fault event, retains the business change point unique identifier, the engineering environment label, the request identifier, and the error reason at that time, and does not write any candidate technical term to the downstream, preventing incomplete results from entering the subsequent sorting and verification link.

[0049] After the candidate technical term list is formed, the intermediate service writes the list together with the business change point unique identifier, the engineering environment label, the model version number, and the generation time into the candidate technical term table, providing input for the next step of business logic category matching, environment matching degree scoring, and business rule verification; the downstream service queries the corresponding candidate technical term list by business change point unique identifier through the inter-service call protocol, realizing the connection between the upstream and downstream.

[0050] In the actual production line, as long as the project has a parsable engineering configuration, a built technical term vector index, and a callable double-branch semantic encoding model, an engineering technician in the field can deploy the intermediate service according to the above process, synchronize the technical term vector index to a node close to the production environment through the daily or pre-online building process, and realize online candidate retrieval of the business change point; preferably, in a typical online environment, the candidate retrieval service can be set to support fifty to one hundred requests per second, each request returns about ten candidate technical terms, and the similarity threshold can be set to zero point seven to zero point eight, and the delay of a single candidate retrieval can be controlled at about two hundred milliseconds; another preferred way is to shard the technical term vector index according to the module label, and route the candidate retrieval requests of different module labels to different nodes, thereby dispersing the load in industry projects with numerous business modules, reducing the pressure on a single node, and independently maintaining the consistent relationship between the engineering environment label and the index data on each shard node.

[0051] S5. Rank and score candidate technical terms based on business logic category and environment matching degree. Verify whether the candidate technical terms meet the business rules in the business terminology library, and remove candidate technical terms that do not meet the verification conditions. Specifically, this is implemented as follows: On a production line for large-scale industry IT projects, this step runs after the candidate technical terminology search, used to further screen and verify candidate technical terms under the same engineering environment label. Business change highlights refer to the structured business change records generated by the preceding steps, including the change object, change type, pre-change value, post-change value, and the referenced business terminology library version number. Business logic category indicates whether the business change highlights belong to categories such as discount rule adjustments, threshold upper and lower limit modifications, or segmented billing structure changes. This is determined by the requirements comparison stage in conjunction with the logical tags of business terms in the business terminology library, serving as the basis for determining the appropriate technical implementation form for this change.

[0052] Environment matching degree refers to the degree of fit between the module label of the candidate technical term and the label of the current engineering environment. It is used to reflect whether the service and module where the candidate technical term is located are consistent with the module where the key points of this business change are located. Business rules refer to the constraint descriptions bound to business terms in the business term library. They are recorded with rule identifiers and rule content and are used to describe order amount ranges, membership level ranges, time ranges and their corresponding discounts or billing methods.

[0053] This step retrieves the candidate technical terminology list from the candidate technical terminology table based on the business change key point identifier and engineering environment label. Each candidate technical terminology record contains at least the technical terminology identifier, semantic similarity score, module label and field meaning, parameter meaning, configuration method, and other information. Based on the business logic category label in the business change key point record, the system determines whether the service and module label of the candidate technical terminology correspond to that category. Candidate technical terms with consistent logical categories are assigned a higher business logic category consistency score, and candidate technical terms that are completely consistent with the module label in the current engineering environment label are assigned a higher environment matching score. The above scores and semantic similarity scores are then combined according to pre-configured weights to form a comprehensive score, and the candidate technical terms are sorted from high to low according to the comprehensive score.

[0054] On the basis of the sorting result, the system retrieves the relevant business rules under the version number of the business terminology library according to the business change point cited business terminology library version number, obtains the rule content through the rule identifier, and compares the sorted candidate technical terms one by one to see if the field meaning, parameter meaning and configuration method can completely express the lower limit of the amount, the upper limit of the amount, the member level, the time range and other conditions involved in the rule. The candidate technical terms that do not have corresponding fields, whose field value range cannot cover the rule requirements, or whose configuration method does not match the rule execution order are marked as not meeting the review conditions, and are removed from the candidate list, and the technical term identifier, the corresponding business rule identifier and the reason for failing to pass are recorded in the log.

[0055] All sorting and checking operations are completed under the same business terminology library version number locking condition to avoid mixing different versions of business rules in the same screening. To ensure order and idempotency, the system uses the combination of the business change point identifier and the model version number as the identifier of this sorting process. When the same combination is triggered multiple times, as long as the version number does not change, the comprehensive score and the passing review result obtained in the first calculation are reused, and the operation is not repeated.

[0056] If the candidate technical term list is empty after the business rule review, the system marks the business change point as requiring manual intervention, and archives the current engineering environment label, model version number, evaluated candidate set and rule matching failure reason to provide a basis for subsequent supplement of technical terms or adjustment of weights. If there is one or more technical terms that pass the review, they are written into a special table as the set of technical terms that pass the review, and the comprehensive score, the cited business terminology library version number and the generation time are recorded to provide a reference for the subsequent step of generating AI coding technical terms.

[0057] This step is suitable for industries such as power, finance and e-commerce that have clear business rules such as amount interval, user group and time period control. As long as the business rules have formed searchable rule identifiers and rule content in the business terminology library, engineering technicians in the field can implement re-screening and compliance review of candidate technical terms according to the above process. Preferably, in the e-commerce billing scenario, the comprehensive score weight can be set so that the semantic similarity score accounts for a large proportion, the business logic category consistency and the environment matching degree each account for a medium proportion, and finally a small number of candidate technical terms with high comprehensive score ranking are retained as the technical terms that pass the review. Another preferred way is to add a manual confirmation link before the business goes online. Experienced developers can select the final technical terms to be used in the sorting results given by the system, and write back the selection record to provide a sample basis for subsequent adjustment of comprehensive score weight and business rule review strategy.

[0058] S6, combine the checked technical terms and corresponding business change points to generate AI coding technical term records, write them into the AI coding technical term list with version number and engineering environment label for code generation component to call, and the specific implementation is: On the production line for large-scale industry informatization projects, this step is used as the last end station next to code generation, and the checked technical terms are combined with the upstream business change points at the same station to form descriptions that can be directly used for automatically generating code. The AI coding technical term record refers to a structured term description for the code generation component, which is used to accurately indicate the functions, interfaces, and data fields that should be used for a certain business change in the current engineering environment. It at least includes technical term text, function name or interface name, parameter name list, related configuration key name, related database field name, and bound business change point identification, business logic category, engineering environment label, and model version number.

[0059] The business change point identification is used to uniquely mark the business change record formed in the previous step, the business logic category is used to indicate that the record belongs to discount rule adjustment, threshold upper and lower limit modification, or segmented billing structure change, etc., the engineering environment label is used to indicate the combination of language label, framework label, and module label, and the model version number is used to indicate the version of the two-branch semantic coding model used when forming the current mapping. The code generation component refers to an automatic generation program deployed in the continuous delivery pipeline or research and development platform, which is used to generate or modify code files in the target engineering according to the AI coding technical term record.

[0060] This step first reads the set of checked technical terms from the checked technical term table according to the business change point identification and engineering environment label. For each technical term in the set, the function name or interface name, parameter name list, related configuration key name, and related database field name are supplemented by indexing design documents and configuration instructions. The field meanings are consistent with the business rule meanings in the previous step to ensure that the amount, time, user group, and other indicators are consistent with the business change point record. At the same time, according to the business logic category, a brief implementation prompt is generated to outline the role of the technical term in the current module in natural language, such as "select different discount values according to order amount intervals in order billing service", and the above content is combined with the business change point identification, business logic category, engineering environment label, model version number to form a complete AI coding technical term record.

[0061] All AI coding technology term records are written into the AI coding technology term list according to the business version number, engineering environment tag, and module tag. The AI coding technology term list refers to a collection of multiple AI coding technology term records under a certain business version and engineering environment. It is stored in the central storage in a documented format or structured record form. The central storage record list generates a time, a person's identity, or an automatically generated identifier. It is associated with the business term library version number and model version number of the previous step, forming an evidence chain from business text to technical terms to code generation basis.

[0062] At the interface level, the system provides a unified service interface for code generation components. The interface uses a text-based request-response approach. The request content includes at least the business version number, module tag, and business change point identification list. The return content is the corresponding AI coding technology term record set and its serial number information. The interface agrees on a set of minimum implementable error codes. For example, zero is used to represent a successful query. A negative value is used to represent a non-existent business version number. Another negative value is used to represent a specified business change point that does not form a technical term mapping. Another negative value is used to represent an unavailable or unverified model version. Another negative value is used to represent internal service failures. Each type of error code and handling suggestion is uniformly registered in the operation and maintenance document.

[0063] To ensure idempotency, this step uses the combination of business change point identification and engineering environment tag as the unique key when writing the AI coding technology term list. Requests for repeated writing of the same combination only update the access log and reference count, without overwriting existing records. To ensure order, the list is written in ascending order of business change point identification number within the same business version, and a serial number field is added to each record. Code generation components can process business change points in a fixed order according to this information. To ensure deduplication, the system compares the fields of the record to be added with the records in the current list before writing. If the technical term text, function name or interface name, parameter name list, configuration key name, and database field name are completely identical and the bound business change point identification, engineering environment tag, and model version number are also identical, the existing record identifier is directly reused, and a new record is not created.

[0064] The step is suitable for informatization projects in industries that need to convert business changes into compilable code in large quantities. As long as the upstream has formed stable business change points and a checked technical term set, an engineering technician in the field can construct an AI coding technical term list in the central storage according to the above process and provide a unified term basis for the code generation component through the interface; preferably, in the e-commerce scenario, the AI coding technical term list can be set to contain ten to one hundred records each time it is updated, the delay of a single query list can be controlled within five hundred milliseconds, and the code generation component reads the list in batches and converts all business change points into compilable code within twenty-four hours; another preferred way is not to use a synchronous query interface, but to use a message queue to transfer term records between the AI coding technical term list and the code generation component, to enhance the stability of the cross-system link through the retry mechanism of the queue, and to carry the business version number, engineering environment label and model version number in each message to realize end-to-end version alignment and safety compliance audit.

[0065] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for generating a mapping from business terms to coded technical terms, characterized in that, include: S1. Collect multi-source business texts, cluster business terms by domain, module, and logic, and generate a business terminology library with version numbers; S2. Align and differentiate the old and new business requirement texts, and use the business terminology library to complete the omitted boundary conditions to generate key business change points with annotations on the object and type of change. S3. Construct pairing samples of key business changes and technical terms, train a dual-branch semantic encoding model composed of business branches and technical branches, and use contrastive learning to establish the correspondence between business semantic vectors and technical semantic vectors. S4. Receive engineering environment tags, filter technical term subsets from the technical term vector index by language and framework, encode key business changes into business semantic vectors, and retrieve candidate technical terms with semantic similarity exceeding the threshold in the subset. S5. Sort and score candidate technical terms according to business logic category and environment matching degree. Verify whether the candidate technical terms meet the business rules in the business term library. Eliminate candidate technical terms that do not meet the verification conditions. S6. Combine the verified technical terms and corresponding business change points to generate an AI coding technical terminology record, and write it into an AI coding technical terminology list with version number and engineering environment label for the code generation component to call.

2. The method for generating a mapping from business terms to coding technology terms according to claim 1, characterized in that, S1 includes: Retrieve business requirements specifications, product configuration specifications, marketing plans, contract terms, and business operation manuals from the enterprise document platform incrementally, based on project identifiers and with source system identifiers and version numbers. Standardize the format of business texts and segment them into sentences and words, filter out colloquial words, count the number of times candidate phrases appear in the same project, and mark phrases that reach the set threshold and refer to customer type, product type, billing cycle and discount method as business terms; Business terms are clustered based on semantic similarity, and business term records are generated by domain labels, module labels, and logical labels. At the end of each extraction cycle, a business term library version number is assigned to the set of business term records and written to the central term library table.

3. The method for generating a mapping from business terms to coding technology terms according to claim 1, characterized in that, S2 include: For new and old business requirement texts with the same module identifier, sentence segments are formed by period and converted into sentence vectors. Within the observation window determined by the old and new version numbers, old sentences are selected for each new sentence based on similarity, and new descriptions and descriptions to be deleted are marked. Aligned sentence pairs are differentiated by words and phrases, and changed fragments containing business terms are identified by terminology and business rules through a business terminology database. When a new requirement text is detected to lack fields for amount boundary, user group, and time range, the boundary conditions are supplemented according to the business rules of the old requirement version. The key points of business changes are composed of the object to be changed, the type of change, the value before the change, the value after the change, and the boundary conditions after the change, and are written into the key points of business changes table.

4. The method for generating a mapping from business terms to coding technology terms according to claim 1, characterized in that, S3 includes: The function name, interface path, configuration key name, table name, and field name are parsed from the interface design document, configuration instructions, database design document, and code repository metadata. The parsing results are combined with the service name, module identifier, file path, parameter meaning, and field meaning to generate technical terminology records and write them into the technical terminology table; Based on historical release records, select business change points that have been implemented in the production environment from the business change point table. Establish a positive pairing relationship between each business change point and the actual technical terms used in the technical terminology table. In the same batch, extract technical terms that are not related to the business change point from the technical terminology table to generate a negative pairing relationship, so as to form a pairing sample set of business change points and technical terms.

5. The method for generating a mapping from business terms to coding technology terms according to claim 1, characterized in that: In the dual-branch semantic coding model, business branches and technical branches are set up. The business branch receives key business change texts, business terminology identifiers, domain labels, module labels and logical labels to generate business semantic vectors. The technical branch receives technical terminology text, the name of the service to which it belongs, the module identifier, and the meaning of the fields to generate a technical semantic vector. Both types of vectors have the same dimension. Using the paired sample set as a unit, the business semantic vectors and technical semantic vectors in positive pairings are set as close vector pairs, and the business semantic vectors and technical semantic vectors in negative pairings are set as far apart vector pairs; By iteratively adjusting the parameters of the business branch and the technical branch, a dual-branch semantic encoding model is obtained that represents the correspondence between key business changes and technical terms.

6. The method for generating a mapping from business terms to coding technology terms according to claim 1, characterized in that, S4 include: Receive the unique identifier of key business changes and the engineering environment label, read the structured business description from the key business change table through the unique identifier of key business changes, and generate a business semantic vector through business branches; In the technical terminology vector index, technical terminology records with consistent language and framework tags are selected based on engineering environment tags to form a subset of technical terms. The semantic similarity between the business semantic vector and each technical semantic vector is calculated in this subset of technical terms. Technical terms with semantic similarity higher than a preset semantic similarity threshold are identified as candidate technical terms, and these candidate technical terms, along with unique identifiers of key business changes and engineering environment labels, are written into the candidate technical term list.

7. The method for generating a mapping from business terms to coding technology terms according to claim 1, characterized in that, S5 include: Retriev candidate technical terms from the candidate technical terminology list by identifying key business changes and engineering environment labels; A comprehensive score is calculated and ranked based on the business logic category of the key business changes, the module tags of candidate technical terms, and semantic similarity scores; Extract business rules from the business terminology database based on the version number of the business terminology database referenced in the key points of business changes; For each of the sorted candidate technical terms, check whether the meaning of the fields, the meaning of the parameters, and the configuration methods fully cover the amount range, membership level, and time range specified in the business rules; Candidate technical terms that are not fully covered are removed, and the remaining candidate technical terms, along with their comprehensive scores and business terminology database version numbers, are written into the approved technical terminology table.

8. The method for generating a mapping from business terms to coding technology terms according to claim 1, characterized in that, S6 include: Read the set of technical terms from the verified technical terminology list according to the key business change indicators and engineering environment labels; Based on the design documents and configuration instructions, supplement each technical term with a list of function names, interface names, parameter names, configuration key names, and database field names, and verify consistency according to the amount, time, and user group criteria in the key points of business changes; The technical terminology text, the aforementioned name information, key business change identifiers, business logic categories, engineering environment tags, and model version numbers are combined to generate an AI coding technical terminology record. This record is then written into the AI ​​coding technical terminology list according to the business version number and engineering environment tags, allowing the code generation component to query the record based on the business version number, module tag, and key business change identifier.

Images