A method and apparatus for integrating multi-electronic channel systems based on a large model

Abstract

This invention provides a method and apparatus for integrating multiple electronic channel systems based on a large model. The integration method includes: acquiring multimodal data from multiple electronic channel systems to be integrated, and constructing a vector knowledge base based on the multimodal data; using retrieval enhancement generation technology and mind chain technology, utilizing an integration model to perform business function identification and architecture analysis on the vector knowledge base, generating integration architecture information; using the integration model to generate target code and data migration scripts for the target framework based on the integration architecture information; constructing an integration system based on the target code and data migration scripts, and using test cases to verify the integration system, obtaining the system integration result. This invention, through an integration model that has been adaptively fine-tuned by the domain, automatically migrates multiple heterogeneous electronic channel systems to a unified platform, constructs an integrated system, shortens the system integration period, reduces costs, and minimizes human error.

Description

Technical Field

[0001] This invention relates to the field of system integration technology, specifically to a method and apparatus for integrating multi-electronic channel systems based on a large model. Background Technology

[0002] In the process of digital transformation, telecom operators often introduce multiple electronic channel systems built by different vendors to meet business needs at different historical periods. These systems include online service halls, mobile apps, and WeChat mini-programs. Over time, these systems have evolved independently, resulting in a siloed system cluster with varying technical architectures, inconsistent data models, and redundant business functions. This heterogeneous coexistence leads to a series of problems, such as wasted internal resources, high maintenance costs, and slow business response.

[0003] For this type of multi-system integration, the industry typically adopts a traditional manual restructuring and integration approach. This approach first requires establishing a dedicated team to manually analyze the business functions, database structures, and interface logic of each system, and then selecting a unified technical framework as a baseline. Subsequently, developers, based on their understanding of the old system logic, must reimplement all functional modules under the new framework almost entirely through manual coding, and finally perform data migration and system cutover.

[0004] However, the aforementioned manual refactoring and integration methods suffer from high labor costs and long project cycles, leading to reliance on senior technical personnel for repetitive code translation work and hindering rapid response to market changes. Furthermore, these solutions carry high implementation risks; manual rewriting can easily introduce new logical errors, and the cutover process is risky, resulting in poor stability of multi-system integration. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method and apparatus for integrating multi-electronic channel systems based on a large model, aiming to solve the problems of high labor costs, long cycles, and high risks in the process of integrating multi-electronic channel systems.

[0006] This invention discloses a method for integrating multi-electronic channel systems based on a large model, comprising: Acquire multimodal data from multiple electronic channel systems to be integrated, and construct a vector knowledge base based on the multimodal data; Based on retrieval enhancement generation technology and thinking chain technology, we use an integrated model to identify business functions and analyze the architecture of the vector knowledge base, and generate integrated architecture information. The large model is a deep convolutional network that has been fine-tuned by domain adaptation. The integration model is used to generate target code and data migration scripts for the target framework based on the integration architecture information. The integration system is built based on the target code and data migration script, and the integration system is verified using test cases to obtain the system integration results.

[0007] Preferably, the multimodal data includes source code files, database scripts, API interface documents, and system configuration documents; A vector knowledge base is built based on multimodal data, including: Multimodal data is preprocessed and vectorized to obtain high-dimensional vectors; High-dimensional vectors and multimodal data are stored in a vector database, and a semantic retrieval index is built for the vector database. The semantic retrieval index is used to characterize the relationship between multimodal data and high-dimensional vectors.

[0008] Preferably, the multimodal data is preprocessed and vectorized to obtain a high-dimensional vector, including: The source code file is parsed to generate an abstract syntax tree, and code feature information is extracted from the abstract syntax tree. The code feature information includes code structure, class definition, method logic and parameter information. Text slicing is performed on code feature information to obtain code text fragments; Extract structured information from database scripts and convert it into database text information. The structured information includes table creation statements, stored procedures, table structure, field types, and index configurations. Extract the API interface description text from the API interface documentation, and perform word segmentation on the interface description text to obtain the key information text; Extract the configuration item names, configuration values, and configuration descriptions from the system configuration document to obtain the configuration text; The code text fragments, database text information, key information text, and configuration text are vectorized to obtain high-dimensional vectors.

[0009] Preferably, the multi-electronic channel system integration method based on a large model further includes: using a pre-trained primary model as the base model, wherein the training data of the primary model consists of general code and general text; Construct a fine-tuning dataset, which includes system analysis question-and-answer data, code transformation sample data, and architecture design task data; The fine-tuning dataset is input into the primary model based on the fine-tuning instructions to fine-tune the primary model and obtain the integrated model.

[0010] Preferably, based on retrieval enhancement generation technology and mind chain technology, an integrated model is used to identify business functions and analyze the architecture of the vector knowledge base, including: Obtain the task instructions from the electronic channel system to be integrated, and convert the task instructions into query vectors. The task instructions include business function identification and architecture analysis. Based on the query vector, a similarity search is performed in the vector knowledge base to obtain the context-related data corresponding to the query vector. The context-related data includes code snippets and document text. Structured prompts are generated based on task instructions, contextual data, and thought chain guide words. These guide words are used to instruct the integrated model on the reasoning process generated using thought chain technology. Input structured prompts into the integration model so that the integration model can generate integration architecture information.

[0011] Preferably, the integrated architecture information includes one or more of the following: a list of business functions, a cross-system function similarity mapping, a system dependency graph, a fused standardized data model, and an API interface design draft.

[0012] Preferably, the target code and data migration script of the target framework are generated based on the integration architecture information using the integration model, including: Source system data is generated based on the multimodal data of the electronic channel system to be integrated. The source system data includes source framework code and source application programming interface. Based on the integrated architecture information, target system data and constraints are generated. The target system data includes the target framework code and the target application programming interface. The constraints include the target framework's programming specifications, coding conventions, and performance requirements. By using retrieval enhancement generation technology, transformation case data related to source system data and target system data are retrieved from the vector knowledge base to obtain reference context information; Based on source system data, target system data, constraints, reference context information, and thought chain guidance words, generate code generation prompt words; Input code generation prompts into the integrated model so that the integrated model can perform distributed reasoning based on the thought chain technology to obtain the target code and data migration script.

[0013] Preferably, an integration system is constructed based on the target code and data migration script, and integration verification is performed on the integration system using test cases, including: The target code is compiled and deployed, and data migration scripts are executed to build the integrated system; Control flow and data flow are generated based on the target code, and test cases are generated based on the control flow and data flow using an integrated model; Run test cases in the integrated system to obtain integration test results, and generate system integration results based on the integration test results.

[0014] Preferably, the method for integrating multiple electronic channel systems based on a large model further includes: executing test cases in the electronic channel systems to be integrated to obtain source test results; A difference analysis was performed on the source test results and the integration test results to obtain the difference data; The integrated model is optimized and updated based on the discrepancy data to improve its parameters.

[0015] This invention discloses a multi-electronic channel system integration device based on a large model, used to execute the aforementioned multi-electronic channel system integration method based on a large model. The multi-electronic channel system integration device includes: The knowledge base construction module is configured to: acquire multimodal data from multiple electronic channel systems to be integrated, and construct a vector knowledge base based on the multimodal data; The integrated architecture module is configured to: use the integrated model to identify business functions and analyze the architecture of the vector knowledge base based on retrieval enhancement generation technology and thinking chain technology, generate integrated architecture information, and the large model is a deep convolutional network that has been fine-tuned by domain adaptation. The code generation module is configured to: use the integration model to generate target code and data migration scripts for the target framework based on the integration architecture information; The result verification module is configured to: build an integrated system based on the target code and data migration script, and use test cases to verify the integrated system to obtain the system integration result.

[0016] Compared with existing technologies, the advantages of this invention are as follows: This invention leverages the integrated model's capabilities for automatic parsing of multimodal data, identification of business functions, and architectural analysis to automate the integration process, significantly reducing labor costs and shortening project cycles. Furthermore, this invention utilizes enhanced generation technology to access historical transformation cases and best practices in a vector knowledge base, combined with mind chain technology to guide the integrated model in step-by-step reasoning, ensuring logical consistency between code generation and data migration. Moreover, by comparing and analyzing test results between the source system and the integrated system, it can quickly pinpoint differences and optimize the model in reverse, further enhancing the stability of system integration. Attached Figure Description

[0017] Figure 1 A flowchart illustrating the multi-electronic channel system integration method based on a large model provided by the present invention; Figure 2 This is a schematic diagram of the structure of the multi-electronic channel system integration device based on a large model provided by the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] The present invention will now be described in further detail with reference to the accompanying drawings.

[0020] This invention provides a method for integrating multi-electronic channel systems based on a large model, such as... Figure 1 As shown, the method for integrating multi-electronic channel systems based on a large model includes the following steps.

[0021] Step S1: Obtain multimodal data from multiple electronic channel systems to be integrated, and construct a vector knowledge base based on the multimodal data.

[0022] In this embodiment of the invention, the multimodal data includes source code files, database scripts, API interface documentation, and system configuration documentation. The source code files encompass business logic code, front-end interaction code, and back-end service code written in different programming languages ​​such as Java and Python; the database scripts contain table creation statements, stored procedures, triggers, and index configurations for databases such as MySQL and Oracle; the API interface documentation includes parameter definitions, return value formats, calling examples, and permission descriptions for RESTful APIs and SOAP interfaces; and the system configuration documentation covers server deployment parameters, middleware configuration, logging policies, and circuit breaker / rate limiting rules.

[0023] When building a vector knowledge base based on multimodal data, the multimodal data is first preprocessed and vectorized to obtain high-dimensional vectors. Then, the high-dimensional vectors and multimodal data are stored in a vector database, and a semantic retrieval index is established for the vector database. The semantic retrieval index is used to characterize the relationship between multimodal data and high-dimensional vectors.

[0024] Specifically, the preprocessing of multimodal data proceeds as follows: Syntax parsing of the source code file generates an abstract syntax tree (AST), and code feature information is extracted from the AST. This code feature information includes code structure, class definitions, method logic, and parameter information. Text slicing is then performed on the code feature information to obtain code text fragments. Structured information is extracted from the database script and converted into database text information, including table creation statements, stored procedures, table structures, field types, and index configurations. Interface description text is extracted from the API documentation and segmented to obtain key information text. Configuration item names, configuration values, and configuration descriptions are extracted from the system configuration documentation to obtain configuration text. Furthermore, the code text fragments, database text information, key information text, and configuration text are vectorized to obtain high-dimensional vectors.

[0025] S2. Based on retrieval enhancement generation technology and mind chain technology, the integrated model is used to identify business functions and analyze the architecture of the vector knowledge base, and generate integrated architecture information.

[0026] In this embodiment of the invention, the ensemble model is a deep convolutional network that has undergone domain-adaptive fine-tuning. Specifically, in constructing the ensemble model, a pre-trained primary model is first used as the base model, and the training data for the primary model consists of general code and general text. Then, a fine-tuning dataset is constructed, which includes system analysis question-answering data, code conversion sample data, and architecture design task data. Finally, based on fine-tuning instructions, the fine-tuning dataset is input into the primary model to fine-tune the primary model, resulting in the ensemble model.

[0027] In this way, the integrated model can accurately understand the business logic and technical specifications of the electronic channel system field, avoiding the "illusion" problem of general models in professional scenarios.

[0028] Furthermore, based on retrieval enhancement generation technology and mind chain technology, an integrated model is used to identify business functions and analyze the architecture of the vector knowledge base. The retrieval enhancement generation technology combines the generation process of the integrated model with the retrieval results of the vector knowledge base to ensure the accuracy and contextual relevance of the output content; the mind chain technology guides the integrated model to simulate the human logical reasoning process, gradually breaking down complex tasks and improving the logic and interpretability of the analysis results.

[0029] In practice, the process begins by acquiring the task instructions from the electronic channel systems to be integrated and converting them into query vectors. These instructions include business function identification and architecture analysis, such as "identify the user login functions of each system and analyze their dependencies." Next, a similarity search is performed in a vector knowledge base based on the query vectors to obtain context-related data. This context-related data includes code snippets and document text. For example, based on the cosine similarity algorithm, context-related data with a similarity higher than a preset threshold (e.g., 0.8) related to the query vectors is retrieved from the vector knowledge base. This includes Java code snippets related to the user login function, MySQL user table creation statements, and RESTful login interface documentation.

[0030] Furthermore, structured prompts are generated based on task instructions, contextual data, and thought chain guides. These thought chain guides instruct the integration model on how to generate reasoning processes using thought chain technology. For example, a structured prompt could be: "First, analyze the code implementation logic of the user login function in each system to be integrated, then analyze the interaction flow of user data between systems, and finally draw a dependency graph between systems."

[0031] Finally, the structured prompts are input into the integration model so that the integration model can generate integration architecture information.

[0032] For example, after inputting structured prompts into the integrated model, the model will first output the reasoning process based on the thought chain guide words, such as "System A's login function is implemented through the UserLoginService class, which depends on the username and password fields of the user table in the database; System B's login function calls a third-party authentication interface and reads user session information from the local cache table; the synchronization of user data between System A and System B depends on a scheduled task calling System C's user synchronization API...". Finally, it generates integrated architecture information such as a list of business functions including user login function, cross-system function similarity mapping (e.g., the similarity between the login functions of System A and System B is 0.75), and inter-system dependency graph.

[0033] In this embodiment of the invention, the integrated architecture information includes one or more of the following: a business function list, a cross-system function similarity mapping, an inter-system dependency graph, a fused standardized data model, and an API interface design draft.

[0034] S3. Utilize the integration model to generate target code and data migration scripts for the target framework based on the integration architecture information.

[0035] In this embodiment of the invention, source system data is generated based on the multimodal data of the electronic channel system to be integrated. The source system data includes source framework code and source application programming interfaces (APIs). Next, target system data and constraints are generated based on the integration architecture information. The target system data includes target framework code and target APIs, and the constraints include the target framework's programming specifications, coding conventions, and performance requirements. Retrieval-enhanced generation technology is used to retrieve transformation case data related to the source system data and target system data from a vector knowledge base to obtain reference context information. Further, code generation prompts are generated based on the source system data, target system data, constraints, reference context information, and thought chain guidance words. Finally, the code generation prompts are input into the integration model, enabling the integration model to perform distributed reasoning based on thought chain technology to obtain the target code and data migration script.

[0036] In this way, by retrieving transformation case data related to the source and target system data from the vector knowledge base using retrieval-enhanced generation techniques, reference context information is obtained, enabling the integrated model to refer to historical successes when generating target code and data migration scripts. The thought chain guiding words then drive the model to break down the transformation task step by step: first, clarifying the core logic of the source system code and its compatibility with the target framework; then, adjusting the code structure based on constraints; and finally, generating target code that conforms to the specifications.

[0037] S4. Build an integrated system based on the target code and data migration script, and use test cases to verify the integrated system and obtain the system integration result.

[0038] In this embodiment of the invention, the target code is compiled and deployed, and a data migration script is executed to build an integrated system. Control flow and data flow are generated based on the target code, and test cases are generated using an integration model based on the control flow and data flow. The test cases are run in the integrated system to obtain integration test results, and the system integration result is generated based on the integration test results. The test cases automatically generated by the integration model can accurately cover the core logic branches and data flow nodes of the target code, avoiding omissions that occur when manually writing test cases.

[0039] For example, for the target code of the user login function, the model will generate test cases including scenarios such as empty parameter input, incorrect password verification, and concurrent login requests, according to the "parameter validation → database query → session generation" process in the control flow. Simultaneously, it will generate test cases for data consistency verification based on the field mapping relationship of user data migrating from the source system to the target system in the data flow, such as verifying the migration accuracy of the "user_id" field in the source system's user table and the "user_identity" field in the target system. After running these test cases in the integrated system, the model will output integration test results including metrics such as function pass rate, data migration accuracy, and interface response time. If the function pass rate reaches 95% or higher and the data migration accuracy is 100%, the system integration result is considered qualified. If there are failed test items, the model will further locate problematic code snippets or data migration logic vulnerabilities, providing a clear direction for subsequent optimization.

[0040] In this embodiment of the invention, test cases are executed in the electronic channel system to be integrated to obtain source test results; the source test results and integration test results are analyzed for differences to obtain difference data; and the integration model is optimized and updated based on the difference data to optimize the model parameters of the integration model.

[0041] In this way, by transforming discrepancy data into feedback signals for model optimization, deviations in the integrated model during business function identification, architecture analysis, or code generation can be corrected in a targeted manner.

[0042] like Figure 2 As shown, the present invention also provides a multi-electronic channel system integration device based on a large model, used to execute a multi-electronic channel system integration method based on a large model. The multi-electronic channel system integration device based on a large model includes: a knowledge base construction module 201, an integration architecture module 202, a code generation module 203, and a result verification module 204.

[0043] The knowledge base construction module 201 is configured to: acquire multimodal data from multiple electronic channel systems to be integrated, and construct a vector knowledge base based on the multimodal data. The integration architecture module 202 is configured to: utilize retrieval enhancement generation technology and thought chain technology, employ an integration model to identify business functions and analyze the architecture of the vector knowledge base, generating integration architecture information. The main model is a deep convolutional network that has undergone domain-adaptive fine-tuning. The code generation module 203 is configured to: use the integration model to generate target code and data migration scripts for the target framework based on the integration architecture information. The result verification module 204 is configured to: construct the integration system based on the target code and data migration scripts, and use test cases to verify the integration system, obtaining the system integration results.

[0044] As can be seen from the above technical solution, this invention provides a method and apparatus for integrating multiple electronic channel systems based on a large model. The integration method includes: acquiring multimodal data of multiple electronic channel systems to be integrated, and constructing a vector knowledge base based on the multimodal data; using retrieval enhancement generation technology and mind chain technology, utilizing the integration model to perform business function identification and architecture analysis on the vector knowledge base, generating integration architecture information; using the integration model to generate target code and data migration scripts for the target framework based on the integration architecture information; constructing the integration system based on the target code and data migration scripts, and using test cases to execute and verify the integration system, obtaining the system integration result. This invention, through an integration model that has been adaptively fine-tuned by the domain, automatically migrates multiple heterogeneous electronic channel systems to a unified platform, constructs an integrated system, shortens the system integration period, reduces costs, and minimizes human error.

[0045] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for integrating multi-electronic channel systems based on a large model, characterized in that, include: Acquire multimodal data from multiple electronic channel systems to be integrated, and construct a vector knowledge base based on the multimodal data; Based on retrieval enhancement generation technology and mind chain technology, the vector knowledge base is used to identify business functions and analyze architecture using an integrated model to generate integrated architecture information. The integrated model is a deep convolutional network that has been fine-tuned by domain adaptation. The integration model is used to generate target code and data migration scripts for the target framework based on the integration architecture information; An integrated system is constructed based on the target code and the data migration script, and the integrated system is verified using test cases to obtain the system integration result.

2. The multi-electronic channel system integration method according to claim 1, characterized in that, The multimodal data includes source code files, database scripts, API interface documents, and system configuration documents; The construction of a vector knowledge base based on the multimodal data includes: The multimodal data is preprocessed and vectorized to obtain a high-dimensional vector; The high-dimensional vector and the multimodal data are stored in the vector database, and a semantic retrieval index is established for the vector database. The semantic retrieval index is used to characterize the association between the multimodal data and the high-dimensional vector.

3. The multi-electronic channel system integration method according to claim 2, characterized in that, The preprocessing and vectorization of the multimodal data to obtain a high-dimensional vector includes: The source code file is parsed to generate an abstract syntax tree, and code feature information is extracted from the abstract syntax tree. The code feature information includes code structure, class definition, method logic, and parameter information. The code feature information is processed by text slicing to obtain code text fragments; Extract the structured information from the database script and convert the structured information into database text information. The structured information includes table creation statements, stored procedures, table structure, field types, and index configurations. Extract the interface description text from the API interface document, and perform word segmentation on the interface description text to obtain the key information text; Extract the configuration item names, configuration values, and configuration descriptions from the system configuration document to obtain the configuration text; The code text fragment, the database text information, the key information text, and the configuration text are vectorized to obtain the high-dimensional vector.

4. The multi-electronic channel system integration method according to claim 1, characterized in that, Also includes: The pre-trained primary model is used as the base model, and the training data of the primary model consists of general code and general text. Construct a fine-tuning dataset, which includes system analysis question-and-answer data, code conversion sample data, and architecture design task data; The fine-tuning dataset is input into the primary model based on the fine-tuning instructions to fine-tune the primary model and obtain the integrated model.

5. The method for integrating multiple electronic channel systems according to claim 1, characterized in that, The method, based on retrieval enhancement generation technology and mind chain technology, utilizes an integrated model to identify business functions and analyze the architecture of the vector knowledge base, generating integrated architecture information, including: Obtain the task instructions from the electronic channel system to be integrated, and convert the task instructions into query vectors. The task instructions include business function identification and architecture analysis. Based on the query vector, a similarity search is performed in the vector knowledge base to obtain the context-related data corresponding to the query vector. The context-related data includes code snippets and document text. Structured prompts are generated based on the task instructions, the context-related data, and the thought chain guide words. The thought chain guide words are used to instruct the integrated model to generate the reasoning process based on thought chain technology. The structured prompts are input into the integration model so that the integration model generates the integration architecture information.

6. The method for integrating multiple electronic channel systems according to claim 1, characterized in that, The integrated architecture information includes one or more of the following: a list of business functions, a cross-system function similarity mapping, a system dependency graph, a standardized data model after integration, and a draft API interface design.

7. The method for integrating multiple electronic channel systems according to claim 1, characterized in that, The step of generating target code and data migration scripts for the target framework based on the integration architecture information using the integration model includes: Source system data is generated based on the multimodal data of the electronic channel system to be integrated. The source system data includes source framework code and source application programming interface. Based on the integrated architecture information, target system data and constraints are generated. The target system data includes target framework code and target application programming interface. The constraints include the target framework's programming specifications, coding conventions, and performance requirements. Using retrieval enhancement generation technology, transformation case data related to the source system data and the target system data are retrieved from the vector knowledge base to obtain reference context information; Based on the source system data, the target system data, the constraints, the reference context information, and the thought chain guidance words, code generation prompt words are generated; The code generation prompts are input into the integrated model, enabling the integrated model to perform distributed reasoning based on the thought chain technology to obtain the target code and the data migration script.

8. The method for integrating multiple electronic channel systems according to claim 1, characterized in that, The step of constructing an integration system based on the target code and the data migration script, and performing integration verification on the integration system using test cases, includes: The target code is compiled and deployed, and the data migration script is executed to build the integrated system; Control flow and data flow are generated based on the target code, and test cases are generated based on the control flow and data flow using the integrated model; The test cases are run in the integrated system to obtain integration test results, and the system integration results are generated based on the integration test results.

9. The multi-electronic channel system integration method according to claim 8, characterized in that, Also includes: The test cases are executed in the electronic channel system to be integrated to obtain the source test results; A difference analysis was performed on the source test results and the integrated test results to obtain difference data; The integrated model is optimized and updated based on the difference data to optimize the model parameters of the integrated model.

10. A multi-electronic channel system integration apparatus based on a large model, used to execute any one of the multi-electronic channel system integration methods based on a large model as described in claims 1-9, characterized in that, include: The knowledge base construction module is configured to: acquire multimodal data from multiple electronic channel systems to be integrated, and construct a vector knowledge base based on the multimodal data; The integrated architecture module is configured to: based on retrieval enhancement generation technology and thinking chain technology, use the integrated model to perform business function identification and architecture analysis on the vector knowledge base, and generate integrated architecture information. The large model is a deep convolutional network that has been domain adaptively fine-tuned. The code generation module is configured to: use the integration model to generate target code and data migration scripts for the target framework based on the integration architecture information; The result verification module is configured to: construct an integrated system based on the target code and the data migration script, and use test cases to verify the integrated system to obtain the system integration result.

Images