A knowledge graph-based software architecture evolution recommendation system and an implementation method thereof

Abstract

The application discloses a software architecture evolution recommendation system based on a knowledge graph and an implementation method thereof, and the system comprises a bottom hardware resource support layer which is used for supporting the persistent storage of software asset data and the high-performance calculation of graph topology through physical computing resources and high-speed communication link support, and the bottom hardware resource support layer comprises a high-density computing cluster, a distributed storage array and a high-speed network switching matrix. All the technical features of the application jointly constitute an organic whole, and aim to solve the software architecture evolution problem under the driving of complex education business.

Description

Technical Field

[0001] This invention relates to the fields of software engineering and artificial intelligence, specifically to a knowledge graph-based software architecture evolution recommendation system and its implementation method. Background Technology

[0002] With the deep integration of information technology and education, educational software systems have evolved from simple functional teaching aids into comprehensive digital ecosystem platforms that support massive amounts of teaching resources, complex teaching logic, and diverse interactive scenarios. In this evolution, software architecture, as the core foundation supporting system stability, scalability, and long-term maintainability, directly determines the lifecycle of educational information services and the user experience through its rational design and flexible evolution.

[0003] In existing software development and operations systems, the evolution of software architecture primarily relies on manual decisions by experienced architects or the use of semi-automated tools based on preset rules and design patterns. Specifically, when business requirements change or system performance faces bottlenecks, technical teams typically use static code analysis to analyze system dependencies and combine this with refactoring tools to decompose, merge, or redefine specific functional modules. These traditional methods are highly standardized and operational when dealing with general software engineering tasks where business logic is relatively stable and requirement changes are infrequent. They effectively solve initial system construction challenges and provide basic structural guidance for subsequent software maintenance.

[0004] However, as the digital transformation of the education industry enters a more complex phase, particularly with the deep integration of complex business logics such as personalized learning path recommendations, multi-dimensional knowledge graph navigation, and intelligent teaching evaluation, some inherent characteristics of the aforementioned technical solutions at the principle level are gradually revealing insurmountable limitations when facing high-frequency and in-depth architectural evolution requirements. At its root, the architectural evolution of educational software systems is not simply code refactoring; it is essentially a dynamic reshaping of the mapping relationship between the underlying complex teaching business domain knowledge and the upper-level technical implementation architecture. Existing rule-based or experience-based evolutionary models often limit their core logic to the physical connections at the technical component level, lacking the ability to capture the deep-seated semantic connections of the business, leading to a non-obvious deep-seated technical contradiction: the contradiction between the strong correlation of business logic and the fragmented distribution of the technical architecture.

[0005] Specifically, educational business knowledge itself possesses strong hierarchical, interconnected, and time-sensitive characteristics. When the curriculum is adjusted, the subject knowledge system is updated, or the assessment model changes, these business-level changes often radiate outwards, affecting multiple heterogeneous modules in the software system, such as the data persistence layer, business logic layer, and front-end interaction layer. Furthermore, due to the lack of a unified semantic model to describe business knowledge and architectural entities, architects often struggle to accurately trace and locate all logical nodes affected by business changes within the vast codebase and complex microservice chains when making evolutionary decisions. This semantic deficiency caused by "information silos" makes the architectural evolution process highly susceptible to "semantic drift," meaning that although the call relationships at the code level are syntactically correct, the business execution logic deviates from its original intent. It may even lead to a break in the global business loop due to adjustments in local modules, resulting in unpredictable architectural collapse risks.

[0006] Furthermore, when dealing with unstructured or semi-structured evolutionary requirements, existing technologies often fail to provide effective knowledge reasoning mechanisms. Their recommended solutions tend to be limited to simple reuse of historical code snippets or pattern matching, making it difficult to evaluate the optimality of the evolutionary path from a global perspective. This is especially true in educational software scenarios with extremely high business continuity requirements. Balancing the "cost of architectural adjustments" with the "match to business needs" has become an insurmountable technical bottleneck for traditional evolutionary solutions. Existing tools often only provide fragmented modification suggestions, failing to offer forward-looking, logically complete, and deeply aligned architecture recommendations that reflect the overall system and the evolving business trends.

[0007] Correspondingly, to maintain the robust operation of educational software systems in dynamic environments, an innovative mechanism is urgently needed that can penetrate the barriers between technical implementation and business semantics to achieve knowledge-driven architectural evolution. This mechanism not only requires multi-dimensional modeling of software assets, development specifications, and domain knowledge, but also necessitates in-depth mining of their inherent implicit connections to achieve intelligent and semantic evaluation of architectural evolution paths. Therefore, how to construct a knowledge graph-based deep reasoning capability to address key challenges such as low architectural evolution efficiency, difficulty in ensuring semantic consistency, and inaccurate evolution path decisions in complex educational scenarios has become a profound issue that urgently needs to be addressed in the fields of software engineering and educational informatization. Summary of the Invention

[0008] To address the technical problems mentioned in the background art, such as the disconnect between business logic and technical architecture, difficulty in ensuring semantic consistency, and lack of global optimality in evolutionary decisions during the architectural evolution of educational software systems, this invention provides a knowledge graph-based software architecture evolution recommendation system and its implementation method.

[0009] This invention provides a knowledge graph-based software architecture evolution recommendation system, the system comprising:

[0010] The underlying hardware resource support layer is used to support the persistent storage of software asset data and high-performance computing of map topology through physical computing resources and high-speed communication links. The underlying hardware resource support layer includes a high-density computing cluster, a distributed storage array, and a high-speed network switching matrix.

[0011] The core data governance layer is logically connected to the underlying hardware resource support layer and is used to extract and integrate multi-dimensional software feature data from heterogeneous data sources. The core data governance layer includes a software asset extraction unit, a business semantic capture unit, and an operation status monitoring unit.

[0012] The knowledge graph construction layer is used to build a unified semantic model covering business logic and technical architecture based on the multi-source heterogeneous data provided by the core data governance layer. The knowledge graph construction layer includes an entity alignment module, a relation extraction module, and an attribute fusion module.

[0013] The evolutionary reasoning decision layer is used to identify evolutionary causes and search paths based on the graph topology generated by the knowledge graph construction layer. The evolutionary reasoning decision layer includes an evolutionary cause identification engine, an influence range propagation model, and an evolutionary path searcher.

[0014] The application interface layer is used to transform the recommended solutions generated by the evolutionary reasoning decision layer into concrete engineering guidelines and to evaluate the evolutionary effect. The application interface layer includes an architecture visualization dashboard, an automatic refactoring script generator, and an evolutionary effect prediction and evaluation tool.

[0015] Furthermore, the software asset extraction unit is configured with a static scanning mechanism. The static scanning mechanism uses an abstract syntax tree parser to access the source code repository, version control system, and container orchestration configuration file of the target system to extract program entities and their explicit dependencies, including class definitions, interface declarations, method calls, package structures, and service discovery metadata in the configuration file.

[0016] The business semantic capture unit connects to the knowledge base and demand tracking matrix of the education business domain, and identifies business domain entities, including subject knowledge points, course models, assessment rules, user permission levels and learning path logic, through semantic annotation technology.

[0017] The operational status monitoring unit collects call chain logs, resource consumption sequences, and exception stack information during system operation through probe components deployed in the production environment, generating a real-time tracking data stream characterizing the dynamic behavior of the system.

[0018] Furthermore, the entity alignment module employs a two-layer filtering mechanism to perform cross-domain association between technical architecture entities and business semantic entities: the first layer is fast matching based on character features to filter low-frequency entities; the second layer is spatial projection matching based on semantic vectors, which eliminates recognition ambiguity caused by namespace differences by calculating the distance between technical entity vectors and business entity vectors in a unified semantic space.

[0019] The relation extraction module constructs multi-dimensional semantic edges based on syntactic dependency analysis and causal logic reasoning, including implementation relations, supporting relations, dependency relations, composition relations and constraint relations. It also performs dynamic enhancements in conjunction with the dynamic call chain provided by the running status monitoring unit. If the frequency of microservice calls triggered by business logic exceeds a preset ratio, the corresponding edge weights are enhanced.

[0020] The attribute fusion module associates static structural information with dynamic operating indicators, assigning nodes in the knowledge graph with multi-dimensional attribute vectors including access frequency, change popularity, coupling degree, and fan-in and fan-out coefficients.

[0021] Furthermore, the evolutionary cause identification engine compares the state changes and performance indicators of business semantic entities in real time. When it detects business logic adjustments or performance indicators falling below a preset threshold, it extracts the evolutionary target feature vector and activates the evolutionary analysis process.

[0022] The impact propagation model utilizes the topological connectivity of the knowledge graph to perform a bidirectional topological search starting from the evolutionary cause node, based on semantic weights, to locate the directly affected nodes and indirectly coupled nodes, outputting a conflict assessment report containing potential change points, and determining the set of technical components affected by the business change based on the coupling strength and propagation probability between nodes. The set of technical components includes API interfaces to be modified, business logic blocks to be rewritten, and database table structures to be adjusted.

[0023] The evolution path searcher performs template matching and combination optimization based on a preset evolution pattern library, and generates an architecture refactoring sequence under multi-dimensional constraints, including business continuity requirements, code change volume, and response speed improvement.

[0024] Furthermore, the architecture visualization dashboard presents the knowledge graph in the form of multi-layered nested charts, and marks the risk level of nodes in a color-coded manner according to the analysis results of the evolutionary reasoning decision layer, where high-risk change points are marked in red, medium-risk points are marked in yellow, and non-impact points are marked in green.

[0025] The automatic refactoring script generator automatically generates automated build and deployment scripts adapted to the continuous integration environment based on the evolution path determined by the evolution path searcher, and each operation step in the script is associated with corresponding rollback logic.

[0026] The evolution effect prediction and evaluation tool constructs a simulated sandbox environment and uses trajectory simulation technology to replay historical service load traffic, thereby quantitatively predicting the improvement of the recommended solution in terms of stability, scalability, and maintenance costs.

[0027] An implementation method for a knowledge graph-based software architecture evolution recommendation system includes the following steps:

[0028] The first step is to initiate a multi-source data acquisition process. The software asset extraction unit traverses the code repository to generate intermediate representation documents, the business semantic capture unit synchronously pulls the business specification and business logic view, and the operation status monitoring unit records the system's operation trajectory within the preset teaching cycle.

[0029] The second step is to construct a knowledge graph of software architecture evolution. Through the knowledge graph construction layer, entity disambiguation, logical connection establishment and attribute injection are performed on the collected raw data to form a knowledge base with both business and technical views, using a graph database as the storage medium.

[0030] The third step is to perform evolution trigger detection. The evolution trigger identification engine compares the logical changes in the business view with the abnormal fluctuations in the operation view. If the teaching business is restructured or the interface response delay increases beyond the preset ratio, the evolution requirement is determined and the evolution target feature vector is extracted.

[0031] The fourth step is to conduct an in-depth analysis of the impact of the architecture. By using the impact propagation model, a multi-hop neighborhood search is performed on the knowledge graph to identify the set of technical components affected by the business changes and the regression risk points. A consistency check is then performed to verify whether there are semantic conflicts between the code segments to be changed and the related business logic.

[0032] The fifth step is to generate a recommended architecture evolution scheme. The evolution path searcher selects applicable templates from the evolution pattern library, constructs multi-objective optimization logic, and generates an architecture evolution operation sequence that includes code modification suggestions, component migration steps and interface adaptation schemes under the constraints of meeting business continuity and reducing code change volume.

[0033] The sixth step is to simulate, verify, and output the evolution path. The evolution effect prediction evaluator runs the evolution operation sequence in the virtual environment and simulates the system's behavior after evolution. If the technical indicators meet expectations, a visual report and automated script are output through the application interface layer.

[0034] Furthermore, in the first step, the multi-source data acquisition process adopts an incremental update strategy, which uses hash fingerprint comparison technology to identify code commit records and business documents that have changed since the last synchronization, and performs local re-parsing only on the changed parts;

[0035] In the second step, the knowledge graph is dynamically expanded through an automatic schema evolution mechanism. When a new technology stack or business paradigm is detected, the new entity type is automatically identified and the ontology definition of the knowledge graph is updated without destroying the existing topology.

[0036] Furthermore, in the fourth step, the consistency check is specifically performed by: searching the knowledge graph to confirm whether the technical components to be deleted or modified are depended on by other existing business logic. If a dependency exists, a mandatory warning is issued in the risk report; and when performing multi-hop neighborhood search, the influence weight of modules involving preset sensitive data is automatically increased.

[0037] Furthermore, in the fifth step, the generated evolutionary operation sequence is atomic, and each sub-step includes pre-checking, execution of changes, and post-verification. When executing multi-objective optimization logic, the evolutionary path seeker comprehensively evaluates evolutionary cost, structural improvement degree, and business fit.

[0038] The evolution costs include the number of lines of code modified, the number of database schemas adjusted, and the estimated downtime.

[0039] The degree of structural improvement includes the reduction in system coupling and the increase in module cohesion.

[0040] The business fit represents how intuitively the evolved architecture maps to the current teaching business logic.

[0041] Furthermore, following the sixth step, a closed-loop feedback learning step is also included:

[0042] The system performance after implementing the evolution plan is continuously tracked by the operation status monitoring unit, and the actual operation feedback is re-injected into the knowledge graph; if the actual evolution effect is better than the predicted value, the confidence weight of the corresponding evolution mode in the current business scenario is increased.

[0043] The system also proactively recommends execution time windows for architecture evolution to users based on the system's stable operation period and peak and off-peak business seasons, combined with real-time monitoring data in the knowledge graph, in order to avoid peak business periods.

[0044] Beneficial Effects: All the technical features of this invention form an organic whole, aiming to solve the challenges of software architecture evolution driven by complex educational business. The setting of each sub-module and the execution of each step are all aimed at achieving the core goal of improving evolution quality, reducing evolution risks, and enhancing business responsiveness. Through the implementation of this invention, developers and operators of educational software systems can gain unprecedented architectural transparency and decision-making accuracy, thereby maintaining a leading position in technical architecture amidst the wave of digital transformation. Attached Figure Description

[0045] Figure 1 This is a schematic diagram of the overall architecture of the system of the present invention.

[0046] The attached diagram is labeled as follows: 1. Underlying hardware resource support layer; 11. High-density computing cluster; 12. Distributed storage array; 13. High-speed network switching matrix; 2. Core data governance layer; 21. Software asset extraction unit; 22. Business semantic capture unit; 23. Operation status monitoring unit; 3. Knowledge graph construction layer; 31. Entity alignment module; 32. Relationship extraction module; 33. Attribute fusion module; 4. Evolutionary reasoning decision layer; 41. Evolutionary cause identification engine; 42. Influence range propagation model; 43. Evolutionary path searcher; 5. Application interface layer; 51. Architecture visualization dashboard; 52. Automatic refactoring script generator; 53. Evolutionary effect prediction and evaluation device. Detailed Implementation

[0047] 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, and 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.

[0048] Please see Figure 1 This invention provides a technical solution: a knowledge graph-based software architecture evolution recommendation system and its implementation method. By deeply integrating software engineering and knowledge engineering technologies, it offers a full lifecycle architecture evolution support solution for complex software systems, especially educational software systems with frequently changing requirements. The system constructs a rigorous logical closed loop from the underlying hardware infrastructure to the top-level application interaction interface.

[0049] The underlying hardware resource support layer 1, serving as the physical foundation of the entire system, directly determines the efficiency and real-time performance of large-scale knowledge graph processing. The high-density computing cluster 11 consists of multiple computing nodes with multi-core, high-frequency characteristics. Each node is equipped with a high-performance central processing unit (CPU) with redundant design, capable of parallel processing of large-scale source code static analysis tasks. To address the intensive random memory access requirements in graph computation, each node is configured with at least 2TB of synchronous dynamic random access memory (DRAM), and error checking and correction techniques are employed to ensure data integrity. The distributed storage array 12 is built on non-volatile, high-speed storage media, utilizing redundant disk array technology to ensure the persistent security of heterogeneous software asset data and evolutionary history. The storage system supports multi-tenant concurrent access, and its actual throughput can meet the needs of rapid retrieval of massive amounts of fine-grained software entities.

[0050] The core data governance layer 2 is responsible for extracting, cleaning, and transforming key information from heterogeneous data sources. The software asset extraction unit 21 integrates deep syntax analyzers for multiple mainstream programming languages, capable of penetrating the surface structure of source code to construct accurate abstract syntax trees. During its operation, this unit not only identifies basic class, interface, and method definitions but also deeply analyzes project configuration files, such as dependency declarations in Maven or Gradle, and deployment manifests under microservice architectures. Through natural language parsing of code comments and naming conventions, the extraction unit can identify the technical intent hidden at the code level and transform it into a structured intermediate representation. The business semantic capture unit 22 focuses on extracting business logic from non-technical documents. By connecting to the teaching syllabus management system, this unit can identify business entities unique to the education field, such as course trees, knowledge point graphs, and evaluation rubrics. Using a pre-defined ontology model, the business semantic capture unit maps descriptive language in documents to standardized business nodes, establishing logical relationships between business domain entities. The runtime monitoring unit 23, acting as a dynamic data collector, deploys non-intrusive probes along various critical paths of the target system.

[0051] The knowledge graph construction layer 3 forms a multi-dimensional, interconnected knowledge base through deep semantic fusion of the governed data. The entity alignment module 31 performs identity verification on entities extracted from different levels. In the education system, the same business concept may have multiple expressions in requirement documents, database table names, and service interface names. This module uses character-level text similarity analysis and contextual semantic cosine similarity calculation, combined with a domain dictionary, to point these heterogeneous expressions to the same globally unique knowledge graph node. The relation extraction module 32, based on the aligned entities, uses causal reasoning and path analysis techniques to establish complex semantic edges. For example, by analyzing injection relationships and calling logic in the code, "implementation relationships" and "call relationships" are established; by analyzing business process diagrams, "support relationships" are established. This multi-dimensional edge structure allows the graph to simultaneously express the system's physical topology and logical intent. The attribute fusion module 33 then mounts the performance indicators collected by the running status monitoring unit 23 onto the nodes in the form of attribute vectors.

[0052] The Evolutionary Reasoning Decision Layer 4 is the core of intelligent recommendation. The Evolutionary Trigger Identification Engine 41 has a multi-dimensional triggering mechanism. When the business side proposes new adjustments to the teaching model, or when the technical side monitors a continuous decline in the throughput of a key link approaching a preset warning level, the engine automatically extracts a snapshot of the current state and compares it with the benchmark model in the knowledge graph. The Influence Scope Propagation Model 42, based on the graph's topology, simulates the diffusion process of changes in the architecture. It starts from the node that triggered the change and performs a multi-step, jump-like search along the dependency edges. During the search, the model calculates the attenuation coefficient of influence based on the edge weights (such as call frequency and coupling strength), thereby determining a set of nodes containing the probability of being affected. The Evolutionary Path Searcher 43 searches for the optimal refactoring solution within a search space composed of preset patterns. It evaluates various evolutionary strategies, such as service merging, interface degradation, and database read / write separation.

[0053] Application Interface Layer 5 transforms complex reasoning processes into intuitive engineering guidance. The Architecture Visualization Dashboard 51 utilizes multi-level nested graph technology, allowing architects to drill down from a macro-level business panorama to a micro-level class call graph, displaying current architectural risks in the form of a heatmap. The Automatic Refactoring Script Generator 52, through deep integration with the continuous integration / continuous deployment toolchain, can directly convert recommended evolutionary operations into automated pipeline scripts. The Evolution Effect Prediction and Evaluation Tool 53, in an independent sandbox environment, uses historical traffic replay technology to pre-simulate the performance of the evolved system, providing quantitative confidence indicators for the final evolutionary decision.

[0054] In one specific embodiment of the present invention, a method for implementing a software architecture evolution recommendation system based on a knowledge graph is executed according to the following steps: First, the system initiates a large-scale, multi-source data synchronization process. The software asset extraction unit traverses all associated code repositories to generate intermediate syntax documents containing millions of nodes. The business semantic capture unit synchronously parses the logical view in the teaching rule engine. The runtime monitoring unit enables traffic mirroring in the production environment and collects runtime data for at least one complete teaching cycle (e.g., one week). Second, a unified evolutionary knowledge foundation is constructed. The knowledge graph construction layer deduplicates and merges the above data to form a complex network covering four layers: business, logic, code, and resources. Third, continuous trigger monitoring is performed. The evolution trigger identification engine compares the offset between the current system representation vector and the target business objective in real time. Fourth, the evolutionary boundary is determined through an influence scope propagation model. This model identifies the core modules directly related to the change requirements and all their upstream dependencies by traversing the knowledge graph. Fifth, a recommendation sequence is generated by the evolutionary path searcher. This sequence not only includes the design blueprint of the target architecture but also details the decomposition steps of each microservice. The sixth step involves the evolution effect prediction and evaluation engine simulating the recommended scheme, outputting an evaluation report containing the expected improvement of various technical indicators, and delivering it through the application interface layer.

[0055] To further demonstrate the superiority of the present invention, this embodiment provides an evolution case for a large-scale smart education platform. The original course selection system of this platform adopted a monolithic architecture, which frequently experienced database response timeouts and system crashes when faced with high-concurrency course selection demands across the entire school.

[0056] In a further preferred embodiment, the software asset extraction unit 21 employs an incremental scanning mechanism when processing source code. This mechanism compares the hash values ​​of commit records in the version control system and only re-parses files that have been changed. This partial update strategy significantly reduces the load on the underlying hardware resource support layer 1. Simultaneously, the extraction unit can identify common "architectural smells" in educational software, such as overly large monolithic service classes. These identification results are injected as attributes into knowledge graph nodes, providing direct reconstruction clues for the evolutionary cause identification engine 41.

[0057] When extracting business rules, the business semantic capture unit 22 incorporates a feedback mechanism from domain experts. Experts can calibrate and weight the identified business entities and their relationships through the application interface layer 5. This "human-machine collaboration" model ensures that the generated knowledge graph possesses both automated efficiency and expert industry insights. In educational scenarios, this is crucial for handling complex administrative logic such as "credit recognition agreements."

[0058] The entity alignment module 31 employs a hierarchical filtering strategy when performing alignment operations. First, at the character representation layer, a fast Bloom filter eliminates absolutely irrelevant entity pairs. Then, at the semantic vector layer, the entity description text is mapped to a high-dimensional continuous vector space, and the semantic similarity of entities is determined by calculating the Euclidean distance between vectors. This strategy effectively solves the computational explosion problem in large-scale entity matching while ensuring alignment accuracy.

[0059] The relation extraction module 32 pays particular attention to cross-language call chains when constructing "supporting relationships." In a heterogeneous microservice environment, the front-end may be written in TypeScript, while the back-end may be in Java or Go. This module successfully penetrates language barriers and establishes an end-to-end business support chain by parsing interface definition files and message queue subscription relationships. This end-to-end perspective is crucial for assessing the impact of front-end page changes on the back-end database load.

[0060] The attribute fusion module 33 also integrates a time-series prediction component. It not only records current performance metrics but also analyzes the historical evolution trends of these metrics. Through smoothing data such as access volume and response time, the module can predict the future technical degradation risks of certain nodes. For example, if the hit rate of a certain cache service shows a slow downward trend, the attribute fusion module will inject this risk signal into the graph, triggering the evolutionary cause identification engine 41 to intervene in advance and recommend cache optimization or expansion solutions.

[0061] The evolution trigger identification engine 41 supports custom threshold alarm rules. For online assessment systems that are extremely sensitive to latency, stricter response time thresholds can be set; while for backend academic scheduling systems, the restrictions can be relaxed. This flexibility allows the system to adapt to the SLA requirements of different business scenarios. In addition, the engine can also identify "logical redundancy" triggers caused by the merging of business rules, prompting architects to streamline the code.

[0062] The impact propagation model 42 incorporates a conflict sensitivity factor when performing multi-hop searches. For modules containing sensitive student data, the model automatically increases their impact weight. This means that any minor changes affecting such modules will be prominently highlighted in the impact assessment report, and a more intensive security audit will be recommended. This design reflects a high level of emphasis on the security of educational data.

[0063] When generating recommended solutions, the evolution path searcher 43 automatically matches mature technology templates from a pre-stored evolution pattern library. These templates cover classic distributed architecture governance solutions such as the "circuit breaker pattern" and the "sidecar pattern." The searcher automatically adapts the specific parameters in the templates based on the target system's technology stack environment.

[0064] The scripts generated by the automatic refactoring script generator 52 have complete atomicity guarantees. Each refactoring step includes three sub-stages: "pre-check," "implement changes," and "post-verification." If the system state is found to be inconsistent with expectations at any stage, the script will automatically trigger a rollback process to restore the system to the backup point before the changes.

[0065] The evolution effect prediction evaluator 53 utilizes a generated digital twin model for simulation. It simulates not only normal business loads but also various abnormal conditions, such as network partitions and single points of failure. By observing the system's fault tolerance and recovery time under extreme conditions, the evaluator provides a scientific basis for demonstrating the robustness of the evolution scheme. This simulation-based prediction transforms traditional "trial-and-error" evolution into "a priori" evolution.

[0066] Furthermore, in a preferred implementation, the system supports continuous learning of evolutionary experience. Whenever an evolutionary path is adopted and successfully implemented, the system automatically compares the actual performance data before and after the evolution. If the evolutionary effect is better than the predicted value, the system increases the confidence weight of the evolutionary pattern in the current business scenario; conversely, it triggers internal debriefing analysis logic to optimize the recommendation algorithm of the evolutionary path searcher 43. This closed-loop mechanism ensures that the system becomes increasingly "intelligent" with increased usage time, accurately capturing the architectural evolution patterns in specific business domains.

[0067] In terms of system deployment and implementation, the underlying hardware resource support layer 1 adopts containerized management. All computing and storage resources are dynamically scheduled through a cloud-native orchestration engine. When the evolutionary inference decision layer 4 needs to perform large-scale parallel search, the system can instantly launch hundreds or thousands of temporary computing nodes to support the computing power demand and quickly release resources after the task is completed. This on-demand allocation model greatly improves the system's economy and scalability.

[0068] The software architecture evolution recommendation system and its implementation method based on knowledge graphs described in this invention are not only a set of technical tools, but also a scientific methodology for architecture governance. By transforming fragmented code, business logic, and operational data into structured knowledge graphs, it achieves comprehensive digital monitoring and intelligent guidance of the software architecture.

[0069] As can be seen from the detailed description of the above embodiments, this invention effectively solves the core technical pain points of existing technologies, such as difficult architectural evolution decisions, high risks, and disconnect from business logic, by constructing a complete system framework that includes underlying support, data governance, knowledge graph construction, reasoning and decision-making, and application interfaces. Its core lies in utilizing the powerful semantic link of knowledge graphs to penetrate the barriers between technical implementation and business logic, transforming software evolution from a non-standard behavior relying on personal experience into a data-driven, logically rigorous, and automated engineering process. This not only enhances the vitality of the software system itself but also provides a successful paradigm for intelligent governance in related fields.

Claims

1. A software architecture evolution recommendation system based on knowledge graphs, characterized in that, The system includes: The underlying hardware resource support layer (1) is used to support the persistent storage of software asset data and high-performance computing of map topology through physical computing resources and high-speed communication links. The underlying hardware resource support layer (1) includes a high-density computing cluster (11), a distributed storage array (12), and a high-speed network switching matrix (13). The core data governance layer (2) is logically connected to the underlying hardware resource support layer (1) and is used to extract and integrate multi-dimensional software feature data from heterogeneous data sources. The core data governance layer (2) includes a software asset extraction unit (21), a business semantic capture unit (22), and a running status monitoring unit (23). The knowledge graph construction layer (3) is used to construct a unified semantic model covering business logic and technical architecture based on the multi-source heterogeneous data provided by the core data governance layer (2). The knowledge graph construction layer (3) includes an entity alignment module (31), a relation extraction module (32), and an attribute fusion module (33). Evolutionary reasoning decision layer (4) is used to identify evolutionary causes and search paths based on the graph topology generated by the knowledge graph construction layer (3). The evolutionary reasoning decision layer (4) includes an evolutionary cause identification engine (41), an influence range propagation model (42), and an evolutionary path searcher (43). The application interface layer (5) is used to transform the recommended scheme generated by the evolutionary reasoning decision layer (4) into a concrete engineering guide and to evaluate the evolutionary effect. The application interface layer (5) includes an architecture visualization dashboard (51), an automatic refactoring script generator (52), and an evolutionary effect prediction and evaluation device (53).

2. The knowledge graph-based software architecture evolution recommendation system according to claim 1, characterized in that, The software asset extraction unit (21) is equipped with a static scanning mechanism. The static scanning mechanism uses an abstract syntax tree parser to access the source code library, version control system and container orchestration configuration file of the target system to extract program entities and their explicit dependencies, including class definitions, interface declarations, method calls, package structures and service discovery metadata in the configuration file. The business semantic capture unit (22) connects to the knowledge base and demand tracking matrix of the education business field, and identifies business field entities including subject knowledge points, course models, assessment rules, user permission levels and learning path logic through semantic annotation technology. The operation status monitoring unit (23) collects call chain logs, resource consumption sequences and exception stack information during system operation through probe components deployed in the production environment, and generates a real-time tracking data stream characterizing the dynamic behavior of the system.

3. The software architecture evolution recommendation system based on knowledge graphs according to claim 1, characterized in that, The entity alignment module (31) uses a two-layer filtering mechanism to perform cross-domain association between technical architecture entities and business semantic entities: the first layer is fast matching based on character features, which is used to filter low-frequency entities; the second layer is spatial projection matching based on semantic vectors, which eliminates the recognition ambiguity caused by namespace differences by calculating the distance between technical entity vectors and business entity vectors in a unified semantic space. The relation extraction module (32) constructs a multi-dimensional semantic edge, including implementation relation, support relation, dependency relation, composition relation and constraint relation, based on syntactic dependency analysis and causal logic reasoning. It also performs dynamic enhancement in conjunction with the dynamic call chain provided by the running status monitoring unit (23). If the frequency of microservice calls triggered by business logic exceeds the preset ratio, the corresponding edge weight is enhanced. The attribute fusion module (33) associates static structural information with dynamic operation indicators, and assigns a multi-dimensional attribute vector to the nodes in the knowledge graph, including access frequency, change popularity, coupling degree and fan-in and fan-out coefficients.

4. The software architecture evolution recommendation system based on knowledge graphs according to claim 1, characterized in that, The evolution trigger identification engine (41) compares the state changes and performance indicators of business semantic entities in real time. When it detects business logic adjustment or performance indicators falling below a preset threshold, it extracts the evolution target feature vector and activates the evolution analysis process. The influence propagation model (42) utilizes the topological connectivity of the knowledge graph to start from the evolution cause node, perform bidirectional topological search according to semantic weight, locate the directly affected nodes and indirectly coupled nodes, output a conflict assessment report containing potential change points, and determine the set of technical components affected by business changes based on the coupling strength and propagation probability between nodes. The set of technical components includes the API interface to be modified, the business logic block to be rewritten, and the database table structure to be adjusted. The evolution path searcher (43) performs template matching and combination optimization based on a preset evolution pattern library, and generates an architecture refactoring sequence under multi-dimensional constraints, including business continuity requirements, code change volume and response speed improvement.

5. A software architecture evolution recommendation system based on knowledge graphs according to claim 1, characterized in that, The architecture visualization dashboard (51) presents the knowledge graph in the form of multi-layer nested charts, and marks the risk level of nodes in a color-coded manner according to the analysis results of the evolutionary reasoning decision layer (4), where high-risk change points are marked in red, medium-risk points are marked in yellow, and non-affected points are marked in green. The automatic refactoring script generator (52) automatically generates an automated build and deployment script adapted to the continuous integration environment based on the evolution path determined by the evolution path searcher (43), and each operation step in the script is associated with corresponding rollback logic. The evolution effect prediction and evaluation tool (53) constructs a simulated sandbox environment and uses trajectory simulation technology to replay historical service load traffic to quantitatively predict the improvement of the recommended solution in terms of stability, scalability and maintenance cost.

6. A method for implementing a software architecture evolution recommendation system based on knowledge graphs, characterized in that, Apply the system described in any one of claims 1-5, Includes the following steps: The first step is to start the multi-source data acquisition process, generate intermediate representation documents by traversing the code repository through the software asset extraction unit (21), pull the business specification and business logic view synchronously through the business semantic capture unit (22), and record the system's running trajectory within the preset teaching cycle through the running status monitoring unit (23). The second step is to construct a knowledge graph of software architecture evolution. Through the knowledge graph construction layer (3), entity disambiguation, logical connection establishment and attribute injection are performed on the collected raw data to form a knowledge base with a dual view of business and technology, using a graph database as the storage medium. The third step is to perform evolution trigger detection. The evolution trigger recognition engine (41) compares the logical changes in the business view with the abnormal fluctuations in the operation view. If the teaching business reconstruction or the interface response delay increases beyond the preset ratio, the evolution requirement is triggered and the evolution target feature vector is extracted. The fourth step is to conduct an in-depth analysis of the impact of the architecture. By using the impact propagation model (42), a multi-hop neighborhood search is performed on the knowledge graph to identify the set of technical components affected by the business changes and the regression risk points. A consistency check is then performed to verify whether there is a semantic conflict between the code segment to be changed and the related business logic. The fifth step is to generate a recommended architecture evolution scheme. The evolution path searcher (43) selects applicable templates from the evolution pattern library, constructs multi-objective optimization logic, and generates an architecture evolution operation sequence that includes code modification suggestions, component migration steps and interface adaptation schemes under the constraints of business continuity and reducing code change volume. The sixth step is to simulate and verify the evolution path and output it. The evolution effect prediction evaluator (53) runs the evolution operation sequence in the virtual environment and simulates the running behavior of the system after evolution. If the technical indicators meet the expectations, the visualization report and automated script are output through the application interface layer (5).

7. The implementation method of a software architecture evolution recommendation system based on knowledge graphs according to claim 6, characterized in that, In the first step, the multi-source data acquisition process adopts an incremental update strategy, which uses hash fingerprint comparison technology to identify code commit records and business documents that have changed since the last synchronization, and performs local re-parsing only on the changed parts; In the second step, the knowledge graph is dynamically expanded through an automatic schema evolution mechanism. When a new technology stack or business paradigm is detected, the new entity type is automatically identified and the ontology definition of the knowledge graph is updated without destroying the existing topology.

8. The implementation method of a knowledge graph-based software architecture evolution recommendation system according to claim 6, characterized in that, In the fourth step, the consistency check is specifically performed by: searching the knowledge graph to confirm whether the technical components to be deleted or modified are depended on by other existing business logic. If a dependency exists, a mandatory warning is issued in the risk report; and when performing multi-hop neighborhood search, the influence weight of modules involving preset sensitive data is automatically increased.

9. The implementation method of a knowledge graph-based software architecture evolution recommendation system according to claim 6, characterized in that, In the fifth step, the generated evolutionary operation sequence is atomic, and each sub-step includes pre-checking, execution of changes, and post-verification. When executing multi-objective optimization logic, the evolutionary path seeker (43) comprehensively evaluates evolutionary cost, structural improvement degree, and business fit. The evolution costs include the number of lines of code modified, the number of database schemas adjusted, and the estimated downtime. The degree of structural improvement includes the reduction in system coupling and the increase in module cohesion. The business fit represents how intuitively the evolved architecture maps to the current teaching business logic.

10. The implementation method of a knowledge graph-based software architecture evolution recommendation system according to claim 6, characterized in that, Following the sixth step, a closed-loop feedback learning step is also included: The system performance after implementing the evolution scheme is continuously tracked by the running status monitoring unit (23), and the actual running feedback is re-injected into the knowledge graph; if the actual evolution effect is better than the predicted value, the confidence weight of the corresponding evolution mode in the current business scenario is increased. The system also proactively recommends execution time windows for architecture evolution to users based on the system's stable operation period and peak and off-peak business seasons, combined with real-time monitoring data in the knowledge graph, in order to avoid peak business periods.

Images