Software Knowledge Graph Construction Method and System for Low-Code Template Recommendation

Abstract

The present invention discloses a method and system for constructing a software knowledge graph for low-code template recommendation, belonging to the technical field of computer software. The method includes: designing a knowledge graph ontology model for low-code template recommendation; wherein, the entities included in the knowledge graph ontology model include: code-level entities and template-level entities bound to specific functions; for a given software system project to be reused, entity extraction is performed based on the knowledge graph ontology model to generate an original knowledge graph; the attribute information of the template-level entities in the original knowledge graph is supplemented based on the keyword information of the template-level entities to obtain the knowledge graph of the software system project. The present invention can strengthen the semantic matching between user requirements and template resources, thereby better realizing the task of low-code template recommendation.

Description

Technical Field

[0001] The present invention belongs to the technical field of computer software, and particularly relates to a method and system for constructing a software knowledge graph for low-code template recommendation. Background Art

[0002] The concept of "low-code" was first clearly proposed by Forrester in 2014, and generally refers to a paradigm that uses visual development tools, predefined components or templates to support software application development. By shielding a series of details of underlying development, low-code development platforms enable personnel without programming backgrounds to engage in software development, thereby improving software development efficiency and reducing development costs. These low-code development platforms can enable developers to carry out software development activities such as data modeling, page design, and business process orchestration based on visual interaction operations and panel property configurations through visual editing interfaces and predefined draggable low-code templates.

[0003] The low-code template resources provided by existing low-code development platforms are limited. When facing new scenarios and new requirements, it is often necessary to use high-code methods or follow the process specifications provided by the platform and rely on a large amount of manual intervention for customized development. The threshold and cost of this process are relatively high; at the same time, the low-code template resources provided by the platform often appear as fragmented independent units, with relatively scarce semantic information and associations. During the use of the platform, it is usually necessary to rely on manual screening to select appropriate templates, increasing the burden on users using the low-code platform.

[0004] In summary, a mechanism for mining and representing low-code template resources is needed to better support low-code template recommendation. Summary of the Invention

[0005] In order to overcome the problems of high customization development cost and scarce semantic information of low-code template resources in low-code development platforms, the present invention provides a method and system for constructing a software knowledge graph for low-code template recommendation, and a tool prototype is implemented based on this. Through the analysis of software systems in specific fields, it mines reusable low-code template resources to complete the construction of the knowledge graph, and supplements information for template entities in the knowledge graph based on the natural language understanding and generation capabilities of large models to strengthen the semantic matching between user requirements and template resources, so as to better implement the low-code template recommendation task.

[0006] To achieve the above object, the technical solution of the present invention includes the following content.

[0007] A method for constructing a software knowledge graph for low-code template recommendation, the method includes:

[0008] Design a knowledge graph ontology model for low-code template recommendation; among them, the entities included in the knowledge graph ontology model are: code-level entities and template-level entities bound to specific functions;

[0009] For a given software system project to be reused, entity extraction is performed based on the knowledge graph ontology model to generate an original knowledge graph;

[0010] Supplement the attribute information of the template-level entities in the original knowledge graph based on the keyword information of the template-level entities to obtain the knowledge graph of the software system project.

[0011] Furthermore, the code-level entities include:

[0012] Data entity classes;

[0013] And,

[0014] Page components and the configuration items and scripts associated with the page components;

[0015] And,

[0016] Classes, methods, and member variables related to backend Java code.

[0017] Furthermore, the template-level entities include: data object type entities, page type entities, and workflow type entities; among them,

[0018] The data object type entities are implemented through the code-level entities of the data entity classes;

[0019] The page type entities contain several components related to the front-end pages and the configuration items and scripts associated with the components;

[0020] The workflow type entities contain classes, methods, member variables related to backend Java code, and the static dependency relationships between the classes, methods, and member variables;

[0021] There are nesting relationships of different granularities between the page type entities, and at the same time, they are bound to specific data object type entities through configuration items or scripts, or associated with workflow type entities related to specific functions;

[0022] There is a call relationship between the workflow type entities, and they are associated with the data object type entities through the code-level entities of the data entity classes.

[0023] Furthermore, performing data object type entity extraction based on the knowledge graph ontology model includes:

[0024] Obtain the annotations in the.java files of the software system project, and the annotations include: @Data annotation and @TableName annotation;

[0025] Filter out the data entity classes in the annotations;

[0026] Complete the extraction of data objects by converting the member variables in the data entity classes into fields in the data table.

[0027] Furthermore, perform page type entity extraction based on the knowledge graph ontology model, including:

[0028] Parse the label keywords of the.vue files in the component directory of the software system project to extract the page component list of the software system project;

[0029] Traverse the.vue files in the page directory of the software system project, identify the script tags in the.vue files, and extract the script content associated with the page under the script tag content;

[0030] Based on the import statements in the script tag content, parse the calls of other page components by the page components corresponding to the script tags to form import dependencies between.vue files;

[0031] Under the template tag content, form the inclusion relationship between page components and pages based on the page component list and the import dependencies between.vue files, and establish the binding relationship between page type entities and data object type entities according to the data associated with the page components included in the page.

[0032] Furthermore, perform workflow type entity extraction based on the knowledge graph ontology model, including:

[0033] Identify the calls to backend services executed in the scripts included in the front-end pages;

[0034] Take the call as the entry point, and extract the corresponding service execution path from the.java files included in the software system project according to the call relationship of the Controller receiving requests, Service service calls, and Repository data interactions sent by the front-end requests;

[0035] Perform parameterization processing on the data transfer and third-party API calls on the service execution path to obtain workflow type entities;

[0036] Establish the association relationship between workflow type entities and data object type entities and page type entities.

[0037] Furthermore, supplement the attribute information of the template-level entities in the original knowledge graph based on the keyword information of the template-level entities, including:

[0038] Extract the keyword information of the template-level entity;

[0039] Use the keyword information as a prompt to input into the large model to generate the functional description of the template-level entity;

[0040] Supplement the functional description into the attribute information of the template-level entity.

[0041] Furthermore, extract the keyword information of the page type entity, including:

[0042] Obtain the functional keywords, where the functional keywords include: the attribute values in the functional label configuration item and the functional prompt words in the page;

[0043] Adopt the regular matching method to extract the functional keywords from the pages of the software system project to obtain the keyword information of the page type entity.

[0044] Furthermore, extract the keyword information of the workflow type entity, including:

[0045] Use the name of the code entity as the node and the static dependencies between the code entities as the relationships to construct a code dependency graph;

[0046] Apply the TextRank algorithm to extract the keywords in the code dependency graph to obtain the keyword information of the workflow type entity.

[0047] A software knowledge graph construction system for low-code template recommendation, the system includes:

[0048] The ontology model design module is used to design the knowledge graph ontology model for low-code template recommendation; among them, the entities included in the knowledge graph ontology model include: code-level entities and template-level entities bound to specific functions;

[0049] The knowledge graph generation module is used to perform entity extraction based on the knowledge graph ontology model for a given software system project to be reused to generate an original knowledge graph;

[0050] The attribute information supplement module is used to supplement the attribute information of the template-level entity in the original knowledge graph based on the keyword information of the template-level entity to obtain the knowledge graph of the software system project.

[0051] Compared with the prior art, the present invention has at least the following beneficial effects.

[0052] When facing the development requirements of new scenarios, the present invention can automatically extract reusable low-code template resources from existing software systems, thereby forming a knowledge graph for low-code template recommendation, covering activities such as data modeling, page layout, and process design in the low-code development process. At the same time, a method for generating function description information of low-code templates is provided, which can help users better understand and apply low-code development templates, and achieve more accurate low-code template recommendation for user business requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 is a flowchart of the present invention.

[0054] Figure 2 is an ontology model of the knowledge graph of the present invention.

[0055] Figure 3 is an example diagram of the knowledge graph of an embodiment.

[0056] Figure 4 is an example diagram of the function description of an embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0057] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below through specific embodiments in conjunction with the accompanying drawings.

[0058] Since the knowledge graph, as a graph-structured knowledge representation form, can effectively organize and manage reusable low-code templates of different types and granularities, store and represent complex association relationships between templates, and can effectively retrieve and recommend low-code templates in multiple scenarios, the present invention intends to implement a software knowledge graph construction method for low-code template recommendation. In the present invention, the ontology model is the conceptual layer representation of the knowledge graph, which is used to guide the construction process of the knowledge graph. Therefore, the present invention first designs a corresponding ontology model for low-code template recommendation; secondly, according to the concept definitions of the ontology model, a scalable plug-in framework is designed to complete the parsing of data to achieve knowledge extraction, forming a preliminary software knowledge graph; finally, for the template-level entities in the knowledge graph, the capabilities of large language models are used to generate supplementary function description information for them, so as to achieve more accurate low-code template recommendation for the given business development requirements of users.

[0059] Figure 1The process framework of the present invention is shown. The present invention takes a software system to be reused as input, which includes multi-source heterogeneous data such as code repositories, software documents, and interface components. Through ontology model design, these data are analyzed to define relevant entity types and relationship types between entities, and knowledge extraction is carried out based on the code static parsing method to complete the construction of a knowledge graph for low-code template recommendation. On this basis, through the entity information supplement method based on keyword extraction, with the help of the capabilities of large language models, function description information is generated for low-code template resources.

[0060] Ontology model design of the knowledge graph for low-code template recommendation.

[0061] Activities such as data modeling, page design, and process design are involved in the low-code development process. The knowledge graph ontology model designed by the present invention should include the above elements and define the association relationships between various resources according to the development scenario and function implementation. The entities in the ontology model are divided into two levels: code level and template level. As Figure 2 shown, it depicts the reusable resources in low-code development according to the different granularities of entities. Code-level entities correspond to code entities at the source code level and the dependency call relationships between entities. That is to say, code-level entities and relationships represent the static relationships of software resources, such as table structures related to data, components related to the front-end page and their associated configuration items, scripts, etc., classes, methods, member variables, etc. related to back-end Java code; Template-level entities include data objects, pages, and workflow types, and further extraction operations are performed on the basis of code-level entities, which are bound to specific low-code development requirement functions to form a knowledge graph subgraph related to specific functions as reusable low-code template resources.

[0062] For data modeling in low-code development, an entity of the "data object" type is defined in the knowledge graph ontology model. This type of entity is usually implemented by a class in back-end Java code and is mapped to the corresponding database table (such as MySQL) through a Mapper;

[0063] For page design in low-code development, an entity of the "page" type is defined in the knowledge graph ontology model, which usually corresponds to the implementation of front-end framework code. A page usually contains several components, and a component is an atomic unit that implements a specific function, such as a button component, a menu component, etc. A component can be associated with a configuration item or a script resource. Through the configuration item, the attributes of the component can be set, and the script defines data transfer and service calls. The component binds relevant data through the association with the script and executes specific function logic. Various components are organized in a certain layout or logical order to finally form a complete page template;

[0064] For process design in low-code development, a "workflow" type entity is defined in the knowledge graph ontology model, corresponding to the functional implementation of the backend code. The present invention defines a static analysis model of Java in the ontology model for the Java programming language, including three types of entities: classes, methods, and member variables, as well as static dependencies between them. The workflow entity is an organization of entity resources such as classes, methods, and member variables based on specific functional requirements.

[0065] For the above template-level entity relationships, there are nested relationships of different granularities between page type entities, and at the same time, they are bound to specific data object type entities through configuration items or scripts, or associated with workflow type entities related to specific functions; and there are calling relationships between workflow type entities, which can be associated with data objects through entity classes.

[0066] Knowledge extraction based on an extensible plugin framework.

[0067] The input of this module is a software project that includes front-end and back-end systems, and the technical development framework used is Vue and Java Spring Boot. For a given software system project to be reused, according to the entity and relationship types defined by the ontology model, the information therein is parsed and sorted based on the static parsing method, and the entities and relationships are extracted to achieve the mapping from the concept layer to the data layer, thereby completing the construction of the knowledge graph. Taking into account the various template resource types involved in the low-code development process and the needs for new scenarios in the future, the present invention implements an extensible plug-in framework, that is, to provide a unified interface for the knowledge extraction algorithm, and implement different knowledge extraction algorithms as different plug-ins. For different data types or software projects, appropriate plug-ins are configured according to the specific data situation and knowledge requirements. The framework will automatically execute these plug-ins in sequence according to specific conventions, and gradually add the knowledge that each plug-in is responsible for acquiring to the knowledge graph, thereby forming the final knowledge graph. The implementation of the knowledge extraction plug-in is specifically introduced below.

[0068] Data object extraction plug-in. Since the data parsed by the present invention is a software project developed using the Java Spring Boot framework, these projects usually use the Mybatis-Plus framework for back-end data persistence. Therefore, this plug-in is based on the Mybatis-Plus framework development specification, and for the .java files in the software project, it uses the annotation discrimination method to identify the annotations such as @Data, @TableName, etc. to filter out the data entity class, and convert the member variables in the entity class into fields in the data table to complete the extraction of the data object.

[0069] Page extraction plugin. The software front-end project selected in the present invention is developed through the Vue framework. Therefore, the object parsed by the page extraction plugin is the.vue files in the project. Combining the analysis of the project architecture, the.vue files in the component directory implement components with single functions, and the.vue files in the page directory are organized based on the developed components to form page templates developed for high-level requirements. For a single.vue file, its code contains <template>The page components identified by the label and <script>标签标识的数据、依赖、脚本等内容。本插件基于标签关键字解析，首先对选定软件项目中component目录下的.vue文件进行解析，抽取出项目的组件列表；然后对page目录下的.vue文件进行遍历，识别出文件中的<script>标签，在该标签内容下抽取出页面关联的脚本内容，包含了页面相关的数据格式、接口调用等信息。同时，识别<script>中的import语句，解析其中对其他页面组件的调用，形成.vue文件之间的import导入依赖；在<template>标签内容下依据前面抽取出的项目组件列表和文件之间的导入依赖，形成组件与页面之间的包含关系，并进而根据页面包含的组件所关联的数据，建立起页面与数据对象之间的绑定关系。

[0070] 工作流抽取插件。工作流是特定功能需求的实现，结合低代码开发流程，本插件采用基于执行路径的工作流抽取方法。具体而言，前端页面包含的脚本中执行了对后端服务的调用，本插件会识别出这类调用作为入口点，依据前端请求发送- Controller接收请求-Service服务调用-Repository数据交互的调用关系，从选定软件项目包含的.java文件中抽取出实现该功能的服务执行路径，并对过程中的数据传递、第三方API调用等进行参数化处理，最终封装为知识图谱中的工作流实体，同时建立起工作流与页面、数据对象的关联关系。

[0071] 基于大语言模型的实体信息补充。

[0072] 对于通过插件框架挖掘得到的低代码模板资源，在实际应用过程中一个主要的问题是这些资源缺少足够的语义信息，与用户输入的业务需求存在明显的语义鸿沟，在面向低代码模板资源推荐的情形下对用户的理解和使用造成了负担。针对基于插件框架构建的知识图谱，本模块利用大语言模型的自然语言理解与生成能力，对知识图谱中的实体节点进行功能描述信息的补充生成。具体而言，本模块基于关键词提取算法抽取出模板级实体的关键词信息作为提示输入到大模型，让大模型能够更多地关注到低代码模板的功能信息，对知识图谱中表示的低代码模板资源进行语义补充，生成相关的功能描述，从而面向用户需求能够更好地进行低代码模板推荐。考虑到模板类型的差异，本模块针对页面模板和工作流模板分别设计相应的关键词抽取方法：

[0073] 页面模板属于前端框架，包含了功能标签以及标签配置项中的属性值，比如<input>标签中的name属性值设置为"momentContent”，表示输入的动态内容文本，同时页面中还涉及相关的功能提示语，如"提交”，"修改”，表示组件执行的动作。本模块采取正则匹配的方法从页面模板中抽取功能关键词（功能标签配置项中的属性值和页面中的功能提示语），在后续过程中提示给大模型，指导其更多关注到页面模板实现的功能需求。

[0074] 工作流模板属于后端逻辑，涉及不同代码实体之间复杂的调用依赖关系。针对这样的特点，本模块根据工作流模板对应知识图谱的表示，以代码实体的名称作为节点，代码实体之间的静态依赖作为关系构建代码依赖图，应用TextRank算法抽取出该图中的关键词，按照和1）中同样的过程将模板和抽取出来的关键词信息提示给大模型让其生成该模板的功能语义描述。

[0075] 为了验证本发明工具的效果，选取了博客领域的4个开源软件项目（aurora、NBlog、DimpleBlog、mogu_blog）作为数据集。对于博客系统而言，通常包含了用户管理、内容管理、动态管理、分类与标签、内容搜索等基本模块。本发明针对给定软件项目数据构建了相应的知识图谱，该知识图谱包含了通用博客系统开发所具备的领域知识，能够面向用户需求进行低代码模板推荐。

[0076] 以博客系统中的博客动态发布模块为例，用户通过文本输入编辑动态内容，点击按钮实现博客动态的发布。

[0077] 该功能模块在本发明构建的知识图谱中的表示，如图3所示。图3中节点内容表示格式为

节点名称：节点类型

[0078] 除了知识图谱自身提供的结构化信息，本发明提出的实体信息补充方法也能够给用户提供语义信息，对实体的功能含义进行描述。例如，对于SetCreateTime方法而言，仅仅从该方法的名称只能获取该方法是用于设置创建时间，但实际的功能场景无从得知。通过本发明的实体信息补充，将代码依赖关系考虑进来，提供了更丰富的上下文信息，最终提示大模型生成该实体的功能描述，如图4所示。

[0079] 本领域技术人员在考虑说明书及实践本公开后，将容易想到本公开的其它实施方案。本公开旨在涵盖本公开的任何变型、用途或者适应性变化，这些变型、用途或者适应性变化遵循本公开的一般性原理并包括本公开未公开的本技术领域中的公知常识或惯用技术手段。说明书和实施例仅被视为示例性的，本公开也并不局限于上面已经描述并在附图中示出的精确结构，并且可以在不脱离其范围进行各种修改和改变。< / script> < / template>

Claims

1. A method for constructing a software knowledge graph for low-code template recommendation, characterized in that: The method comprises: Design a knowledge graph ontology model for low-code template recommendation; wherein the entities included in the knowledge graph ontology model include: code-level entities and template-level entities bound to specific functions, the code-level entities include: data entity classes; page components and configuration items, scripts and classes, methods and member variables related to the back-end Java code associated with the page components; the template-level entities include: data object type entities, page type entities and workflow type entities; the data object type entities are implemented through the code-level entities of the data entity class; the page type entity contains several components related to the front-end page and configuration items and scripts associated with the components; the workflow type entity contains classes, methods and member variables related to the back-end Java code and static dependencies between classes, methods and member variables; there are nested relationships of different granularities between the page type entities, and at the same time, they are bound to specific data object type entities through configuration items or scripts, or associated with workflow type entities related to specific functions; there are calling relationships between the workflow type entities, and they are associated to the data object type entities through the code-level entities of the data entity class; For a given software system project to be reused, entity extraction is performed based on the knowledge graph ontology model to generate an original knowledge graph; wherein workflow type entity extraction is performed based on the knowledge graph ontology model, including: Identify calls to backend services executed in scripts included in frontend pages; Taking the call as the entry point, and based on the call relationship between the Controller receiving request, the Service service call and the Repository data interaction sent by the front-end request, the corresponding service execution path is extracted from the .java file contained in the software system project; Parameterizing the data transmission and third-party API calls on the service execution path to obtain a workflow type entity; Establishing associations between workflow type entities, data object type entities, and page type entities; Based on the keyword information of the template-level entity, the attribute information of the template-level entity in the original knowledge graph is supplemented to obtain the knowledge graph of the software system project.

2. The method according to claim 1, characterized in that Extracting data object type entities based on the knowledge graph ontology model includes: Obtain annotations in the .java file of the software system project, including: @Data annotation and @TableName annotation; Filter out the data entity classes in the annotations; The data object is extracted by converting the member variables in the data entity class into fields in the data table.

3. The method according to claim 1, characterized in that Extracting page type entities based on the knowledge graph ontology model includes: Perform tag keyword parsing on the .vue files in the component directory of the software system project to extract a page component list of the software system project; Traverse the .vue files in the page directory of the software system project, identify the script tags in the .vue files, and extract the script content associated with the page under the script tag content; Based on the import statements in the script tag content, parse the calls of the page component corresponding to the script tag to other page components to form import dependencies between .vue files; Under the template tag content, based on the import dependency between the page component list and the .vue file, an inclusion relationship between the page component and the page is formed, and based on the data associated with the page component contained in the page, a binding relationship between the page type entity and the data object type entity is established.

4. The method according to claim 1, characterized in that: The attribute information of the template-level entity in the original knowledge graph is supplemented based on the keyword information of the template-level entity, including: Extract keyword information of template-level entities; Inputting the keyword information as prompts into the large model to generate a functional description of the template-level entity; The function description is added to the attribute information of the template-level entity.

5. The method according to claim 4, characterized in that Extract keyword information of page type entities, including: Acquire function keywords, where the function keywords include: attribute values ​​in function tag configuration items and function prompts in the page; The function keyword is extracted from the page of the software system project by using a regular matching method to obtain keyword information of the page type entity.

6. The method according to claim 4, characterized in that Extract keyword information of workflow type entities, including: The code dependency graph is constructed with the names of code entities as nodes and the static dependencies between code entities as relationships; The TextRank algorithm is applied to extract the keywords in the code dependency graph to obtain the keyword information of the workflow type entity.

7. A software knowledge graph construction system for low-code template recommendation, characterized in that: The system comprises: An ontology model design module is used to design a knowledge graph ontology model for low-code template recommendations; wherein the entities contained in the knowledge graph ontology model include: code-level entities and template-level entities bound to specific functions, wherein the code-level entities include: data entity classes; page components and configuration items, scripts, and classes, methods, and member variables related to the back-end Java code associated with the page components; the template-level entities include: data object type entities, page type entities, and workflow type entities; the data object type entities are implemented through code-level entities of data entity classes; the page type entities contain several components related to the front-end pages and configuration items and scripts associated with the components; the workflow type entities contain classes, methods, and member variables related to the back-end Java code, as well as static dependencies between classes, methods, and member variables; there are nested relationships of different granularities between the page type entities, and at the same time, they are bound to specific data object type entities through configuration items or scripts, or associated with workflow type entities related to specific functions; there are calling relationships between the workflow type entities, and they are associated to the data object type entities through code-level entities of the data entity class; The knowledge graph generation module is used to extract entities based on the knowledge graph ontology model for a given software system project to be reused, so as to generate an original knowledge graph; wherein the workflow type entity extraction based on the knowledge graph ontology model includes: Identify calls to backend services executed in scripts included in frontend pages; Taking the call as the entry point, and based on the call relationship between the Controller receiving request, the Service service call and the Repository data interaction sent by the front-end request, the corresponding service execution path is extracted from the .java file contained in the software system project; Parameterizing the data transmission and third-party API calls on the service execution path to obtain a workflow type entity; Establishing associations between workflow type entities, data object type entities, and page type entities; The attribute information supplement module is used to supplement the attribute information of the template-level entity in the original knowledge graph based on the keyword information of the template-level entity to obtain the knowledge graph of the software system project.

Images