| There are domain entities and the relationship between domain entities in the domain knowledge graph,which can be imported into many application scenarios as domain knowledge.Code generation is used to generate code that can be executed on the domain application platform,so that developers can complete the development of domain related services or applications faster based on the existing domain platform.Domain knowledge in domain knowledge graph can be applied to the process of code generation for a domain platform.However,there are some problems in the generation of domain specific code based on domain knowledge graph.First,the construction of domain knowledge graph needs more domain related data sources.For data coming from multiple data sources,data fusion is needed to solve the heterogeneity of data.Second,constructing domain knowledge graph artificially is very time-consuming,and it is very difficult to update and maintain.As the core work of constructing domain knowledge graph,it is necessary to automatically extract domain entities and the relationship between domain entities.Third,in the process of code generation,because of the text expression of the requirement description and the incompleteness of platform related transformation,domain knowledge needs to be introduced to implement the parsing and transformation of domain terms and business logic from the description of the requirements.In this paper,the domain related data in the open data is used to extract domain entities and the relationship between domain entities,which are used to constructed a domain knowledge graph.Based on the constructed domain knowledge graph,the code of the domain target platform is generated.The work of this paper mainly includes the following aspects:(1)A fusion method for heterogeneous open data is proposed to fuse data items in different data documents.The data sources are domain related documents in open data that contain domain entities.The credibility score of a candidate domain entity in a data document is calculated based on the correlation between data item and the credibility of a data document,which implement extraction of domain entities from domain related open data.(2)A method of constructing domain knowledge graph is proposed.In this paper,domain related data in open data are used to extract domain entities and domain entities relationship.Based on the collection of the extracted domain entities,combined with the application of domain knowledge in the generation of domain specific code,we extract the attribute relations,synonymous relations and multivariate relations among domain entities to construct the domain knowledge graph.(3)A domain related code generation method based on domain knowledge graph is proposed.In this paper,based on domain knowledge graph constructed for quantitative trading,we transform strategy description text to platform independent strategy logic model.Then,the strategy logic model is transformed into a platform specific strategy model of the target platform based on the transformation corresponding relationship of the platform.In the end,code templates and template engines are used to generate the strategy code which can be executed in selected quantitative trading backtesting platform.(4)A code generation system used in quantitative trading backtesting platform is constructed.According to the method framework proposed in this paper,the knowledge graph in the field of financial analysis is constructed by using the relevant open data.The effectiveness of the method is introduced in the context of the code generation of the quantitative trading backtesting platform.In this paper,domain knowledge graph is used as domain knowledge in domain specific code generation process,which fills the gap between description and the code of target platform,and reduces the threshold for users to use domain related application platforms.At the same time,a more abstract description of business logic instead of coding following specific platform rules can also facilitate the transplantation of code across different domains. |
