Abstract:Aiming at the problems of strong heterogeneity of data sources, complex data semantics, and variable testing requirements in enterprise test data management, a dynamic construction method of test data model based on knowledge graph and intent recognition was studied. Multi-source data adapters were employed to extract structural metadata, natural language processing technology was utilized to generate semantic metadata, and both were fused to construct a cross-source metadata graph, forming a unified semantic association network. A test intent parsing mechanism was designed to support both graphical and natural language input modes, transforming user requirements into structured test intent descriptors. Through the dynamic semantic association engine, intelligent matching between test intent and metadata graph was achieved, and cross-source data entities and their associations were automatically discovered based on confidence calculation. A test intent-driven model synthesizer was used to select and combine operators from a predefined operator library to dynamically generate data processing logic conforming to test scenarios. Experimental tests in real enterprise environments showed that test data preparation efficiency was improved by 78.0 % compared with traditional manual SQL methods, data quality score was increased by 42.7 %, and cross-source data association accuracy reached 89.6 %, realizing automated construction of test data models in multi-source heterogeneous environments. The practical application demonstrated that the method meets the requirements of cross-source data association, dynamic requirement adaptation, and automatic data processing logic generation in enterprise test data management.
