Generating a Software Testing Knowledge Graph with a Large Language Model

In 2016 the author manually created a comprehensive software testing panorama that attracted 15,000 reads and was printed on paper. Seeking a modern approach, they used the Claude 4 Sonnet Think large language model to generate a software testing knowledge graph.

Core characteristics of the generated knowledge graph

Entity hierarchy

Core testing entities: TestCase, TestSuite, TestPlan, TestExecution – the basic building blocks of testing activities.

Professional test types: UnitTest, IntegrationTest, PerformanceTest, etc. – representing different test dimensions.

Supporting entities: TestData, TestEnvironment, TestTool – the infrastructure for test execution.

Quality entities: TestMetrics, Coverage, TestReport – metrics and feedback on test effectiveness.

(The core entities are correct; the professional test types are partially inaccurate because “unit test, integration test” are usually grouped with “system test, acceptance test” to reflect test levels, while test types should include functional, performance, security, compatibility, reliability, etc.)

Key business relationships traced_to: bidirectional traceability from requirements to test cases. executes: dynamic link between test execution and test cases. discovers: quality‑detection relationship where execution discovers defects. supports: tools supporting testing activities. measures: quantitative relationship measuring the testing process.

(These relationships capture traceability, dynamic execution, defect detection, tool support, and quantification.)

The author then zooms in on the central TestCase entity, showing seven input lines (requirements, test personnel, execution, automation, test plan, coverage) and three output lines (belonging test suite, using test data, executing in test environment).

The model’s output aligns with most testers’ intuition that test cases (including scripts) are central, but the author notes a missing emphasis on test analysis—the source of test scope, risk, and strategy—which is not well represented despite being part of a test plan.

After providing feedback, the model produced a simplified graph that hides entity attributes, which the author judges as an improvement for clarity.

Additional versions were generated, including a vertical layout, to make the graph easier to read.

Looking ahead, the team plans to refine the knowledge graph with more domain expertise, store it in a graph database, and use it as a knowledge base to enhance the large model’s capabilities for software testing education and practical applications.