생성형 AI를 이용한 소프트웨어 특허 문서로부터 가장 작은 기능적 요소 추출 메커니즘

초록

Software verification requires extensive testing across various aspects. Reverse engineering-based testing generates and executes test cases from intermediate outputs. These artifacts include software patents. However, research on test case generation from software patents is very few. Because patent documents have a different format than typical requirements specifications, they require a different approach than traditional analysis techniques. Also processes such as patent sentence interpretation and component classification are difficult to automate. To solve this problem, we proposes a mechanism that extracts the smallest functional elements from software patent sentences using generative AI. 1) Text data is extracted from software patent documents. 2) The extracted text is classified, and text containing algorithms is identified. 3) The sentences are normalized. 4) Sentences describing software functions are identified. 5) The sentences are simplified. 6) The simplified sentences are converted into the smallest functional components. This method classifies and identifies functional elements from patent natural language sentences. Furthermore, it can generate basic model for test case generation. Testing costs can be reduced because test cases are automatically generated from intermediate outputs.

키워드