This paper presents a method for decomposing long, complex consumer health questions. Our approach largely decomposes questions using their syntactic structure, recognizing independent questions embedded in clauses, as well as coordinations and exemplifying phrases. Additionally, we identify elements specific to disease-related consumer health questions, such as the focus disease and background information. To achieve this, our approach combines rank-and-filter machine learning methods with rule-based methods. Our results demonstrate significant improvements over the heuristic methods typically employed for question decomposition mat rely only on the syntactic parse tree.
展开▼
