Abstract
AbstractSpan extraction is an important task of machine reading comprehension (MRC). Traditionally, for a question, the answer is expected to be a single text span from the given context. However, in practice, the answer to a question may exist in multiple spans. Besides, multiple answers to a question may be of different granularity and hierarchically related to each other. In this paper, we propose a simple but effective pipeline to solve the hierarchical multi-answer questions with conditions, an evaluation task introduced in CCSK 2022. The pipeline mainly contains two components: the answer span detection and relation classification. Answer span detection focuses on finding multiple answer spans while relation classification aims to determine whether there is a relationship between two answers. In addition, some helpful strategies are also introduced. Finally, our pipeline achieved an F1 score of 0.759 on online testing data and ranking second among 181 teams.KeywordsSpan extractionMRCRelation classification
Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
