A Prompt-based Few-shot Machine Reading Comprehension Model for Intelligent Bridge Management

Luyi Zhang, Ren Li, Qiao Xiao

Abstract: Bridge inspection reports are an important data source in the bridge management process, and they contain a large amount of fine-grained information. However, the research on machine reading comprehension (MRC) methods for this field is insufficient, and annotating large scale domain-specific corpus is time-consuming. This paper presented a novel prompt-based few-shot MRC approach for intelligent bridge management. The proposed model uses the pretrained model MacBERT as backbone. The prompt templates are designed based on some domain-specific heuristic rules. The experimental results show that our model outperforms the baseline models in different few-shot settings. The proposed model can provide technical support for the construction of automatic question answering system in the field of bridge management.