Few-Shot Multihop Question Answering over Knowledge Base

KBQA is a task that requires to answer questions by using semantic structured information in knowledge base. Previous work this area has been restricted due the lack of large parsing dataset and exponential growth searching space with increasing hops relation paths. In paper, we propose an efficient pipeline method equipped pretrained language model. By adopting beam search algorithm, will not be subgraph 3 hops. Besides, data generation strategy, which enables our model generalize well from few training samples. We evaluate on open-domain complex Chinese question answering CCKS2019 achieve F1-score 62.55% test dataset. addition, order few-shot learning capability model, randomly select 10% primary train result shows can still achieves 58.54%, verifies process advantage learning.