Few-Shot Multihop Question Answering over Knowledge Base

Table 3

We evaluate our model on primary training datasets, where created samples are excluded.

Data	Train	Valid	Test
10%	82.90	84.31	80.13
10% + created data	87.51	89.54	82.75
50%	94.95	93.99	88.50
50% + created data	95.12	93.72	89.41
100%	97.39	95.42	88.76
100% + created data	99.09	95.45	91.11