Open Relation Extraction in Patent Claims with a Hybrid Network

Table 4

Results of ORE dataset experiments on models.

	(%)	(%)	F1-score (%)
Bi-LSTM^1	89.1542%	90.2017%	89.6749%
Bi-LSTM+CNN+CRF^2	93.9204%	93.1501%	93.5337%
Bi-LSTM+CNN+CRF (label embedding)^3	97.0457%	96.1206%	96.5809%
Bi-LSTM+CNN+CRF (label embedding+attention)^4	97.3074%	97.7388%	97.5226%