Table 5.

F1-Score results of models pre-trained on SemEval-2016 and fine-tuned on the concussion dataset. The original performance of the not pre-trained model is also illustrated for comparison. In addition, the top-performing models are bolded

Model	Layer/Filter Sizes	F1 (%) Non-Pre-trained	F1 (%) Pre-trained SemEval-2016
Single layer CNN	[1, 2] [1, 2, 3]	62.01 61.35	60.76 61.39
Multi-layer CNN	[1, 2]	61.21	60.93
LSTM	[255]	61.38	61.17
Bi-Dir LSTM	[255]	60.78	60.66

Bold values indicates the top-performing models