4.2. NEURAL MODELS VS. CRFSNEXT, WE CONDUCTED EXPERIMENTS USING...
5.4.2. Neural models vs. CRFs
Next, we conducted experiments using neural models to compare with the CRF
model. Precision, recall, and the F
1
scores (on average of all entity types) of neural
extraction models are shown in Table 8. Our first observation is that all the variants
of the neural models outperformed the CRF model by a large margin. This shows the
power and effectiveness of the neural models with automatically learned features for
the task. Our best model using bidirectional LSTM and CRFs with both automatically
learned and handcrafted features achieved 88.11% in the F
1