These sentences contained in total 253 neuron properties, of which 207 could be recognized by neuroNER resulting in a 0.82 recall. Only 3 properties were incorrectly extracted, resulting in a 0.98 precision.
Here's some ideas:
consider all terms within the same sentence phrase of a neuron trigger (i.e., cortical cells recorded in layer 2/3) as part of a neuron mention, even if there isn't an explicit rule for that.
more part of speech info - terms which seem to modify the neuron trigger
consider applying leon french's rules for normalization (e.g., somatostatin and martinotti cells = 2 neuron mentions)
These sentences contained in total 253 neuron properties, of which 207 could be recognized by neuroNER resulting in a 0.82 recall. Only 3 properties were incorrectly extracted, resulting in a 0.98 precision.
Here's some ideas: