Fasttext1 results include multiple selected expert Action Types while only one Action Type could be selected by the skipgram model, so results are better than reflected here. Results for Transfer may reflect its marginal utility. Results for Prescribe, Perform Therapeutic Procedure, and Test are excellent given a limited data set of 316 recommendations and 13 categories.
Action Type |
TP |
FP |
FN |
Precision |
Recall |
F1 |
Prepare |
32 |
7 |
3 |
0.821 |
0.914 |
0.865 |
Monitor |
15 |
4 |
6 |
0.789 |
0.714 |
0.750 |
Perform Therapeutic Procedure |
31 |
4 |
6 |
0.886 |
0.838 |
0.861 |
Test |
44 |
6 |
9 |
0.880 |
0.830 |
0.854 |
Educate/Counsel |
37 |
12 |
7 |
0.755 |
0.841 |
0.796 |
Prescribe |
56 |
5 |
3 |
0.918 |
0.949 |
0.933 |
Examine |
5 |
1 |
1 |
0.833 |
0.833 |
0.833 |
Conclude |
15 |
4 |
0 |
0.789 |
1.000 |
0.882 |
Inquire |
10 |
1 |
4 |
0.909 |
0.714 |
0.800 |
Advocate |
5 |
1 |
2 |
0.833 |
0.714 |
0.769 |
Refer/Consult |
12 |
2 |
2 |
0.857 |
0.857 |
0.857 |
Prevent |
7 |
0 |
2 |
1.000 |
0.778 |
0.875 |
Transfer |
0 |
0 |
2 |
|
0.000 |
0.000 |
[1] A. Joulin, E. Grave, P. Bojanowski, T. Mikolov, Bag of Tricks for Efficient Text Classification