Distant Supervision with Transductive Learning for Adverse Drug Reaction Identification from Electronic Medical Records
Table 5
The effectiveness comparison on fivefold cross-validation of text transformation across three types of document representation using MIL-iEM with soft decision making (MIL-iEM-S) and hard decision making (MIL-iEM-H).
Models
Soft decision making
Hard decision making
B
TF
TFIDF
B
TF
TFIDF
P
R
F1
P
R
F1
P
R
F1
P
R
F1
P
R
F1
P
R
F1
MIL-iEM-S-T
MIL-iEM-H-T
S–P
0.858
0.308
0.454
0.858
0.308
0.454
0.857
0.307
0.452
0.856
0.327
0.473
0.856
0.327
0.473
0.858
0.330
0.477
S–W
0.879
0.599
0.712
0.873
0.600
0.711
0.871
0.589
0.703
0.871
0.651
0.745
0.863
0.642
0.736
0.873
0.650
0.745
L–P
0.890
0.460
0.606
0.890
0.460
0.606
0.882
0.451
0.597
0.887
0.500
0.639
0.887
0.500
0.639
0.881
0.498
0.636
L–W
0.868
0.609
0.716
0.863
0.611
0.716
0.873
0.604
0.714
0.866
0.659
0.748
0.863
0.653
0.744
0.868
0.662
0.752
BOW
0.755
0.624
0.683
0.780
0.628
0.696
0.765
0.624
0.687
0.744
0.642
0.690
0.730
0.646
0.685
0.726
0.642
0.682
MIL-iEM-S-T
MIL-iEM-H-T
S–P
0.845
0.836
0.840
0.844
0.838
0.841
0.846
0.830
0.838
0.620
0.985
0.761
0.620
0.985
0.761
0.621
0.985
0.762
S–W
0.783
0.792
0.788
0.784
0.801
0.792
0.787
0.743
0.764
0.686
0.971
0.804
0.683
0.969
0.802
0.691
0.967
0.806
L–P
0.840
0.816
0.828
0.840
0.816
0.828
0.836
0.799
0.817
0.651
0.974
0.781
0.652
0.974
0.781
0.651
0.973
0.780
L–W
0.785
0.797
0.791
0.796
0.799
0.798
0.777
0.714
0.744
0.693
0.962
0.805
0.689
0.960
0.802
0.696
0.960
0.807
BOW
0.692
0.927
0.793
0.749
0.850
0.797
0.735
0.861
0.793
0.632
0.973
0.766
0.646
0.962
0.773
0.649
0.960
0.774
MIL-iEM-S-T
MIL-iEM-H-T
S–P
0.840
0.836
0.838
0.842
0.834
0.838
0.841
0.819
0.830
0.645
0.755
0.696
0.644
0.754
0.695
0.630
0.757
0.688
S–W
0.773
0.790
0.782
0.782
0.792
0.787
0.782
0.732
0.756
0.666
0.825
0.737
0.649
0.834
0.730
0.640
0.856
0.732
L–P
0.833
0.830
0.832
0.833
0.828
0.831
0.835
0.801
0.818
0.668
0.814
0.734
0.668
0.814
0.734
0.656
0.841
0.737
L–W
0.783
0.796
0.789
0.799
0.805
0.802
0.778
0.710
0.742
0.657
0.827
0.732
0.642
0.823
0.722
0.641
0.863
0.736
BOW
0.691
0.900
0.782
0.748
0.834
0.789
0.734
0.841
0.784
0.655
0.746
0.698
0.666
0.801
0.727
0.675
0.794
0.730
B: binary frequency; TF: term frequency; TFIDF: term frequency-inverse document frequency.
