Skip to main content
. 2025 Aug 14;9:e64536. doi: 10.2196/64536

Table 2.

Results of the baseline approach (n=463) in terms of true positive, false positive, false negative, true negative, and runtime for each model.

Model TPa, n (%) FPb, n (%) FNc, n (%) TNd, n (%) Runtime (s)
Word_Tokenize (n=463)
BiLSTMe 248 (53.6) 36 (7.8) 26 (5.6) 153 (33.0) 389.01
GRUf 249 (53.8) 40 (8.6) 25 (5.4) 149 (32.2) 303.88
CNNg 252 (54.4) 43 (9.3) 22 (4.8) 146 (31.5) 138.72
DeepCut (n=463)
BiLSTM 258 (55.7) 47 (10.2) 16 (3.5) 142 (30.7) 476.81
GRU 249 (53.8) 46 (9.9) 25 (5.4) 143 (30.9) 411.99
CNN 258 (55.7) 49 (10.6) 16 (3.5) 140 (30.2) 152.16
AttaCut (n=463)
BiLSTM 260 (56.2) 46 (9.9) 14 (3.0) 143 (30.9) 467.10
GRU 251 (54.2) 40 (8.6) 23 (5.0) 149 (32.2) 387.65
CNN 248 (53.6) 35 (7.6) 26 (5.6) 154 (33.3) 146.02

aTP: true positive.

bFP: false positive.

cFN: false negative.

dTN: true negative.

eBiLSTM: bidirectional long short-term memory.

fGRU: gated recurrent unit.

gCNN: convolutional neural network.