Skip to main content

View full-text article in PMC

. 2020 Aug 12;22(8):e17478. doi: 10.2196/17478

Table 6.

Performance of relevance classifiers.

Classifiers	Area under the receiver operating characteristic curve (95% CI)	Precision	Recall	F1
Logistic regression	0.84 (0.78-0.89)	0.80	1.00	0.92
Random forest	0.95 (0.93-0.98)	0.93	0.97	0.98
Support vector machine	0.92 (0.88-0.96)	0.91	0.97	0.95
Naive Bayes	0.88 (0.83-0.93)	0.88	0.99	0.93
CNN^a (vaping-related word vectors)	0.94 (0.91-0.97)	0.90	0.97	0.98
LSTM^b (vaping-related word vectors)	0.91 (0.88-0.95)	0.89	0.98	0.96
LSTM-CNN (vaping-related word vectors)	0.89 (0.85-0.93)	0.93	0.87	0.95
BiLSTM^c (vaping-related word vectors)	0.89 (0.85-0.94)	0.90	0.96	0.94
CNN (GloVe^d word vectors)	0.95 (0.92-0.97)	0.93	0.95	0.98
LSTM (GloVe word vectors)	0.95 (0.92-0.98)	0.95	0.95	0.98
LSTM-CNN (GloVe word vectors)	0.96 (0.93-0.98)	0.96	0.93	0.98
BiLSTM (GloVe word vectors)	0.95 (0.93-0.98)	0.92	0.96	0.98

^aCNN: convolutional neural network.

^bLSTM: long short-term memory.

^cBiLSTM: bidirectional long short-term memory.

^dGloVe: Global Vectors for Word Representation.