Table 2 Classification results of the compared methods for the TwiMed corpus

Feature-rich SVM [24]

0.799

0.681

0.728±0.100

0.752

0.810

0.778±0.047

IAN [24]

0.878

0.738

0.792±0.016

0.836

0.813

0.824±0.042

CNN-based method [42]

0.849

0.831

0.835±0.060

0.739

0.788

0.761±0.061

multichannel CNN [43]

0.861

0.780

0.816±0.072

0.738

0.841

0.780±0.054

Joint AB-LSTM [44]

0.817

0.856

0.831±0.040

0.701

0.828

0.754±0.072

BiLSTM+MSAM+position

0.858

0.852

0.853±0.057

0.748

0.856

0.799±0.046