As suggested in title I am trying to combine CNN and HAN for text classification purpose. Individually I am getting F1-score of 0.82 and 0.83 respectively. However, when combining I am getting f1-score of 0.42. So the model is performing badly. Right now I have first applied CNN and then the output of CNN is fed to word attention network and then HAN implementation. Could anyone suggest me what other architectures should I try? I am literally running out of ideas.