Hello, i train model3v2 on SemEval 2010 task8 dataset, but cannot get the 84% f1?

[tangm00]

what can i do to reach the performance? thank you

[Sshanu]

You may reach that accuracy by applying these steps:

1. First, feed the data in the model using bootstrap
2. Use dropout in LSTM
3. I have not optimized the hyperparameters well
4. Train the POS and Dependency embeddings by Skip Gram Model or any other technique.
   Then use these embeddings for initializing the embeddings in the model.
   Try these it will help in increasing the accuracy.

[tangm00]

Thank you very much! You mean using bootstrap resampling methods for the training data in each batch?

[Sshanu]

Yeah

[tangm00]

Thank you.

[damao666]

what can i do to reach the performance? thank you

Can you provide me with a working code? Every time I run, I report an error.

Thank you very much.