These are chat archives for beniz/deepdetect

11th
Nov 2018
zakhariarestu
@zakhariarestu
Nov 11 2018 15:54
Hi @beniz I want to ask, I'm doing research to detect hate speech tweets (only text) on my country's twitter, I use deepdetect with MLP templates, my question is how do I determine the value of parameters when I do training with my dataset so the prediction is accurate? Is it by trying to replace 1 per 1 parameter value or is there another way that is more efficient? One more, is there a journal reference about this deepdetect? I'm sorry if my English isn't good, thanks
Emmanuel Benazera
@beniz
Nov 11 2018 17:08
Hi, you can use your test set and metrics as a witness and change the number and size of layers. Usually you'll find a good configuration in just a few trials. That is unless you have a lot of data and then for tweets, I'd recommend using the vdcnn templates for character-based models.
zakhariarestu
@zakhariarestu
Nov 11 2018 18:27
Okay, thank you, but is there a journal reference that I can read to increase my knowledge about Deepdetect? And is there a minimum amount of data training if using vdcnn template?
Emmanuel Benazera
@beniz
Nov 11 2018 18:32