Moumeneb1/LIVRABLE_ABOUMADA

A tweet crisis classification and psychilogical tweets classification

Livrable STAGE

This directory contains the implementation of the easy_nlp pipeline applied to the field of crisis and psychology.

To use the pipeline, you have to install the package easy_nlp

$ git clone https://github.com/Moumeneb1/IRIT_INTERNSHIP.git
$ pip install IRIT_INTERNSHIP/

.
├── ...
├── Crisis
│   ├── Bert_Base.ipynb
│   ├── Flaubert_base.ipynb
│   ├── Camembert_base.ipynb
│   ├── Bert+features.ipynb
│   ├── Flaubert+Features.ipynb
│   ├── Flaubert_base_adapted.ipynb
│   ├── Bert_CrisisNLP.ipynb
│   ├── Flaubert_LSTM.ipynb
│   ├── Flaubert_CNN.ipynb
│   ├── Bert Multitask.ipynb
│   ├── Flaubert_Multitask.ipynb 
|   └── Flaubert_base+ Focal Loss.ipynb
├── Psycho
│   ├── Bert_Base.ipynb
│   ├── Flaubert_LSTM.ipynb
|   └── Flaubert_Multitask.ipynb
└── ...

EcologyCrisis

These results showcase some of the models used on the RANDOM-SAMPLING CONFIGURATION

Model	Binary	Three class	Multiclass
Bert_base	0.824	0.742	0.586
Flaubert_base	0.841	0.765	0.617
Camembert_base	0.812	0.7427	0.5587
Flaubert+Features	0.834	0.834	0.613
Flaubert_base_adapted	0.853	0.767	0.654
Bert_CrisisNLP	0.822	0.742	0.591
Flaubert_LSTM.ipynb	0.848	0.7637	0.6713
Flaubert_CNN.ipynb	0.8515	0.7656	0.6654
Flaubert_Multitask.ipynb	0.847	0.769	0.625
Flaubert_base+ Focal Loss.ipynb	0.853	0.7804	0.66

Psycho

These results showcase the results of using these models on the psycho corpus

Model	Binary	Three class
Bert_base	0.7096	0.5610
Flaubert_LSTM	0.7836	0.6972
Flaubert_Multitask	0.7607	0.6972

You can find the adapted Language models weights on these links :

• flaubert_base_cased_psycho
• bert_base_multilingual_cased_psycho
• flaubert_base_cased_crisis
• bert_base_multilingual_cased_crisis

Notice : Adapted models are adapted Language models weights that can be used with any combination, You can train them using the LM_Training_script from the pipeline and saved them on a folder that you can reference after, for instance if you wanna use the flaubert_base_cased_psycho.

1. Download the LM
2. Unzip the folder
3. Reference the folder when creating instantiating your model

>>> from transformers import AutoModel
>>> from easy_nlp.models import BasicBertForClassification,
>>> base_model = AutoModel.from_pretrained(#Path too your Folder)
>>> model = BasicBertForClassification(base_model,n_class)