reab5555/perBERT

This model is a fine-tuned version of bert-base-uncased on MyPersonality essays and Pandora datasets.
It predicts the Big Five Traits from textual content.

It achieves the following results on the validation set:

• Val Loss: 0.574
• F1: 0.64
• Precision: 0.65

Description

This model is fine-tuned version of bert-base-uncased.
Trained on both MyPersonality dataset and Pandora dataset (Reddit posts written by users that were taken a Big Five questionnaires).

It includes the following classes:

• Extraversion (outgoing/energetic vs. solitary/reserved)
• Neuroticism (sensitive/nervous vs. resilient/confident)
• Agreeableness (friendly/compassionate vs. critical/judgmental)
• Conscientiousness (efficient/organized vs. extravagant/careless)
• Openness to experience (inventive/curious vs. consistent/cautious)

Training

model-finetuning: google-bert/bert-base-uncased

The following hyperparameters were used during training:

Training set: 90% - 29930
Validation set: 10% - 7238

seed: 42
learning_rate: 1e-5
train_batch_size: 16
val_batch_size: 16
optimizer: AdamW
num_epochs: 3

Training results

Epoch	Training Loss	Validation Loss	F1 Score
1.0	0.6149	0.5989	0.4928
2.0	0.5721	0.5811	0.6315
3.0	0.5129	0.5740	0.6381

Classification Report

          precision    recall  f1-score   support

    cEXT       0.61      0.45      0.51      1225
    cNEU       0.60      0.58      0.59      1430
    cAGR       0.59      0.46      0.52      1211
    cCON       0.60      0.52      0.56       990
    cOPN       0.76      0.89      0.82      2382