Fake Job Predictor
Data
- Data trained comes from this Kaggle repository: https://www.kaggle.com/datasets/shivamb/real-or-fake-fake-jobposting-prediction
- Original data size is around 18k samples. To avoid the class imbalacing problem, it was undersampled the majority class (true jobs).
- Final dataset used to train has a size of 4k sample.
Model
- Multi-head neural network. One head is used for each feature (description, requirements, and benefits of the job).
- Best metrics achieved (over validation data-split): Precision: 0.83, Recall: 0.65, F1-score: 0.71
- Code used for training comes from this GitHub repo: https://github.com/sebassaras02/AdvancedDLCourse/blob/master/02_transformers_nlp/bert.ipynb
Components:
Text Encoder: distilbert-base-uncased is used to encode the textual input into a dense vector.
Future work:
Train over larger datasets and with more computer resources