Abstract

Online recruitments have triggered a massive crime problem of online recruitment scams with many people posting their job online. Fraudsters take advantage of the situation and use deceptive job adverts and steal information or get money. The first challenge is the imbalance of classes, and existing systems utilize classical algorithms of machine learning instead of which there are boundaries. These restrictions on fake job ads see a significant rise in imbalanced forecasts. These weaknesses are addressed in the proposed method which uses deep learning models such as BERT and RoBERTa and special datasets, such as unrealistic job postings and Pakistani and US job postings. Also, the proposed solution uses SMOTE (Synthetic Minority Oversampling Techniques) to overcome so-called class imbalance in the dataset. The intended method aims at preventing biased predictions and maximizing accuracy and recall.

Authors

Chandra Sekhar Sanaboina
University College of Engineering Kakinada, India

Keywords

Deep Learning, Natural Language Processing, Bidirectional Encoder Representations from Transformer, Robustly Optimized BERT, Synthetic Minority Oversampling Technique, Synthetic Minority Oversampling Borderline