Online Recruitment Fraud (ORF) Detection Using Deep Learning Approaches

Aim:

         To propose an advanced fraud detection system for online job postings by utilizing a transformer-based machine learning model, BERT, to enhance the detection of fraudulent job listings and improve the security of online recruitment platforms.

Abstract:

         The rise of digital platforms for job recruitment has brought with it a growing threat of fraudulent job postings, undermining the trust and safety of online hiring systems. This paper proposes an advanced fraud detection system based on the BERT (Bidirectional Encoder Representations from Transformers) model to identify fraudulent job postings. The system will utilize a dataset created by merging job postings from multiple sources to better capture both legitimate and fraudulent job listings. A comprehensive pre-processing pipeline, including data cleaning and feature engineering, will be applied to prepare the data for model training.

        The system will incorporate various techniques, such as SMOTE (Synthetic Minority Over-sampling Technique), to address class imbalance. The BERT model will then be fine-tuned to classify job postings, and the system will be evaluated using standard performance metrics like accuracy, precision, recall, and F1-score. The goal is to provide an effective and scalable solution for detecting fraudulent job postings in real-time, enhancing the overall security and trustworthiness of online recruitment platforms.

Existing System:

          Current fraud detection systems for job postings primarily rely on traditional machine learning models, such as Long Short-Term Memory (LSTM) networks. While these systems have shown promise, they are often plagued by issues such as false positives and false negatives, making them unreliable in distinguishing between legitimate and fraudulent job listings. These challenges highlight the need for more advanced techniques capable of improving classification accuracy and reducing errors in predictions. The inconsistencies in performance also suggest that current systems are insufficient for handling the complexity and nuance of online job postings.

Problem Definition:

          Fraudulent job postings pose a significant risk to job seekers and undermine the integrity of online recruitment platforms. Existing methods, including those based on LSTM, are limited by high rates of false positives and false negatives, resulting in unreliable predictions. There is an urgent need for a more accurate and reliable fraud detection system that can effectively differentiate between legitimate and fraudulent job postings in real-time, fostering greater trust in online job platforms and protecting users from potential harm.

Proposed System:

      This proposed system aims to address the limitations of current fraud detection methods by leveraging the BERT model, a transformer-based deep learning architecture known for its ability to understand the context of words in a sentence. The system will utilize a dataset composed of job postings gathered from diverse sources, ensuring a broad representation of both legitimate and fraudulent job listings.

       The data will undergo a series of pre-processing steps, including data cleaning (removing irrelevant content), handling missing values, and renaming columns for better clarity. The BERT model will be fine-tuned to process the text of the job postings, learning contextual embeddings that improve classification accuracy.

        The system will be capable of classifying job postings in real-time, processing incoming job listing data and applying the pre-trained BERT model to detect fraudulent posts. A web-based user interface, developed using Flask, will allow users to submit job postings for classification and view the results. The system will also include user authentication and session management features for secure operation. MySQL will be used to manage user data and session information, ensuring that the application runs smoothly and securely.

Algorithm:

          The proposed system utilizes the BERT (Bidirectional Encoder Representations from Transformers) model, a state-of-the-art deep learning technique that excels in understanding contextual relationships in text. The model will be fine-tuned on a dataset of job postings to classify them as fraudulent or legitimate. BERT’s ability to learn from the context of words, phrases, and sentences allows it to better understand the nuances of job postings and make more accurate predictions.