How subh-ds/fake-job-internship-prediction works
This repository is a research project for detecting fraudulent job postings from text and posting screenshots. Its workflow uses a shared dataset in two experiments: Experiment 1 compares classical baselines and a custom multimodal model; Experiment 2 uses a pretrained ResNet-50, validation-based checkpointing and threshold selection, then evaluates predictions. Report-generation code summarizes the experiments and results. Evidence is limited to the sampled source excerpts and report text; the experiment notebooks and dataset wiring were not inspected directly.
Main entry points: Experiment_2_Multimodal_ResNet50.ipynb Experiment 2 · build_report.py Report generator · Experiment_1.ipynb Experiment 1 · make_figs.py Results figures
Components 9
Research data
- Posting dataset
Dataset/fake_job_postings.csv - Posting screenshots
Dataset/images
Detection experiments
- Experiment 1
Experiment_1/Experiment_1.ipynb - Experiment 2
Experiment_2/Experiment_2_Multimodal_ResNet50.ipynb - Notebook builder
Experiment_2/build_nb.py
Results reporting
- Results figures
report_figs/make_figs.py - Report generator
report_figs/build_report.py - Experiment report
Docs/Fake_Job_Fraud_Detection_Experiments_Report.docx - Detection report
Docs/Fake_Job_Fraud_Detection_Report.docx
Connections 10
- Posting dataset → Experiment 1: provides postings
- Posting screenshots → Experiment 1: provides screenshots
- Posting dataset → Experiment 2: provides postings
- Posting screenshots → Experiment 2: provides screenshots
- Notebook builder → Experiment 2: generates notebook
Experiment_2/build_nb.py - Experiment 1 → Results figures: supplies results
report_figs/make_figs.py - Experiment 2 → Results figures: supplies results
report_figs/make_figs.py - Results figures → Report generator: adds figures
report_figs/build_report.py - Report generator → Experiment report: generates report
- Report generator → Detection report: generates report