Transformers Based Lie Detectors: Deceptive Opinions Detection Using Transformers on Single and Multiple Domains
Lie_Detector_NLP_transformers
The project was made for the 097215 Natural Language Processing course at the Technion - Israel Institute of Technology.
lie_detector.py - running the experiment mention in our paper: "Transformers Based Lie Detectors: Deceptive Opinions Detection Using Transformers on Single and Multiple Domains". the results of this experiment are output into txt files which we parse into an Excel (csv) file using parser_results.py . create_mix_files.py - scripts for creating our datasets.
in order to run the experiment by your self:
- run create_mix_files.py once (choose which datasets to begin with and which output datasets you will get)
- run lie_detector.py - choose which datasets and hyperparameters to run it with (we recommend to split the run for different datasets since it may take a long time to run all of them in one run)
- if you wish to collect the experiment outputs in a convenient way - run parser_results.py which will collect all the data into a single csv file (which can be easily read using Office-Excel)
notice we added an Excel file called results_all_parsed4084.xlsx that sum-up the cases we experimented in our paper