The data provided for this competition is structured in a simple way. There's a text excerpt from a scientific paper and an indication of whether this text is "fake" (probably generated) or real, i.e. human-written. The data comes from both published and retracted Scopus papers. 

## Files

*   **fake_papers_train_part_public.csv** - the training set (5327 records)
*   **fake_papers_test_public.csv** - the test set (21310 records)
*   **sample_submission.csv** - a sample submission file in the correct format (21310 records)

## Columns

*   `text` - text input, excerpt from a scientific paper
*   `fake` - whether the text is generated or not, i.e. 1 stands for the generated content and 0 stands for human-written content (only present in the training set)