# TLDR

The data used in this Challenge come from many different [The Messenger] (https://messenger.eso.org) journals. Importantly, *both* texts in each sample (fake and real) have been significantly modified using LLMs.

Your task is to find which text in each pair is considered **Real**.

## List of directories and files

- **train** - directory
- **test** - directory
- **train.csv** - file with ground truth for samples in `train` directory

The data in the folder are divided into `train` and `test` directory. There is also file `train.csv` with ground truth for each sample from `train` directory. 

In both folders, `train` and `test` there are different articles, randomly numbered. For each of such articles, there are two versions provided `file_1.txt` and `file_2.txt`, one is treated as `Real` and one is `Fake`, they are also randomly placed as number 1 and number 2. 

In `train.csv` file there is ground truth provided, first column is an `id` of an article, second is `real_text_id`, so which of the files consists of `Real` sample (number 1 or 2). Sample format of submission is shown in [Overview](https://www.kaggle.com/competitions/fake-or-real-impostor-hunt/overview) tab.

## More details


### Preprocessing

The dataset is a highly preprocessed version of the raw [The Messenger] (https://messenger.eso.org) journal articles (both 
**Real** and **Fake**). For the preprocessing, different techniques were used. LLMs were also used.