{
  "fetched_at": "2026-10-19T18:56:00Z",
  "source_url": "https://www.kaggle.com/competitions/artificial-text-detection-homework-3/data",
  "pages": [
    {
      "name": "data-description",
      "content": "## Files\n\n*   **dev.csv** - the development set\n*   **test.csv** - the test set\n\n\n## Columns\n\n* ID -- the ID \n* Text  -- the text \n* Class -- the target class, `H` stands for human-written texts, `M` for machine-generated text",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Description",
      "content": "This assignment extends from the hands-on session on artificial text detection.\n\n* We use the **IMDB** dataset as a source of human-written texts\n* Eight text generation models and three decoding strategies are used to generate the testing data\n* The specific text generation models will remain undisclosed until the homework deadline.\n* We will only reveal that *GPT2-large* is one of the eight models under consideration.\n\nThis assignment is a binary classification task, wherein you are required to train a classifier that predicts whether a text is human-written (`H`) or machine-generated (`M`). \n\n\n\n\n\n",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Evaluation",
      "content": "Accuracy is used as a main evaluation metric.\n\n",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "rules",
      "content": "- You are allowed to use any additional materials and any pre-trained models, with the exception of direct markup of the test set and search on the Internet\n- You should work solely on this task\n- You will get five points for surpassing the **weak baseline **\n- Additionally, you will earn bonus points if your submission ranks among the top five. The calculation for bonus points would be $$5 - \\text{your rank} + 1$$",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "abstract",
      "content": "**Artificial text detection** \n\nMotivated by the urgent need for artificial text detection, we suggest you develop your own classifier capable of distinguishing between human-written and machine-generated texts.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Baselines",
      "content": "We provide three baselines for reference:\n* The random baseline randomly assign one of two labels to a text (`H` or `M`)\n* The weak baseline re-uses the model, trained in the practical session \n* The strong baseline improves upon weak baseline by incorporating data from various text generation models\n* The super-strong baseline outperforms other baselines by integrating data from all text generation models employed in the test set",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Submission Format",
      "content": "For every Id in the test set, submission files should contain two columns: Id and Class.\n\nThe file should contain a header and have the following format:\n```\nID,Class\n0,H\n1,M\n2,M\n```",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Data",
      "content": "- We will provide development data only as an only labeled set \n- We will not disclose the text generation models used to create the test set\n- You are welcome to utilize any text generation model and any sources of your choice for training data",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Grading",
      "content": "You will receive:\n- 1 point if your top submission surpasses the random baseline\n- +1 points if your leading submission outperforms the weak baseline\n- +1 points if your top submission outperforms the strong baseline\n- +1 bonus point if your best submission outperforms the super strong baseline\n\nGrading will be determined by the scores achieved on the **private** leaderboard.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Reference ",
      "content": "- Use [this Kaggle tutorial](https://www.kaggle.com/docs/competitions#making-a-submission) to learn how to get started \n- Refer to [this Quora discussion](https://www.quora.com/What-is-the-difference-between-public-and-private-leaderboard-in-Kaggle) to familiarize yourself with the concept of public/private leaderboards\n",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    }
  ]
}