{
  "fetched_at": "2026-10-24T17:00:45Z",
  "source_url": "https://www.kaggle.com/competitions/demolab-news-popularity-prediction/data",
  "pages": [
    {
      "name": "data-description",
      "content": "# File descriptions:\n`train.csv:` Contains 61 columns. The primary key that represents each new article is id. The target variable that you need to predict is `shares` which contains a continuous value (integer). This data is what you are going to use for EDA, cleaning, feature engineering, model building, model evaluation, model selection, and finally model training.\n\n`test.csv:` Contains 60 columns, doesn't contain the target variable `shares`. You will apply all the necessary preprocessing steps to get this data into the right format and then use the model trained using the `train.csv` file to make predictions with this. This is unseen data! Only the competition hosts know the actual values of the target feature for this data and therefore your submissions will be evaluated on how well your model does with this dataset.\n\n`sample.csv:` This contains the format in which you need to submit the solutions to Kaggle. The id column in this dataset exactly the same as the id column in `test.csv`. You will make your predictions on the `test.csv` data and store them in a submission file that has the same format as this file. Check the `Overview>Evaluation` tab for mode details.\n\n`data_dictionary.csv:` This contains the definitions for the various columns/features that you will need for the modeling exercise. This competition provides some numeric features for each of the articles and their metadata, as well as some features extracted by different models such as topic models (LDA), sentiment analysis models, subjectivity analysis models etc.\n\n---------\n\n# Dataset \nThis dataset summarizes a heterogeneous set of features about articles published by Mashable in a period of two years. The goal is to predict the number of shares in social networks (popularity). The features provided to you broadly come from 5 areas - \n\n1. Word features: Features that summarize statistics about the words being used in the article\n2. Links & References features: Features that summarize the statistics about the links and references being used in the article\n3. Digital Media features: Features that include information about the images and videos in the article.\n4. Temporal features: Time-based features related to when the article was published.\n5. Keyword features: Features that talk about the keywords (hashtags for an article) and categories (data channels) associated with the article (e.g. arts, science etc)\n6. NLP features: Features that come from a variety of models that were run on the articles, including Topic models (LDA), Sentiment analysis models, Subjectivity analysis models etc.\n\n![features](https://i.imgur.com/BBm7VLQ.png)\n\n>Note:\n>\n>1. Detailed definitions for each of the variables are provided in the `data_dictionary.csv`\n>\n>2. The starter notebook will also help guide you on how to work with the features provided.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Description",
      "content": "# Problem Statement\n\nIn this competition, you will build Regression models which accurately predict the popularity of new articles (the number of times it will be shared online) based on ~60 features provided to you. \n\n### Importance of online news \nThe consumption of online news is expediting day by day due to the extensive adoption of smartphones and the rise of social networks. Note that online news content comprises various key properties. For example, it is easily produced and small in size; its lifespan is short and the cost is low. Such Qualities make news content more effective to be consumed on social sharing platforms. More interestingly, this sort of content can capture the eye of a significant amount of Internet users within a brief period of your time. As a result, the main target on the analysis of online news content like predicting the recognition of stories articles, demonstrating the decay of interest over time to know the world of online news has greatly increased since it has so many practical meanings.\n\n### Why news popularity prediction?\nThere are multiple areas of applications for online news popularity prediction. Some of these benefits include gaining better insights into the audience consuming online news content. Consequently, it increases the ability of news organizations to deliver more relevant and appealing content in a proactive manner as well as the company can allocate resources more wisely to prepare stories over their life cycle. Moreover, prediction of news content is also beneficial for trend forecasting, understanding the collective human behavior, advertisers to propose more profitable monetization techniques, and readers to filter the huge amount of information quickly and efficiently.\n\n>**Note: Do check the Data section to understand, in detail, the features that are provided to you for this competition.**",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Evaluation",
      "content": "Goal\nIt is your job to predict how many times an article will be shared online (target variable name: \"shares\"), given the ~60 columns containing a summary about the content of the article, temporal features, features associated with metadata of the article, and features pulled from other NLP models run on the data.\n\nMetric\nSubmissions are evaluated on Root-Mean-Squared-Error (RMSE) between the value of the predicted value and the actual value of # of shares for each article.\n\nRMSE formula\n\nThe public leaderboard is going to rank your submission against other users while the competition is active, however, once the competition is ended, the final ranks will be calculated on the private leaderboard.\n\nSubmission file format\nThe file should contain a header and have the following format (CSV):\n\nid,shares\n29733,3262.0\n29734,802.0\n29735,953.0\netc.\nA sample file can be found attached in the Data section of this competition.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "rules",
      "content": "## One account per participant\nYou cannot sign up to Kaggle from multiple accounts and therefore you cannot submit from multiple accounts.\n\n## No private sharing\nPrivately sharing code or data outside of teams is not permitted.\n\n## Individual Participation\nEach participant must take the challenge individually. You can find a starter notebook in the code section to get you started.\n\n## Submission Limits\n- You may submit a maximum of 10 entries per day.\n- You may select up to 2 final submissions for judging.\n\n## Competition Timeline\n- *Start Date: _ 2022 11.59 PM , \n- *End Date: _ 2022 11.59 PM\n\n*Due to the restricted nature of the data, this competition does not count towards Kaggle ranking points.*\n\nWe ask that you respect the spirit of the competition and do not cheat. **Hand-labeling is forbidden.**\n\nHave fun!",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    }
  ]
}