{
  "fetched_at": "2026-10-19T20:51:13Z",
  "source_url": "https://www.kaggle.com/competitions/ai-against-modern-slavery-twiml-competition/data",
  "pages": [
    {
      "name": "data-description",
      "content": "There are two files that you need:\n\n- **train.csv** - containing the 650 examples of training data\n- **test.csv** - containing the 331 examples of competition data, with the LABELs that you need to predict.\n\nThe sample_submission.csv is an example of correctly formatted submission file.\n\nThe training data contains the extracted texts from companies' modern slavery statements. These statements are originally published in PDF (both scanned and digital) and HTML formats. A variety of methods has been used to get the raw text including OCR and HTML scraping techniques which is why sometimes you might see some noise or misaligned titles.\n\nThe target metric LABEL_MSA_training_binary essentially answers the question: Does the company provide training specifically to address modern slavery and human trafficking? \n\nThis training can take place as a standalone course on modern slavery, or be embedded in a wider training programme on supply chains or human rights. It can cover many different groups- procurement specialists, and those who work in recruitment or in leadership positions.\n\nTraining can be provided to:\n- Employees (all)\n- Procurement / purchasing\n- Recruitment / HR\n- Leadership\n- Suppliers\n\nThe train.csv includes the ground truth 1/0 values for each statement and your task is to estimate the probability for all the documents in the test set whether they contain information about the provision of training.\n\nIt's possible that the statement mentions the provision of training but:\na) it's not specified to whom or about what, or;\nb) some training methodology is only being created, that is, it's \"in development\"\nin those instances the target value is 0 because companies can provide any types of training while we care about specifically modern slavery related training programs. Additionally, training programs that are in development are not creating any impactful change in the world.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Description",
      "content": "<h1>Welcome to the TWIMLfest AI for Good Hackathon!</h1>\n<h2>Contribute to Modern Slavery Research</h2>\n<p>In 2018, the Global Slavery Index found that there were 40.3 M people in modern slavery, of whom 25M were in forced labor producing computers, clothing, agricultural products, raw materials, etc and 15M were in forced marriage. In order to facilitate measures to end modern slavery the United Nations asks companies to take immediate measures and state clear policies. The Future Society is an independent nonprofit think-and-do tank curating an up-to-date repository of >16K of those statements. Use your NLP-Skills predict whether those statements meet the required standards!</p>\n\n<p>The data for this competition consists of 981 of those statements from different companies with an expert-reviewed annotation. This annotation is used to determine whether the company is providing training to their employees and supply chain workers regarding addressing modern slavery and human trafficking. Training can be provided to\n<ul>\n  <li>Employees (all)</li>\n  <li>Procurement / purchasing</li>\n  <li>Recruitment / HR</li>\n  <li>Leadership</li>\n  <li>Suppliers</li>\n  <li>or it might not be mentioned who exactly</li>\n</ul>\n\nYou can use any NLP-Method on the text of those statements and try to predict if the training is sufficiently addressed or not.</p>\n\n<h2>Evaluation</h2>\n<p>From the 981 data points of the data set 288 are used for evaluating your models. These 288 statements are evenly split into a public and a private leaderboard. For this binary classification problem we're evaluating with binary cross entropy/ logloss:</p>\n\n<p>\n\\[\n\\text{score} = - \\frac{1}{N}\\sum_{i=1}^{N} \\left[ y_{i} \\log(\\hat{y}_{i}) + (1 - y_{i}) \\log(1 - \\hat{y}_{i})\\right]\n\\]\n</p>\n\n<h2>Submission Format</h2>\nYour task is to provide <b>probability estimates (NOT final classifications)</b> for the test set whether they classify as \"Training Provided\" or not. All statements in the test set have a unique ID. The predictions on the statements of the test set should be submitted in the following format as a comma separated csv file:\n<pre>\nID, LABEL_MSA_training_binary\n1, 0.0\n2, 0.0\n3, 0.0\n...\n</pre>\n\n<h2>Timeline</h2>\n<p>The hackathon starts on Wednesday, October 14th after the introduction session and we will finish on Wednesday, October 28th at 4:30 PM CEST (Register <a href=\"https://twimlai.com/twimlfest/sessions/ai-for-good-hackathon-end-modern-slavery/\" target=\"_blank\">here</a>). During TWIMLfest we will have a weekly checking, you can find the dates <a href=\"https://twimlai.com/twimlfest/sessions/ai-for-good-hackathon-end-modern-slavery/\" target=\"_blank\">here</a>.\n</p>\n\n<h2>Prizes and Karma</h2>\nThis Hackathon is intended to help the global research community improve their work on screening Modern Slavery Statements to hold global corporations accountable. As a small incentive there will be the following prizes:\n<ul>\n  <li>1. Place: $250</li>\n  <li>2. Place: $100</li>\n  <li>3. Place: $50</li>\n</ul>\n<p>The Prizes can be taken as credit for your favourite cloud provider or as a credit for TWIML swag. </p>\n<p>You are not allowed to use external data sources for this competition than what is provided through this page. Your solution should also be able to scale over tens of thousands of modern slavery statements so manually reading and classifying the statements does not qualify here. We ask you to be fair and not cheat. Cheating at an AI for Good Hackathon probably brings some pretty bad karma :-) As a winner you should be prepared to showcase your approach, share your code so it can be reproduced!</p>\n",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "rules",
      "content": "This is an AI for Good Hackathon, so please bear in mind that only authentic submissions drive AI for actual Good :)\n\nThe rules are:\n---\n- Manual statement classifications don't count. The winning solution should demonstrate an automated method classify thousands of novel documents as they become available. \n- Using external data sources specifically related to modern slavery statements benchmarking (e.g. from Wikirate) to leak into your model is not allowed. It will immediately become obvious when we try to reproduce your results.\n- Using pre-trained models trained on large corpora is allowed.\n- The winning solution must be open sourced.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    }
  ]
}