{
  "id": 350643,
  "title": "Only a single user from test set also in training set",
  "url": "/competitions/goodreads-books-reviews-290312/discussion/350643",
  "author_name": "vedyakm",
  "post_date": "2022-09-06T15:05:32.250000",
  "votes": 0,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Hi,</p>\n<p>I was excited to look into this competition because I'm interested in link prediction problems like Netflix, MovieLens, Yahoo Music, or Goodreads. However, after looking into the data, the training test split seems strange.</p>\n<ul>\n<li>Number of users in training data <code>#USER_TRAINING</code>: 12188</li>\n<li>Number of users in test data <code>#USER_TEST</code>: 6705</li>\n<li>Number of users in test data but not in training data <code>#USER_TRAINING diff #USER_TEST</code>: 6704</li>\n</ul>\n<p>The overlap between the test data and the training data is a single user.</p>\n<p>This basically rules out collaborative filtering because there is no data basis to identify similar users or getting ratings for similar items for these cold start users.  This basically leaves a regression problem where the user information might as well not exist.</p>\n<p>Is this intentional or is this a mistake with the train/test-split?</p>\n<p>Best regards</p>",
  "messages": [
    {
      "id": 1928682,
      "postDate": "2022-09-06T15:05:32.250Z",
      "content": "<p>Hi,</p>\n<p>I was excited to look into this competition because I'm interested in link prediction problems like Netflix, MovieLens, Yahoo Music, or Goodreads. However, after looking into the data, the training test split seems strange.</p>\n<ul>\n<li>Number of users in training data <code>#USER_TRAINING</code>: 12188</li>\n<li>Number of users in test data <code>#USER_TEST</code>: 6705</li>\n<li>Number of users in test data but not in training data <code>#USER_TRAINING diff #USER_TEST</code>: 6704</li>\n</ul>\n<p>The overlap between the test data and the training data is a single user.</p>\n<p>This basically rules out collaborative filtering because there is no data basis to identify similar users or getting ratings for similar items for these cold start users.  This basically leaves a regression problem where the user information might as well not exist.</p>\n<p>Is this intentional or is this a mistake with the train/test-split?</p>\n<p>Best regards</p>",
      "rawMarkdown": "Hi,\n\nI was excited to look into this competition because I'm interested in link prediction problems like Netflix, MovieLens, Yahoo Music, or Goodreads. However, after looking into the data, the training test split seems strange.\n\n* Number of users in training data `#USER_TRAINING`: 12188\n* Number of users in test data `#USER_TEST`: 6705\n* Number of users in test data but not in training data `#USER_TRAINING diff #USER_TEST`: 6704\n\nThe overlap between the test data and the training data is a single user.\n\nThis basically rules out collaborative filtering because there is no data basis to identify similar users or getting ratings for similar items for these cold start users.  This basically leaves a regression problem where the user information might as well not exist.\n\nIs this intentional or is this a mistake with the train/test-split?\n\nBest regards"
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "1928682": "Hi,\n\nI was excited to look into this competition because I'm interested in link prediction problems like Netflix, MovieLens, Yahoo Music, or Goodreads. However, after looking into the data, the training test split seems strange.\n\n* Number of users in training data `#USER_TRAINING`: 12188\n* Number of users in test data `#USER_TEST`: 6705\n* Number of users in test data but not in training data `#USER_TRAINING diff #USER_TEST`: 6704\n\nThe overlap between the test data and the training data is a single user.\n\nThis basically rules out collaborative filtering because there is no data basis to identify similar users or getting ratings for similar items for these cold start users.  This basically leaves a regression problem where the user information might as well not exist.\n\nIs this intentional or is this a mistake with the train/test-split?\n\nBest regards"
  }
}