{
  "id": 9745,
  "title": "Use of is_proved and close_hours",
  "url": "/competitions/avito-prohibited-content/discussion/9745",
  "author_name": "",
  "post_date": "2014-07-14T16:29:11.390Z",
  "votes": null,
  "comment_count": 2,
  "views": 1547,
  "content": "<p>The fields is_proved and close_hours are available in the training data only. &nbsp;According to the description of the fields, they&nbsp;could potentially indicate records that&nbsp;are more or less likely to actually contain illicit content&nbsp;(more in case of is_proved + is_blocked, less in case of high number of close_hours +&nbsp;not is_blocked).</p>\n<p>How could these be useful&nbsp;since we are not expected to&nbsp;predict actually illicit ads&nbsp;but rather the is_blocked field, which means whether or not the illicit ad is detected as such by moderators?</p>\n<p>Or will the private leaderboard&nbsp;be based on ads&nbsp;that are proven to actually contain or not contain illicit content?</p>",
  "messages": [
    {
      "id": "50478",
      "postDate": "07/14/2014 16:29:11",
      "content": "<p>The fields is_proved and close_hours are available in the training data only. &nbsp;According to the description of the fields, they&nbsp;could potentially indicate records that&nbsp;are more or less likely to actually contain illicit content&nbsp;(more in case of is_proved + is_blocked, less in case of high number of close_hours +&nbsp;not is_blocked).</p>\n<p>How could these be useful&nbsp;since we are not expected to&nbsp;predict actually illicit ads&nbsp;but rather the is_blocked field, which means whether or not the illicit ad is detected as such by moderators?</p>\n<p>Or will the private leaderboard&nbsp;be based on ads&nbsp;that are proven to actually contain or not contain illicit content?</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "50479",
      "postDate": "07/14/2014 16:35:28",
      "content": "<p>You can use these to calculate *how* illicit an advertisement is, instead of just binary &quot;is&quot; or &quot;is not&quot;.</p>\n<p>Then instead of training a classification model on 0's and 1's, you may train a regression model on an &quot;illicit score&quot; ranging between 0 and (for example) 75.</p>\n<p>Sometimes the predictions from your regression model are more accurate in ranking illicit / not-illicit content, because it uses more information than just the binary labels.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "51806",
      "postDate": "08/10/2014 02:00:58",
      "content": "<p>Since there is an error in the close hours as discussed here:&nbsp;http://www.kaggle.com/c/avito-prohibited-content/forums/t/9729/close-hours-value-range</p>\n<p>which&nbsp;haven't been fixed yet, they are basically useless from my point of view.</p>",
      "rawMarkdown": "",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 50479,
      "author_name": "triskelion",
      "author_url": "",
      "post_date": "07/14/2014 16:35:28",
      "content": "<p>You can use these to calculate *how* illicit an advertisement is, instead of just binary &quot;is&quot; or &quot;is not&quot;.</p>\n<p>Then instead of training a classification model on 0's and 1's, you may train a regression model on an &quot;illicit score&quot; ranging between 0 and (for example) 75.</p>\n<p>Sometimes the predictions from your regression model are more accurate in ranking illicit / not-illicit content, because it uses more information than just the binary labels.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 51806,
      "author_name": "chenglongchen",
      "author_url": "",
      "post_date": "08/10/2014 02:00:58",
      "content": "<p>Since there is an error in the close hours as discussed here:&nbsp;http://www.kaggle.com/c/avito-prohibited-content/forums/t/9729/close-hours-value-range</p>\n<p>which&nbsp;haven't been fixed yet, they are basically useless from my point of view.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "50478": "",
    "50479": "",
    "51806": ""
  },
  "source": "meta"
}