{
  "id": 13990,
  "title": "Trainbids",
  "url": "/competitions/facebook-recruiting-iv-human-or-bot/discussion/13990",
  "author_name": "",
  "post_date": "2015-05-04T15:06:42.750Z",
  "votes": 3,
  "comment_count": 12,
  "views": 3396,
  "content": "<p>Is it possible that some of the human bidders in trainbids can be bot ?</p>",
  "messages": [
    {
      "id": "76607",
      "postDate": "05/04/2015 15:06:42",
      "content": "<p>Is it possible that some of the human bidders in trainbids can be bot ?</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "76703",
      "postDate": "05/04/2015 21:33:27",
      "content": "<p>I would be surprised if it is not the case.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "78015",
      "postDate": "05/12/2015 07:30:33",
      "content": "<p>I am absolutely convinced that there are bots among the training data. In the attached file I plotted the time history (x-axis) of two 'humans' (according to the training set) vs the auctions they are involved in (hashed to ints).&nbsp;The bids are extremely frequent and all over the place, I doubt that a human can do that. What do you think?</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "78040",
      "postDate": "05/12/2015 10:45:34",
      "content": "<p>[quote=thorstenkurth;78015]</p>\n<p>I am absolutely convinced that there are bots among the training data. In the attached file I plotted the time history (x-axis) of two 'humans' (according to the training set) vs the auctions they are involved in (hashed to ints).&nbsp;The bids are extremely frequent and all over the place, I doubt that a human can do that. What do you think?</p>\n<p>[/quote]</p>\n\n<p>I got to the same conclusion, for example,&nbsp;bidder 'f5b2bbad20d1d7ded3ed960393bec0f40u6hn' has 5 bids at the same time stamp from 5 different devices&nbsp;on&nbsp;auctions. I don't think that's humanly &nbsp;possible unless: 1) time stamp is&nbsp;at a higher level of grain than &quot;second&quot; &nbsp;as I'm assuming it is or 2) one can&nbsp;schedule something to bid on its&nbsp;behalf that is not really considered a bot?</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "78067",
      "postDate": "05/12/2015 13:55:49",
      "content": "<p>I have few thoughts about that:</p>\n<p>1) humans may share same account (a crowd acts like&nbsp;a bot)<br>2) there may be &quot;good&quot; bots - refer to eBay's autobidding (http://pages.ebay.com/help/buy/bidding-overview.html#auto)<br>3) the data might be noisy (whatever it&nbsp;means) &nbsp;and yes, we have to deal with all the errors</p>\n<p>4) it's easier to think in abstract terms like &quot;this user is either 1 or 0&quot; instead of &quot;bot or human&quot;</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "78197",
      "postDate": "05/13/2015 12:50:27",
      "content": "<p>I think that we are trying to 'learn' someone's bot detection algorithm - instead of really identifying bots/humans from hand-picked cases.</p>\n<p>Think of it this way. If I'm right, then it should be possible to get 1 auc score by implementing the exact same algorithm..</p>\n<p>:)</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "78237",
      "postDate": "05/13/2015 16:54:16",
      "content": "<p>[quote=CihanBaran;78197]</p>\n<p>I think that we are trying to 'learn' someone's bot detection algorithm - instead of really identifying bots/humans from hand-picked cases.</p>\n<p>Think of it this way. If I'm right, then it should be possible to get 1 auc score by implementing the exact same algorithm..</p>\n<p>:)</p>\n<p>[/quote]</p>\n<p>But we just have a few labeled users, if its detection algorithm has been trained with a lot of labeled data it is going to be impossible to get 1 auc score.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "78256",
      "postDate": "05/13/2015 17:41:30",
      "content": "<p>&quot;It is a human&quot; is right 94.87% of the time.</p>\n<p>Skew stew.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "78453",
      "postDate": "05/15/2015 01:06:24",
      "content": "<p>[quote=CihanBaran;78197]</p>\n<p>I think that we are trying to 'learn' someone's bot detection algorithm - instead of really identifying bots/humans from hand-picked cases.</p>\n<p>Think of it this way. If I'm right, then it should be possible to get 1 auc score by implementing the exact same algorithm..</p>\n<p>:)</p>\n<p>[/quote]<br><br>Except you don't have the same training sample!</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "78586",
      "postDate": "05/15/2015 22:41:37",
      "content": "<p>A few things to clarify. The results were half hand labeled, half stats-based. There are two types of &quot;bots&quot; with different levels of proof:</p>\n<p>1. Bidders who are identified as bots/fraudulent&nbsp;with clear proof.&nbsp;Their accounts were banned by the auction site.&nbsp;</p>\n<p>2. Bidder&nbsp;who may have just started their business/clicks&nbsp;or their stats exceed&nbsp;from system wide average. There are no clear proof that they are bots.&nbsp;</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "78590",
      "postDate": "05/15/2015 23:04:59",
      "content": "<p>What about humans in training set? Was there any proof that they're humans? Or humans in training set are those who wasn't marked as robots during labeling phase (both hand- and stats-based)?</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "78625",
      "postDate": "05/16/2015 10:27:48",
      "content": "<p>Thanks for the info Wendy. Can you clarify what you mean by &quot;Bidder who may have just started their business/clicks&quot;?&nbsp;</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "78677",
      "postDate": "05/16/2015 17:02:07",
      "content": "<p>@mario86: Humans are the ones that aren't marked as robots.&nbsp;</p>\n<p>@Jared: We don't really know. My interpretation is that the evil data scientists&nbsp;behind the bots just got started with their evil bidding activities, so their accounts aren't banned yet.&nbsp;</p>\n\n<p>Since neither Kaggle or Facebook collected this dataset, our knowledge isn't very thorough, so I'm going to stop answering these detailed data source questions. Please focus on making every use of the data &quot;as is&quot;. Thanks and good luck!!</p>",
      "rawMarkdown": "",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 76703,
      "author_name": "andreskull",
      "author_url": "",
      "post_date": "05/04/2015 21:33:27",
      "content": "<p>I would be surprised if it is not the case.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 78015,
      "author_name": "tkurth",
      "author_url": "",
      "post_date": "05/12/2015 07:30:33",
      "content": "<p>I am absolutely convinced that there are bots among the training data. In the attached file I plotted the time history (x-axis) of two 'humans' (according to the training set) vs the auctions they are involved in (hashed to ints).&nbsp;The bids are extremely frequent and all over the place, I doubt that a human can do that. What do you think?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 78040,
      "author_name": "dmenin",
      "author_url": "",
      "post_date": "05/12/2015 10:45:34",
      "content": "<p>[quote=thorstenkurth;78015]</p>\n<p>I am absolutely convinced that there are bots among the training data. In the attached file I plotted the time history (x-axis) of two 'humans' (according to the training set) vs the auctions they are involved in (hashed to ints).&nbsp;The bids are extremely frequent and all over the place, I doubt that a human can do that. What do you think?</p>\n<p>[/quote]</p>\n\n<p>I got to the same conclusion, for example,&nbsp;bidder 'f5b2bbad20d1d7ded3ed960393bec0f40u6hn' has 5 bids at the same time stamp from 5 different devices&nbsp;on&nbsp;auctions. I don't think that's humanly &nbsp;possible unless: 1) time stamp is&nbsp;at a higher level of grain than &quot;second&quot; &nbsp;as I'm assuming it is or 2) one can&nbsp;schedule something to bid on its&nbsp;behalf that is not really considered a bot?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 78067,
      "author_name": "zaborskiy",
      "author_url": "",
      "post_date": "05/12/2015 13:55:49",
      "content": "<p>I have few thoughts about that:</p>\n<p>1) humans may share same account (a crowd acts like&nbsp;a bot)<br>2) there may be &quot;good&quot; bots - refer to eBay's autobidding (http://pages.ebay.com/help/buy/bidding-overview.html#auto)<br>3) the data might be noisy (whatever it&nbsp;means) &nbsp;and yes, we have to deal with all the errors</p>\n<p>4) it's easier to think in abstract terms like &quot;this user is either 1 or 0&quot; instead of &quot;bot or human&quot;</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 78197,
      "author_name": "cihanb",
      "author_url": "",
      "post_date": "05/13/2015 12:50:27",
      "content": "<p>I think that we are trying to 'learn' someone's bot detection algorithm - instead of really identifying bots/humans from hand-picked cases.</p>\n<p>Think of it this way. If I'm right, then it should be possible to get 1 auc score by implementing the exact same algorithm..</p>\n<p>:)</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 78237,
      "author_name": "robertod",
      "author_url": "",
      "post_date": "05/13/2015 16:54:16",
      "content": "<p>[quote=CihanBaran;78197]</p>\n<p>I think that we are trying to 'learn' someone's bot detection algorithm - instead of really identifying bots/humans from hand-picked cases.</p>\n<p>Think of it this way. If I'm right, then it should be possible to get 1 auc score by implementing the exact same algorithm..</p>\n<p>:)</p>\n<p>[/quote]</p>\n<p>But we just have a few labeled users, if its detection algorithm has been trained with a lot of labeled data it is going to be impossible to get 1 auc score.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 78256,
      "author_name": "charlesdavis",
      "author_url": "",
      "post_date": "05/13/2015 17:41:30",
      "content": "<p>&quot;It is a human&quot; is right 94.87% of the time.</p>\n<p>Skew stew.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 78453,
      "author_name": "ccphillippi",
      "author_url": "",
      "post_date": "05/15/2015 01:06:24",
      "content": "<p>[quote=CihanBaran;78197]</p>\n<p>I think that we are trying to 'learn' someone's bot detection algorithm - instead of really identifying bots/humans from hand-picked cases.</p>\n<p>Think of it this way. If I'm right, then it should be possible to get 1 auc score by implementing the exact same algorithm..</p>\n<p>:)</p>\n<p>[/quote]<br><br>Except you don't have the same training sample!</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 78586,
      "author_name": "wendykan",
      "author_url": "",
      "post_date": "05/15/2015 22:41:37",
      "content": "<p>A few things to clarify. The results were half hand labeled, half stats-based. There are two types of &quot;bots&quot; with different levels of proof:</p>\n<p>1. Bidders who are identified as bots/fraudulent&nbsp;with clear proof.&nbsp;Their accounts were banned by the auction site.&nbsp;</p>\n<p>2. Bidder&nbsp;who may have just started their business/clicks&nbsp;or their stats exceed&nbsp;from system wide average. There are no clear proof that they are bots.&nbsp;</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 78590,
      "author_name": "podlipensky",
      "author_url": "",
      "post_date": "05/15/2015 23:04:59",
      "content": "<p>What about humans in training set? Was there any proof that they're humans? Or humans in training set are those who wasn't marked as robots during labeling phase (both hand- and stats-based)?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 78625,
      "author_name": "jturkewitz",
      "author_url": "",
      "post_date": "05/16/2015 10:27:48",
      "content": "<p>Thanks for the info Wendy. Can you clarify what you mean by &quot;Bidder who may have just started their business/clicks&quot;?&nbsp;</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 78677,
      "author_name": "wendykan",
      "author_url": "",
      "post_date": "05/16/2015 17:02:07",
      "content": "<p>@mario86: Humans are the ones that aren't marked as robots.&nbsp;</p>\n<p>@Jared: We don't really know. My interpretation is that the evil data scientists&nbsp;behind the bots just got started with their evil bidding activities, so their accounts aren't banned yet.&nbsp;</p>\n\n<p>Since neither Kaggle or Facebook collected this dataset, our knowledge isn't very thorough, so I'm going to stop answering these detailed data source questions. Please focus on making every use of the data &quot;as is&quot;. Thanks and good luck!!</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "76607": "",
    "76703": "",
    "78015": "",
    "78040": "",
    "78067": "",
    "78197": "",
    "78237": "",
    "78256": "",
    "78453": "",
    "78586": "",
    "78590": "",
    "78625": "",
    "78677": ""
  },
  "source": "meta"
}