{
  "id": 1881,
  "title": "Anybody can give some hints on how to model the new users?",
  "url": "/competitions/kddcup2012-track2/discussion/1881",
  "author_name": "",
  "post_date": "2012-05-09T11:18:21.930Z",
  "votes": null,
  "comment_count": 6,
  "views": 7628,
  "content": "<p>Anybody can give some hints on how to model the new users that are in the test set but not the training set?</p>\r\n<p>Many thanks in advance!</p>",
  "messages": [
    {
      "id": "10893",
      "postDate": "05/09/2012 11:18:21",
      "content": "<p>Anybody can give some hints on how to model the new users that are in the test set but not the training set?</p>\r\n<p>Many thanks in advance!</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "10894",
      "postDate": "05/09/2012 12:16:48",
      "content": "",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "10909",
      "postDate": "05/10/2012 03:43:14",
      "content": "<p>You can predict them using the average rating of users as it is done in the benchmark dataset. Using raw averages would overfit bad, so you can use some maximum likelihood estimator (log likelihood) to define the mean value.\r\n</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "10916",
      "postDate": "05/10/2012 12:34:33",
      "content": "<p>Hi&nbsp;<a class=\"x_x_profilelink\" title=\"View Leustagos's profile\" href=\"http://www.kddcup2012.org/users/24266/leustagos\">Leustagos</a>,</p>\r\n<p>Thanks for your reply!~ Sorry for that I didn't fully get your advice.</p>\r\n<p>I only know It's possible to estimate the user click-throught rate (#click/#impression&nbsp;) for the common users between training set and test set, but how to do this for the&nbsp;uncommon users, thanks a lot!</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "11024",
      "postDate": "05/14/2012 19:28:58",
      "content": "<p>using raw averages overfits badly. </p>\r\n<p>=right, btw, what do you mean by shrinking?</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "11038",
      "postDate": "05/15/2012 12:12:17",
      "content": "<p>http://en.wikipedia.org/wiki/Shrinkage_(statistics)</p>\r\n<p>Basicly it is to use a maximum likelihood estimator to minimize the negative log of a function. Try to look for resources on the web. In a quick search I found this paper: http://mercury.bio.uaf.edu/courses/wlf625/readings/MLEstimation.PDF<br>\r\nBut there ar many others out there!</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "55140",
      "postDate": "09/24/2014 13:32:13",
      "content": "<p>[quote=Leustagos;11038]</p>\n<p>http://en.wikipedia.org/wiki/Shrinkage_(statistics)</p>\n<p>Basicly it is to use a maximum likelihood estimator to minimize the negative log of a function. Try to look for resources on the web. In a quick search I found this paper: http://mercury.bio.uaf.edu/courses/wlf625/readings/MLEstimation.PDF<br> But there ar many others out there!</p>\n<p>[/quote]</p>\n\n<p>verrrrry old post, but :) .... wondering about this advice for estimating new users. Isnt the MLE of a Bernoulli rv just the sample proportion?</p>",
      "rawMarkdown": "",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 10894,
      "author_name": "",
      "author_url": "",
      "post_date": "05/09/2012 12:16:48",
      "content": "",
      "votes": null,
      "replies": []
    },
    {
      "id": 10909,
      "author_name": "leustagos",
      "author_url": "",
      "post_date": "05/10/2012 03:43:14",
      "content": "<p>You can predict them using the average rating of users as it is done in the benchmark dataset. Using raw averages would overfit bad, so you can use some maximum likelihood estimator (log likelihood) to define the mean value.\r\n</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 10916,
      "author_name": "sidav40256",
      "author_url": "",
      "post_date": "05/10/2012 12:34:33",
      "content": "<p>Hi&nbsp;<a class=\"x_x_profilelink\" title=\"View Leustagos's profile\" href=\"http://www.kddcup2012.org/users/24266/leustagos\">Leustagos</a>,</p>\r\n<p>Thanks for your reply!~ Sorry for that I didn't fully get your advice.</p>\r\n<p>I only know It's possible to estimate the user click-throught rate (#click/#impression&nbsp;) for the common users between training set and test set, but how to do this for the&nbsp;uncommon users, thanks a lot!</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 11024,
      "author_name": "sidav40256",
      "author_url": "",
      "post_date": "05/14/2012 19:28:58",
      "content": "<p>using raw averages overfits badly. </p>\r\n<p>=right, btw, what do you mean by shrinking?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 11038,
      "author_name": "leustagos",
      "author_url": "",
      "post_date": "05/15/2012 12:12:17",
      "content": "<p>http://en.wikipedia.org/wiki/Shrinkage_(statistics)</p>\r\n<p>Basicly it is to use a maximum likelihood estimator to minimize the negative log of a function. Try to look for resources on the web. In a quick search I found this paper: http://mercury.bio.uaf.edu/courses/wlf625/readings/MLEstimation.PDF<br>\r\nBut there ar many others out there!</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 55140,
      "author_name": "inspector",
      "author_url": "",
      "post_date": "09/24/2014 13:32:13",
      "content": "<p>[quote=Leustagos;11038]</p>\n<p>http://en.wikipedia.org/wiki/Shrinkage_(statistics)</p>\n<p>Basicly it is to use a maximum likelihood estimator to minimize the negative log of a function. Try to look for resources on the web. In a quick search I found this paper: http://mercury.bio.uaf.edu/courses/wlf625/readings/MLEstimation.PDF<br> But there ar many others out there!</p>\n<p>[/quote]</p>\n\n<p>verrrrry old post, but :) .... wondering about this advice for estimating new users. Isnt the MLE of a Bernoulli rv just the sample proportion?</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "10893": "",
    "10894": "",
    "10909": "",
    "10916": "",
    "11024": "",
    "11038": "",
    "55140": ""
  },
  "source": "meta"
}