{
  "id": 15005,
  "title": "Your Machine-Learning-Algorithm ...",
  "url": "/competitions/avito-context-ad-clicks/discussion/15005",
  "author_name": "",
  "post_date": "2015-07-02T13:23:16.913Z",
  "votes": null,
  "comment_count": 2,
  "views": 1012,
  "content": "<p>Hi there,</p>\n<p>there is little discussion about the mathematics/algorithms for that competition so far, so I wonder if I am missing something.&nbsp;</p>\n<p>My approach is to build a features for each ad-impression. This includes features about the ad (e.g. price-range), about the search (e.g. day of week), but also about the user (e.g. click-ratio on ads). The last point will be the hard one as it means to build up a user-history.</p>\n<p>I will ignore any search-params, -queries, etc in order to avoid issues with the russion char-set. Also I assume the search is good enough so that ads will always be a good match to the search.</p>\n<p>On that features-set I will try different ML-algorithms. I guess I have to sample at some point because my hardware will not be able to handle that data.</p>\n<p>So what kind of (other) aproaches do you follow?</p>\n<p>Regards Olli</p>",
  "messages": [
    {
      "id": "83224",
      "postDate": "07/02/2015 13:23:16",
      "content": "<p>Hi there,</p>\n<p>there is little discussion about the mathematics/algorithms for that competition so far, so I wonder if I am missing something.&nbsp;</p>\n<p>My approach is to build a features for each ad-impression. This includes features about the ad (e.g. price-range), about the search (e.g. day of week), but also about the user (e.g. click-ratio on ads). The last point will be the hard one as it means to build up a user-history.</p>\n<p>I will ignore any search-params, -queries, etc in order to avoid issues with the russion char-set. Also I assume the search is good enough so that ads will always be a good match to the search.</p>\n<p>On that features-set I will try different ML-algorithms. I guess I have to sample at some point because my hardware will not be able to handle that data.</p>\n<p>So what kind of (other) aproaches do you follow?</p>\n<p>Regards Olli</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "83239",
      "postDate": "07/02/2015 14:55:39",
      "content": "<p>I guess you described the generic approach taken here...</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "83396",
      "postDate": "07/04/2015 06:09:01",
      "content": "<p>[quote=Oliver Meyfarth;83224]</p>\n<p>about the search (e.g. day of week)</p>\n<p>[/quote]</p>\n<p>Just in case, May 1 - May 4 and May 9 - May 11 were public holidays in Russia. Lot of&nbsp;people took their vacation on May 5 - May 8 as well.</p>",
      "rawMarkdown": "",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 83239,
      "author_name": "leustagos",
      "author_url": "",
      "post_date": "07/02/2015 14:55:39",
      "content": "<p>I guess you described the generic approach taken here...</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 83396,
      "author_name": "nightbat",
      "author_url": "",
      "post_date": "07/04/2015 06:09:01",
      "content": "<p>[quote=Oliver Meyfarth;83224]</p>\n<p>about the search (e.g. day of week)</p>\n<p>[/quote]</p>\n<p>Just in case, May 1 - May 4 and May 9 - May 11 were public holidays in Russia. Lot of&nbsp;people took their vacation on May 5 - May 8 as well.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "83224": "",
    "83239": "",
    "83396": ""
  },
  "source": "meta"
}