{
  "id": 56117,
  "title": "How to extract feature from user_id",
  "url": "/competitions/avito-demand-prediction/discussion/56117",
  "author_name": "",
  "post_date": "2018-05-06T11:17:37.099845100Z",
  "votes": null,
  "comment_count": 6,
  "views": 0,
  "content": "<p>Hi guys,</p>\n\n<p>I have try hash encoding user_id and k-means cluster for user_id, but these not work will in my solution.</p>\n\n<p>Would you like share some other trick method for extracting features from user_id ?</p>",
  "messages": [
    {
      "id": "323836",
      "postDate": "05/06/2018 11:17:37",
      "content": "<p>Hi guys,</p>\n\n<p>I have try hash encoding user_id and k-means cluster for user_id, but these not work will in my solution.</p>\n\n<p>Would you like share some other trick method for extracting features from user_id ?</p>",
      "rawMarkdown": "Hi guys,\n\nI have try hash encoding user_id and k-means cluster for user_id, but these not work will in my solution.\n\nWould you like share some other trick method for extracting features from user_id ?",
      "votes": null
    },
    {
      "id": "323837",
      "postDate": "05/06/2018 11:19:39",
      "content": "<p>Or embedding with deep learning ?</p>",
      "rawMarkdown": "Or embedding with deep learning ?",
      "votes": null
    },
    {
      "id": "324050",
      "postDate": "05/07/2018 02:56:20",
      "content": "<p>I tried to add user_id as embedding, but it makes my model worse. So far, I think user_id is an unused feature for my model.</p>",
      "rawMarkdown": "I tried to add user_id as embedding, but it makes my model worse. So far, I think user_id is an unused feature for my model.",
      "votes": null
    },
    {
      "id": "324054",
      "postDate": "05/07/2018 03:15:11",
      "content": "<p>I don't think you can use it directly, but you can aggregate with it. Maybe try things like:</p>\n\n<ul>\n<li>How many items has this user posted?</li>\n<li>How frequently does this user post?</li>\n<li>How long has it been since this user last posted?</li>\n</ul>\n\n<p>Not all of these ideas have worked for me, but some of these ideas have given me a combined 0.0002 boost in my leaderboard placement.</p>",
      "rawMarkdown": "I don't think you can use it directly, but you can aggregate with it. Maybe try things like:\n\n * How many items has this user posted?\n * How frequently does this user post?\n * How long has it been since this user last posted?\n\nNot all of these ideas have worked for me, but some of these ideas have given me a combined 0.0002 boost in my leaderboard placement.",
      "votes": null
    },
    {
      "id": "324284",
      "postDate": "05/07/2018 14:24:08",
      "content": "<p>I extract UV for each categorical columns, it improve my model.</p>",
      "rawMarkdown": "I extract UV for each categorical columns, it improve my model.",
      "votes": null
    },
    {
      "id": "324450",
      "postDate": "05/07/2018 17:55:25",
      "content": "<p>My suspicion was that this wouldn't be possible because it was only over an 8-day period, far too little to provide any signal, I thought.  Anyways, I will try these feature; thanks for sharing.</p>",
      "rawMarkdown": "My suspicion was that this wouldn't be possible because it was only over an 8-day period, far too little to provide any signal, I thought.  Anyways, I will try these feature; thanks for sharing.",
      "votes": null
    },
    {
      "id": "324916",
      "postDate": "05/08/2018 01:37:45",
      "content": "<p>And I'm trying  cat2vec at user_id colum, I haven't know it effect</p>",
      "rawMarkdown": "And I'm trying  cat2vec at user_id colum, I haven't know it effect",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 323837,
      "author_name": "classtag",
      "author_url": "",
      "post_date": "05/06/2018 11:19:39",
      "content": "<p>Or embedding with deep learning ?</p>",
      "votes": null,
      "replies": [
        {
          "id": 324050,
          "author_name": "ngxbac",
          "author_url": "",
          "post_date": "05/07/2018 02:56:20",
          "content": "<p>I tried to add user_id as embedding, but it makes my model worse. So far, I think user_id is an unused feature for my model.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 324054,
      "author_name": "peterhurford",
      "author_url": "",
      "post_date": "05/07/2018 03:15:11",
      "content": "<p>I don't think you can use it directly, but you can aggregate with it. Maybe try things like:</p>\n\n<ul>\n<li>How many items has this user posted?</li>\n<li>How frequently does this user post?</li>\n<li>How long has it been since this user last posted?</li>\n</ul>\n\n<p>Not all of these ideas have worked for me, but some of these ideas have given me a combined 0.0002 boost in my leaderboard placement.</p>",
      "votes": null,
      "replies": [
        {
          "id": 324284,
          "author_name": "classtag",
          "author_url": "",
          "post_date": "05/07/2018 14:24:08",
          "content": "<p>I extract UV for each categorical columns, it improve my model.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 324450,
          "author_name": "matthewa313",
          "author_url": "",
          "post_date": "05/07/2018 17:55:25",
          "content": "<p>My suspicion was that this wouldn't be possible because it was only over an 8-day period, far too little to provide any signal, I thought.  Anyways, I will try these feature; thanks for sharing.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 324916,
          "author_name": "classtag",
          "author_url": "",
          "post_date": "05/08/2018 01:37:45",
          "content": "<p>And I'm trying  cat2vec at user_id colum, I haven't know it effect</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "323836": "Hi guys,\n\nI have try hash encoding user_id and k-means cluster for user_id, but these not work will in my solution.\n\nWould you like share some other trick method for extracting features from user_id ?",
    "323837": "Or embedding with deep learning ?",
    "324050": "I tried to add user_id as embedding, but it makes my model worse. So far, I think user_id is an unused feature for my model.",
    "324054": "I don't think you can use it directly, but you can aggregate with it. Maybe try things like:\n\n * How many items has this user posted?\n * How frequently does this user post?\n * How long has it been since this user last posted?\n\nNot all of these ideas have worked for me, but some of these ideas have given me a combined 0.0002 boost in my leaderboard placement.",
    "324284": "I extract UV for each categorical columns, it improve my model.",
    "324450": "My suspicion was that this wouldn't be possible because it was only over an 8-day period, far too little to provide any signal, I thought.  Anyways, I will try these feature; thanks for sharing.",
    "324916": "And I'm trying  cat2vec at user_id colum, I haven't know it effect"
  },
  "source": "meta"
}