{
  "id": 6334,
  "title": "Dwell time of a click at the end of session log",
  "url": "/competitions/yandex-personalized-web-search-challenge/discussion/6334",
  "author_name": "",
  "post_date": "2013-11-19T05:18:43.280Z",
  "votes": null,
  "comment_count": 5,
  "views": 1474,
  "content": "<p>From the evaluation page, dwell time is defined as follow:</p>\n<p style=\"padding-left: 30px\">Dwell time is the time passed between the click on the document and the next click or the next query.</p>\n<p>Based on my understanding, the dwell time of a click is computed as&nbsp;</p>\n<p style=\"padding-left: 30px\">timePassed of the next action - timePassed of the click</p>\n<p>What happened if the click is at the end of session (last action of the session)? How do you determine whether the url is not relevant, relevant or highly relevant?</p>\n<p>Thank you.</p>",
  "messages": [
    {
      "id": "34875",
      "postDate": "11/19/2013 05:18:43",
      "content": "<p>From the evaluation page, dwell time is defined as follow:</p>\n<p style=\"padding-left: 30px\">Dwell time is the time passed between the click on the document and the next click or the next query.</p>\n<p>Based on my understanding, the dwell time of a click is computed as&nbsp;</p>\n<p style=\"padding-left: 30px\">timePassed of the next action - timePassed of the click</p>\n<p>What happened if the click is at the end of session (last action of the session)? How do you determine whether the url is not relevant, relevant or highly relevant?</p>\n<p>Thank you.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "34882",
      "postDate": "11/19/2013 07:44:52",
      "content": "<p>Hello,</p>\n<p>As defined on the&nbsp;<a href=\"https://www.kaggle.com/c/yandex-personalized-web-search-challenge/details/evaluation\">evaluation page</a>,&nbsp;the `highly relevant` grade corresponds to documents with clicks and dwell time not shorter than 400 time units and to the documents with the last click in the search session.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "34884",
      "postDate": "11/19/2013 08:28:07",
      "content": "<p>&quot;...&nbsp;and to the documents with the last click in the search session.&quot;</p>\n<p>Oh, I missed that part. Thank you very much for clarification.</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "34895",
      "postDate": "11/19/2013 12:07:52",
      "content": "<p>What about this situation&#65306; the click c1 of query q1 has another query q2 follows it, but user didn't click any page in q2, then c1 becomes the last click of this session ?</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "34914",
      "postDate": "11/19/2013 17:04:26",
      "content": "<p>Sorry for the ambiguity in the description. We'll fix it asap.</p>\n<p>The relevance level of 2 is assigned to the clicks which have dwell time not smaller than 400 time units and to the clicks that are&nbsp;<strong>the last actions in the corresponding sessions.</strong></p>\n<p>In other words, if the dwell time can be calculated then the relevance is assigned according to it. If the click is the last action in the session (in other words, the next action is performed after a long time window separating two sessions) then the relevance is set to 2.</p>\n<p>We believe this definition is more consistent.&nbsp;</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "34927",
      "postDate": "11/19/2013 18:42:36",
      "content": "<p>Thanks for raising that question ...&nbsp;</p>\n<p>I labelled all these clicks as not relevant. I hope fixing this bug in my code will improve my results :-)</p>\n<p>&nbsp;</p>",
      "rawMarkdown": "",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 34882,
      "author_name": "eugene1751",
      "author_url": "",
      "post_date": "11/19/2013 07:44:52",
      "content": "<p>Hello,</p>\n<p>As defined on the&nbsp;<a href=\"https://www.kaggle.com/c/yandex-personalized-web-search-challenge/details/evaluation\">evaluation page</a>,&nbsp;the `highly relevant` grade corresponds to documents with clicks and dwell time not shorter than 400 time units and to the documents with the last click in the search session.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 34884,
      "author_name": "philipskokoh",
      "author_url": "",
      "post_date": "11/19/2013 08:28:07",
      "content": "<p>&quot;...&nbsp;and to the documents with the last click in the search session.&quot;</p>\n<p>Oh, I missed that part. Thank you very much for clarification.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 34895,
      "author_name": "wangjing1",
      "author_url": "",
      "post_date": "11/19/2013 12:07:52",
      "content": "<p>What about this situation&#65306; the click c1 of query q1 has another query q2 follows it, but user didn't click any page in q2, then c1 becomes the last click of this session ?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 34914,
      "author_name": "eugene1751",
      "author_url": "",
      "post_date": "11/19/2013 17:04:26",
      "content": "<p>Sorry for the ambiguity in the description. We'll fix it asap.</p>\n<p>The relevance level of 2 is assigned to the clicks which have dwell time not smaller than 400 time units and to the clicks that are&nbsp;<strong>the last actions in the corresponding sessions.</strong></p>\n<p>In other words, if the dwell time can be calculated then the relevance is assigned according to it. If the click is the last action in the session (in other words, the next action is performed after a long time window separating two sessions) then the relevance is set to 2.</p>\n<p>We believe this definition is more consistent.&nbsp;</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 34927,
      "author_name": "quentinjerome",
      "author_url": "",
      "post_date": "11/19/2013 18:42:36",
      "content": "<p>Thanks for raising that question ...&nbsp;</p>\n<p>I labelled all these clicks as not relevant. I hope fixing this bug in my code will improve my results :-)</p>\n<p>&nbsp;</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "34875": "",
    "34882": "",
    "34884": "",
    "34895": "",
    "34914": "",
    "34927": ""
  },
  "source": "meta"
}