{
  "id": 71408,
  "title": "Counterintuitive Feature?",
  "url": "/competitions/quora-insincere-questions-classification/discussion/71408",
  "author_name": "JohnM",
  "post_date": "2018-11-13T12:51:32.179000",
  "votes": 17,
  "comment_count": 7,
  "views": 0,
  "content": "<p>Insincere questions are much more likely to start with \"Why\". Nearly 4 in 10 compared to 1 in 10 for sincere questions. \n<img src=\"https://i.imgur.com/71QlWsr.png\" alt=\"crosstab\"></p>",
  "messages": [
    {
      "id": 420316,
      "postDate": "2018-11-13T12:51:32.180Z",
      "content": "<p>Insincere questions are much more likely to start with \"Why\". Nearly 4 in 10 compared to 1 in 10 for sincere questions. \n<img src=\"https://i.imgur.com/71QlWsr.png\" alt=\"crosstab\"></p>",
      "rawMarkdown": "Insincere questions are much more likely to start with \"Why\". Nearly 4 in 10 compared to 1 in 10 for sincere questions. \n![crosstab][1]\n\n\n  [1]: https://i.imgur.com/71QlWsr.png",
      "votes": 17
    },
    {
      "id": 421192,
      "postDate": "2018-11-14T17:36:05.570Z",
      "content": "<p>This makes sense, since questions that start with \"why\" are likely to be rhetorical questions that are actually just stating a controversial idea. I found a few other counterintuitive features. For example, 17.4% of questions that end with a period are insincere (versus 6.2% for all questions). Also, insincere questions have a mean length of 19.6 words, versus a mean length of 14.1 words for sincere questions. Only 3.8% of questions under 10 words are insincere, and 18.1% of questions over 50 words are insincere.</p>",
      "rawMarkdown": "This makes sense, since questions that start with \"why\" are likely to be rhetorical questions that are actually just stating a controversial idea. I found a few other counterintuitive features. For example, 17.4% of questions that end with a period are insincere (versus 6.2% for all questions). Also, insincere questions have a mean length of 19.6 words, versus a mean length of 14.1 words for sincere questions. Only 3.8% of questions under 10 words are insincere, and 18.1% of questions over 50 words are insincere.",
      "votes": 8
    },
    {
      "id": 438754,
      "postDate": "2018-12-14T06:04:34.663Z",
      "content": "<p>I think this feature could be easy for RNNs to catch.</p>",
      "rawMarkdown": "I think this feature could be easy for RNNs to catch.",
      "votes": 2,
      "replies": [
        {
          "id": 440372,
          "postDate": "2018-12-17T13:15:08.603Z",
          "content": "<p>I agree it should. Have you found any kernel which could show this?</p>",
          "rawMarkdown": "I agree it should. Have you found any kernel which could show this?"
        }
      ]
    },
    {
      "id": 420563,
      "postDate": "2018-11-13T20:20:25.767Z",
      "content": "<p>Great and simple feature, thanks! Maybe interesting to look more at the word distributions across categories and in different parts of sentences. Reminds me of reading about similar types of results for sarcasm detection in tweets or such. Just cannot remember the exact material now, hmm...</p>",
      "rawMarkdown": "Great and simple feature, thanks! Maybe interesting to look more at the word distributions across categories and in different parts of sentences. Reminds me of reading about similar types of results for sarcasm detection in tweets or such. Just cannot remember the exact material now, hmm...",
      "votes": 2,
      "replies": [
        {
          "id": 420797,
          "postDate": "2018-11-14T06:07:10.330Z",
          "content": "<p>I seem to remember Professor Frink on the Simpsons had built a sarcasm detector.... Seriously though, that's a great parallel. I'm seeing a lot of relevant work on Github and elsewhere that seems worth looking into.</p>",
          "rawMarkdown": "I seem to remember Professor Frink on the Simpsons had built a sarcasm detector.... Seriously though, that's a great parallel. I'm seeing a lot of relevant work on Github and elsewhere that seems worth looking into."
        }
      ]
    },
    {
      "id": 421218,
      "postDate": "2018-11-14T18:32:37.747Z",
      "content": "<p>Great insights, also looking at some words involving particular topics/peoples might also help in finding these distributions.</p>",
      "rawMarkdown": "Great insights, also looking at some words involving particular topics/peoples might also help in finding these distributions."
    },
    {
      "id": 420914,
      "postDate": "2018-11-14T10:11:58.923Z",
      "rawMarkdown": "",
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 421192,
      "author_name": "Emily Boyajian",
      "author_url": "",
      "post_date": "2018-11-14T17:36:05.570000",
      "content": "<p>This makes sense, since questions that start with \"why\" are likely to be rhetorical questions that are actually just stating a controversial idea. I found a few other counterintuitive features. For example, 17.4% of questions that end with a period are insincere (versus 6.2% for all questions). Also, insincere questions have a mean length of 19.6 words, versus a mean length of 14.1 words for sincere questions. Only 3.8% of questions under 10 words are insincere, and 18.1% of questions over 50 words are insincere.</p>",
      "votes": 8,
      "replies": []
    },
    {
      "id": 438754,
      "author_name": "mr007rin",
      "author_url": "",
      "post_date": "2018-12-14T06:04:34.663000",
      "content": "<p>I think this feature could be easy for RNNs to catch.</p>",
      "votes": 2,
      "replies": [
        {
          "id": 440372,
          "author_name": "Andrzej Kuro",
          "author_url": "",
          "post_date": "2018-12-17T13:15:08.603000",
          "content": "<p>I agree it should. Have you found any kernel which could show this?</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 420563,
      "author_name": "averagemn",
      "author_url": "",
      "post_date": "2018-11-13T20:20:25.767000",
      "content": "<p>Great and simple feature, thanks! Maybe interesting to look more at the word distributions across categories and in different parts of sentences. Reminds me of reading about similar types of results for sarcasm detection in tweets or such. Just cannot remember the exact material now, hmm...</p>",
      "votes": 2,
      "replies": [
        {
          "id": 420797,
          "author_name": "JohnM",
          "author_url": "",
          "post_date": "2018-11-14T06:07:10.330000",
          "content": "<p>I seem to remember Professor Frink on the Simpsons had built a sarcasm detector.... Seriously though, that's a great parallel. I'm seeing a lot of relevant work on Github and elsewhere that seems worth looking into.</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 421218,
      "author_name": "Ankit Sati",
      "author_url": "",
      "post_date": "2018-11-14T18:32:37.747000",
      "content": "<p>Great insights, also looking at some words involving particular topics/peoples might also help in finding these distributions.</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 420914,
      "author_name": "",
      "author_url": "",
      "post_date": "2018-11-14T10:11:58.923000",
      "content": "",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "420316": "Insincere questions are much more likely to start with \"Why\". Nearly 4 in 10 compared to 1 in 10 for sincere questions. \n![crosstab][1]\n\n\n  [1]: https://i.imgur.com/71QlWsr.png",
    "421192": "This makes sense, since questions that start with \"why\" are likely to be rhetorical questions that are actually just stating a controversial idea. I found a few other counterintuitive features. For example, 17.4% of questions that end with a period are insincere (versus 6.2% for all questions). Also, insincere questions have a mean length of 19.6 words, versus a mean length of 14.1 words for sincere questions. Only 3.8% of questions under 10 words are insincere, and 18.1% of questions over 50 words are insincere.",
    "438754": "I think this feature could be easy for RNNs to catch.",
    "420563": "Great and simple feature, thanks! Maybe interesting to look more at the word distributions across categories and in different parts of sentences. Reminds me of reading about similar types of results for sarcasm detection in tweets or such. Just cannot remember the exact material now, hmm...",
    "421218": "Great insights, also looking at some words involving particular topics/peoples might also help in finding these distributions.",
    "420914": ""
  }
}