{
  "id": 76409,
  "title": "What kind of post-processing we can do?",
  "url": "/competitions/quora-insincere-questions-classification/discussion/76409",
  "author_name": "",
  "post_date": "2019-01-02T15:03:06.398397200Z",
  "votes": 2,
  "comment_count": 2,
  "views": 0,
  "content": "<p>I'm a newbie in ML and I tried to find some papers about post-processing in text tasks, but I couldn't find anything.\nMaybe somebody could help with this and give a direction?</p>",
  "messages": [
    {
      "id": "449029",
      "postDate": "01/02/2019 15:03:06",
      "content": "<p>I'm a newbie in ML and I tried to find some papers about post-processing in text tasks, but I couldn't find anything.\nMaybe somebody could help with this and give a direction?</p>",
      "rawMarkdown": "I'm a newbie in ML and I tried to find some papers about post-processing in text tasks, but I couldn't find anything.\nMaybe somebody could help with this and give a direction?",
      "votes": null
    },
    {
      "id": "450099",
      "postDate": "01/04/2019 10:00:05",
      "content": "<p>Agreed, most of them concentrate on pre-processing like cleaning, tokenizing and stemming. <br>\nThen comes the main processing which is usually embedding and attention. \nWhere do we land then?</p>",
      "rawMarkdown": "Agreed, most of them concentrate on pre-processing like cleaning, tokenizing and stemming.  \nThen comes the main processing which is usually embedding and attention. \nWhere do we land then?",
      "votes": null
    },
    {
      "id": "450144",
      "postDate": "01/04/2019 11:47:04",
      "content": "<p>Once we feed our data into the model and get the prediction (which is what we want), it is then not clear what is the purpose of post-processing, if any. </p>\n\n<p>For other NLP tasks like <strong>text generation</strong>, it may make sense to do the post processing e.g. grammar correction.</p>\n\n<p>Here, perhaps what we can do after getting predictions is to do some kind of post error analysis?? (see the common errors and rethink how to improve) --- if you agree, in that case my kernel may be of little help.</p>",
      "rawMarkdown": "Once we feed our data into the model and get the prediction (which is what we want), it is then not clear what is the purpose of post-processing, if any. \n\nFor other NLP tasks like **text generation**, it may make sense to do the post processing e.g. grammar correction.\n\nHere, perhaps what we can do after getting predictions is to do some kind of post error analysis?? (see the common errors and rethink how to improve) --- if you agree, in that case my kernel may be of little help.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 450099,
      "author_name": "akuropatwinski",
      "author_url": "",
      "post_date": "01/04/2019 10:00:05",
      "content": "<p>Agreed, most of them concentrate on pre-processing like cleaning, tokenizing and stemming. <br>\nThen comes the main processing which is usually embedding and attention. \nWhere do we land then?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 450144,
      "author_name": "ratthachat",
      "author_url": "",
      "post_date": "01/04/2019 11:47:04",
      "content": "<p>Once we feed our data into the model and get the prediction (which is what we want), it is then not clear what is the purpose of post-processing, if any. </p>\n\n<p>For other NLP tasks like <strong>text generation</strong>, it may make sense to do the post processing e.g. grammar correction.</p>\n\n<p>Here, perhaps what we can do after getting predictions is to do some kind of post error analysis?? (see the common errors and rethink how to improve) --- if you agree, in that case my kernel may be of little help.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "449029": "I'm a newbie in ML and I tried to find some papers about post-processing in text tasks, but I couldn't find anything.\nMaybe somebody could help with this and give a direction?",
    "450099": "Agreed, most of them concentrate on pre-processing like cleaning, tokenizing and stemming.  \nThen comes the main processing which is usually embedding and attention. \nWhere do we land then?",
    "450144": "Once we feed our data into the model and get the prediction (which is what we want), it is then not clear what is the purpose of post-processing, if any. \n\nFor other NLP tasks like **text generation**, it may make sense to do the post processing e.g. grammar correction.\n\nHere, perhaps what we can do after getting predictions is to do some kind of post error analysis?? (see the common errors and rethink how to improve) --- if you agree, in that case my kernel may be of little help."
  },
  "source": "meta"
}