{
  "id": 190588,
  "title": "Questions and Lecture tags",
  "url": "/competitions/riiid-test-answer-prediction/discussion/190588",
  "author_name": "",
  "post_date": "2020-10-12T13:24:11.277638800Z",
  "votes": 5,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Any ideas on how to cluster the list of tags??<br>\nSo that we could make some reliable validation scheme.</p>",
  "messages": [
    {
      "id": "1047315",
      "postDate": "10/12/2020 13:24:11",
      "content": "<p>Any ideas on how to cluster the list of tags??<br>\nSo that we could make some reliable validation scheme.</p>",
      "rawMarkdown": "Any ideas on how to cluster the list of tags??\nSo that we could make some reliable validation scheme.",
      "votes": null
    },
    {
      "id": "1052590",
      "postDate": "10/18/2020 01:20:44",
      "content": "<p>You can use character level n-gram embeddings such as those used by <a href=\"https://fasttext.cc/\" target=\"_blank\">Facebook's fastText</a>. If you can uniquely map each tag to a single character, you can construct a 'word' whose 'letters' are the tags. <a href=\"https://www.kaggle.com/tuckerarrants/riiid-fasttext-embeddings\" target=\"_blank\">I've explored this</a> but am not quite ready to use these embeddings in a sequential model. </p>",
      "rawMarkdown": "You can use character level n-gram embeddings such as those used by [Facebook's fastText](https://fasttext.cc/). If you can uniquely map each tag to a single character, you can construct a 'word' whose 'letters' are the tags. [I've explored this](https://www.kaggle.com/tuckerarrants/riiid-fasttext-embeddings) but am not quite ready to use these embeddings in a sequential model.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1052590,
      "author_name": "tuckerarrants",
      "author_url": "",
      "post_date": "10/18/2020 01:20:44",
      "content": "<p>You can use character level n-gram embeddings such as those used by <a href=\"https://fasttext.cc/\" target=\"_blank\">Facebook's fastText</a>. If you can uniquely map each tag to a single character, you can construct a 'word' whose 'letters' are the tags. <a href=\"https://www.kaggle.com/tuckerarrants/riiid-fasttext-embeddings\" target=\"_blank\">I've explored this</a> but am not quite ready to use these embeddings in a sequential model. </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1047315": "Any ideas on how to cluster the list of tags??\nSo that we could make some reliable validation scheme.",
    "1052590": "You can use character level n-gram embeddings such as those used by [Facebook's fastText](https://fasttext.cc/). If you can uniquely map each tag to a single character, you can construct a 'word' whose 'letters' are the tags. [I've explored this](https://www.kaggle.com/tuckerarrants/riiid-fasttext-embeddings) but am not quite ready to use these embeddings in a sequential model."
  },
  "source": "meta"
}