{
  "id": 70769,
  "title": "Is Topic Modeling still relevant?",
  "url": "/competitions/quora-insincere-questions-classification/discussion/70769",
  "author_name": "",
  "post_date": "2018-11-07T06:02:06.082795300Z",
  "votes": 1,
  "comment_count": 4,
  "views": 0,
  "content": "<p>Does anyone have thoughts on using LDA or other topic modeling type techniques for data like this? I've had success with it in the past. Now that word embeddings are well-developed, I wonder if/how it might still be helpful.</p>",
  "messages": [
    {
      "id": "416700",
      "postDate": "11/07/2018 06:02:06",
      "content": "<p>Does anyone have thoughts on using LDA or other topic modeling type techniques for data like this? I've had success with it in the past. Now that word embeddings are well-developed, I wonder if/how it might still be helpful.</p>",
      "rawMarkdown": "Does anyone have thoughts on using LDA or other topic modeling type techniques for data like this? I've had success with it in the past. Now that word embeddings are well-developed, I wonder if/how it might still be helpful.",
      "votes": null
    },
    {
      "id": "417146",
      "postDate": "11/07/2018 20:29:23",
      "content": "<p>No free lunch, right? </p>\n\n<p>My guess is that embeddings and RNNs significantly outperform LDA.   </p>",
      "rawMarkdown": "No free lunch, right? \n\nMy guess is that embeddings and RNNs significantly outperform LDA.",
      "votes": null
    },
    {
      "id": "417191",
      "postDate": "11/07/2018 23:15:43",
      "content": "<p>Right. There may be a place for it in an ensemble. Shhh, let us never speak of it again :)</p>",
      "rawMarkdown": "Right. There may be a place for it in an ensemble. Shhh, let us never speak of it again :)",
      "votes": null
    },
    {
      "id": "417270",
      "postDate": "11/08/2018 02:44:45",
      "content": "<p>I use LDA for clustering in this <a href=\"https://www.kaggle.com/konohayui/topic-modeling-on-quora-insincere-questions\">kernel</a>. Topics number may be a good feature for non NN models.</p>",
      "rawMarkdown": "I use LDA for clustering in this [kernel][1]. Topics number may be a good feature for non NN models.\n\n  [1]: https://www.kaggle.com/konohayui/topic-modeling-on-quora-insincere-questions",
      "votes": null
    },
    {
      "id": "417288",
      "postDate": "11/08/2018 03:12:35",
      "content": "<p>Great idea! Looking at your kernel, maybe t-SNE coordinates also.</p>",
      "rawMarkdown": "Great idea! Looking at your kernel, maybe t-SNE coordinates also.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 417146,
      "author_name": "mihaskalic",
      "author_url": "",
      "post_date": "11/07/2018 20:29:23",
      "content": "<p>No free lunch, right? </p>\n\n<p>My guess is that embeddings and RNNs significantly outperform LDA.   </p>",
      "votes": null,
      "replies": [
        {
          "id": 417191,
          "author_name": "jpmiller",
          "author_url": "",
          "post_date": "11/07/2018 23:15:43",
          "content": "<p>Right. There may be a place for it in an ensemble. Shhh, let us never speak of it again :)</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 417270,
      "author_name": "konohayui",
      "author_url": "",
      "post_date": "11/08/2018 02:44:45",
      "content": "<p>I use LDA for clustering in this <a href=\"https://www.kaggle.com/konohayui/topic-modeling-on-quora-insincere-questions\">kernel</a>. Topics number may be a good feature for non NN models.</p>",
      "votes": null,
      "replies": [
        {
          "id": 417288,
          "author_name": "jpmiller",
          "author_url": "",
          "post_date": "11/08/2018 03:12:35",
          "content": "<p>Great idea! Looking at your kernel, maybe t-SNE coordinates also.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "416700": "Does anyone have thoughts on using LDA or other topic modeling type techniques for data like this? I've had success with it in the past. Now that word embeddings are well-developed, I wonder if/how it might still be helpful.",
    "417146": "No free lunch, right? \n\nMy guess is that embeddings and RNNs significantly outperform LDA.",
    "417191": "Right. There may be a place for it in an ensemble. Shhh, let us never speak of it again :)",
    "417270": "I use LDA for clustering in this [kernel][1]. Topics number may be a good feature for non NN models.\n\n  [1]: https://www.kaggle.com/konohayui/topic-modeling-on-quora-insincere-questions",
    "417288": "Great idea! Looking at your kernel, maybe t-SNE coordinates also."
  },
  "source": "meta"
}