{
  "id": 324534,
  "title": "how do you use word2vec in this competition?",
  "url": "/competitions/h-and-m-personalized-fashion-recommendations/discussion/324534",
  "author_name": "Radek Osmulski",
  "post_date": "2022-05-12T06:56:54.468000",
  "votes": 5,
  "comment_count": 0,
  "views": 0,
  "content": "<p>First of all, huge congrats to top place finishers 🥳 And thank you very much for posting the write-ups, they are a wonderful read 🙂</p>\n<p>Could I please ask you how did you frame the problem to train a word2vec model? (this has been a component of many top submissions)</p>\n<p>My guess is that you used the Skipgram with Negative Sampling formulation, but how did you generate the dataset? Would items bought by a single customer fall into \"one sentence\" and constitute \"neighboring words\"?</p>\n<p>If a customer bought (in that chronological order) <em>x</em> <em>a</em> <em>b</em> <em>c</em> <em>y</em> and we used a window size of 3, for <em>b</em> would <em>a</em> and <em>c</em> be positive examples and <em>x</em> and <em>y</em> negative? Or did you use some other formulation?</p>\n<p>Also, what did you use to train this? Did you develop your own code and used Pytorch or Tensorflow or did you use an off the shelf package?</p>\n<p>Thanks so much for sharing your insights on this 🙂</p>",
  "messages": [
    {
      "id": 1785522,
      "postDate": "2022-05-12T06:56:54.470Z",
      "content": "<p>First of all, huge congrats to top place finishers 🥳 And thank you very much for posting the write-ups, they are a wonderful read 🙂</p>\n<p>Could I please ask you how did you frame the problem to train a word2vec model? (this has been a component of many top submissions)</p>\n<p>My guess is that you used the Skipgram with Negative Sampling formulation, but how did you generate the dataset? Would items bought by a single customer fall into \"one sentence\" and constitute \"neighboring words\"?</p>\n<p>If a customer bought (in that chronological order) <em>x</em> <em>a</em> <em>b</em> <em>c</em> <em>y</em> and we used a window size of 3, for <em>b</em> would <em>a</em> and <em>c</em> be positive examples and <em>x</em> and <em>y</em> negative? Or did you use some other formulation?</p>\n<p>Also, what did you use to train this? Did you develop your own code and used Pytorch or Tensorflow or did you use an off the shelf package?</p>\n<p>Thanks so much for sharing your insights on this 🙂</p>",
      "rawMarkdown": "First of all, huge congrats to top place finishers 🥳 And thank you very much for posting the write-ups, they are a wonderful read 🙂\n\nCould I please ask you how did you frame the problem to train a word2vec model? (this has been a component of many top submissions)\n\nMy guess is that you used the Skipgram with Negative Sampling formulation, but how did you generate the dataset? Would items bought by a single customer fall into \"one sentence\" and constitute \"neighboring words\"?\n\nIf a customer bought (in that chronological order) _x_ _a_ _b_ _c_ _y_ and we used a window size of 3, for _b_ would _a_ and _c_ be positive examples and _x_ and _y_ negative? Or did you use some other formulation?\n\nAlso, what did you use to train this? Did you develop your own code and used Pytorch or Tensorflow or did you use an off the shelf package?\n\nThanks so much for sharing your insights on this 🙂",
      "votes": 5
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "1785522": "First of all, huge congrats to top place finishers 🥳 And thank you very much for posting the write-ups, they are a wonderful read 🙂\n\nCould I please ask you how did you frame the problem to train a word2vec model? (this has been a component of many top submissions)\n\nMy guess is that you used the Skipgram with Negative Sampling formulation, but how did you generate the dataset? Would items bought by a single customer fall into \"one sentence\" and constitute \"neighboring words\"?\n\nIf a customer bought (in that chronological order) _x_ _a_ _b_ _c_ _y_ and we used a window size of 3, for _b_ would _a_ and _c_ be positive examples and _x_ and _y_ negative? Or did you use some other formulation?\n\nAlso, what did you use to train this? Did you develop your own code and used Pytorch or Tensorflow or did you use an off the shelf package?\n\nThanks so much for sharing your insights on this 🙂"
  }
}