{
  "id": 308917,
  "title": "Embedding the products",
  "url": "/competitions/h-and-m-personalized-fashion-recommendations/discussion/308917",
  "author_name": "",
  "post_date": "2022-02-20T23:04:40.756593100Z",
  "votes": 21,
  "comment_count": 1,
  "views": 0,
  "content": "<p>One part of the data that hasn't received a lot of attention (so far, anyway ;-) seems to be the product themselves - despite the fact there is a quite a lot of information there. Depending on how you look at it, you can come up with a representation from different angles:</p>\n<ul>\n<li>based on the purchase history</li>\n<li>using the images</li>\n<li>using text info (the detailed description)</li>\n</ul>\n<p>If think having diverse embeddings might come in real handy: some products do not have images, while others might only have limited purchase history (hence methods like collaborative filtering might run into a cold start problem)).</p>\n<p>The notebook with (ongoing) analysis is here: <a href=\"https://www.kaggle.com/konradb/product-embeddings\" target=\"_blank\">https://www.kaggle.com/konradb/product-embeddings</a></p>\n<p>If you want to start playing with the dataset, have fun: <a href=\"https://www.kaggle.com/konradb/embedding-the-products\" target=\"_blank\">https://www.kaggle.com/konradb/embedding-the-products</a></p>\n<p>Looking forward to feedback on this one - atm there is a small version for embeddings based on product history, I want to handle images next, then text descriptions (and perhaps the category structure as well).</p>",
  "messages": [
    {
      "id": "1699094",
      "postDate": "02/20/2022 23:04:40",
      "content": "<p>One part of the data that hasn't received a lot of attention (so far, anyway ;-) seems to be the product themselves - despite the fact there is a quite a lot of information there. Depending on how you look at it, you can come up with a representation from different angles:</p>\n<ul>\n<li>based on the purchase history</li>\n<li>using the images</li>\n<li>using text info (the detailed description)</li>\n</ul>\n<p>If think having diverse embeddings might come in real handy: some products do not have images, while others might only have limited purchase history (hence methods like collaborative filtering might run into a cold start problem)).</p>\n<p>The notebook with (ongoing) analysis is here: <a href=\"https://www.kaggle.com/konradb/product-embeddings\" target=\"_blank\">https://www.kaggle.com/konradb/product-embeddings</a></p>\n<p>If you want to start playing with the dataset, have fun: <a href=\"https://www.kaggle.com/konradb/embedding-the-products\" target=\"_blank\">https://www.kaggle.com/konradb/embedding-the-products</a></p>\n<p>Looking forward to feedback on this one - atm there is a small version for embeddings based on product history, I want to handle images next, then text descriptions (and perhaps the category structure as well).</p>",
      "rawMarkdown": "One part of the data that hasn't received a lot of attention (so far, anyway ;-) seems to be the product themselves - despite the fact there is a quite a lot of information there. Depending on how you look at it, you can come up with a representation from different angles:\n\n* based on the purchase history\n* using the images\n* using text info (the detailed description)\n\nIf think having diverse embeddings might come in real handy: some products do not have images, while others might only have limited purchase history (hence methods like collaborative filtering might run into a cold start problem)).\n\nThe notebook with (ongoing) analysis is here: https://www.kaggle.com/konradb/product-embeddings\n\nIf you want to start playing with the dataset, have fun: https://www.kaggle.com/konradb/embedding-the-products\n\nLooking forward to feedback on this one - atm there is a small version for embeddings based on product history, I want to handle images next, then text descriptions (and perhaps the category structure as well).",
      "votes": null
    },
    {
      "id": "2168787",
      "postDate": "03/04/2023 14:12:10",
      "content": "<p>Is there any follow-up operation?</p>",
      "rawMarkdown": "Is there any follow-up operation?",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2168787,
      "author_name": "liuwenshuo",
      "author_url": "",
      "post_date": "03/04/2023 14:12:10",
      "content": "<p>Is there any follow-up operation?</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1699094": "One part of the data that hasn't received a lot of attention (so far, anyway ;-) seems to be the product themselves - despite the fact there is a quite a lot of information there. Depending on how you look at it, you can come up with a representation from different angles:\n\n* based on the purchase history\n* using the images\n* using text info (the detailed description)\n\nIf think having diverse embeddings might come in real handy: some products do not have images, while others might only have limited purchase history (hence methods like collaborative filtering might run into a cold start problem)).\n\nThe notebook with (ongoing) analysis is here: https://www.kaggle.com/konradb/product-embeddings\n\nIf you want to start playing with the dataset, have fun: https://www.kaggle.com/konradb/embedding-the-products\n\nLooking forward to feedback on this one - atm there is a small version for embeddings based on product history, I want to handle images next, then text descriptions (and perhaps the category structure as well).",
    "2168787": "Is there any follow-up operation?"
  },
  "source": "meta"
}