{
  "id": 315571,
  "title": "Product launches and relevant articles",
  "url": "/competitions/h-and-m-personalized-fashion-recommendations/discussion/315571",
  "author_name": "",
  "post_date": "2022-03-28T22:15:50.664702100Z",
  "votes": 10,
  "comment_count": 2,
  "views": 0,
  "content": "<p>I tried to narrow down the number of articles to consider for each prediction. I found that most of the time the articles sold in the last three weeks + the new articles of the week to be predicted should account for &gt;99% of all unit sales. So instead of considering 105k articles for a prediction it should normally be okay to only consider about 20k articles.</p>\n<p>More details here: <a href=\"https://www.kaggle.com/tbierhance/product-launches-and-relevant-articles\" target=\"_blank\">https://www.kaggle.com/tbierhance/product-launches-and-relevant-articles</a></p>",
  "messages": [
    {
      "id": "1737945",
      "postDate": "03/28/2022 22:15:50",
      "content": "<p>I tried to narrow down the number of articles to consider for each prediction. I found that most of the time the articles sold in the last three weeks + the new articles of the week to be predicted should account for &gt;99% of all unit sales. So instead of considering 105k articles for a prediction it should normally be okay to only consider about 20k articles.</p>\n<p>More details here: <a href=\"https://www.kaggle.com/tbierhance/product-launches-and-relevant-articles\" target=\"_blank\">https://www.kaggle.com/tbierhance/product-launches-and-relevant-articles</a></p>",
      "rawMarkdown": "I tried to narrow down the number of articles to consider for each prediction. I found that most of the time the articles sold in the last three weeks + the new articles of the week to be predicted should account for >99% of all unit sales. So instead of considering 105k articles for a prediction it should normally be okay to only consider about 20k articles.\n\nMore details here: https://www.kaggle.com/tbierhance/product-launches-and-relevant-articles",
      "votes": null
    },
    {
      "id": "1742157",
      "postDate": "04/01/2022 14:13:10",
      "content": "<p>Just added an update that also includes the same kind of analysis for customers. Compared to the articles it's harder to cut the customers down. Customers that bought something in the last 4 weeks are about 273k (20% of all customers) and account for 70% of the total sales in one week. Still this would be 20k * 273k = 5.4b records and I guess I need to get this further down.</p>",
      "rawMarkdown": "Just added an update that also includes the same kind of analysis for customers. Compared to the articles it's harder to cut the customers down. Customers that bought something in the last 4 weeks are about 273k (20% of all customers) and account for 70% of the total sales in one week. Still this would be 20k * 273k = 5.4b records and I guess I need to get this further down.",
      "votes": null
    },
    {
      "id": "1742752",
      "postDate": "04/02/2022 08:57:44",
      "content": "<p>Nice Analysis<br>\nIt may be better change our prediction model when predict different cluster (like bought something in the last 4 weeks or not)</p>",
      "rawMarkdown": "Nice Analysis\nIt may be better change our prediction model when predict different cluster (like bought something in the last 4 weeks or not)",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1742157,
      "author_name": "tbierhance",
      "author_url": "",
      "post_date": "04/01/2022 14:13:10",
      "content": "<p>Just added an update that also includes the same kind of analysis for customers. Compared to the articles it's harder to cut the customers down. Customers that bought something in the last 4 weeks are about 273k (20% of all customers) and account for 70% of the total sales in one week. Still this would be 20k * 273k = 5.4b records and I guess I need to get this further down.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1742752,
      "author_name": "deepkun1995",
      "author_url": "",
      "post_date": "04/02/2022 08:57:44",
      "content": "<p>Nice Analysis<br>\nIt may be better change our prediction model when predict different cluster (like bought something in the last 4 weeks or not)</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1737945": "I tried to narrow down the number of articles to consider for each prediction. I found that most of the time the articles sold in the last three weeks + the new articles of the week to be predicted should account for >99% of all unit sales. So instead of considering 105k articles for a prediction it should normally be okay to only consider about 20k articles.\n\nMore details here: https://www.kaggle.com/tbierhance/product-launches-and-relevant-articles",
    "1742157": "Just added an update that also includes the same kind of analysis for customers. Compared to the articles it's harder to cut the customers down. Customers that bought something in the last 4 weeks are about 273k (20% of all customers) and account for 70% of the total sales in one week. Still this would be 20k * 273k = 5.4b records and I guess I need to get this further down.",
    "1742752": "Nice Analysis\nIt may be better change our prediction model when predict different cluster (like bought something in the last 4 weeks or not)"
  },
  "source": "meta"
}