{
  "id": 314511,
  "title": "Serve top 12 recommendations or only serve above a certain threshold?",
  "url": "/competitions/h-and-m-personalized-fashion-recommendations/discussion/314511",
  "author_name": "",
  "post_date": "2022-03-23T00:48:44.786807200Z",
  "votes": -2,
  "comment_count": 5,
  "views": 0,
  "content": "<p>What would get better results: serve the top 12 recommendations per customer regardless of the score, or only serve a recommendation if purchase confidence is high?</p>",
  "messages": [
    {
      "id": "1732055",
      "postDate": "03/23/2022 00:48:44",
      "content": "<p>What would get better results: serve the top 12 recommendations per customer regardless of the score, or only serve a recommendation if purchase confidence is high?</p>",
      "rawMarkdown": "What would get better results: serve the top 12 recommendations per customer regardless of the score, or only serve a recommendation if purchase confidence is high?",
      "votes": null
    },
    {
      "id": "1732068",
      "postDate": "03/23/2022 01:12:52",
      "content": "<p>Hi <a href=\"https://www.kaggle.com/coenvalk\" target=\"_blank\">@coenvalk</a> </p>\n<p>it would be better to serve 12 recs rather than less. because there is no penalty to predict 12 recommendations per customer when the cust only buy 11 item or less.</p>\n<p>if you serve more than 12, the submission only counts the first 12 recs </p>",
      "rawMarkdown": "Hi @coenvalk \n\nit would be better to serve 12 recs rather than less. because there is no penalty to predict 12 recommendations per customer when the cust only buy 11 item or less.\n\nif you serve more than 12, the submission only counts the first 12 recs",
      "votes": null
    },
    {
      "id": "1732070",
      "postDate": "03/23/2022 01:15:06",
      "content": "<p>Thanks, I'll serve the top 12 recs then!</p>",
      "rawMarkdown": "Thanks, I'll serve the top 12 recs then!",
      "votes": null
    },
    {
      "id": "1733063",
      "postDate": "03/24/2022 01:24:20",
      "content": "<p>We want to submit 12 recommendations for each user. Another strategy is for your model to suggest more than 12 (for example 24), then add additional logic or model which removes (and/or rearranges order) some recommendations from those 24 and submits the top 12 to Kaggle.</p>",
      "rawMarkdown": "We want to submit 12 recommendations for each user. Another strategy is for your model to suggest more than 12 (for example 24), then add additional logic or model which removes (and/or rearranges order) some recommendations from those 24 and submits the top 12 to Kaggle.",
      "votes": null
    },
    {
      "id": "1734000",
      "postDate": "03/24/2022 21:14:07",
      "content": "<p>From the transactions dataset os possible to see that there are customers that  buy only one item  and others that buy more than 12. Does this metric make sense? If a customer prediction os only 1 it 2 itens does it make any sense to predict other values just to fill the 12? </p>",
      "rawMarkdown": "From the transactions dataset os possible to see that there are customers that  buy only one item  and others that buy more than 12. Does this metric make sense? If a customer prediction os only 1 it 2 itens does it make any sense to predict other values just to fill the 12?",
      "votes": null
    },
    {
      "id": "1734002",
      "postDate": "03/24/2022 21:21:43",
      "content": "<p>yes, because if a customer buys only 1 item and you predict 12, then you increase the chance of getting at least one of them right.</p>",
      "rawMarkdown": "yes, because if a customer buys only 1 item and you predict 12, then you increase the chance of getting at least one of them right.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1732068,
      "author_name": "hervind",
      "author_url": "",
      "post_date": "03/23/2022 01:12:52",
      "content": "<p>Hi <a href=\"https://www.kaggle.com/coenvalk\" target=\"_blank\">@coenvalk</a> </p>\n<p>it would be better to serve 12 recs rather than less. because there is no penalty to predict 12 recommendations per customer when the cust only buy 11 item or less.</p>\n<p>if you serve more than 12, the submission only counts the first 12 recs </p>",
      "votes": null,
      "replies": [
        {
          "id": 1732070,
          "author_name": "coenvalk",
          "author_url": "",
          "post_date": "03/23/2022 01:15:06",
          "content": "<p>Thanks, I'll serve the top 12 recs then!</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1733063,
      "author_name": "cdeotte",
      "author_url": "",
      "post_date": "03/24/2022 01:24:20",
      "content": "<p>We want to submit 12 recommendations for each user. Another strategy is for your model to suggest more than 12 (for example 24), then add additional logic or model which removes (and/or rearranges order) some recommendations from those 24 and submits the top 12 to Kaggle.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1734000,
      "author_name": "eduardocatarino",
      "author_url": "",
      "post_date": "03/24/2022 21:14:07",
      "content": "<p>From the transactions dataset os possible to see that there are customers that  buy only one item  and others that buy more than 12. Does this metric make sense? If a customer prediction os only 1 it 2 itens does it make any sense to predict other values just to fill the 12? </p>",
      "votes": null,
      "replies": [
        {
          "id": 1734002,
          "author_name": "atulverma",
          "author_url": "",
          "post_date": "03/24/2022 21:21:43",
          "content": "<p>yes, because if a customer buys only 1 item and you predict 12, then you increase the chance of getting at least one of them right.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1732055": "What would get better results: serve the top 12 recommendations per customer regardless of the score, or only serve a recommendation if purchase confidence is high?",
    "1732068": "Hi @coenvalk \n\nit would be better to serve 12 recs rather than less. because there is no penalty to predict 12 recommendations per customer when the cust only buy 11 item or less.\n\nif you serve more than 12, the submission only counts the first 12 recs",
    "1732070": "Thanks, I'll serve the top 12 recs then!",
    "1733063": "We want to submit 12 recommendations for each user. Another strategy is for your model to suggest more than 12 (for example 24), then add additional logic or model which removes (and/or rearranges order) some recommendations from those 24 and submits the top 12 to Kaggle.",
    "1734000": "From the transactions dataset os possible to see that there are customers that  buy only one item  and others that buy more than 12. Does this metric make sense? If a customer prediction os only 1 it 2 itens does it make any sense to predict other values just to fill the 12?",
    "1734002": "yes, because if a customer buys only 1 item and you predict 12, then you increase the chance of getting at least one of them right."
  },
  "source": "meta"
}