{
  "id": 596975,
  "title": "dealing with large group sizes",
  "url": "/competitions/aeroclub-recsys-2025/discussion/596975",
  "author_name": "",
  "post_date": "2025-08-05T23:24:38.659214800Z",
  "votes": null,
  "comment_count": 1,
  "views": 0,
  "content": "<p>has anyone been able to share some insights on how to deal with the large group sizes? looking at the hitrate@3 from the example notebooks it has been dismal. ive tried negative sampling and mining and running a pre-classifier to downsample the group sizes to the ranker but the performance seems to be worse than if i just feed the entire dataset to the ltr ranker</p>",
  "messages": [
    {
      "id": "3263952",
      "postDate": "08/05/2025 23:24:38",
      "content": "<p>has anyone been able to share some insights on how to deal with the large group sizes? looking at the hitrate@3 from the example notebooks it has been dismal. ive tried negative sampling and mining and running a pre-classifier to downsample the group sizes to the ranker but the performance seems to be worse than if i just feed the entire dataset to the ltr ranker</p>",
      "rawMarkdown": "has anyone been able to share some insights on how to deal with the large group sizes? looking at the hitrate@3 from the example notebooks it has been dismal. ive tried negative sampling and mining and running a pre-classifier to downsample the group sizes to the ranker but the performance seems to be worse than if i just feed the entire dataset to the ltr ranker",
      "votes": null
    },
    {
      "id": "3265509",
      "postDate": "08/07/2025 15:43:50",
      "content": "<p>That is one of the most interesting question in the competition. Hope we'll continue to discover it very soon 😊</p>",
      "rawMarkdown": "That is one of the most interesting question in the competition. Hope we'll continue to discover it very soon 😊",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3265509,
      "author_name": "samvelkoch",
      "author_url": "",
      "post_date": "08/07/2025 15:43:50",
      "content": "<p>That is one of the most interesting question in the competition. Hope we'll continue to discover it very soon 😊</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3263952": "has anyone been able to share some insights on how to deal with the large group sizes? looking at the hitrate@3 from the example notebooks it has been dismal. ive tried negative sampling and mining and running a pre-classifier to downsample the group sizes to the ranker but the performance seems to be worse than if i just feed the entire dataset to the ltr ranker",
    "3265509": "That is one of the most interesting question in the competition. Hope we'll continue to discover it very soon 😊"
  },
  "source": "meta"
}