{
  "id": 522214,
  "title": "32nd place solution",
  "url": "/competitions/uspto-explainable-ai/discussion/522214",
  "author_name": "tuna",
  "post_date": "2024-07-25T01:30:16.305000",
  "votes": 12,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Thank you to the hosts and participants for hosting this competition! </p>\n<h2>Main Approach</h2>\n<p><code>OR</code> query consisting of single word(title) and cpc. </p>\n<h3>How to construct an <code>OR</code> query</h3>\n<p>I retrieved candidates of single word(title) and cpc.<br>\n　- single word(title): using TF-IDF top-100<br>\n　- cpc: all cpc with 50 documents</p>\n<p>I checked the relevance by running queries for this candidates.<br>\n　- ap@50<br>\n　- which of the 50 documents were hit</p>\n<p>I used Greedy Algorithm for construct an <code>OR</code> query<br>\n　- score using which of the 50 documents were hit</p>\n<p>(# of words: 24)</p>\n<h2>Tried Other Approaches</h2>\n<ul>\n<li>title2query with T5: <a href=\"https://www.kaggle.com/code/tinatuna/uspto-t5-based-doc2query-titles\" target=\"_blank\">https://www.kaggle.com/code/tinatuna/uspto-t5-based-doc2query-titles</a> (LB: 0.16~)</li>\n<li>using LPProblem: <a href=\"https://www.kaggle.com/code/tinatuna/uspto-using-lpproblem\" target=\"_blank\">https://www.kaggle.com/code/tinatuna/uspto-using-lpproblem</a> (LB: 0.34~)</li>\n<li>using BeamSearch: (I was unable to resolve Score Submission Error)</li>\n<li>using abstract (LB: 0.43)</li>\n</ul>",
  "messages": [
    {
      "id": 2935106,
      "postDate": "2024-07-25T01:30:16.307Z",
      "content": "<p>Thank you to the hosts and participants for hosting this competition! </p>\n<h2>Main Approach</h2>\n<p><code>OR</code> query consisting of single word(title) and cpc. </p>\n<h3>How to construct an <code>OR</code> query</h3>\n<p>I retrieved candidates of single word(title) and cpc.<br>\n　- single word(title): using TF-IDF top-100<br>\n　- cpc: all cpc with 50 documents</p>\n<p>I checked the relevance by running queries for this candidates.<br>\n　- ap@50<br>\n　- which of the 50 documents were hit</p>\n<p>I used Greedy Algorithm for construct an <code>OR</code> query<br>\n　- score using which of the 50 documents were hit</p>\n<p>(# of words: 24)</p>\n<h2>Tried Other Approaches</h2>\n<ul>\n<li>title2query with T5: <a href=\"https://www.kaggle.com/code/tinatuna/uspto-t5-based-doc2query-titles\" target=\"_blank\">https://www.kaggle.com/code/tinatuna/uspto-t5-based-doc2query-titles</a> (LB: 0.16~)</li>\n<li>using LPProblem: <a href=\"https://www.kaggle.com/code/tinatuna/uspto-using-lpproblem\" target=\"_blank\">https://www.kaggle.com/code/tinatuna/uspto-using-lpproblem</a> (LB: 0.34~)</li>\n<li>using BeamSearch: (I was unable to resolve Score Submission Error)</li>\n<li>using abstract (LB: 0.43)</li>\n</ul>",
      "rawMarkdown": "Thank you to the hosts and participants for hosting this competition! \n\n## Main Approach\n`OR` query consisting of single word(title) and cpc. \n\n### How to construct an `OR` query\nI retrieved candidates of single word(title) and cpc.\n　- single word(title): using TF-IDF top-100\n　- cpc: all cpc with 50 documents\n\nI checked the relevance by running queries for this candidates.\n　- ap@50\n　- which of the 50 documents were hit\n\nI used Greedy Algorithm for construct an `OR` query\n　- score using which of the 50 documents were hit\n\n(# of words: 24)\n\n## Tried Other Approaches\n- title2query with T5: https://www.kaggle.com/code/tinatuna/uspto-t5-based-doc2query-titles (LB: 0.16~)\n- using LPProblem: https://www.kaggle.com/code/tinatuna/uspto-using-lpproblem (LB: 0.34~)\n- using BeamSearch: (I was unable to resolve Score Submission Error)\n- using abstract (LB: 0.43)",
      "votes": 12
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "2935106": "Thank you to the hosts and participants for hosting this competition! \n\n## Main Approach\n`OR` query consisting of single word(title) and cpc. \n\n### How to construct an `OR` query\nI retrieved candidates of single word(title) and cpc.\n　- single word(title): using TF-IDF top-100\n　- cpc: all cpc with 50 documents\n\nI checked the relevance by running queries for this candidates.\n　- ap@50\n　- which of the 50 documents were hit\n\nI used Greedy Algorithm for construct an `OR` query\n　- score using which of the 50 documents were hit\n\n(# of words: 24)\n\n## Tried Other Approaches\n- title2query with T5: https://www.kaggle.com/code/tinatuna/uspto-t5-based-doc2query-titles (LB: 0.16~)\n- using LPProblem: https://www.kaggle.com/code/tinatuna/uspto-using-lpproblem (LB: 0.34~)\n- using BeamSearch: (I was unable to resolve Score Submission Error)\n- using abstract (LB: 0.43)"
  }
}