{
  "id": 367870,
  "title": "We can try more NLP methods in sequence recomendation, like ....",
  "url": "/competitions/otto-recommender-system/discussion/367870",
  "author_name": "KKY",
  "post_date": "2022-11-22T13:55:00.387000",
  "votes": 10,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Since this is a user-anonymized dataset, it's very important for us to get good representation of the action/item sequence. So maybe some NLP ideas will help us for this competition:</p>\n<ol>\n<li>word2vec train item/action embedding,  average the item/action embedding sequence as the representation. then we can find the most similar item as recall candidates.</li>\n<li>MLM pretrained transformer,  then finetune.</li>\n<li>Generation Model like GPT-format, predicts the next n step as the prediction. </li>\n<li>Tfidf feature.</li>\n<li>…</li>\n</ol>",
  "messages": [
    {
      "id": 2039754,
      "postDate": "2022-11-22T13:55:00.387Z",
      "content": "<p>Since this is a user-anonymized dataset, it's very important for us to get good representation of the action/item sequence. So maybe some NLP ideas will help us for this competition:</p>\n<ol>\n<li>word2vec train item/action embedding,  average the item/action embedding sequence as the representation. then we can find the most similar item as recall candidates.</li>\n<li>MLM pretrained transformer,  then finetune.</li>\n<li>Generation Model like GPT-format, predicts the next n step as the prediction. </li>\n<li>Tfidf feature.</li>\n<li>…</li>\n</ol>",
      "rawMarkdown": "Since this is a user-anonymized dataset, it's very important for us to get good representation of the action/item sequence. So maybe some NLP ideas will help us for this competition:\n1. word2vec train item/action embedding,  average the item/action embedding sequence as the representation. then we can find the most similar item as recall candidates.\n2. MLM pretrained transformer,  then finetune.\n3. Generation Model like GPT-format, predicts the next n step as the prediction. \n4. Tfidf feature.\n5. ...",
      "votes": 10
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "2039754": "Since this is a user-anonymized dataset, it's very important for us to get good representation of the action/item sequence. So maybe some NLP ideas will help us for this competition:\n1. word2vec train item/action embedding,  average the item/action embedding sequence as the representation. then we can find the most similar item as recall candidates.\n2. MLM pretrained transformer,  then finetune.\n3. Generation Model like GPT-format, predicts the next n step as the prediction. \n4. Tfidf feature.\n5. ..."
  }
}