{
  "id": 447103,
  "title": "333rd place...sharing my findings",
  "content": "<p>Firstly, huge thanks for organizing this competition and everyone who has shared their ideas in the notebooks and discussion forums!</p>\n<p><strong>Approach</strong></p>\n<p>My solution is built upon <a href=\"https://www.kaggle.com/tsunotsuno\" target=\"_blank\">@tsunotsuno</a>'s effective idea of stacking Debertav3base with LGBM. I mostly played around with adding more nlp features and tweaking with the model.</p>\n<p><strong>What worked</strong></p>\n<p>Freezing the first 4 encoder layers gave the best results<br>\nUsing CLS token output for regression was effective<br>\nAdding the 6 features from <a href=\"https://www.kaggle.com/competitions/feedback-prize-english-language-learning/overview\" target=\"_blank\">FB3 commonlit competition</a> gave a good boost<br>\nAdding textstat features such as <code>automated_readability_index, smog_index, dale_chall_readability_score, gunning_fog, reading_time, difficult_words</code> showed improvement<br>\nAddding cosine similarity between prompt text and text embeddings also helped to a small extent<br>\nRemoving <code>length_ratio</code> also showed small improvement</p>\n<p><strong>What didn't work</strong></p>\n<p>Freezing embedding layers deteriorated performance<br>\nMean pooling/ GeM pooling<br>\nAutocorrect didn't improve score<br>\nTried other models such as Roberta, Electra which were not as good as Debertav3base. Even ensembling Debertav3base with them couldn't help<br>\nTraining separate models for each target was not effective. In fact, content scores improved when trained together with wording scores</p>\n<p>I wanted to include <code>prompt_text</code> and also try using Debertav3large, however it didn't turn out well this time :(</p>\n<p>Although I couldn't earn a medal, my learnings through this competition have been immense and I'm very happy to have come this far. I've already learnt a lot from the teams who have published their solutions and I look forward to participating in the next competitions!</p>",
  "comments": [],
  "comment_count": 0,
  "messages": [
    {
      "id": 2481682,
      "postDate": "2023-10-14T11:19:49.983Z",
      "content": "<p>Firstly, huge thanks for organizing this competition and everyone who has shared their ideas in the notebooks and discussion forums!</p>\n<p><strong>Approach</strong></p>\n<p>My solution is built upon <a href=\"https://www.kaggle.com/tsunotsuno\" target=\"_blank\">@tsunotsuno</a>'s effective idea of stacking Debertav3base with LGBM. I mostly played around with adding more nlp features and tweaking with the model.</p>\n<p><strong>What worked</strong></p>\n<p>Freezing the first 4 encoder layers gave the best results<br>\nUsing CLS token output for regression was effective<br>\nAdding the 6 features from <a href=\"https://www.kaggle.com/competitions/feedback-prize-english-language-learning/overview\" target=\"_blank\">FB3 commonlit competition</a> gave a good boost<br>\nAdding textstat features such as <code>automated_readability_index, smog_index, dale_chall_readability_score, gunning_fog, reading_time, difficult_words</code> showed improvement<br>\nAddding cosine similarity between prompt text and text embeddings also helped to a small extent<br>\nRemoving <code>length_ratio</code> also showed small improvement</p>\n<p><strong>What didn't work</strong></p>\n<p>Freezing embedding layers deteriorated performance<br>\nMean pooling/ GeM pooling<br>\nAutocorrect didn't improve score<br>\nTried other models such as Roberta, Electra which were not as good as Debertav3base. Even ensembling Debertav3base with them couldn't help<br>\nTraining separate models for each target was not effective. In fact, content scores improved when trained together with wording scores</p>\n<p>I wanted to include <code>prompt_text</code> and also try using Debertav3large, however it didn't turn out well this time :(</p>\n<p>Although I couldn't earn a medal, my learnings through this competition have been immense and I'm very happy to have come this far. I've already learnt a lot from the teams who have published their solutions and I look forward to participating in the next competitions!</p>",
      "rawMarkdown": "Firstly, huge thanks for organizing this competition and everyone who has shared their ideas in the notebooks and discussion forums!\n\n**Approach**\n\nMy solution is built upon @tsunotsuno's effective idea of stacking Debertav3base with LGBM. I mostly played around with adding more nlp features and tweaking with the model.\n\n**What worked**\n\nFreezing the first 4 encoder layers gave the best results\nUsing CLS token output for regression was effective\nAdding the 6 features from [FB3 commonlit competition](https://www.kaggle.com/competitions/feedback-prize-english-language-learning/overview) gave a good boost\nAdding textstat features such as `automated_readability_index, smog_index, dale_chall_readability_score, gunning_fog, reading_time, difficult_words` showed improvement\nAddding cosine similarity between prompt text and text embeddings also helped to a small extent\nRemoving `length_ratio` also showed small improvement\n\n**What didn't work**\n\nFreezing embedding layers deteriorated performance\nMean pooling/ GeM pooling\nAutocorrect didn't improve score\nTried other models such as Roberta, Electra which were not as good as Debertav3base. Even ensembling Debertav3base with them couldn't help\nTraining separate models for each target was not effective. In fact, content scores improved when trained together with wording scores\n\nI wanted to include `prompt_text` and also try using Debertav3large, however it didn't turn out well this time :(\n\nAlthough I couldn't earn a medal, my learnings through this competition have been immense and I'm very happy to have come this far. I've already learnt a lot from the teams who have published their solutions and I look forward to participating in the next competitions!",
      "votes": 3
    }
  ],
  "complete": true,
  "source": "live",
  "fetched_at": "2026-10-19T23:06:20Z",
  "competition_id": 53482,
  "competition_slug": "commonlit-evaluate-student-summaries",
  "team_id": 10681638,
  "solution_url": "/competitions/commonlit-evaluate-student-summaries/writeups/rush24-333rd-place-sharing-my-findings"
}