Has anyone tried BERTweet here? I'm going to try it, but was curious if others had tried already. It performs on par with RoBERTa in the Tweet Sentiment Extraction.
\n\nIt isn't multi-lingual (trained only on English Tweets), so you'll have to use the translated validation and test data. Twitter removes toxic tweets, so my intuition is that it won't perform well.
\n\nBERTweet paper:\nhttps://arxiv.org/abs/2005.10200\nhttps://github.com/VinAIResearch/BERTweet
\n\nSee discussion here: https://www.kaggle.com/c/tweet-sentiment-extraction/discussion/152861\nSee implementation in Tweet Sentiment Extraction here: https://www.kaggle.com/davelo12/bertweet (we can use internet in this competition, so loading the packages will be more straightforward)
", "messages": [ { "id": "872734", "postDate": "06/03/2020 13:09:58", "content": "Has anyone tried BERTweet here? I'm going to try it, but was curious if others had tried already. It performs on par with RoBERTa in the Tweet Sentiment Extraction.
\n\nIt isn't multi-lingual (trained only on English Tweets), so you'll have to use the translated validation and test data. Twitter removes toxic tweets, so my intuition is that it won't perform well.
\n\nBERTweet paper:\nhttps://arxiv.org/abs/2005.10200\nhttps://github.com/VinAIResearch/BERTweet
\n\nSee discussion here: https://www.kaggle.com/c/tweet-sentiment-extraction/discussion/152861\nSee implementation in Tweet Sentiment Extraction here: https://www.kaggle.com/davelo12/bertweet (we can use internet in this competition, so loading the packages will be more straightforward)
", "rawMarkdown": "Has anyone tried BERTweet here? I'm going to try it, but was curious if others had tried already. It performs on par with RoBERTa in the Tweet Sentiment Extraction. \n\nIt isn't multi-lingual (trained only on English Tweets), so you'll have to use the translated validation and test data. Twitter removes toxic tweets, so my intuition is that it won't perform well. \n\nBERTweet paper:\nhttps://arxiv.org/abs/2005.10200\nhttps://github.com/VinAIResearch/BERTweet\n\nSee discussion here: https://www.kaggle.com/c/tweet-sentiment-extraction/discussion/152861\nSee implementation in Tweet Sentiment Extraction here: https://www.kaggle.com/davelo12/bertweet (we can use internet in this competition, so loading the packages will be more straightforward)", "votes": null }, { "id": "875742", "postDate": "06/06/2020 06:09:17", "content": "Thank you for sharing. \nCould you tell me the result?
", "rawMarkdown": "Thank you for sharing. \nCould you tell me the result?", "votes": null }, { "id": "878362", "postDate": "06/08/2020 13:25:10", "content": "I haven't had success with BERTweet for this competition, but I'll let you know if something changes!
\n\nI haven't spent much time digging, but I think this goes back to BERTweet being trained on tweets, and Twitter removes toxic tweets from the platform (i.e., BERTweet would not see the same sort of toxic language we're trying to identify here).
", "rawMarkdown": "I haven't had success with BERTweet for this competition, but I'll let you know if something changes! \n\nI haven't spent much time digging, but I think this goes back to BERTweet being trained on tweets, and Twitter removes toxic tweets from the platform (i.e., BERTweet would not see the same sort of toxic language we're trying to identify here).", "votes": null } ], "comments": [ { "id": 875742, "author_name": "kannelliu", "author_url": "", "post_date": "06/06/2020 06:09:17", "content": "Thank you for sharing. \nCould you tell me the result?
", "votes": null, "replies": [ { "id": 878362, "author_name": "davelo12", "author_url": "", "post_date": "06/08/2020 13:25:10", "content": "I haven't had success with BERTweet for this competition, but I'll let you know if something changes!
\n\nI haven't spent much time digging, but I think this goes back to BERTweet being trained on tweets, and Twitter removes toxic tweets from the platform (i.e., BERTweet would not see the same sort of toxic language we're trying to identify here).
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "872734": "Has anyone tried BERTweet here? I'm going to try it, but was curious if others had tried already. It performs on par with RoBERTa in the Tweet Sentiment Extraction. \n\nIt isn't multi-lingual (trained only on English Tweets), so you'll have to use the translated validation and test data. Twitter removes toxic tweets, so my intuition is that it won't perform well. \n\nBERTweet paper:\nhttps://arxiv.org/abs/2005.10200\nhttps://github.com/VinAIResearch/BERTweet\n\nSee discussion here: https://www.kaggle.com/c/tweet-sentiment-extraction/discussion/152861\nSee implementation in Tweet Sentiment Extraction here: https://www.kaggle.com/davelo12/bertweet (we can use internet in this competition, so loading the packages will be more straightforward)", "875742": "Thank you for sharing. \nCould you tell me the result?", "878362": "I haven't had success with BERTweet for this competition, but I'll let you know if something changes! \n\nI haven't spent much time digging, but I think this goes back to BERTweet being trained on tweets, and Twitter removes toxic tweets from the platform (i.e., BERTweet would not see the same sort of toxic language we're trying to identify here)." }, "source": "meta" }