Looks like an out-and-out NLP-based competition. As we've recently seen BERT has been the go-to approach, but before jumping into that I think we should discuss more possible approaches. Thoughts?
", "messages": [ { "id": "1250433", "postDate": "03/24/2021 03:27:59", "content": "Looks like an out-and-out NLP-based competition. As we've recently seen BERT has been the go-to approach, but before jumping into that I think we should discuss more possible approaches. Thoughts?
", "rawMarkdown": "Looks like an out-and-out NLP-based competition. As we've recently seen BERT has been the go-to approach, but before jumping into that I think we should discuss more possible approaches. Thoughts?", "votes": null }, { "id": "1250467", "postDate": "03/24/2021 04:21:44", "content": "By what criteria do you think BERT is close to the only approach worth mentioning? I'm new to NLP and can't tell heads from tails when reading a news article about BERT research.
\nAlso, do competitors often come up with their own approaches to the problem? Will we be judged negatively if we don't employ state-of-the-art research?
", "rawMarkdown": "By what criteria do you think BERT is close to the only approach worth mentioning? I'm new to NLP and can't tell heads from tails when reading a news article about BERT research. \n\nAlso, do competitors often come up with their own approaches to the problem? Will we be judged negatively if we don't employ state-of-the-art research?", "votes": null }, { "id": "1250567", "postDate": "03/24/2021 06:01:55", "content": "I don't think BERT is the only approach worth mentioning, that's precisely why I'm asking for the opinion of the community :) You should start reading about the basics of BERT before trying to wrap your head around BERT research I think
\nNot everyone here wants to 'win' competitions, a lot of them are here to learn from each other π
\nAnd you won't be judged at all if you don't use SOTA so don't worry about that
Pretrained NLP Transformers will be the way to go definitely.
", "rawMarkdown": "Pretrained NLP Transformers will be the way to go definitely.", "votes": null }, { "id": "1250963", "postDate": "03/24/2021 11:52:51", "content": "Day 2 of the competition and the winning methodology so far is searching for string literals with upto a 100% score!
\nWhen all you have is a BERT, everything looks like an Ernie and you forget about the Count.
", "rawMarkdown": "Day 2 of the competition and the winning methodology so far is searching for string literals with upto a 100% score! \n- https://www.kaggle.com/jamesmcguigan/coleridge-string-literals\n- https://www.kaggle.com/josephassaker/coleridge-initiative-eda-na-ve-submission\n\nWhen all you have is a BERT, everything looks like an Ernie and you forget about the Count.", "votes": null }, { "id": "1250979", "postDate": "03/24/2021 12:03:55", "content": "Haha, waiting for them to fix the competition
", "rawMarkdown": "Haha, waiting for them to fix the competition", "votes": null }, { "id": "1251661", "postDate": "03/25/2021 02:18:23", "content": "If someone can get a gold medal without using bert, he will have one more loyal fan.π
", "rawMarkdown": "If someone can get a gold medal without using bert, he will have one more loyal fan.π", "votes": null }, { "id": "1251886", "postDate": "03/25/2021 08:25:04", "content": "Its either bert or some other derivation of it or some other type of transformer for sure.
\nIf someone wins without these they will have at least 1 huge fan π
XLNet should also give good results as per this article like BERT, kindly let us know your thoughts.
\nhttps://towardsdatascience.com/what-is-xlnet-and-why-it-outperforms-bert-8d8fce710335
", "rawMarkdown": "XLNet should also give good results as per this article like BERT, kindly let us know your thoughts.\n\nhttps://towardsdatascience.com/what-is-xlnet-and-why-it-outperforms-bert-8d8fce710335", "votes": null }, { "id": "1252309", "postDate": "03/25/2021 15:17:04", "content": "Bert is an encoder only pretrained LM. You may need to add a decoder if you use it here. Or just go to Seq2Seq pretrained Transformers.
", "rawMarkdown": "Bert is an encoder only pretrained LM. You may need to add a decoder if you use it here. Or just go to Seq2Seq pretrained Transformers.", "votes": null }, { "id": "1252371", "postDate": "03/25/2021 16:17:13", "content": "My goal in the competition is to learn NLP, Bert or other models. I don't mindπ
", "rawMarkdown": "My goal in the competition is to learn NLP, Bert or other models. I don't mindπ", "votes": null }, { "id": "1260462", "postDate": "04/02/2021 06:22:19", "content": "Sorry, I don't know what is your meansβ¦ :(
", "rawMarkdown": "Sorry, I don't know what is your means... :(", "votes": null }, { "id": "1260493", "postDate": "04/02/2021 07:06:21", "content": "He means treating this task as text generation, rather than NER or answer extraction.
", "rawMarkdown": "He means treating this task as text generation, rather than NER or answer extraction.", "votes": null }, { "id": "1260521", "postDate": "04/02/2021 07:34:37", "content": "oh, mlm is perfect.
", "rawMarkdown": "oh, mlm is perfect.", "votes": null }, { "id": "1309837", "postDate": "05/16/2021 10:41:15", "content": "Im a little confused as to how one would implement a BERT model without using the internet. (Yes I'm a beginner)
", "rawMarkdown": "Im a little confused as to how one would implement a BERT model without using the internet. (Yes I'm a beginner)", "votes": null } ], "comments": [ { "id": 1250467, "author_name": "algebralearner", "author_url": "", "post_date": "03/24/2021 04:21:44", "content": "By what criteria do you think BERT is close to the only approach worth mentioning? I'm new to NLP and can't tell heads from tails when reading a news article about BERT research.
\nAlso, do competitors often come up with their own approaches to the problem? Will we be judged negatively if we don't employ state-of-the-art research?
", "votes": null, "replies": [ { "id": 1250567, "author_name": "arka47", "author_url": "", "post_date": "03/24/2021 06:01:55", "content": "I don't think BERT is the only approach worth mentioning, that's precisely why I'm asking for the opinion of the community :) You should start reading about the basics of BERT before trying to wrap your head around BERT research I think
\nNot everyone here wants to 'win' competitions, a lot of them are here to learn from each other π
\nAnd you won't be judged at all if you don't use SOTA so don't worry about that
Pretrained NLP Transformers will be the way to go definitely.
", "votes": null, "replies": [] }, { "id": 1250963, "author_name": "jamesmcguigan", "author_url": "", "post_date": "03/24/2021 11:52:51", "content": "Day 2 of the competition and the winning methodology so far is searching for string literals with upto a 100% score!
\nWhen all you have is a BERT, everything looks like an Ernie and you forget about the Count.
", "votes": null, "replies": [ { "id": 1250979, "author_name": "arka47", "author_url": "", "post_date": "03/24/2021 12:03:55", "content": "Haha, waiting for them to fix the competition
", "votes": null, "replies": [] } ] }, { "id": 1251661, "author_name": "xiaowangiiiii", "author_url": "", "post_date": "03/25/2021 02:18:23", "content": "If someone can get a gold medal without using bert, he will have one more loyal fan.π
", "votes": null, "replies": [ { "id": 1252309, "author_name": "serigne", "author_url": "", "post_date": "03/25/2021 15:17:04", "content": "Bert is an encoder only pretrained LM. You may need to add a decoder if you use it here. Or just go to Seq2Seq pretrained Transformers.
", "votes": null, "replies": [] }, { "id": 1260462, "author_name": "xiaowangiiiii", "author_url": "", "post_date": "04/02/2021 06:22:19", "content": "Sorry, I don't know what is your means⦠:(
", "votes": null, "replies": [] }, { "id": 1260493, "author_name": "zzy990106", "author_url": "", "post_date": "04/02/2021 07:06:21", "content": "He means treating this task as text generation, rather than NER or answer extraction.
", "votes": null, "replies": [] }, { "id": 1260521, "author_name": "xiaowangiiiii", "author_url": "", "post_date": "04/02/2021 07:34:37", "content": "oh, mlm is perfect.
", "votes": null, "replies": [] } ] }, { "id": 1251886, "author_name": "manabendrarout", "author_url": "", "post_date": "03/25/2021 08:25:04", "content": "Its either bert or some other derivation of it or some other type of transformer for sure.
\nIf someone wins without these they will have at least 1 huge fan π
XLNet should also give good results as per this article like BERT, kindly let us know your thoughts.
\nhttps://towardsdatascience.com/what-is-xlnet-and-why-it-outperforms-bert-8d8fce710335
", "votes": null, "replies": [] }, { "id": 1252371, "author_name": "yangxiaoshuai", "author_url": "", "post_date": "03/25/2021 16:17:13", "content": "My goal in the competition is to learn NLP, Bert or other models. I don't mindπ
", "votes": null, "replies": [] }, { "id": 1309837, "author_name": "blinglegend", "author_url": "", "post_date": "05/16/2021 10:41:15", "content": "Im a little confused as to how one would implement a BERT model without using the internet. (Yes I'm a beginner)
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1250433": "Looks like an out-and-out NLP-based competition. As we've recently seen BERT has been the go-to approach, but before jumping into that I think we should discuss more possible approaches. Thoughts?", "1250467": "By what criteria do you think BERT is close to the only approach worth mentioning? I'm new to NLP and can't tell heads from tails when reading a news article about BERT research. \n\nAlso, do competitors often come up with their own approaches to the problem? Will we be judged negatively if we don't employ state-of-the-art research?", "1250567": "I don't think BERT is the only approach worth mentioning, that's precisely why I'm asking for the opinion of the community :) You should start reading about the basics of BERT before trying to wrap your head around BERT research I think\n\nNot everyone here wants to 'win' competitions, a lot of them are here to learn from each other π\nAnd you won't be judged at all if you don't use SOTA so don't worry about that", "1250651": "Pretrained NLP Transformers will be the way to go definitely.", "1250963": "Day 2 of the competition and the winning methodology so far is searching for string literals with upto a 100% score! \n- https://www.kaggle.com/jamesmcguigan/coleridge-string-literals\n- https://www.kaggle.com/josephassaker/coleridge-initiative-eda-na-ve-submission\n\nWhen all you have is a BERT, everything looks like an Ernie and you forget about the Count.", "1250979": "Haha, waiting for them to fix the competition", "1251661": "If someone can get a gold medal without using bert, he will have one more loyal fan.π", "1251886": "Its either bert or some other derivation of it or some other type of transformer for sure.\nIf someone wins without these they will have at least 1 huge fan π", "1252296": "XLNet should also give good results as per this article like BERT, kindly let us know your thoughts.\n\nhttps://towardsdatascience.com/what-is-xlnet-and-why-it-outperforms-bert-8d8fce710335", "1252309": "Bert is an encoder only pretrained LM. You may need to add a decoder if you use it here. Or just go to Seq2Seq pretrained Transformers.", "1252371": "My goal in the competition is to learn NLP, Bert or other models. I don't mindπ", "1260462": "Sorry, I don't know what is your means... :(", "1260493": "He means treating this task as text generation, rather than NER or answer extraction.", "1260521": "oh, mlm is perfect.", "1309837": "Im a little confused as to how one would implement a BERT model without using the internet. (Yes I'm a beginner)" }, "source": "meta" }