I tried my best in preprocessing but it always hurt cv score.Some public kernels use preprocessing which doesn't make any sense,like seperating punctuations and then correcting contractions.\nWhat could be the reason behind this and is it good to do just for better scoring ?
", "messages": [ { "id": "462148", "postDate": "01/27/2019 19:17:52", "content": "I tried my best in preprocessing but it always hurt cv score.Some public kernels use preprocessing which doesn't make any sense,like seperating punctuations and then correcting contractions.\nWhat could be the reason behind this and is it good to do just for better scoring ?
", "rawMarkdown": "I tried my best in preprocessing but it always hurt cv score.Some public kernels use preprocessing which doesn't make any sense,like seperating punctuations and then correcting contractions.\nWhat could be the reason behind this and is it good to do just for better scoring ?", "votes": null }, { "id": "462157", "postDate": "01/27/2019 19:41:58", "content": "The correcting contractions is done to help find entries in the word embeddings
\n\nI think the idea around putting spaces around punctuation rather than removing them is because we know punctuation is important in the guidelines Quora have for sincere questions. There's a bit more about that on this thread https://www.kaggle.com/c/quora-insincere-questions-classification/discussion/77691
", "rawMarkdown": "The correcting contractions is done to help find entries in the word embeddings\n\nI think the idea around putting spaces around punctuation rather than removing them is because we know punctuation is important in the guidelines Quora have for sincere questions. There's a bit more about that on this thread https://www.kaggle.com/c/quora-insincere-questions-classification/discussion/77691", "votes": null }, { "id": "462165", "postDate": "01/27/2019 19:57:54", "content": "No,what I am saying is haven't is seperated as haven ' t and then haven't is searched to be replaced with have not which it will never match.
", "rawMarkdown": "No,what I am saying is haven't is seperated as haven ' t and then haven't is searched to be replaced with have not which it will never match.", "votes": null }, { "id": "462176", "postDate": "01/27/2019 20:04:19", "content": "Oh yeah, I think that's a bug in one of the first publicly shared kernels which has been forked over and over again without correction
", "rawMarkdown": "Oh yeah, I think that's a bug in one of the first publicly shared kernels which has been forked over and over again without correction", "votes": null }, { "id": "462241", "postDate": "01/27/2019 23:54:29", "content": "Yes, it is probably a bug. But you would think that fixing that bug helps the score. Except this isn't the case. It performs better with the bug.
", "rawMarkdown": "Yes, it is probably a bug. But you would think that fixing that bug helps the score. Except this isn't the case. It performs better with the bug.", "votes": null }, { "id": "462583", "postDate": "01/28/2019 14:33:42", "content": "you are right，but why ? I've been puzzled by this question for a long time.
", "rawMarkdown": "you are right，but why ? I've been puzzled by this question for a long time.", "votes": null } ], "comments": [ { "id": 462157, "author_name": "hamishdickson", "author_url": "", "post_date": "01/27/2019 19:41:58", "content": "The correcting contractions is done to help find entries in the word embeddings
\n\nI think the idea around putting spaces around punctuation rather than removing them is because we know punctuation is important in the guidelines Quora have for sincere questions. There's a bit more about that on this thread https://www.kaggle.com/c/quora-insincere-questions-classification/discussion/77691
", "votes": null, "replies": [ { "id": 462165, "author_name": "mayurnewase", "author_url": "", "post_date": "01/27/2019 19:57:54", "content": "No,what I am saying is haven't is seperated as haven ' t and then haven't is searched to be replaced with have not which it will never match.
", "votes": null, "replies": [] }, { "id": 462176, "author_name": "hamishdickson", "author_url": "", "post_date": "01/27/2019 20:04:19", "content": "Oh yeah, I think that's a bug in one of the first publicly shared kernels which has been forked over and over again without correction
", "votes": null, "replies": [] }, { "id": 462241, "author_name": "julius6", "author_url": "", "post_date": "01/27/2019 23:54:29", "content": "Yes, it is probably a bug. But you would think that fixing that bug helps the score. Except this isn't the case. It performs better with the bug.
", "votes": null, "replies": [] }, { "id": 462583, "author_name": "elias8888", "author_url": "", "post_date": "01/28/2019 14:33:42", "content": "you are right，but why ? I've been puzzled by this question for a long time.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "462148": "I tried my best in preprocessing but it always hurt cv score.Some public kernels use preprocessing which doesn't make any sense,like seperating punctuations and then correcting contractions.\nWhat could be the reason behind this and is it good to do just for better scoring ?", "462157": "The correcting contractions is done to help find entries in the word embeddings\n\nI think the idea around putting spaces around punctuation rather than removing them is because we know punctuation is important in the guidelines Quora have for sincere questions. There's a bit more about that on this thread https://www.kaggle.com/c/quora-insincere-questions-classification/discussion/77691", "462165": "No,what I am saying is haven't is seperated as haven ' t and then haven't is searched to be replaced with have not which it will never match.", "462176": "Oh yeah, I think that's a bug in one of the first publicly shared kernels which has been forked over and over again without correction", "462241": "Yes, it is probably a bug. But you would think that fixing that bug helps the score. Except this isn't the case. It performs better with the bug.", "462583": "you are right，but why ? I've been puzzled by this question for a long time." }, "source": "meta" }