Based on the labels of the train set, how can we convert this into NER B-I-O (I-O-B) tagging. How many entities will be present in the created label tags . Is it only 1 or we can have multiples tags?
\nWhat kind of custom features we can create as part of this activity? Some known examples include POS of current tag, POS of previous tag, last 4 character of a word, etc.
\nPlease provide some informative ideas to make it fun and knowledge.
", "messages": [ { "id": 1312437, "postDate": "2021-05-18T04:02:20.600Z", "content": "Emmm，i used some functions from this kernal https://www.kaggle.com/tungmphung/coleridge-matching-bert-ner to convert sentence into BIO tag.
\nAs to the custom features, i used pos ,context,prefix，suffix，and one feature that whether a word in a abbreviation。like this “Science, Technology and Society STS“。you can also learn some features from this article :http://www.oegai.at/konvens2012/proceedings/17_tkachenko12o/17_tkachenko12o.pdf.
Unfortunately,all this features doesn't help me get a better score than baseline.Actually,my model get a not bad score in validation sets.So i think this is because my model relies too much on word matching.Now i am tring to use bilstm+crf in this task.
\nBy the way,you can get many useful resources in discussion with most votes.
Based on the labels of the train set, how can we convert this into NER B-I-O (I-O-B) tagging. How many entities will be present in the created label tags . Is it only 1 or we can have multiples tags?
\nWhat kind of custom features we can create as part of this activity? Some known examples include POS of current tag, POS of previous tag, last 4 character of a word, etc.
\nPlease provide some informative ideas to make it fun and knowledge.
", "rawMarkdown": "Based on the labels of the train set, how can we convert this into NER B-I-O (I-O-B) tagging. How many entities will be present in the created label tags . Is it only 1 or we can have multiples tags?\n\nWhat kind of custom features we can create as part of this activity? Some known examples include POS of current tag, POS of previous tag, last 4 character of a word, etc. \n\nPlease provide some informative ideas to make it fun and knowledge. ", "votes": 2 }, { "id": 1322832, "postDate": "2021-05-25T18:02:53.867Z", "rawMarkdown": "", "isDeleted": true }, { "id": 1313861, "postDate": "2021-05-18T18:47:03.277Z", "content": "Thank you @coffeebear98.
", "rawMarkdown": "Thank you @coffeebear98." } ], "comments": [ { "id": 1312437, "author_name": "Coffeebear98", "author_url": "", "post_date": "2021-05-18T04:02:20.600000", "content": "Emmm，i used some functions from this kernal https://www.kaggle.com/tungmphung/coleridge-matching-bert-ner to convert sentence into BIO tag.
\nAs to the custom features, i used pos ,context,prefix，suffix，and one feature that whether a word in a abbreviation。like this “Science, Technology and Society STS“。you can also learn some features from this article :http://www.oegai.at/konvens2012/proceedings/17_tkachenko12o/17_tkachenko12o.pdf.
Unfortunately,all this features doesn't help me get a better score than baseline.Actually,my model get a not bad score in validation sets.So i think this is because my model relies too much on word matching.Now i am tring to use bilstm+crf in this task.
\nBy the way,you can get many useful resources in discussion with most votes.
Thank you @coffeebear98.
", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "1312437": "Emmm，i used some functions from this kernal https://www.kaggle.com/tungmphung/coleridge-matching-bert-ner to convert sentence into BIO tag.\nAs to the custom features, i used pos ,context,prefix，suffix，and one feature that whether a word in a abbreviation。like this “Science, Technology and Society STS“。you can also learn some features from this article :http://www.oegai.at/konvens2012/proceedings/17_tkachenko12o/17_tkachenko12o.pdf.\n\nUnfortunately,all this features doesn't help me get a better score than baseline.Actually,my model get a not bad score in validation sets.So i think this is because my model relies too much on word matching.Now i am tring to use bilstm+crf in this task.\nBy the way,you can get many useful resources in discussion with most votes.\n \n", "1312132": "Based on the labels of the train set, how can we convert this into NER B-I-O (I-O-B) tagging. How many entities will be present in the created label tags . Is it only 1 or we can have multiples tags?\n\nWhat kind of custom features we can create as part of this activity? Some known examples include POS of current tag, POS of previous tag, last 4 character of a word, etc. \n\nPlease provide some informative ideas to make it fun and knowledge. ", "1322832": "", "1313861": "Thank you @coffeebear98." } }