I think maybe how to build the correlation among grapheme root, vowel diacritics, and consonant diacritics is the key to improvements.\nhow about solving it as a multi-label problem?\nI hope that someone share ideas about correlation mining.\nThanks!
", "messages": [ { "id": 740384, "postDate": "2020-02-09T10:58:38.417Z", "content": "I think maybe how to build the correlation among grapheme root, vowel diacritics, and consonant diacritics is the key to improvements.\nhow about solving it as a multi-label problem?\nI hope that someone share ideas about correlation mining.\nThanks!
", "rawMarkdown": "I think maybe how to build the correlation among grapheme root, vowel diacritics, and consonant diacritics is the key to improvements.\nhow about solving it as a multi-label problem?\nI hope that someone share ideas about correlation mining.\nThanks!", "votes": 2 }, { "id": 740973, "postDate": "2020-02-10T03:16:56.950Z", "content": "Is there someone using GCN on this problem?
", "rawMarkdown": "Is there someone using GCN on this problem?" }, { "id": 740604, "postDate": "2020-02-09T15:26:03.943Z", "content": "I tried doing it @zhaorenjie but am afraid that it might lead to over-fitting.\nThe dataset is quite sparse at a grapheme level to be able to do this, and it might be wiser to do this (corellation mining) on an external dataset which is much larger and then port results for this competition.
\n\nI would love to see if anyone has implemented this.
", "rawMarkdown": "I tried doing it @zhaorenjie but am afraid that it might lead to over-fitting.\nThe dataset is quite sparse at a grapheme level to be able to do this, and it might be wiser to do this (corellation mining) on an external dataset which is much larger and then port results for this competition.\n\nI would love to see if anyone has implemented this.", "replies": [ { "id": 740970, "postDate": "2020-02-10T03:15:27.760Z", "content": "Thanks for you comment! Have you tried that on an external dataset? And, how about considering the combination about vowel and consonant first, then mining the correlation between the combination(v and c) and the grapheme? Maybe it can relieve the sparse problem to some extent.
", "rawMarkdown": "Thanks for you comment! Have you tried that on an external dataset? And, how about considering the combination about vowel and consonant first, then mining the correlation between the combination(v and c) and the grapheme? Maybe it can relieve the sparse problem to some extent." }, { "id": 740974, "postDate": "2020-02-10T03:20:16.670Z", "content": "Good idea @zhaorenjie , let me try and report back.
", "rawMarkdown": "Good idea @zhaorenjie , let me try and report back." }, { "id": 740990, "postDate": "2020-02-10T03:54:49.350Z", "content": "Keep communication!:)
", "rawMarkdown": "Keep communication!:)" } ] }, { "id": 740969, "postDate": "2020-02-10T03:14:43.113Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 740973, "author_name": "Tobepellucid", "author_url": "", "post_date": "2020-02-10T03:16:56.950000", "content": "Is there someone using GCN on this problem?
", "votes": 0, "replies": [] }, { "id": 740604, "author_name": "Rohit Agarwal", "author_url": "", "post_date": "2020-02-09T15:26:03.943000", "content": "I tried doing it @zhaorenjie but am afraid that it might lead to over-fitting.\nThe dataset is quite sparse at a grapheme level to be able to do this, and it might be wiser to do this (corellation mining) on an external dataset which is much larger and then port results for this competition.
\n\nI would love to see if anyone has implemented this.
", "votes": 0, "replies": [ { "id": 740970, "author_name": "Tobepellucid", "author_url": "", "post_date": "2020-02-10T03:15:27.760000", "content": "Thanks for you comment! Have you tried that on an external dataset? And, how about considering the combination about vowel and consonant first, then mining the correlation between the combination(v and c) and the grapheme? Maybe it can relieve the sparse problem to some extent.
", "votes": 0, "replies": [] }, { "id": 740974, "author_name": "Rohit Agarwal", "author_url": "", "post_date": "2020-02-10T03:20:16.670000", "content": "Good idea @zhaorenjie , let me try and report back.
", "votes": 0, "replies": [] }, { "id": 740990, "author_name": "Tobepellucid", "author_url": "", "post_date": "2020-02-10T03:54:49.350000", "content": "Keep communication!:)
", "votes": 0, "replies": [] } ] }, { "id": 740969, "author_name": "", "author_url": "", "post_date": "2020-02-10T03:14:43.113000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "740384": "I think maybe how to build the correlation among grapheme root, vowel diacritics, and consonant diacritics is the key to improvements.\nhow about solving it as a multi-label problem?\nI hope that someone share ideas about correlation mining.\nThanks!", "740973": "Is there someone using GCN on this problem?", "740604": "I tried doing it @zhaorenjie but am afraid that it might lead to over-fitting.\nThe dataset is quite sparse at a grapheme level to be able to do this, and it might be wiser to do this (corellation mining) on an external dataset which is much larger and then port results for this competition.\n\nI would love to see if anyone has implemented this.", "740969": "" } }