1st PLACE - WINNER SOLUTION - Chenglong Chen
\n\nFirst, thanks to the Kaggle team and CrowdFlower for such great competition.
\n\nI have put the code to generate the winning solution (also the best single model using XGBoost with linear booster) on my github: https://github.com/ChenglongChen/Kaggle_CrowdFlower
\n\nFor the approach, please find details in ./Doc/Kaggle_CrowdFlower_ChenglongChen.pdf
\n\nYou can also find it in the attachment.
\n\n(I apologize that I am not very good at visualization and making figures, and also my poor English.)
\n\nBest,
\n\nChenglong
\n\nUpdated: Sorry about uploading the file twice, please download the second one.
", "comments": [ { "id": 85557, "author_name": "ChenglongChen", "author_url": "", "post_date": "2015-07-15T16:40:22.570000", "content": "[quote=sirnodin;85477]
\n\nBesides,\none question, have you ever tried word embedding methods, such as Word2Vector, Sentence2Vector (which can be found, for example from Gensim) or DNN?\nI got the conclusion that all the features above is no better than simple tf-idf, which to me is a bit of surprise.
\n\n[/quote]
\n\nI didn't spend much time exploring word embedding method. But I think it worth a try. In fact, as I have mentioned in Sec. 8 of my solution doc, the author of [1] used Word Mover’s Distance (WMD) metric together with word2vec embeddings to measure the distance between text documents. This metric is\nshown to have superior performance than BOW and TF-IDF features.
\n\n[1] Matt J. Kusner, Yu Sun, Nicholas I. Kolkin, and Kilian Q. Weinberger. From word embeddings to document distances. In the 32ndInternational Conference on Machine Learning, 2015.
", "votes": 4, "replies": [] }, { "id": 87205, "author_name": "Ana Lina Morelia", "author_url": "", "post_date": "2015-07-27T20:23:29.957000", "content": "Chenglong's write-up on his approach is now live on the blog. You can find it here! Great details on the work he did during each stage of the competition, tips for those looking to get started ensembling, and why XGBoost is becoming everyone's favorite. Thanks again for the excellent post!
", "votes": 2, "replies": [] }, { "id": 85501, "author_name": "Julian de Wit", "author_url": "", "post_date": "2015-07-15T07:12:34.093000", "content": "Congrats..\nAs a long-time "fan" of you (like in sports! ) I'd say that you had this win coming for a long time.\nAs an extra plus you are always very open about your approaches during the competitions.
", "votes": 2, "replies": [] }, { "id": 86868, "author_name": "YingnanWang", "author_url": "", "post_date": "2015-07-24T19:52:40.060000", "content": "Proud of you !! As your name :) And thanks for your sharing.
", "votes": 0, "replies": [] }, { "id": 86772, "author_name": "Chen Zhang", "author_url": "", "post_date": "2015-07-24T07:36:50.467000", "content": "[quote=Chenglong Chen;85557]
\n\n[quote=sirnodin;85477]
\n\nBesides,\none question, have you ever tried word embedding methods, such as Word2Vector, Sentence2Vector (which can be found, for example from Gensim) or DNN?\nI got the conclusion that all the features above is no better than simple tf-idf, which to me is a bit of surprise.
\n\n[/quote]
\n\nI didn't spend much time exploring word embedding method. But I think it worth a try. In fact, as I have mentioned in Sec. 8 of my solution doc, the author of [1] used Word Mover’s Distance (WMD) metric together with word2vec embeddings to measure the distance between text documents. This metric is\nshown to have superior performance than BOW and TF-IDF features.
\n\n[1] Matt J. Kusner, Yu Sun, Nicholas I. Kolkin, and Kilian Q. Weinberger. From word embeddings to document distances. In the 32ndInternational Conference on Machine Learning, 2015.
\n\n[/quote]
\n\nThank you very much for your reply!
", "votes": 0, "replies": [] }, { "id": 85477, "author_name": "Chen Zhang", "author_url": "", "post_date": "2015-07-15T01:15:58.043000", "content": "Really really great job！\nI have learnt a lot just from the beginning of your pdf, and of course I will go on with it.
\n\nBesides,\none question, have you ever tried word embedding methods, such as Word2Vector, Sentence2Vector (which can be found, for example from Gensim) or DNN?\nI got the conclusion that all the features above is no better than simple tf-idf, which to me is a bit of surprise.
\n\nAlso I noticed that you processed spelling correction and synonym replacement at first, and even used no external data.\nThose tasks should, in theory, be done better with word embedding methods, but it seemed not to be the truth.
\n\nWhy word embedding not work well?\nOr should I just try to use them better?
", "votes": 0, "replies": [] }, { "id": 84317, "author_name": "Soumya Smruti Mishra", "author_url": "", "post_date": "2015-07-13T22:50:56.453000", "content": "Just wow
", "votes": 0, "replies": [] }, { "id": 84298, "author_name": "Jerry Lin", "author_url": "", "post_date": "2015-07-13T17:45:58.073000", "content": "Thanks so much for sharing your solutions!!
", "votes": 0, "replies": [] } ], "comment_count": 8, "messages": [ { "id": 84196, "postDate": "2015-07-12T06:58:26.520Z", "content": "1st PLACE - WINNER SOLUTION - Chenglong Chen
\n\nFirst, thanks to the Kaggle team and CrowdFlower for such great competition.
\n\nI have put the code to generate the winning solution (also the best single model using XGBoost with linear booster) on my github: https://github.com/ChenglongChen/Kaggle_CrowdFlower
\n\nFor the approach, please find details in ./Doc/Kaggle_CrowdFlower_ChenglongChen.pdf
\n\nYou can also find it in the attachment.
\n\n(I apologize that I am not very good at visualization and making figures, and also my poor English.)
\n\nBest,
\n\nChenglong
\n\nUpdated: Sorry about uploading the file twice, please download the second one.
", "rawMarkdown": "**1st PLACE - WINNER SOLUTION - Chenglong Chen**\r\n\r\nFirst, thanks to the Kaggle team and CrowdFlower for such great competition.\r\n\r\nI have put the code to generate the winning solution (also the best single model using XGBoost with linear booster) on my github: https://github.com/ChenglongChen/Kaggle_CrowdFlower\r\n\r\nFor the approach, please find details in ./Doc/Kaggle_CrowdFlower_ChenglongChen.pdf\r\n\r\nYou can also find it in the attachment.\r\n\r\n(I apologize that I am not very good at visualization and making figures, and also my poor English.)\r\n\r\nBest,\r\n\r\nChenglong\r\n\r\nUpdated: Sorry about uploading the file twice, please download the second one.", "votes": 51 }, { "id": 85557, "postDate": "2015-07-15T16:40:22.570Z", "content": "[quote=sirnodin;85477]
\n\nBesides,\none question, have you ever tried word embedding methods, such as Word2Vector, Sentence2Vector (which can be found, for example from Gensim) or DNN?\nI got the conclusion that all the features above is no better than simple tf-idf, which to me is a bit of surprise.
\n\n[/quote]
\n\nI didn't spend much time exploring word embedding method. But I think it worth a try. In fact, as I have mentioned in Sec. 8 of my solution doc, the author of [1] used Word Mover’s Distance (WMD) metric together with word2vec embeddings to measure the distance between text documents. This metric is\nshown to have superior performance than BOW and TF-IDF features.
\n\n[1] Matt J. Kusner, Yu Sun, Nicholas I. Kolkin, and Kilian Q. Weinberger. From word embeddings to document distances. In the 32ndInternational Conference on Machine Learning, 2015.
", "rawMarkdown": "[quote=sirnodin;85477]\r\n\r\nBesides,\r\none question, have you ever tried word embedding methods, such as Word2Vector, Sentence2Vector (which can be found, for example from Gensim) or DNN?\r\nI got the conclusion that all the features above is no better than simple tf-idf, which to me is a bit of surprise.\r\n\r\n[/quote]\r\n\r\nI didn't spend much time exploring word embedding method. But I think it worth a try. In fact, as I have mentioned in Sec. 8 of my solution doc, the author of [1] used Word Mover’s Distance (WMD) metric together with word2vec embeddings to measure the distance between text documents. This metric is\r\nshown to have superior performance than BOW and TF-IDF features.\r\n\r\n[1] Matt J. Kusner, Yu Sun, Nicholas I. Kolkin, and Kilian Q. Weinberger. From word embeddings to document distances. In the 32ndInternational Conference on Machine Learning, 2015.", "votes": 4 }, { "id": 87205, "postDate": "2015-07-27T20:23:29.957Z", "content": "Chenglong's write-up on his approach is now live on the blog. You can find it here! Great details on the work he did during each stage of the competition, tips for those looking to get started ensembling, and why XGBoost is becoming everyone's favorite. Thanks again for the excellent post!
", "rawMarkdown": "Chenglong's write-up on his approach is now live on the blog. You can find it [here][1]! Great details on the work he did during each stage of the competition, tips for those looking to get started ensembling, and why XGBoost is becoming everyone's favorite. Thanks again for the excellent post!\r\n\r\n\r\n [1]: http://blog.kaggle.com/2015/07/27/crowdflower-winners-interview-1st-place-chenglong-chen/", "votes": 2 }, { "id": 85501, "postDate": "2015-07-15T07:12:34.093Z", "content": "Congrats..\nAs a long-time "fan" of you (like in sports! ) I'd say that you had this win coming for a long time.\nAs an extra plus you are always very open about your approaches during the competitions.
", "rawMarkdown": "Congrats..\r\nAs a long-time \"fan\" of you (like in sports! ) I'd say that you had this win coming for a long time.\r\nAs an extra plus you are always very open about your approaches during the competitions.\r\n\r\n", "votes": 2 }, { "id": 86868, "postDate": "2015-07-24T19:52:40.060Z", "content": "Proud of you !! As your name :) And thanks for your sharing.
", "rawMarkdown": "Proud of you !! As your name :) And thanks for your sharing. " }, { "id": 86772, "postDate": "2015-07-24T07:36:50.467Z", "content": "[quote=Chenglong Chen;85557]
\n\n[quote=sirnodin;85477]
\n\nBesides,\none question, have you ever tried word embedding methods, such as Word2Vector, Sentence2Vector (which can be found, for example from Gensim) or DNN?\nI got the conclusion that all the features above is no better than simple tf-idf, which to me is a bit of surprise.
\n\n[/quote]
\n\nI didn't spend much time exploring word embedding method. But I think it worth a try. In fact, as I have mentioned in Sec. 8 of my solution doc, the author of [1] used Word Mover’s Distance (WMD) metric together with word2vec embeddings to measure the distance between text documents. This metric is\nshown to have superior performance than BOW and TF-IDF features.
\n\n[1] Matt J. Kusner, Yu Sun, Nicholas I. Kolkin, and Kilian Q. Weinberger. From word embeddings to document distances. In the 32ndInternational Conference on Machine Learning, 2015.
\n\n[/quote]
\n\nThank you very much for your reply!
", "rawMarkdown": "[quote=Chenglong Chen;85557]\r\n\r\n[quote=sirnodin;85477]\r\n\r\nBesides,\r\none question, have you ever tried word embedding methods, such as Word2Vector, Sentence2Vector (which can be found, for example from Gensim) or DNN?\r\nI got the conclusion that all the features above is no better than simple tf-idf, which to me is a bit of surprise.\r\n\r\n[/quote]\r\n\r\nI didn't spend much time exploring word embedding method. But I think it worth a try. In fact, as I have mentioned in Sec. 8 of my solution doc, the author of [1] used Word Mover’s Distance (WMD) metric together with word2vec embeddings to measure the distance between text documents. This metric is\r\nshown to have superior performance than BOW and TF-IDF features.\r\n\r\n[1] Matt J. Kusner, Yu Sun, Nicholas I. Kolkin, and Kilian Q. Weinberger. From word embeddings to document distances. In the 32ndInternational Conference on Machine Learning, 2015.\r\n\r\n[/quote]\r\n\r\nThank you very much for your reply!" }, { "id": 85477, "postDate": "2015-07-15T01:15:58.043Z", "content": "Really really great job！\nI have learnt a lot just from the beginning of your pdf, and of course I will go on with it.
\n\nBesides,\none question, have you ever tried word embedding methods, such as Word2Vector, Sentence2Vector (which can be found, for example from Gensim) or DNN?\nI got the conclusion that all the features above is no better than simple tf-idf, which to me is a bit of surprise.
\n\nAlso I noticed that you processed spelling correction and synonym replacement at first, and even used no external data.\nThose tasks should, in theory, be done better with word embedding methods, but it seemed not to be the truth.
\n\nWhy word embedding not work well?\nOr should I just try to use them better?
", "rawMarkdown": "Really really great job！\r\nI have learnt a lot just from the beginning of your pdf, and of course I will go on with it.\r\n\r\nBesides,\r\none question, have you ever tried word embedding methods, such as Word2Vector, Sentence2Vector (which can be found, for example from Gensim) or DNN?\r\nI got the conclusion that all the features above is no better than simple tf-idf, which to me is a bit of surprise.\r\n\r\nAlso I noticed that you processed spelling correction and synonym replacement at first, and even used no external data.\r\nThose tasks should, in theory, be done better with word embedding methods, but it seemed not to be the truth.\r\n\r\nWhy word embedding not work well?\r\nOr should I just try to use them better?" }, { "id": 84317, "postDate": "2015-07-13T22:50:56.453Z", "content": "Just wow
", "rawMarkdown": "Just wow" }, { "id": 84298, "postDate": "2015-07-13T17:45:58.073Z", "content": "Thanks so much for sharing your solutions!!
", "rawMarkdown": "Thanks so much for sharing your solutions!!" } ], "complete": true, "source": "live", "fetched_at": "2026-10-19T20:59:25Z", "competition_id": 4407, "competition_slug": "crowdflower-search-relevance", "team_id": 181206, "solution_url": "/competitions/crowdflower-search-relevance/writeups/chenglong-chen-1st-place-winner-solution-chenglong" }