In data description, it said "Your goal of this competition is to predict the booking outcome (hotel cluster) for a user event, based on their search and other attributes associated with that user event." and \n"Training data includes all the users in the logs, including both click events and booking events. Test data only includes booking events."
\n\nIn data description, it said "Your goal of this competition is to predict the booking outcome (hotel cluster) for a user event, based on their search and other attributes associated with that user event." and \n"Training data includes all the users in the logs, including both click events and booking events. Test data only includes booking events."
\n\nI recently had to book a hotel using Expedia, navigating their website will help answer the questions.
\n\nWe are trying to predict which hotel cluster the user will book. Hotels were clustered(or grouped) by there attributes so that similar hotels will be in the same cluster(see data page). As a hypothetical example all hotels that cost less then 50 USD night, have a 2.5 or lower star rating and located more then 2 miles from the city center might be one cluster.
Click probably means viewing a hotel. After a searching I was presented with a list of hotels with short descriptions, and a price offer. click is following a link to a specific hotel/offer. The option to book the hotel at the given price is then given. So there can be a click without a booking, but there will almost always be a click before a booking event.
Adam
", "rawMarkdown": "1. The holdout data contains only bookings and the goal is to predict bookings based on previous bookings and clicks.\r\n2. A click means that a user clicked a link to see hotel details on a hotel infosite page.\r\n\r\nAdam", "votes": null }, { "id": "115499", "postDate": "04/19/2016 00:06:40", "content": "Thanks for your reply! I think I got it.
", "rawMarkdown": "Thanks for your reply! I think I got it.", "votes": null }, { "id": "115864", "postDate": "04/20/2016 14:40:59", "content": "The submission file contains three columns with two labels id, hotel_cluster,\nMy guess is :\nsample_submittion.id equals test.id , where id is the user search id index [0,..,2528242] .\nsample_submittion.hotel_cluster – this is what you need to predict, based on the test/train.hotel_cluster, varies between 0-99.
\n\nHowever what is the third number always set to 1 in the example?
", "rawMarkdown": "The submission file contains three columns with two labels id, hotel_cluster,\r\nMy guess is :\r\nsample_submittion.id equals test.id , where id is the user search id index [0,..,2528242] .\r\nsample_submittion.hotel_cluster – this is what you need to predict, based on the test/train.hotel_cluster, varies between 0-99.\r\n\r\nHowever what is the third number always set to 1 in the example?", "votes": null }, { "id": "115894", "postDate": "04/20/2016 16:33:17", "content": "My read of the requirements is as follows...
\n\nThe submission file should contain two columns, the first being the id in the Test set and second being an ordered list of up to 5 hotel_clusters separated by spaces (i.e. "99 0 20 32 60"). This is a recommender problem and the evaluation metric implies that it is a MAP@5 result which is a ranking of the top 5 hotel clusters recommended for a given row in the Test data.
\n\nSomeone correct me if I'm wrong.
", "rawMarkdown": "My read of the requirements is as follows...\r\n\r\nThe submission file should contain two columns, the first being the id in the Test set and second being an ordered list of up to 5 hotel_clusters separated by spaces (i.e. \"99 0 20 32 60\"). This is a recommender problem and the evaluation metric implies that it is a MAP@5 result which is a ranking of the top 5 hotel clusters recommended for a given row in the Test data.\r\n\r\nSomeone correct me if I'm wrong.", "votes": null }, { "id": "117831", "postDate": "05/01/2016 05:23:11", "content": "I still have a few questions I'm hoping someone can clear up... Consider user_id 494510, an extreme case of a user in the test set with multiple "user events" (i.e. records in the test set) - 273 to be exact.
\n\n1) Can a single booking be associated to multiple "user_event"s? For example, user_id 494510 created multiple events on the 8th and 9th of January. Suppose he books a hotel on the 11th. Would that booking be associated with each of the previous events or just the most recent one?
\n\n2) Can a single "user_event" lead to multiple bookings? In other words, for id 2315018, I am going to make some sort of submission of 5 hotel clusters like "99 0 20 32 60". Could it be the case that the user makes bookings for BOTH clusters 99 and 20, which are both associated to this single event? I assume not, even though the evaluation metric MAP@5 is kind of designed to support this type of behavior.
", "rawMarkdown": "I still have a few questions I'm hoping someone can clear up... Consider user_id 494510, an extreme case of a user in the test set with multiple \"user events\" (i.e. records in the test set) - 273 to be exact.\r\n\r\n1) Can a single booking be associated to multiple \"user_event\"s? For example, user_id 494510 created multiple events on the 8th and 9th of January. Suppose he books a hotel on the 11th. Would that booking be associated with each of the previous events or just the most recent one?\r\n\r\n2) Can a single \"user_event\" lead to multiple bookings? In other words, for id 2315018, I am going to make some sort of submission of 5 hotel clusters like \"99 0 20 32 60\". Could it be the case that the user makes bookings for BOTH clusters 99 and 20, which are both associated to this single event? I assume not, even though the evaluation metric MAP@5 is kind of designed to support this type of behavior.", "votes": null }, { "id": "117936", "postDate": "05/01/2016 20:31:13", "content": "[quote=eran st;115864]
\n\nThe submission file contains three columns with two labels id, hotel_cluster,\nMy guess is :\nsample_submittion.id equals test.id , where id is the user search id index [0,..,2528242] .\nsample_submittion.hotel_cluster – this is what you need to predict, based on the test/train.hotel_cluster, varies between 0-99.
\n\nHowever what is the third number always set to 1 in the example?
\n\nThanks, it took some time for me to understand what this id is, among so many ids, so its just row number in test.
\n\n[/quote]
", "rawMarkdown": "[quote=eran st;115864]\r\n\r\nThe submission file contains three columns with two labels id, hotel_cluster,\r\nMy guess is :\r\nsample_submittion.id equals test.id , where id is the user search id index [0,..,2528242] .\r\nsample_submittion.hotel_cluster – this is what you need to predict, based on the test/train.hotel_cluster, varies between 0-99.\r\n\r\nHowever what is the third number always set to 1 in the example? \r\n\r\nThanks, it took some time for me to understand what this id is, among so many ids, so its just row number in test.\r\n\r\n[/quote]", "votes": null } ], "comments": [ { "id": 115426, "author_name": "devinanzelmo", "author_url": "", "post_date": "04/18/2016 16:55:33", "content": "I recently had to book a hotel using Expedia, navigating their website will help answer the questions.
\n\nWe are trying to predict which hotel cluster the user will book. Hotels were clustered(or grouped) by there attributes so that similar hotels will be in the same cluster(see data page). As a hypothetical example all hotels that cost less then 50 USD night, have a 2.5 or lower star rating and located more then 2 miles from the city center might be one cluster.
Click probably means viewing a hotel. After a searching I was presented with a list of hotels with short descriptions, and a price offer. click is following a link to a specific hotel/offer. The option to book the hotel at the given price is then given. So there can be a click without a booking, but there will almost always be a click before a booking event.
Adam
", "votes": null, "replies": [] }, { "id": 115499, "author_name": "lintseju", "author_url": "", "post_date": "04/19/2016 00:06:40", "content": "Thanks for your reply! I think I got it.
", "votes": null, "replies": [] }, { "id": 115864, "author_name": "eranstr", "author_url": "", "post_date": "04/20/2016 14:40:59", "content": "The submission file contains three columns with two labels id, hotel_cluster,\nMy guess is :\nsample_submittion.id equals test.id , where id is the user search id index [0,..,2528242] .\nsample_submittion.hotel_cluster – this is what you need to predict, based on the test/train.hotel_cluster, varies between 0-99.
\n\nHowever what is the third number always set to 1 in the example?
", "votes": null, "replies": [] }, { "id": 115894, "author_name": "rrgiles", "author_url": "", "post_date": "04/20/2016 16:33:17", "content": "My read of the requirements is as follows...
\n\nThe submission file should contain two columns, the first being the id in the Test set and second being an ordered list of up to 5 hotel_clusters separated by spaces (i.e. "99 0 20 32 60"). This is a recommender problem and the evaluation metric implies that it is a MAP@5 result which is a ranking of the top 5 hotel clusters recommended for a given row in the Test data.
\n\nSomeone correct me if I'm wrong.
", "votes": null, "replies": [] }, { "id": 117831, "author_name": "ben519", "author_url": "", "post_date": "05/01/2016 05:23:11", "content": "I still have a few questions I'm hoping someone can clear up... Consider user_id 494510, an extreme case of a user in the test set with multiple "user events" (i.e. records in the test set) - 273 to be exact.
\n\n1) Can a single booking be associated to multiple "user_event"s? For example, user_id 494510 created multiple events on the 8th and 9th of January. Suppose he books a hotel on the 11th. Would that booking be associated with each of the previous events or just the most recent one?
\n\n2) Can a single "user_event" lead to multiple bookings? In other words, for id 2315018, I am going to make some sort of submission of 5 hotel clusters like "99 0 20 32 60". Could it be the case that the user makes bookings for BOTH clusters 99 and 20, which are both associated to this single event? I assume not, even though the evaluation metric MAP@5 is kind of designed to support this type of behavior.
", "votes": null, "replies": [] }, { "id": 117936, "author_name": "adamwoz", "author_url": "", "post_date": "05/01/2016 20:31:13", "content": "[quote=eran st;115864]
\n\nThe submission file contains three columns with two labels id, hotel_cluster,\nMy guess is :\nsample_submittion.id equals test.id , where id is the user search id index [0,..,2528242] .\nsample_submittion.hotel_cluster – this is what you need to predict, based on the test/train.hotel_cluster, varies between 0-99.
\n\nHowever what is the third number always set to 1 in the example?
\n\nThanks, it took some time for me to understand what this id is, among so many ids, so its just row number in test.
\n\n[/quote]
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "115407": "In data description, it said \"Your goal of this competition is to predict the booking outcome (hotel cluster) for a user event, based on their search and other attributes associated with that user event.\" and \r\n\"Training data includes all the users in the logs, including both click events and booking events. Test data only includes booking events.\"\r\n\r\n 1. So we have to predict which hotel is most likely to be clicked in testing data? \r\n 2. What is \"click\" mean? Does it meaning booking outcome=True iff click event=1?", "115426": "I recently had to book a hotel using Expedia, navigating their website will help answer the questions. \r\n\r\n1. We are trying to predict which hotel cluster the user will book. Hotels were clustered(or grouped) by there attributes so that similar hotels will be in the same cluster(see [data page][1]). As a hypothetical example all hotels that cost less then 50 USD night, have a 2.5 or lower star rating and located more then 2 miles from the city center might be one cluster. \r\n\r\n2. Click probably means viewing a hotel. After a searching I was presented with a list of hotels with short descriptions, and a price offer. `click` is following a link to a specific hotel/offer. The option to book the hotel at the given price is then given. So there can be a click without a booking, but there will almost always be a click before a booking event. \r\n\r\n\r\n [1]: https://www.kaggle.com/c/expedia-hotel-recommendations/data", "115457": "1. The holdout data contains only bookings and the goal is to predict bookings based on previous bookings and clicks.\r\n2. A click means that a user clicked a link to see hotel details on a hotel infosite page.\r\n\r\nAdam", "115499": "Thanks for your reply! I think I got it.", "115864": "The submission file contains three columns with two labels id, hotel_cluster,\r\nMy guess is :\r\nsample_submittion.id equals test.id , where id is the user search id index [0,..,2528242] .\r\nsample_submittion.hotel_cluster – this is what you need to predict, based on the test/train.hotel_cluster, varies between 0-99.\r\n\r\nHowever what is the third number always set to 1 in the example?", "115894": "My read of the requirements is as follows...\r\n\r\nThe submission file should contain two columns, the first being the id in the Test set and second being an ordered list of up to 5 hotel_clusters separated by spaces (i.e. \"99 0 20 32 60\"). This is a recommender problem and the evaluation metric implies that it is a MAP@5 result which is a ranking of the top 5 hotel clusters recommended for a given row in the Test data.\r\n\r\nSomeone correct me if I'm wrong.", "117831": "I still have a few questions I'm hoping someone can clear up... Consider user_id 494510, an extreme case of a user in the test set with multiple \"user events\" (i.e. records in the test set) - 273 to be exact.\r\n\r\n1) Can a single booking be associated to multiple \"user_event\"s? For example, user_id 494510 created multiple events on the 8th and 9th of January. Suppose he books a hotel on the 11th. Would that booking be associated with each of the previous events or just the most recent one?\r\n\r\n2) Can a single \"user_event\" lead to multiple bookings? In other words, for id 2315018, I am going to make some sort of submission of 5 hotel clusters like \"99 0 20 32 60\". Could it be the case that the user makes bookings for BOTH clusters 99 and 20, which are both associated to this single event? I assume not, even though the evaluation metric MAP@5 is kind of designed to support this type of behavior.", "117936": "1. It depends how you interpret the data. For example, if events on the 8th and 9th of January are in the context of the same search destination as the booking on the 11th, then I would assume they are related.\r\n2. No, a single booking/click event (i.e. a record in the train or holdout data) it always associated with a single hotel.", "118949": "[quote=eran st;115864]\r\n\r\nThe submission file contains three columns with two labels id, hotel_cluster,\r\nMy guess is :\r\nsample_submittion.id equals test.id , where id is the user search id index [0,..,2528242] .\r\nsample_submittion.hotel_cluster – this is what you need to predict, based on the test/train.hotel_cluster, varies between 0-99.\r\n\r\nHowever what is the third number always set to 1 in the example? \r\n\r\nThanks, it took some time for me to understand what this id is, among so many ids, so its just row number in test.\r\n\r\n[/quote]" }, "source": "meta" }