Does anybody know (I have not seen anything in the existing discussions) if the app id column represents the id of the application thats being advertised, or, if its the id of the app that hosted the advertisement?
", "messages": [ { "id": 309999, "postDate": "2018-04-06T11:20:00.823Z", "content": "Does anybody know (I have not seen anything in the existing discussions) if the app id column represents the id of the application thats being advertised, or, if its the id of the app that hosted the advertisement?
", "rawMarkdown": "Does anybody know (I have not seen anything in the existing discussions) if the app id column represents the id of the application thats being advertised, or, if its the id of the app that hosted the advertisement?", "votes": 7 }, { "id": 312302, "postDate": "2018-04-11T14:42:54.473Z", "content": "There is only one app in the data, so my understanding is that app is what advertised via channel, and when app is downloaded and installed, it calls TalkingData sdk to report the fact it was installed along with time.\nIf advertising is displayed within other app, I think this app should be considered as a channel.
", "rawMarkdown": "There is only one app in the data, so my understanding is that app is what advertised via channel, and when app is downloaded and installed, it calls TalkingData sdk to report the fact it was installed along with time.\nIf advertising is displayed within other app, I think this app should be considered as a channel.", "votes": 1, "replies": [ { "id": 312411, "postDate": "2018-04-11T17:45:37.690Z", "content": "Interesting...
\n\nWhat do you mean by :\n\"If advertising is displayed within other app, I think this app should be considered as a channel.\"
\n\nYou mean that if an ad for Pepsi is displayed in Youtube App and Twitter App, then youtube and twitter should be considered as a channel ? And the app id would correspond to Pepsi ?
", "rawMarkdown": "Interesting... \n\nWhat do you mean by :\n\"If advertising is displayed within other app, I think this app should be considered as a channel.\"\n\nYou mean that if an ad for Pepsi is displayed in Youtube App and Twitter App, then youtube and twitter should be considered as a channel ? And the app id would correspond to Pepsi ?" }, { "id": 312431, "postDate": "2018-04-11T18:15:39.947Z", "content": "Yes, provided that Pepsi is app. And we do not know that Youtube or Twitter are apps, we see them as channels in the data.
", "rawMarkdown": "Yes, provided that Pepsi is app. And we do not know that Youtube or Twitter are apps, we see them as channels in the data.", "votes": 3 }, { "id": 312432, "postDate": "2018-04-11T18:17:26.287Z", "content": "Thanks for confirming !
", "rawMarkdown": "Thanks for confirming !" }, { "id": 313812, "postDate": "2018-04-13T21:30:14.437Z", "content": "@alexander,
\n\nHow do you explain in this case that App_id has so much importance for LGBM model ? And channel not so much ?
\n\nI was thinking that if people use some kind of robot to make money, they would use their own apps, and run a bot to always click on the same channel ? So that would make the channel have an artificially high click rate ? And that should be easy to detect no ?
\n\nIf you look at the confidence rate on this kernel : \nhttps://www.kaggle.com/nanomathias/feature-engineering-importance-testing
\n\nIt seems that the app has a lot bigger impact than channel. Which I can explain by the fact that some app might have very attractive add, but then very bad content, so people don't end up downloading it.
\n\nI'm just curious why the channel doesn't play a bigger role in this case for LightGBM ? Any idea ?
", "rawMarkdown": "@alexander,\n\nHow do you explain in this case that App_id has so much importance for LGBM model ? And channel not so much ? \n\nI was thinking that if people use some kind of robot to make money, they would use their own apps, and run a bot to always click on the same channel ? So that would make the channel have an artificially high click rate ? And that should be easy to detect no ?\n\nIf you look at the confidence rate on this kernel : \nhttps://www.kaggle.com/nanomathias/feature-engineering-importance-testing\n\nIt seems that the app has a lot bigger impact than channel. Which I can explain by the fact that some app might have very attractive add, but then very bad content, so people don't end up downloading it.\n\nI'm just curious why the channel doesn't play a bigger role in this case for LightGBM ? Any idea ?" }, { "id": 313827, "postDate": "2018-04-13T22:19:29.230Z", "content": "Antoine, I shared my understanding based on the competition data and description. If there were two types of apps (advertized app and advertizing app), it would be explicitely described on data page. We have only one.
\n\nAs for app vs channel importance, I think we should focus on real users, not on bots. Users are interested (or not interested) in the app, but they do not care about channel. Channel does not provide any value itself for the user. I think this explains your question, you also gave nice example.
\n\nThe goal of the bots is to behave as close to the users as possible, and not just click to own ads. Even if we deal with such a primitive bot, we see mixture of bot and user activity, which makes our task harder especially considering statement from organizers that 90% activities are potentially fraudulent.
", "rawMarkdown": "Antoine, I shared my understanding based on the competition data and description. If there were two types of apps (advertized app and advertizing app), it would be explicitely described on data page. We have only one.\n\nAs for app vs channel importance, I think we should focus on real users, not on bots. Users are interested (or not interested) in the app, but they do not care about channel. Channel does not provide any value itself for the user. I think this explains your question, you also gave nice example.\n\nThe goal of the bots is to behave as close to the users as possible, and not just click to own ads. Even if we deal with such a primitive bot, we see mixture of bot and user activity, which makes our task harder especially considering statement from organizers that 90% activities are potentially fraudulent.", "votes": 2 }, { "id": 313847, "postDate": "2018-04-14T00:16:33.373Z", "content": "Thanks Alexander for clarifying this ! You are right it makes more sense that app_id would be the id the app to be downloaded. Apparently I was wrong to think channel might be a good indicator to identify robot click !
", "rawMarkdown": "Thanks Alexander for clarifying this ! You are right it makes more sense that app_id would be the id the app to be downloaded. Apparently I was wrong to think channel might be a good indicator to identify robot click !" }, { "id": 313932, "postDate": "2018-04-14T05:11:18.480Z", "content": "Also, I would say that we have to pay per click on a platform like Google or Facebook. Same would be the case for Talking Data. Hence, deploying bots to fraudulently click on our App would increase the cost per effective click. Thus, no one would want to deploy a bot to click on their app through any channel. So the importance of channel is decreased through this thought process as well.
", "rawMarkdown": "Also, I would say that we have to pay per click on a platform like Google or Facebook. Same would be the case for Talking Data. Hence, deploying bots to fraudulently click on our App would increase the cost per effective click. Thus, no one would want to deploy a bot to click on their app through any channel. So the importance of channel is decreased through this thought process as well. " } ] }, { "id": 310011, "postDate": "2018-04-06T11:56:50.607Z", "content": "That is really important question. I am also looking for an answer.
", "rawMarkdown": "That is really important question. I am also looking for an answer.", "replies": [ { "id": 310014, "postDate": "2018-04-06T12:02:29.797Z", "content": "As it happens I may just have found the answer here : http://doc.talkingdata.com/posts/163 \nIt is described as the unique id of the application being advertised within their ad tracking system.
", "rawMarkdown": "As it happens I may just have found the answer here : http://doc.talkingdata.com/posts/163 \nIt is described as the unique id of the application being advertised within their ad tracking system.", "votes": 2 }, { "id": 310029, "postDate": "2018-04-06T12:31:37.413Z", "content": "Google Translate only says \"Unique identification of APP in Ad Tracking system\".
", "rawMarkdown": "Google Translate only says \"Unique identification of APP in Ad Tracking system\"." }, { "id": 310084, "postDate": "2018-04-06T15:20:45.580Z", "content": "@Ahmet - What is your current assumption : ' App on which the Ad popped up for a click' vs ' App that was advertised in the Ad'.
\n\nIf the ID is about what was advertised then it need not necessarily be an app right? The ad could have been about any other product or service. In that context it should be the app that hosted the ad that led to the click.
", "rawMarkdown": "@Ahmet - What is your current assumption : ' App on which the Ad popped up for a click' vs ' App that was advertised in the Ad'. \n\nIf the ID is about what was advertised then it need not necessarily be an app right? The ad could have been about any other product or service. In that context it should be the app that hosted the ad that led to the click. " }, { "id": 310469, "postDate": "2018-04-07T16:33:05.890Z", "content": "app: app id for marketing.\nchannel: channel id of mobile ad publisher
\n\nIncluding the above information and reading the translated version of http://doc.talkingdata.com/posts/163
\n\nWhat I have understood is following:\napp: id for an app that is being advertised.\nchannel: id for a channel through which this app is being advertised.
\n\nLet say I want to place an advertisement of my app named XYZ on youtube. \nThen youtube is channel while XYZ is an app!
", "rawMarkdown": "app: app id for marketing.\nchannel: channel id of mobile ad publisher\n\nIncluding the above information and reading the translated version of http://doc.talkingdata.com/posts/163\n\nWhat I have understood is following:\napp: id for an app that is being advertised.\nchannel: id for a channel through which this app is being advertised. \n\nLet say I want to place an advertisement of my app named XYZ on youtube. \nThen youtube is channel while XYZ is an app!", "votes": 2 }, { "id": 310943, "postDate": "2018-04-09T04:40:26.577Z", "content": "Agree, looks like ADNETNAME in the doc is the name of place where app was advertised.
", "rawMarkdown": "Agree, looks like ADNETNAME in the doc is the name of place where app was advertised.", "votes": 1 }, { "id": 312273, "postDate": "2018-04-11T13:48:43.017Z", "content": "That would be great to have a confirmation on this...
", "rawMarkdown": "That would be great to have a confirmation on this...", "votes": 1 }, { "id": 312486, "postDate": "2018-04-11T20:40:51.950Z", "content": "is_attributed: the target that is to be predicted, indicating the app was downloaded
\n\n\"if app was downloaded\" this clearly states what app is
", "rawMarkdown": "is_attributed: the target that is to be predicted, indicating the app was downloaded\n\n\"if app was downloaded\" this clearly states what app is", "votes": 1 } ] }, { "id": 310007, "postDate": "2018-04-06T11:50:59.583Z", "content": "I assumed it to be the app on which the ad showed up - Basically the ad that generated the click.
", "rawMarkdown": "I assumed it to be the app on which the ad showed up - Basically the ad that generated the click. ", "replies": [ { "id": 310009, "postDate": "2018-04-06T11:52:48.243Z", "content": "I assumed the same but was worried it was not as straight forward as it seemed.
", "rawMarkdown": "I assumed the same but was worried it was not as straight forward as it seemed." } ] }, { "id": 310026, "postDate": "2018-04-06T12:29:08.970Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 312302, "author_name": "Alexander Firsov", "author_url": "", "post_date": "2018-04-11T14:42:54.473000", "content": "There is only one app in the data, so my understanding is that app is what advertised via channel, and when app is downloaded and installed, it calls TalkingData sdk to report the fact it was installed along with time.\nIf advertising is displayed within other app, I think this app should be considered as a channel.
", "votes": 1, "replies": [ { "id": 312411, "author_name": "Antoine", "author_url": "", "post_date": "2018-04-11T17:45:37.690000", "content": "Interesting...
\n\nWhat do you mean by :\n\"If advertising is displayed within other app, I think this app should be considered as a channel.\"
\n\nYou mean that if an ad for Pepsi is displayed in Youtube App and Twitter App, then youtube and twitter should be considered as a channel ? And the app id would correspond to Pepsi ?
", "votes": 0, "replies": [] }, { "id": 312431, "author_name": "Alexander Firsov", "author_url": "", "post_date": "2018-04-11T18:15:39.947000", "content": "Yes, provided that Pepsi is app. And we do not know that Youtube or Twitter are apps, we see them as channels in the data.
", "votes": 3, "replies": [] }, { "id": 312432, "author_name": "Antoine", "author_url": "", "post_date": "2018-04-11T18:17:26.287000", "content": "Thanks for confirming !
", "votes": 0, "replies": [] }, { "id": 313812, "author_name": "Antoine", "author_url": "", "post_date": "2018-04-13T21:30:14.437000", "content": "@alexander,
\n\nHow do you explain in this case that App_id has so much importance for LGBM model ? And channel not so much ?
\n\nI was thinking that if people use some kind of robot to make money, they would use their own apps, and run a bot to always click on the same channel ? So that would make the channel have an artificially high click rate ? And that should be easy to detect no ?
\n\nIf you look at the confidence rate on this kernel : \nhttps://www.kaggle.com/nanomathias/feature-engineering-importance-testing
\n\nIt seems that the app has a lot bigger impact than channel. Which I can explain by the fact that some app might have very attractive add, but then very bad content, so people don't end up downloading it.
\n\nI'm just curious why the channel doesn't play a bigger role in this case for LightGBM ? Any idea ?
", "votes": 0, "replies": [] }, { "id": 313827, "author_name": "Alexander Firsov", "author_url": "", "post_date": "2018-04-13T22:19:29.230000", "content": "Antoine, I shared my understanding based on the competition data and description. If there were two types of apps (advertized app and advertizing app), it would be explicitely described on data page. We have only one.
\n\nAs for app vs channel importance, I think we should focus on real users, not on bots. Users are interested (or not interested) in the app, but they do not care about channel. Channel does not provide any value itself for the user. I think this explains your question, you also gave nice example.
\n\nThe goal of the bots is to behave as close to the users as possible, and not just click to own ads. Even if we deal with such a primitive bot, we see mixture of bot and user activity, which makes our task harder especially considering statement from organizers that 90% activities are potentially fraudulent.
", "votes": 2, "replies": [] }, { "id": 313847, "author_name": "Antoine", "author_url": "", "post_date": "2018-04-14T00:16:33.373000", "content": "Thanks Alexander for clarifying this ! You are right it makes more sense that app_id would be the id the app to be downloaded. Apparently I was wrong to think channel might be a good indicator to identify robot click !
", "votes": 0, "replies": [] }, { "id": 313932, "author_name": "Gaurav Arora", "author_url": "", "post_date": "2018-04-14T05:11:18.480000", "content": "Also, I would say that we have to pay per click on a platform like Google or Facebook. Same would be the case for Talking Data. Hence, deploying bots to fraudulently click on our App would increase the cost per effective click. Thus, no one would want to deploy a bot to click on their app through any channel. So the importance of channel is decreased through this thought process as well.
", "votes": 0, "replies": [] } ] }, { "id": 310011, "author_name": "Ahmet Erdem", "author_url": "", "post_date": "2018-04-06T11:56:50.607000", "content": "That is really important question. I am also looking for an answer.
", "votes": 0, "replies": [ { "id": 310014, "author_name": "Garry", "author_url": "", "post_date": "2018-04-06T12:02:29.797000", "content": "As it happens I may just have found the answer here : http://doc.talkingdata.com/posts/163 \nIt is described as the unique id of the application being advertised within their ad tracking system.
", "votes": 2, "replies": [] }, { "id": 310029, "author_name": "Ahmet Erdem", "author_url": "", "post_date": "2018-04-06T12:31:37.413000", "content": "Google Translate only says \"Unique identification of APP in Ad Tracking system\".
", "votes": 0, "replies": [] }, { "id": 310084, "author_name": "Shanth", "author_url": "", "post_date": "2018-04-06T15:20:45.580000", "content": "@Ahmet - What is your current assumption : ' App on which the Ad popped up for a click' vs ' App that was advertised in the Ad'.
\n\nIf the ID is about what was advertised then it need not necessarily be an app right? The ad could have been about any other product or service. In that context it should be the app that hosted the ad that led to the click.
", "votes": 0, "replies": [] }, { "id": 310469, "author_name": "Gaurav Arora", "author_url": "", "post_date": "2018-04-07T16:33:05.890000", "content": "app: app id for marketing.\nchannel: channel id of mobile ad publisher
\n\nIncluding the above information and reading the translated version of http://doc.talkingdata.com/posts/163
\n\nWhat I have understood is following:\napp: id for an app that is being advertised.\nchannel: id for a channel through which this app is being advertised.
\n\nLet say I want to place an advertisement of my app named XYZ on youtube. \nThen youtube is channel while XYZ is an app!
", "votes": 2, "replies": [] }, { "id": 310943, "author_name": "Fei", "author_url": "", "post_date": "2018-04-09T04:40:26.577000", "content": "Agree, looks like ADNETNAME in the doc is the name of place where app was advertised.
", "votes": 1, "replies": [] }, { "id": 312273, "author_name": "Antoine", "author_url": "", "post_date": "2018-04-11T13:48:43.017000", "content": "That would be great to have a confirmation on this...
", "votes": 1, "replies": [] }, { "id": 312486, "author_name": "Gaurav Arora", "author_url": "", "post_date": "2018-04-11T20:40:51.950000", "content": "is_attributed: the target that is to be predicted, indicating the app was downloaded
\n\n\"if app was downloaded\" this clearly states what app is
", "votes": 1, "replies": [] } ] }, { "id": 310007, "author_name": "Shanth", "author_url": "", "post_date": "2018-04-06T11:50:59.583000", "content": "I assumed it to be the app on which the ad showed up - Basically the ad that generated the click.
", "votes": 0, "replies": [ { "id": 310009, "author_name": "Garry", "author_url": "", "post_date": "2018-04-06T11:52:48.243000", "content": "I assumed the same but was worried it was not as straight forward as it seemed.
", "votes": 0, "replies": [] } ] }, { "id": 310026, "author_name": "", "author_url": "", "post_date": "2018-04-06T12:29:08.970000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "309999": "Does anybody know (I have not seen anything in the existing discussions) if the app id column represents the id of the application thats being advertised, or, if its the id of the app that hosted the advertisement?", "312302": "There is only one app in the data, so my understanding is that app is what advertised via channel, and when app is downloaded and installed, it calls TalkingData sdk to report the fact it was installed along with time.\nIf advertising is displayed within other app, I think this app should be considered as a channel.", "310011": "That is really important question. I am also looking for an answer.", "310007": "I assumed it to be the app on which the ad showed up - Basically the ad that generated the click. ", "310026": "" } }