This might be a stupid question, since I personally do not use a mobile device.\nI want to put each column of the data into more context, which might eventually help in feature engineering.
\n\nWhat is a fraudulent click?
\n\nIn terms of data, it is simply defined as \"a history showing a click on an ad but no follow-up downloads (of what the ad points to? I presume?)\".\nThat is well-defined by the data, but what is it in the real world?
\n\nI can think of a few possibilities.
\n\nA user who intentionally clicks on the ad without the intention to follow up.
A user who clicks and finds out that it's not what she really wants.
A malicious program which behaves like a user doing either of the above.
An (ill-designed) app/os/device that misguides users to click on things they don't really want.
It seems like knowing how this actually works will help us finding the better combination of features.
\n\nOr, maybe it is the other way around.\nMaybe the only way to answer my question is to first find out what features provide the best prediction, and use them to derive what happens in the real world?
", "messages": [ { "id": "307149", "postDate": "03/31/2018 23:24:23", "content": "This might be a stupid question, since I personally do not use a mobile device.\nI want to put each column of the data into more context, which might eventually help in feature engineering.
\n\nWhat is a fraudulent click?
\n\nIn terms of data, it is simply defined as \"a history showing a click on an ad but no follow-up downloads (of what the ad points to? I presume?)\".\nThat is well-defined by the data, but what is it in the real world?
\n\nI can think of a few possibilities.
\n\nA user who intentionally clicks on the ad without the intention to follow up.
A user who clicks and finds out that it's not what she really wants.
A malicious program which behaves like a user doing either of the above.
An (ill-designed) app/os/device that misguides users to click on things they don't really want.
It seems like knowing how this actually works will help us finding the better combination of features.
\n\nOr, maybe it is the other way around.\nMaybe the only way to answer my question is to first find out what features provide the best prediction, and use them to derive what happens in the real world?
", "rawMarkdown": "This might be a stupid question, since I personally do not use a mobile device.\nI want to put each column of the data into more context, which might eventually help in feature engineering.\n\n**What is a fraudulent click?**\n\nIn terms of data, it is simply defined as \"*a history showing a click on an ad but no follow-up downloads (of what the ad points to? I presume?)*\".\nThat is well-defined by the data, but what is it **in the real world**?\n\nI can think of a few possibilities.\n\n1. A user who intentionally clicks on the ad without the intention to follow up.\n\n2. A user who clicks and finds out that it's not what she really wants.\n\n3. A malicious program which behaves like a user doing either of the above.\n\n4. An (ill-designed) app/os/device that misguides users to click on things they don't really want.\n\nIt seems like knowing how this actually works will help us finding the better combination of features.\n\nOr, maybe it is the other way around.\nMaybe the only way to answer my question is to first find out what features provide the best prediction, and use them to derive what happens in the real world?", "votes": null }, { "id": "307244", "postDate": "04/01/2018 07:07:53", "content": "There are a LOT of \"normal, legal\" clicks that don't result in a conversion. That's just the way things are. Imagine youself going shopping for a t-shirt. Are you going to buy every single shirt you see?\nFraudant clicks in my understanding are those coming from bots (a milicious programs). There are advertisers that pay per click with the idea that the more clicks - the more conversions. So I guess you can see where things go wrong and why people want to limit those clicks that are not \"real'. Next to that, I doubt the whole competition and the goal is really to understand the fraudant clicks. In the bigger picture it is all about being able to predict if a click will end up in conversion. If it is not a conversion, you (the advertiser) does not really care much if it is a fraudant or not. (a small * is that, they do care a bit of course, if there are a lot of clicks that don't convert and are not fraudant, that might mean that the ad by itself is deceiving or the product is not good and therefore people are interested and click, but then are dissapointed and dont convert)
", "rawMarkdown": "There are a LOT of \"normal, legal\" clicks that don't result in a conversion. That's just the way things are. Imagine youself going shopping for a t-shirt. Are you going to buy every single shirt you see?\nFraudant clicks in my understanding are those coming from bots (a milicious programs). There are advertisers that pay per click with the idea that the more clicks - the more conversions. So I guess you can see where things go wrong and why people want to limit those clicks that are not \"real'. Next to that, I doubt the whole competition and the goal is really to understand the fraudant clicks. In the bigger picture it is all about being able to predict if a click will end up in conversion. If it is not a conversion, you (the advertiser) does not really care much if it is a fraudant or not. (a small * is that, they do care a bit of course, if there are a lot of clicks that don't convert and are not fraudant, that might mean that the ad by itself is deceiving or the product is not good and therefore people are interested and click, but then are dissapointed and dont convert)", "votes": null } ], "comments": [ { "id": 307244, "author_name": "asparuhhristov", "author_url": "", "post_date": "04/01/2018 07:07:53", "content": "There are a LOT of \"normal, legal\" clicks that don't result in a conversion. That's just the way things are. Imagine youself going shopping for a t-shirt. Are you going to buy every single shirt you see?\nFraudant clicks in my understanding are those coming from bots (a milicious programs). There are advertisers that pay per click with the idea that the more clicks - the more conversions. So I guess you can see where things go wrong and why people want to limit those clicks that are not \"real'. Next to that, I doubt the whole competition and the goal is really to understand the fraudant clicks. In the bigger picture it is all about being able to predict if a click will end up in conversion. If it is not a conversion, you (the advertiser) does not really care much if it is a fraudant or not. (a small * is that, they do care a bit of course, if there are a lot of clicks that don't convert and are not fraudant, that might mean that the ad by itself is deceiving or the product is not good and therefore people are interested and click, but then are dissapointed and dont convert)
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "307149": "This might be a stupid question, since I personally do not use a mobile device.\nI want to put each column of the data into more context, which might eventually help in feature engineering.\n\n**What is a fraudulent click?**\n\nIn terms of data, it is simply defined as \"*a history showing a click on an ad but no follow-up downloads (of what the ad points to? I presume?)*\".\nThat is well-defined by the data, but what is it **in the real world**?\n\nI can think of a few possibilities.\n\n1. A user who intentionally clicks on the ad without the intention to follow up.\n\n2. A user who clicks and finds out that it's not what she really wants.\n\n3. A malicious program which behaves like a user doing either of the above.\n\n4. An (ill-designed) app/os/device that misguides users to click on things they don't really want.\n\nIt seems like knowing how this actually works will help us finding the better combination of features.\n\nOr, maybe it is the other way around.\nMaybe the only way to answer my question is to first find out what features provide the best prediction, and use them to derive what happens in the real world?", "307244": "There are a LOT of \"normal, legal\" clicks that don't result in a conversion. That's just the way things are. Imagine youself going shopping for a t-shirt. Are you going to buy every single shirt you see?\nFraudant clicks in my understanding are those coming from bots (a milicious programs). There are advertisers that pay per click with the idea that the more clicks - the more conversions. So I guess you can see where things go wrong and why people want to limit those clicks that are not \"real'. Next to that, I doubt the whole competition and the goal is really to understand the fraudant clicks. In the bigger picture it is all about being able to predict if a click will end up in conversion. If it is not a conversion, you (the advertiser) does not really care much if it is a fraudant or not. (a small * is that, they do care a bit of course, if there are a lot of clicks that don't convert and are not fraudant, that might mean that the ad by itself is deceiving or the product is not good and therefore people are interested and click, but then are dissapointed and dont convert)" }, "source": "meta" }