New ask Hacker News story: Ask HN: Why do recommendation algorithms suck?
Ask HN: Why do recommendation algorithms suck?
6 by nassimBreeze | 4 comments on Hacker News.
I see plenty of discussions about recommendation algorithms sucking (ex: YouTube). When I hear about recommendation algorithms someone always brings up Machine Learning. I've been thinking about how to make a better recommendation algorithm, here's my idea: First, we ask the user to select topics he likes from a given list. Second, we ask the user to select topics he dislikes from the same given list. Then we present recommendations of 4 types : (Familiar) : Content from topics the user clearly likes. (Fresh): Content from topics that the user likes that intersects with topics that user does not dislike. (Novel): Content from topics that the user does not dislike. (Hit-or-miss): Content from liked topics that intersects with disliked topics or content from liked topics that intersects with not-disliked topics that intersects with disliked topics. (more brefly: [like and disliked] or [liked and not-disliked and disliked]) We present those 4 types with the following ratio : (Familiar) 50% (Fresh) 30% (Novel) 18% (Hit-or-miss) 2% Each time the user watches a video from type (Fresh),(Novel) and (Hit-or-miss) we give the opportunity to the user to add the topics of the video that are in doesn't disliked or disliked to the liked category or not-disliked category. If the user does not do anything it stays in their respective categories. At any time the user can change his preferences in the settings. I wonder if others think that this is a good idea for handling recommendations? Do you think that it is better to do it that way as opposed to relying on machine learning or the tiktok 5 sec rule. Was inspired by this discussion specifically : https://ift.tt/3j0C4Kh
6 by nassimBreeze | 4 comments on Hacker News.
I see plenty of discussions about recommendation algorithms sucking (ex: YouTube). When I hear about recommendation algorithms someone always brings up Machine Learning. I've been thinking about how to make a better recommendation algorithm, here's my idea: First, we ask the user to select topics he likes from a given list. Second, we ask the user to select topics he dislikes from the same given list. Then we present recommendations of 4 types : (Familiar) : Content from topics the user clearly likes. (Fresh): Content from topics that the user likes that intersects with topics that user does not dislike. (Novel): Content from topics that the user does not dislike. (Hit-or-miss): Content from liked topics that intersects with disliked topics or content from liked topics that intersects with not-disliked topics that intersects with disliked topics. (more brefly: [like and disliked] or [liked and not-disliked and disliked]) We present those 4 types with the following ratio : (Familiar) 50% (Fresh) 30% (Novel) 18% (Hit-or-miss) 2% Each time the user watches a video from type (Fresh),(Novel) and (Hit-or-miss) we give the opportunity to the user to add the topics of the video that are in doesn't disliked or disliked to the liked category or not-disliked category. If the user does not do anything it stays in their respective categories. At any time the user can change his preferences in the settings. I wonder if others think that this is a good idea for handling recommendations? Do you think that it is better to do it that way as opposed to relying on machine learning or the tiktok 5 sec rule. Was inspired by this discussion specifically : https://ift.tt/3j0C4Kh
Comments
Post a Comment