🕸📝Fergus Duniho wrote on Wed, Sep 27, 2017 03:10 PM UTC:
I am thinking about how to use the favorites data for making recommendations. This might be broken down into different things to calculate. One is similarity between the favorites of two members.
The simplest way is to just count up how many games they have favorited in common. But this doesn't take into account the favorites they don't have in common, and it allows somone to become someone else's most similar neighbor by favoriting lots of games. So I want a method that takes into account both the number they have both favorited and the total number of games favorited by either of them.
One simple way to calculate this would be as the total number of games they have in common divided by the total number of games they have each favorited. This would count each game only once, giving a percentage of how similar two lists of favorites are.
A variation on this method is to count a game each time it is favorited. This would double the numerator and increase the denominator. In this case, the denominator would be twice as large only if they favorited all the same games. This would count games they have favorited more than games only one has favorited.
Another option that would do this even more is to square the games they have in common for the numerator and use the total number of games they have each favorited in the numerator. This would amplify the significance of having favorite games in common.
I am thinking about how to use the favorites data for making recommendations. This might be broken down into different things to calculate. One is similarity between the favorites of two members.
The simplest way is to just count up how many games they have favorited in common. But this doesn't take into account the favorites they don't have in common, and it allows somone to become someone else's most similar neighbor by favoriting lots of games. So I want a method that takes into account both the number they have both favorited and the total number of games favorited by either of them.
One simple way to calculate this would be as the total number of games they have in common divided by the total number of games they have each favorited. This would count each game only once, giving a percentage of how similar two lists of favorites are.
A variation on this method is to count a game each time it is favorited. This would double the numerator and increase the denominator. In this case, the denominator would be twice as large only if they favorited all the same games. This would count games they have favorited more than games only one has favorited.
Another option that would do this even more is to square the games they have in common for the numerator and use the total number of games they have each favorited in the numerator. This would amplify the significance of having favorite games in common.
What do the mathematicians here think?