Let's say I have 5 users, each with 5 boolean attributes, which could look like this:
| A B C D E
--------------------------
User 0 | 1 1 0 1 0
User 1 | 0 1 0 1 0
User 2 | 0 0 1 0 1
User 3 | 1 1 0 0 0
User 4 | 0 0 0 1 0
Now what would be the best approach to get a list of the top x users with the most "trues" in common. So in the example above the ranking should look like his:
Top 1: Users 0 (most true attributes)
Top 2: Users 0 and 1 OR Users 0 and 3 (both pairs have 2 attributes in common)
Top 3: Users 0, 1 and 3
Top 4: Users 0, 1, 3 and 4
Top 5: Users 0, 1, 2, 3, 4
I know there are metrics and distance measures to tell how similar two users are, but i want a list of most similar ones. Should i use some kind of clustering algorithm? But which one would consider multiple binary attributes and how could I implement it (preferably in C#)?
Since I haven't taken any classes on data mining, the literature on this topic is kinda overwhelming, so any help is highly appreciated.
User mostTrueUser = Users
.OrderByDescending(u => (u.A?1:0) + (u.B?1:0) + (u.C?1:0) + (u.D?1:0) + (u.E?1:0))
.First();
var groups = Users.GroupBy(u => ((u.A && mostTrueUser.A)?1:0)
+((u.B && mostTrueUser.B)?1:0)
+((u.C && mostTrueUser.C)?1:0)
+((u.D && mostTrueUser.D)?1:0)
+((u.E && mostTrueUser.E)?1:0)
,u => u).OrderByDescending(g => g.Key);
foreach(var group in groups)
{
Console.WriteLine("{0} // following have {0} 'true' in common with {1}",
group.Key,
mostTrueUser.ID);
foreach(var g in group)
{
Console.WriteLine(" " + g.ID);
}
}
This gives me the following:
3 // following have 3 'true' in common with 0
0
2 // following have 2 'true' in common with 0
1
3
1 // following have 1 'true' in common with 0
4
0 // following have 0 'true' in common with 0
2
I used u.A?1:0 so true becomes 1 and false becomes 0.
I then got the User with most true using OrderByDescending([sum of trues]).
Then the GroupBy is used to group all Users on the number of true in common with the mostTrueUser.
Your ranking seems a little bit more complicated, but you can start with this to solve it.
I wrote a little tweak:
public class UserRank
{
public User UserA{get;set;}
public User UserB{get;set;}
public int Compare{
get{return ((UserA.A && UserB.A)?1:0)
+((UserA.B && UserB.B)?1:0)
+((UserA.C && UserB.C)?1:0)
+((UserA.D && UserB.D)?1:0)
+((UserA.E && UserB.E)?1:0);}
}
}
and then:
List<UserRank> userRanks = new List<UserRank>();
for(int i=0;i<Users.Count;i++)
{
for(int j=i;j<Users.Count;j++)
{
userRanks.Add(new UserRank
{
UserA = Users[i],
UserB = Users[j]
});
}
}
var groups = userRanks.GroupBy(u => u.Compare, u => u).OrderByDescending(g => g.Key);
foreach(var group in groups)
{
Console.WriteLine("{0} in common:",group.Key);
foreach(var u in group)
{
Console.WriteLine(" {0}-{1}",u.UserA.ID,u.UserB.ID);
}
}
gives me:
3 in common:
0-0
2 in common:
0-1
0-3
1-1
2-2
3-3
1 in common:
0-4
1-3
1-4
4-4
0 in common:
0-2
1-2
2-3
2-4
3-4