NDCG Exercise
- System1 MAP : (1/1 + 2/3 + 3/9 + 4/10) * 1/4
- System2 MAP : (1/2 + 2/5 + 3/6 + 4/7) * 1/4
System1 has higher MAP than System2 - System1 DCG : 1 + 1/log3 + 1/log9 + 1/log10
- System2 DCG : 1 + 1/log5 + 1/log6 + 1/log7
- MaxDCG ( R 을 전부 앞으로 옮긴 후 계산 ) : 1 + 1/log2 + 1/log3 + 1/log4
- NDCG = DCG / MaxDCG
Kendall Tau Distance
X is agree key / Y is disagree key
- X + Y is size of benchmark data
–> Kendall tau distance : (X-Y) / (X+Y)
1 is Maximum value and -1 is Minimun value
Example)
p = { (1,2) , (1,3) , (1,4) , (2,3) , (2,4) , (3,4) } (1,2) means 1 is better than 2 A = ( 1 , 3 , 2 , 4 ) ==> X = 5 , Y = 1 ( p 에서 맞는 것 5개 틀린 것 1개 ) Kendall tau distance : (X-Y) / (X+Y) = 4 / 6 = 2/3