Research Article
Improving Top-N Recommendation Performance Using Missing Data
Table 2
Performance of different approaches on EM.
| ā | NDCG | Recall | 1-call | COV | CIL | NDCG+ | ā | 1 | 3 | 5 | 1 | 3 | 5 |
| UserCF | 0.010 | 0.014 | 0.019 | 0.002 | 0.008 | 0.017 | 0.12 | 174.0 | 116.2 | 0.65 | Slope-one | 0.035 | 0.034 | 0.034 | 0.008 | 0.019 | 0.027 | 0.15 | 11.0 | 7.0 | 0.71 | SVD++ | 0.112 | 0.091 | 0.088 | 0.031 | 0.058 | 0.080 | 0.28 | 30.6 | 12.6 | 0.83 | OrdRec | 0.063 | 0.057 | 0.049 | 0.015 | 0.040 | 0.044 | 0.19 | 10.6 | 6.8 | 0.66 | Pure | 0.074 | 0.071 | 0.071 | 0.018 | 0.039 | 0.052 | 0.27 | 38.8 | 15.0 | 0.69 | AllRank | 0.111 | 0.104 | 0.103 | 0.031 | 0.073 | 0.103 | 0.37 | 13.4 | 4.4 | 0.83 | WSVD++ | 0.081 | 0.098 | 0.100 | 0.020 | 0.065 | 0.090 | 0.36 | 56.0 | 26.0 | 0.80 | RSSVD++ | 0.095 | 0.096 | 0.100 | 0.027 | 0.061 | 0.086 | 0.34 | 52.8 | 23.6 | 0.79 | NSSVD++ | 0.153 | 0.148 | 0.150 | 0.042 | 0.102 | 0.143 | 0.50 | 58.2 | 23.8 | 0.81 |
|
|