Hyperparameter ranges for all considered baseline models

Algorithms topK shrink Similarity Normalize α β ℓ2 norm
kNN algorithms [5 to 1000] [0 to 1000] [cosine, jaccard, asymmetric, dice, tversky] [True, False] — — —
Graph based algorithms [5 to 1000] — — — [0 to 2], strategy: uniform distribution [0 to 2], strategy: uniform distribution —
EASER — — — — — — [1 to 1e7], strategy: log-uniform distribution


Optimal hyperparameter values

Algorithms topK shrink Similarity Normalize α β ℓ2 norm
ItemkNN 144 1000 cosine True — — —
UserkNN 146 1000 cosine True — — —
P3α 496 — — False 0.7681732734954694 — —
RP3β 496 — — True 0.44477903655656115 0.5968193614337285 —
EASER — — — — — — 1e3

Statistics of the provided train-test splits

Users Items Interactions Density Data leakage Interactions per usermin Interactions per itemmin Interactions per usermax Interactions per itemmax Testing users
78,578 77,801 2,240,156 0.00062 No 16 10 13313 1907 52,639


Experimental Results

Methods R@20 R@40 N@20 N@40 T-Time(s) P-Time(s) Average Perdiction Time Per User(ms)
Random 0.0004 0.0006 0.0002 0.0003 0 209.023 2.660
TopPopular 0.009 0.016 0.007 0.009 0.036 186.625 2.375
UserKNN 0.126 0.172 0.104 0.119 386.199 154.089 1.960
ItemKNN 0.152 0.205 0.127 0.144 374.143 154.483 1.965
P3α 0.133 0.188 0.107 0.124 168.419 167.800 2.135
RP3β 0.141 0.190 0.117 0.133 183.405 177.312 2.250
EASE — — — — — — —
DCCF (Code) 0.092 0.138 0.071 0.086 - - -
DCCF (Paper) 0.088 0.134 0.068 0.082

Optimal hyperparameter values

Algorithms topK Shrink Similarity Normalize α β ℓ2 norm
ItemkNN 144 1000 cosine True — — —
UserkNN 146 1000 cosine True — — —
P3α 496 — — False 0.7681732734954694 — —
RP3β 496 — — True 0.44477903655656115 0.5968193614337285 —
EASER — — — — — — 1e3

Statistics of the provided train-test splits

Users Items Interactions Density Data leakage Interactions per usermin Interactions per itemmin Interactions per usermax Interactions per itemmax Testing users
50,821 41,390 1,172,425 0.00040 No 5 1 1037 2513 50,821


Experimental Results

Methods R@20 R@40 N@20 N@40 T-Time(s) P-Time(s) Average Perdiction Time Per User(ms)
Random 0.0004 0.0007 0.0001 0.0003 0 86.736 1.705
TopPopular 0.025 0.037 0.013 0.017 0.013 71.108 1.399
UserKNN 0.213 0.295 0.132 0.153 124.524 66.588 1.310
ItemKNN 0.235 0.321 0.146 0.168 169.421 80.366 1.581
P3α 0.231 0.317 0.141 0.164 130.398 74.102 1.458
RP3β 0.245 0.337 0.149 0.173 134.385 73.912 1.454
EASE 0.218 0.303 0.132 0.154 2497.719 134.283 2.642
DCCF (Code) 0.189 0.266 0.112 0.132 - - -
DCCF (Paper) 0.187 0.264 0.112 0.132

Optimal hyperparameter values

Algorithms topK shrink Similarity Normalize α β ℓ2 norm
ItemkNN 144 1000 cosine True — — —
UserkNN 146 1000 cosine True — — —
P3α 496 — — False 0.7681732734954694 — —
RP3β 496 — — True 0.44477903655656115 0.5968193614337285 —
EASER — — — — — — 1e3

Statistics of the provided train-test splits

Users Items Interactions Density Data leakage Interactions per usermin Interactions per itemmin Interactions per usermax Interactions per itemmax Testing users
47,939 41,390 2,357,450 0.00119 No 20 3 251 2408 47,528


Experimental Results>
Methods R@20 R@40 N@20 N@40 T-Time(s) P-Time(s) Average Perdiction Time Per User(ms)
Random 0.0005 0.0010 0.0003 0.0005 0 92.281 1.924
TopPopular 0.010 0.017 0.007 0.009 0.051 82.401 1.718
UserKNN 0.087 0.131 0.063 0.079 225.227 86.014 1.794
ItemKNN 0.110 0.163 0.080 0.099 232.223 90.064 1.878
P3α 0.109 0.162 0.080 0.099 80.009 93.111 1.942
RP3β 0.116 0.173 0.084 0.104 70.928 96.728 2.017
EASE 0.093 0.143 0.067 0.084 662.472 158.241 3.300
DCCF (Code) 0.064 0.101 0.045 0.058
DCCF (Paper) 0.066 0.104 0.046 0.059