Skip to content

🔎 VBPR — VLM & CNN (Top-2)

Dataset: MovieLens 1M | Model: VBPR | Top-N: 2

Each section below corresponds to a different visual encoder configuration. Rows represent the modality/fusion variant used: audio, visual, text, pca_90, cca_40.


🧩 10 Epochs

Visual CLIP + Audio (BLF) MMTF + Text RAG-Plus (Llama LLM)

Modality NDCG@N Novelty Recall@N Fairness Diversity Coverage@N Cold Start Rate Popularity Bias Calibration Bias Hit Rate Precision MAP
audio 0.1131 11.077 0.0651 0.8350 0.7322 0.5230 0.0000 0.0213 8.3038 0.7434 0.3717 0.0560
visual 0.1160 11.159 0.0672 0.8334 0.7419 0.4504 0.0000 0.0170 8.4118 0.7484 0.3742 0.0581
text 0.1089 10.905 0.0623 0.8201 0.7466 0.3970 0.0000 0.0236 8.3405 0.7266 0.3633 0.0529
pca_90 0.1144 10.647 0.0665 0.8143 0.7367 0.5111 0.0000 0.0325 8.4925 0.7360 0.3680 0.0568
cca_40 0.1182 10.630 0.0689 0.8093 0.6891 0.6163 0.0000 0.0292 8.5439 0.7655 0.3828 0.0590

Visual Dino-v2-base + Audio (BLF) MMTF + Text RAG-Plus (Llama LLM)

Modality NDCG@N Novelty Recall@N Fairness Diversity Coverage@N Cold Start Rate Popularity Bias Calibration Bias Hit Rate Precision MAP
audio 0.1162 11.462 0.0676 0.8780 0.7280 0.5363 0.0000 0.0176 8.3820 0.7555 0.3777 0.0575
visual 0.1165 10.724 0.0674 0.8039 0.7422 0.4415 0.0000 0.0290 8.5095 0.7506 0.3753 0.0579
text 0.1125 10.739 0.0641 0.8105 0.7431 0.4607 0.0000 0.0267 8.2205 0.7405 0.3703 0.0551
pca_90 0.1122 10.461 0.0643 0.7868 0.7497 0.5630 0.0000 0.0339 8.5108 0.7284 0.3642 0.0555
cca_40 0.1136 11.065 0.0664 0.8471 0.7143 0.6459 0.0000 0.0247 8.5956 0.7346 0.3673 0.0568

Visual Dino-v2-large + Audio (BLF) MMTF + Text RAG-Plus (Llama LLM)

Modality NDCG@N Novelty Recall@N Fairness Diversity Coverage@N Cold Start Rate Popularity Bias Calibration Bias Hit Rate Precision MAP
audio 0.1059 10.927 0.0600 0.8165 0.7466 0.5170 0.0000 0.0194 8.4265 0.7047 0.3524 0.0514
visual 0.1308 10.767 0.0790 0.8179 0.7300 0.5541 0.0000 0.0287 8.2404 0.8016 0.4008 0.0675
text 0.1160 10.653 0.0659 0.7949 0.7379 0.3585 0.0000 0.0260 8.2381 0.7699 0.3850 0.0569
pca_90 0.1170 10.556 0.0680 0.7976 0.7300 0.4874 0.0000 0.0325 8.5240 0.7491 0.3745 0.0586
cca_40 0.1178 10.866 0.0685 0.8257 0.7061 0.7111 0.0000 0.0299 8.6945 0.7434 0.3717 0.0591

Visual SigLip-Base + Audio (BLF) MMTF + Text RAG-Plus (Llama LLM)

Modality NDCG@N Novelty Recall@N Fairness Diversity Coverage@N Cold Start Rate Popularity Bias Calibration Bias Hit Rate Precision MAP
audio 0.1228 11.597 0.0753 0.8913 0.7331 0.5206 0.0000 0.0161 8.1601 0.7032 0.3516 0.0650
visual 0.1355 11.337 0.0845 0.8650 0.7244 0.5206 0.0000 0.0220 8.1422 0.7455 0.3727 0.0732
text 0.1246 11.839 0.0783 0.9196 0.7378 0.5864 0.0000 0.0126 7.9893 0.6865 0.3433 0.0669
pca_90 0.1282 11.157 0.0796 0.8439 0.7384 0.5514 0.0000 0.0207 8.2217 0.7122 0.3561 0.0689
cca_40 0.1266 11.576 0.0786 0.8941 0.7190 0.6728 0.0000 0.0183 8.3766 0.7020 0.3510 0.0679

Visual Sig2Lip-Base + Audio (BLF) MMTF + Text RAG-Plus (Llama LLM)

Modality NDCG@N Novelty Recall@N Fairness Diversity Coverage@N Cold Start Rate Popularity Bias Calibration Bias Hit Rate Precision MAP
audio 0.1173 10.894 0.0693 0.8278 0.7199 0.5511 0.0000 0.0244 8.5302 0.7541 0.3771 0.0585
visual 0.1276 11.020 0.0744 0.8362 0.7240 0.5289 0.0000 0.0245 8.4039 0.7986 0.3993 0.0649
text 0.1063 10.945 0.0605 0.8124 0.7552 0.4415 0.0000 0.0193 8.3051 0.7073 0.3536 0.0518
pca_90 0.1071 11.193 0.0614 0.8569 0.7167 0.4163 0.0000 0.0232 8.2746 0.7089 0.3545 0.0525
cca_40 0.1092 10.765 0.0634 0.8172 0.7280 0.5941 0.0000 0.0262 8.6837 0.7130 0.3565 0.0542

Visual OpenCLIP + Audio (BLF) MMTF + Text RAG-Plus (Llama LLM)

Modality NDCG@N Novelty Recall@N Fairness Diversity Coverage@N Cold Start Rate Popularity Bias Calibration Bias Hit Rate Precision MAP
audio 0.1140 11.100 0.0652 0.8365 0.7352 0.5259 0.0000 0.0199 8.5179 0.7469 0.3734 0.0562
visual 0.1185 10.809 0.0688 0.8077 0.7128 0.4504 0.0000 0.0231 8.3700 0.7644 0.3822 0.0592
text 0.1227 10.543 0.0721 0.7982 0.7350 0.4370 0.0000 0.0331 8.3156 0.7793 0.3897 0.0620
pca_90 0.1081 10.652 0.0618 0.8053 0.7327 0.4756 0.0000 0.0304 8.4065 0.7136 0.3568 0.0529
cca_40 0.1229 11.007 0.0733 0.8335 0.7065 0.7200 0.0000 0.0235 8.5275 0.7546 0.3773 0.0630

🧩 2 Epochs

Visual CLIP + Audio (BLF) MMTF + Text RAG-Plus (Llama LLM)

Modality NDCG@N Novelty Recall@N Fairness Diversity Coverage@N Cold Start Rate Popularity Bias Calibration Bias Hit Rate Precision MAP
audio 0.0962 10.587 0.0532 0.7730 0.7535 0.2444 0.0000 0.0233 8.5580 0.6614 0.3307 0.0460
visual 0.0958 11.406 0.0546 0.8624 0.7934 0.2163 0.0000 0.0148 8.5207 0.6631 0.3315 0.0457
text 0.0859 10.247 0.0480 0.7483 0.7733 0.1526 0.0000 0.0327 8.6328 0.5972 0.2986 0.0408
pca_90 0.0853 10.276 0.0470 0.7489 0.7533 0.1911 0.0000 0.0335 8.8760 0.6001 0.3001 0.0401
cca_40 0.0915 11.190 0.0514 0.8429 0.8143 0.2444 0.0000 0.0201 9.2061 0.6364 0.3182 0.0431

Visual Dino-v2-base + Audio (BLF) MMTF + Text RAG-Plus (Llama LLM)

Modality NDCG@N Novelty Recall@N Fairness Diversity Coverage@N Cold Start Rate Popularity Bias Calibration Bias Hit Rate Precision MAP
audio 0.0951 10.426 0.0525 0.7423 0.7850 0.2415 0.0000 0.0231 8.6112 0.6631 0.3315 0.0450
visual 0.0888 10.305 0.0483 0.7372 0.7481 0.1496 0.0000 0.0254 8.3521 0.6221 0.3111 0.0419
text 0.0883 10.172 0.0487 0.7546 0.7666 0.1259 0.0000 0.0371 8.3107 0.6129 0.3064 0.0420
pca_90 0.0887 10.791 0.0499 0.8141 0.7778 0.1763 0.0000 0.0296 8.5930 0.6098 0.3049 0.0423
cca_40 0.0963 10.841 0.0536 0.8059 0.7897 0.2711 0.0000 0.0244 8.9431 0.6627 0.3314 0.0461

Visual Dino-v2-large + Audio (BLF) MMTF + Text RAG-Plus (Llama LLM)

Modality NDCG@N Novelty Recall@N Fairness Diversity Coverage@N Cold Start Rate Popularity Bias Calibration Bias Hit Rate Precision MAP
audio 0.0992 10.517 0.0553 0.7566 0.7597 0.2311 0.0000 0.0225 8.5493 0.6789 0.3394 0.0475
visual 0.0971 10.892 0.0538 0.8220 0.7404 0.1541 0.0000 0.0276 8.7179 0.6648 0.3324 0.0467
text 0.0932 11.326 0.0508 0.8600 0.7722 0.1496 0.0000 0.0184 8.1285 0.6382 0.3191 0.0442
pca_90 0.0885 10.615 0.0487 0.7925 0.7936 0.2015 0.0000 0.0327 9.0053 0.6125 0.3063 0.0418
cca_40 0.0954 10.073 0.0527 0.7412 0.7643 0.1778 0.0000 0.0327 8.9691 0.6515 0.3257 0.0457

Visual SigLip-base + Audio (BLF) MMTF + Text RAG-Plus (Llama LLM)

Modality NDCG@N Novelty Recall@N Fairness Diversity Coverage@N Cold Start Rate Popularity Bias Calibration Bias Hit Rate Precision MAP
audio 0.1052 10.746 0.0631 0.7979 0.7546 0.2058 0.0000 0.0308 8.6544 0.6164 0.3082 0.0542
visual 0.1151 11.068 0.0699 0.8310 0.7491 0.1996 0.0000 0.0261 8.2956 0.6501 0.3251 0.0609
text 0.1009 10.566 0.0609 0.7765 0.7448 0.1399 0.0000 0.0313 8.9049 0.5874 0.2937 0.0522
pca_90 0.1151 10.944 0.0695 0.8195 0.7501 0.2675 0.0000 0.0276 8.4699 0.6558 0.3279 0.0603
cca_40 0.1165 11.485 0.0717 0.8779 0.8075 0.2469 0.0000 0.0175 8.6999 0.6642 0.3321 0.0613

Visual SigLip2-base + Audio (BLF) MMTF + Text RAG-Plus (Llama LLM)

Modality NDCG@N Novelty Recall@N Fairness Diversity Coverage@N Cold Start Rate Popularity Bias Calibration Bias Hit Rate Precision MAP
audio 0.0917 10.029 0.0507 0.7161 0.7394 0.1911 0.0000 0.0289 8.6536 0.6387 0.3194 0.0436
visual 0.1071 11.304 0.0602 0.8582 0.7400 0.2593 0.0000 0.0180 8.5123 0.7165 0.3582 0.0521
text 0.0923 10.402 0.0506 0.7778 0.7773 0.1422 0.0000 0.0335 8.3093 0.6382 0.3191 0.0439
pca_90 0.0816 10.307 0.0453 0.7530 0.7867 0.1659 0.0000 0.0348 8.9329 0.5887 0.2943 0.0376
cca_40 0.0931 10.167 0.0520 0.7495 0.7820 0.2193 0.0000 0.0349 8.9537 0.6475 0.3237 0.0442

Visual CNN MMTF + Audio (BLF) MMTF + Text RAG-Plus (Llama LLM) — 10 Epochs

Modality NDCG@N Novelty Recall@N Fairness Diversity Coverage@N Cold Start Rate Popularity Bias Calibration Bias Hit Rate Precision MAP
audio 0.0947 10.815 0.0523 0.7898 0.7817 0.1867 0.0000 0.0198 8.5521 0.6491 0.3246 0.0454
visual 0.1046 10.689 0.0590 0.7990 0.7226 0.2741 0.0000 0.0264 8.2722 0.7029 0.3514 0.0509
text 0.0938 11.220 0.0523 0.8374 0.7571 0.1807 0.0000 0.0167 8.1464 0.6374 0.3187 0.0450
pca_90 0.0839 9.695 0.0459 0.7293 0.7368 0.1733 0.0000 0.0468 8.9406 0.5932 0.2966 0.0391
cca_40 0.0986 10.297 0.0545 0.7545 0.7877 0.2281 0.0000 0.0273 8.6382 0.6769 0.3384 0.0470