Skip to content

🔎 VBPR — VLM & CNN (Top-10)

Dataset: MovieLens 1M | Model: VBPR | Top-N: 10

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


🧩 10 Epochs

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

Modality NDCG@N Novelty Recall@N Fairness Diversity Coverage@N Cold Start Rate Pop. Bias Cal. Bias Hit Rate Precision MAP
audio 0.2315 11.2937 0.2167 0.3271 0.7667 0.7481 0.0000 0.0187 2.9594 2.6859 0.2686 0.1162
visual 0.2496 11.0438 0.2273 0.2804 0.7604 0.7304 0.0000 0.0233 2.9439 2.7842 0.2784 0.1305
text 0.2412 10.9604 0.2210 0.2719 0.7785 0.6652 0.0000 0.0261 2.9302 2.7341 0.2734 0.1239
pca_90 0.2366 11.2486 0.2174 0.3108 0.7683 0.7793 0.0000 0.0218 2.9552 2.6831 0.2683 0.1205

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

Modality NDCG@N Novelty Recall@N Fairness Diversity Coverage@N Cold Start Rate Pop. Bias Cal. Bias Hit Rate Precision MAP
audio 0.2443 11.1350 0.2262 0.2985 0.7647 0.8652 0.0000 0.0239 3.0158 2.6701 0.2670 0.1269
visual 0.2501 11.0575 0.2285 0.2868 0.7671 0.7259 0.0000 0.0259 2.9942 2.7877 0.2788 0.1306
text 0.2434 11.0611 0.2234 0.2592 0.7833 0.6874 0.0000 0.0219 2.8584 2.7284 0.2728 0.1248
pca_90 0.2243 11.0128 0.2077 0.2715 0.7802 0.8178 0.0000 0.0283 2.8460 2.5499 0.2550 0.1121
cca_40 0.2398 10.9527 0.2202 0.3043 0.7490 0.8563 0.0000 0.0272 3.2483 2.7209 0.2721 0.1233

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

Modality NDCG@N Novelty Recall@N Fairness Diversity Coverage@N Cold Start Rate Pop. Bias Cal. Bias Hit Rate Precision MAP
audio 0.2634 11.6951 0.2626 0.3988 0.7677 0.8704 0.0000 0.0176 2.7733 2.4827 0.2483 0.1441
visual 0.2691 11.7901 0.2620 0.4269 0.7656 0.7675 0.0000 0.0151 2.9011 2.5515 0.2551 0.1490
text 0.2397 11.8323 0.2390 0.4436 0.7761 0.7778 0.0000 0.0141 2.8981 2.3303 0.2330 0.1265
pca_90 0.2424 11.7128 0.2400 0.3979 0.7714 0.8519 0.0000 0.0148 2.8187 2.3583 0.2358 0.1287
cca_40 0.2676 11.6929 0.2615 0.4263 0.7488 0.9177 0.0000 0.0176 3.1253 2.5540 0.2554 0.1480

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 Pop. Bias Cal. Bias Hit Rate Precision MAP
audio 0.2480 10.9619 0.2293 0.2631 0.7660 0.8444 0.0000 0.0265 2.9799 2.7259 0.2726 0.1285
visual 0.2480 11.0000 0.2261 0.2564 0.7762 0.6652 0.0000 0.0262 2.8660 2.7798 0.2780 0.1293
text 0.2456 10.9186 0.2233 0.2653 0.7738 0.7156 0.0000 0.0258 2.7431 2.7618 0.2762 0.1275
pca_90 0.2343 10.9903 0.2126 0.2688 0.7798 0.7867 0.0000 0.0280 2.8005 2.6275 0.2627 0.1193
cca_40 0.2503 11.0393 0.2291 0.3053 0.7366 0.8756 0.0000 0.0275 3.3575 2.8008 0.2801 0.1305

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 Pop. Bias Cal. Bias Hit Rate Precision MAP
audio 0.2358 11.0975 0.2145 0.2702 0.7663 0.7570 0.0000 0.0224 2.9252 2.6678 0.2668 0.1203
visual 0.2425 10.9873 0.2242 0.2631 0.7767 0.7096 0.0000 0.0268 2.8832 2.7205 0.2721 0.1248
text 0.2404 10.9118 0.2208 0.2591 0.7825 0.6859 0.0000 0.0269 2.7389 2.7104 0.2710 0.1234
pca_90 0.2367 10.9338 0.2155 0.2592 0.7764 0.7289 0.0000 0.0283 2.8834 2.6579 0.2658 0.1215
cca_40 0.2456 11.1145 0.2259 0.3381 0.7410 0.8933 0.0000 0.0256 3.3349 2.7581 0.2758 0.1277

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

Modality NDCG@N Novelty Recall@N Fairness Diversity Coverage@N Cold Start Rate Pop. Bias Cal. Bias Hit Rate Precision MAP
audio 0.2374 11.1817 0.2150 0.2984 0.7648 0.7481 0.0000 0.0215 2.9852 2.6923 0.2692 0.1214
visual 0.2536 11.1786 0.2349 0.2935 0.7700 0.7852 0.0000 0.0228 2.8731 2.8030 0.2803 0.1328
text 0.2306 11.0898 0.2107 0.2667 0.7860 0.6400 0.0000 0.0221 2.5969 2.6300 0.2630 0.1165
pca_90 0.2323 10.9821 0.2134 0.2680 0.7732 0.8237 0.0000 0.0278 2.9539 2.6014 0.2601 0.1180
cca_40 0.2334 11.1568 0.2155 0.3214 0.7408 0.8785 0.0000 0.0241 3.3436 2.6787 0.2679 0.1189

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

Modality NDCG@N Novelty Recall@N Fairness Diversity Coverage@N Cold Start Rate Pop. Bias Cal. Bias Hit Rate Precision MAP
audio 0.2333 11.1032 0.2166 0.2972 0.7676 0.8296 0.0000 0.0244 2.9771 2.6263 0.2626 0.1176
visual 0.2220 11.3707 0.2030 0.3339 0.7738 0.6267 0.0000 0.0196 2.8274 2.5857 0.2586 0.1109
text 0.2444 11.0139 0.2260 0.2627 0.7747 0.7126 0.0000 0.0273 2.8429 2.7482 0.2748 0.1259
pca_90 0.2279 11.0740 0.2089 0.3051 0.7565 0.7674 0.0000 0.0289 3.3164 2.6327 0.2633 0.1146

🧩 2 Epochs

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

Modality NDCG@N Novelty Recall@N Fairness Diversity Coverage@N Cold Start Rate Pop. Bias Cal. Bias Hit Rate Precision MAP
audio 0.1985 10.6393 0.1753 0.2446 0.7790 0.4756 0.0000 0.0286 3.2023 2.3549 0.2355 0.0965
visual 0.1511 10.9327 0.1364 0.2250 0.8229 0.2948 0.0000 0.0251 3.2146 1.9237 0.1924 0.0673
text 0.1804 10.4595 0.1621 0.2118 0.8087 0.3244 0.0000 0.0334 3.0101 2.1720 0.2172 0.0845
cca_40 0.2181 11.0609 0.1983 0.2681 0.7779 0.6756 0.0000 0.0247 3.2498 2.5459 0.2546 0.1088
pca_95 0.1784 10.7542 0.1589 0.2318 0.8082 0.3748 0.0000 0.0331 3.0514 2.1542 0.2154 0.0838