Generative Recommendation on Book-Crossing (test)
1.77H@10CausalDPO
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| CausalDPObase model=Llama-3.1-8B, epochs=3, negative sampling=random sampling, distribution=IID2026.03 | 1.77 | 99 | 3.28 | 1.18 | |
| SASRecembedding dimension=32, dropout rate=0.1, distribution=IID2026.03 | 1.7 | 90 | 3.12 | 1.02 | |
| BIGRecbase model=Llama-3.1-8B, epochs=3, distribution=IID2026.03 | 1.64 | 73 | 2.19 | 0.87 | |
| DMPObase model=Llama-3.1-8B, epochs=3, negative samples=single, distribution=IID2026.03 | 1.61 | 79 | 2.25 | 0.91 | |
| RosePObase model=Llama-3.1-8B, epochs=3, negative samples=single, distribution=IID2026.03 | 1.6 | 77 | 2 | 0.91 | |
| SDPObase model=Llama-3.1-8B, epochs=3, negative samples=single, distribution=IID2026.03 | 1.59 | 75 | 2.17 | 0.88 | |
| RWbase model=Llama-3.1-8B, epochs=3, distribution=IID2026.03 | 1.52 | 81 | 1.89 | 0.95 | |
| SPRecbase model=Llama-3.1-8B, epochs=3, negative samples=single-iteration SFT output, distribution=IID2026.03 | 1.52 | 70 | 2.31 | 0.9 | |
| D3base model=Llama-3.1-8B, epochs=3, distribution=IID2026.03 | 1.48 | 71 | 2.03 | 0.82 |