SASRec - ml-100k (Finetuned with MODPO)
Model Description
This model is fine-tuned from viberec/ML-100k-SASRec using Multi-Objective Direct Preference Optimization (MODPO).
Training Results
Baseline (Before Finetune)
- ndcg@10: 0.0681
- hit@10: 0.148
- averagepopularity@10: 164.4179
- giniindex@10: 0.7077
- itemcoverage@10: 0.7247
- shannonentropy@10: 0.0081
- tailpercentage@10: 0.0108
Best Valid Results (MODPO)
- ndcg@10: 0.0672
- hit@10: 0.1416
- averagepopularity@10: 178.2734
- giniindex@10: 0.8001
- itemcoverage@10: 0.5231
- shannonentropy@10: 0.0105
- tailpercentage@10: 0.0112
Test Results (MODPO)
- ndcg@10: 0.071
- hit@10: 0.1523
- averagepopularity@10: 176.2649
- giniindex@10: 0.8004
- itemcoverage@10: 0.532
- shannonentropy@10: 0.0103
- tailpercentage@10: 0.0113
RL Hyperparameters
- Alpha: 0.8
- KL Beta: 0.08
- Group Size: 8
- Learning Rate: 5e-05