Upload README.md with huggingface_hub
Browse files
README.md
ADDED
|
@@ -0,0 +1,48 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
tags:
|
| 3 |
+
- recommender-system
|
| 4 |
+
- sasrec
|
| 5 |
+
- reinforcement-learning
|
| 6 |
+
- modpo
|
| 7 |
+
datasets:
|
| 8 |
+
- ml-100k
|
| 9 |
+
---
|
| 10 |
+
|
| 11 |
+
# SASRec - ml-100k (Finetuned with MODPO)
|
| 12 |
+
|
| 13 |
+
## Model Description
|
| 14 |
+
This model is fine-tuned from [viberec/ML-100k-SASRec](https://huggingface.co/viberec/ML-100k-SASRec) using Multi-Objective Direct Preference Optimization (MODPO).
|
| 15 |
+
|
| 16 |
+
## Training Results
|
| 17 |
+
### Baseline (Before Finetune)
|
| 18 |
+
- **ndcg@10**: 0.0549
|
| 19 |
+
- **hit@10**: 0.1171
|
| 20 |
+
- **averagepopularity@10**: 236.3066
|
| 21 |
+
- **giniindex@10**: 0.9185
|
| 22 |
+
- **itemcoverage@10**: 0.2144
|
| 23 |
+
- **shannonentropy@10**: 0.0216
|
| 24 |
+
- **tailpercentage@10**: 0.0002
|
| 25 |
+
|
| 26 |
+
### Best Valid Results (MODPO)
|
| 27 |
+
- **ndcg@10**: 0.0508
|
| 28 |
+
- **hit@10**: 0.1044
|
| 29 |
+
- **averagepopularity@10**: 242.5457
|
| 30 |
+
- **giniindex@10**: 0.9246
|
| 31 |
+
- **itemcoverage@10**: 0.1868
|
| 32 |
+
- **shannonentropy@10**: 0.0243
|
| 33 |
+
- **tailpercentage@10**: 0.0001
|
| 34 |
+
|
| 35 |
+
### Test Results (MODPO)
|
| 36 |
+
- **ndcg@10**: 0.0532
|
| 37 |
+
- **hit@10**: 0.1129
|
| 38 |
+
- **averagepopularity@10**: 247.405
|
| 39 |
+
- **giniindex@10**: 0.9324
|
| 40 |
+
- **itemcoverage@10**: 0.1711
|
| 41 |
+
- **shannonentropy@10**: 0.0259
|
| 42 |
+
- **tailpercentage@10**: 0.0001
|
| 43 |
+
|
| 44 |
+
## RL Hyperparameters
|
| 45 |
+
- **Alpha**: 0.9
|
| 46 |
+
- **KL Beta**: 0.1
|
| 47 |
+
- **Group Size**: 16
|
| 48 |
+
- **Learning Rate**: 5e-05
|