Score prediction on PRISM held-out-user (test)
23.33RMSEPEBS (EB-shrunk)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| PEBS (EB-shrunk)Arm=EB-shrunk (ours), Base Model=Qwen2.5-7B-Instruct, Validation=k=5-fold cross-validation2026.06 | 23.33 | 23.12 | -8.58 | 3.8 | |
| Per-user OLSArm=Per-user OLS, Base Model=Qwen2.5-7B-Instruct, Validation=k=5-fold cross-validation2026.06 | 23.73 | 23.36 | -7.02 | 1.2 | |
| Population slopeArm=Population slope, Base Model=Qwen2.5-7B-Instruct, Validation=k=5-fold cross-validation2026.06 | 25.52 | 25.26 | — | — | |
| No calibrationArm=No calibration, Base Model=Qwen2.5-7B-Instruct, Validation=k=5-fold cross-validation2026.06 | 27.13 | 27.13 | 6.3 | — |