Average Rating Prediction on Research Paper Rating Dataset (test)
0.999MSETF-1l-8h
Evaluation results
| Method | Links | ||
|---|---|---|---|
| TF-1l-8hInput Features=Author Identity, Capability + Abstract, Fusion Mechanism=1-layer 8-head Transformer encoder2026.01 | 0.999 | 0.776 | |
| SA-1Input Features=Author-Capability-Idea (ACI), Fusion Mechanism=Self-attention with residual connections (SA1)2026.01 | 1.013 | 0.789 | |
| SA-2Input Features=Author-Capability-Idea (ACI), Fusion Mechanism=Self-attention without layer normalization (SA2)2026.01 | 1.013 | 0.794 | |
| R1Input Features=Author-Capability-Idea (ACI), Fusion Mechanism=Residual Linear Fusion2026.01 | 1.018 | 0.794 | |
| Gated Ave.Input Features=Author-Capability-Idea (ACI), Fusion Mechanism=Weighted averaging with gating2026.01 | 1.024 | 0.795 | |
| TF-1l-1hInput Features=Author-Capability-Idea (ACI), Fusion Mechanism=1-layer 1-head Transformer encoder2026.01 | 1.033 | 0.799 | |
| TF-1l-8hInput Features=Author-Capability-Idea (ACI), Fusion Mechanism=1-layer 8-head Transformer encoder2026.01 | 1.036 | 0.799 | |
| bertInput Features=Author-Capability-Idea (ACI)2026.01 | 1.169 | 0.853 | |
| bert-largeInput Features=Author-Capability-Idea (ACI)2026.01 | 1.169 | 0.852 | |
| bertInput Features=Title + Abstract2026.01 | 1.171 | 0.855 | |
| bert-largeInput Features=Title + Abstract2026.01 | 1.172 | 0.853 | |
| average2026.01 | 1.233 | 0.885 | |
| Gemini-3-previewEvaluation Protocol=Zero-shot2026.01 | 1.573 | 0.977 | |
| ChatGLM-4.7Evaluation Protocol=Zero-shot2026.01 | 1.681 | 1.024 | |
| Deepseek-V3.2Evaluation Protocol=Zero-shot2026.01 | 1.709 | 1.082 | |
| Claude-sonnet-4.5Evaluation Protocol=Zero-shot2026.01 | 1.83 | 1.088 | |
| GPT-4oEvaluation Protocol=Zero-shot2026.01 | 1.859 | 1.085 |