Information Retrieval on Academic Search Computer Science
20.4MAP@100DistilBERT-Houlsby
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| DistilBERT-HoulsbyTraining Paradigm=Knowledge Distillation (Oracle), Model Architecture=DistilBERT, Adapter Type=Houlsby2026.02 | 20.4 | 60.9 | 31.5 | |
| DistilBERT-HoulsbyTraining Paradigm=DRAMA, Model Architecture=DistilBERT, Adapter Type=Houlsby2026.02 | 20.4 | 60.9 | 31.4 | |
| BERT-TTraining Paradigm=Full Model, Model Architecture=BERT2026.02 | 20.3 | 60.9 | 31.4 | |
| DistilBERT-LoRATraining Paradigm=Knowledge Distillation (Oracle), Model Architecture=DistilBERT, Adapter Type=LoRA2026.02 | 19.7 | 60.2 | 30.7 | |
| DistilBERT-LoRATraining Paradigm=DRAMA, Model Architecture=DistilBERT, Adapter Type=LoRA2026.02 | 19.7 | 60.2 | 30.7 | |
| MiniLM_SBackbone=MiniLM, Training Paradigm=Full Model, Training Scope=S2026.02 | 19.3 | 59.7 | 30 | |
| MiniLM_HoulsbyBackbone=MiniLM, Training Paradigm=Knowledge Distillation (Oracle), Adapter Type=Houlsby2026.02 | 19.3 | 59.7 | 30 | |
| MiniLM_HoulsbyBackbone=MiniLM, Training Paradigm=DRAMA, Adapter Type=Houlsby2026.02 | 19.3 | 59.5 | 30 | |
| DistilBERT-STraining Paradigm=Full Model, Model Architecture=DistilBERT2026.02 | 19.3 | 60 | 30.4 | |
| MiniLM_LoRABackbone=MiniLM, Training Paradigm=Knowledge Distillation (Oracle), Adapter Type=LoRA2026.02 | 19 | 59.5 | 29.4 | |
| MiniLM_LoRABackbone=MiniLM, Training Paradigm=DRAMA, Adapter Type=LoRA2026.02 | 19 | 59.5 | 29.3 | |
| DistilBERT-ALLTraining Paradigm=Full Model, Model Architecture=DistilBERT2026.02 | 19 | 59.5 | 30 | |
| MiniLM_ALLBackbone=MiniLM, Training Paradigm=Full Model, Training Scope=ALL2026.02 | 18.9 | 59.3 | 29.2 | |
| MiniLM_SArchitecture=CrossEncoder, Training Paradigm=Full Model, Model Variant=Specialized2026.02 | 17.7 | 58.9 | 28.9 | |
| MiniLM_HoulsbyArchitecture=CrossEncoder, Training Paradigm=Knowledge Distillation (Oracle), Adapter Type=Houlsby2026.02 | 17.7 | 58.9 | 28.9 | |
| DRAMA (MiniLM_Houlsby)Architecture=CrossEncoder, Training Paradigm=DRAMA, Adapter Type=Houlsby2026.02 | 17.4 | 58.4 | 28.3 | |
| MiniLM_LoRAArchitecture=CrossEncoder, Training Paradigm=Knowledge Distillation (Oracle), Adapter Type=LoRA2026.02 | 16.4 | 56.7 | 27.6 | |
| DRAMA (MiniLM_LoRA)Architecture=CrossEncoder, Training Paradigm=DRAMA, Adapter Type=LoRA2026.02 | 16.4 | 56.7 | 27.6 | |
| MiniLM_ALLArchitecture=CrossEncoder, Training Paradigm=Full Model, Model Variant=General2026.02 | 16.3 | 56.4 | 27.3 | |
| BM252026.02 | 12.3 | 48.9 | 22.5 | |
| BM25Type=Unsupervised Baseline2026.02 | 12.3 | 48.9 | 22.5 | |
| BM25Training Paradigm=Lexical Baseline2026.02 | 12.3 | 48.9 | 22.5 |