MUSS is compared against the following selection methods (see the README for the full project overview).
| Method | Where | Notes |
|---|---|---|
| Random | baselines.run_random_selection |
Uniform sampling |
| MMR | baselines.run_experiment_MMR |
Greedy MMR over the full corpus |
| Clustering | baselines.run_experiment_clustering |
FAISS k-means, top-quality per cluster |
| k-DPP | baselines.run_experiment_DPP |
Exact k-DPP via dppy |
| DGDS | baselines.run_DGDS |
Random partitioning + within-partition MMR (no cluster selection) |
| MUSS (full) | helpers.run_MUSS |
Stage 1+2+3 with FAISS clustering |
| MUSS (rand. clusters) | ablation_muss.run_MUSS_random_cluster_selection |
Stage 2 replaced by random selection |
| MUSS (rand. partition) | ablation_muss.run_MUSS_rand_partition |
Stage 1 replaced by random partitioning |