Reliable Evaluation Protocol for Low-Precision Retrieval
This addresses evaluation reliability for researchers and practitioners using low-precision retrieval, but it is incremental as it focuses on improving protocols rather than retrieval methods.
The paper tackles the problem of unreliable evaluation in low-precision retrieval systems due to spurious ties from reduced granularity, proposing a protocol that reduces score variation and recovers expected metric values in experiments on multiple models and datasets.
Lowering the numerical precision of model parameters and computations is widely adopted to improve the efficiency of retrieval systems. However, when computing relevance scores between the query and documents in low-precision, we observe spurious ties due to the reduced granularity. This introduces high variability in the results based on tie resolution, making the evaluation less reliable. To address this, we propose a more robust retrieval evaluation protocol designed to reduce score variation. It consists of: (1) High-Precision Scoring (HPS), which upcasts the final scoring step to higher precision to resolve tied candidates with minimal computational cost; and (2) Tie-aware Retrieval Metrics (TRM), which report expected scores, range, and bias to quantify order uncertainty of tied candidates. Our experiments test multiple models with three scoring functions on two retrieval datasets to demonstrate that HPS dramatically reduces tie-induced instability, and TRM accurately recovers expected metric values. This combination enables a more consistent and reliable evaluation system for lower-precision retrievals.