One-Shot Labeling for Automatic Relevance Estimation

Dealing with unjudged documents ("holes") in relevance assessments is a perennial problem when evaluating search systems with offline experiments. Holes can reduce the apparent effectiveness of retrieval systems during evaluation and introduce biases in models trained with incomplete data. In this work, we explore whether large language models can help us fill such holes to improve offline evaluations. We examine an extreme, albeit common, evaluation setting wherein only a single known relevant document per query is available for evaluation. We then explore various approaches for predicting the relevance of unjudged documents with respect to a query and the known relevant document, including nearest neighbor, supervised, and prompting techniques. We find that although the predictions of these One-Shot Labelers (1SLs) frequently disagree with human assessments, the labels they produce yield a far more reliable ranking of systems than the single labels do alone. Specifically, the strongest approaches can consistently reach system ranking correlations of over 0.85 with the full rankings over a variety of measures. Meanwhile, the approach substantially reduces the false positive rate of t-tests due to holes in relevance assessments (from 15-30% down to under 5%), giving researchers more confidence in results they find to be significant.