Likert or Not: LLM Absolute Relevance Judgments on Fine-Grained Ordinal Scales

TL;DR

It is found that the gap between pointwise scoring and listwise ranking shrinks when pointwise scoring is implemented using a sufficiently large ordinal relevance label space, becoming statistically insignificant for many LLM-benchmark dataset combinations (where ``significant'' means ``95\% confidence that listwise ranking improves NDCG@10'').

Abstract

Large language models (LLMs) obtain state of the art zero shot relevance ranking performance on a variety of information retrieval tasks. The two most common prompts to elicit LLM relevance judgments are pointwise scoring (a.k.a. relevance generation), where the LLM sees a single query-document pair and outputs a single relevance score, and listwise ranking (a.k.a. permutation generation), where the LLM sees a query and a list of documents and outputs a permutation, sorting the documents in decreasing order of relevance. The current research community consensus is that listwise ranking yields superior performance, and significant research effort has been devoted to crafting LLM listwise ranking algorithms. The underlying hypothesis is that LLMs are better at making relative relevance judgments than absolute ones. In tension with this hypothesis, we find that the gap between pointwise scoring and listwise ranking shrinks when pointwise scoring is implemented using a sufficiently large ordinal relevance label space, becoming statistically insignificant for many LLM-benchmark dataset combinations (where ``significant'' means ``95\% confidence that listwise ranking improves NDCG@10''). Our evaluations span four LLMs, eight benchmark datasets from the BEIR and TREC-DL suites, and two proprietary datasets with relevance labels collected after the training cut-off of all LLMs evaluated.

Figures (7)

• Figure 1: Comparison of Pointwise Scoring (left), Ranking (middle) and Scoring and Ranking (right). Here $i_1, i_2, \dots, i_N$ is a permutation of $1, 2, \dots, N$ and $r_1, r_2, \dots, r_N$ are relevance scores.
• Figure 2: Observed differences in NDCG@10 for pointwise scoring and ranking on benchmark datasets. Bootstrap 95% confidence intervals (see \ref{['sec:quantifying-confidence']}). For absolute values see see \ref{['fig:ndcg10sweep']}.
• Figure 3: Comparing relevance classification performance of pointwise scoring vs. simultaneous ranking+scoring on benchmark datasets. Here we use an 11-point ordinal relevance scale.
• Figure 4: Observed differences in NDCG@10 for ranking+scoring vs. pure permutation generation listwise ranking (bubble-sort implementation) on benchmark datasets. Bootstrap 95% confidence intervals (see \ref{['sec:quantifying-confidence']}). For absolute NDCG@10 see \ref{['fig:ndcg10sweep']}.
• Figure 5: Observed differences in NDCG@10 for pointwise scoring and listwise ranking methods on Tax&Accounting {Full Document, Passage} datasets. Bootstrap 95% confidence intervals (see \ref{['sec:quantifying-confidence']}). For absolute NDCG@10 see \ref{['fig:ndcg10_proprietary']}.