Systematic Evaluation of Neural Retrieval Models on the Touch\'e 2020 Argument Retrieval Subset of BEIR

TL;DR

This study systematically evaluates neural retrieval models on the Touché 2020 argument retrieval dataset, revealing biases towards short passages and data quality issues, and demonstrates that data denoising improves neural model performance but BM25 remains superior.

Contribution

It provides a detailed reproducibility analysis of neural retrieval models on Touché 2020, highlighting biases and data issues, and proposes data denoising to improve neural retrieval effectiveness.

Findings

Neural models favor short passages, often non-argumentative.

Many neural model results are based on unjudged data.

Denoising improves neural model performance by up to 0.52 in nDCG@10.

Abstract

The zero-shot effectiveness of neural retrieval models is often evaluated on the BEIR benchmark -- a combination of different IR evaluation datasets. Interestingly, previous studies found that particularly on the BEIR subset Touch\'e 2020, an argument retrieval task, neural retrieval models are considerably less effective than BM25. Still, so far, no further investigation has been conducted on what makes argument retrieval so "special". To more deeply analyze the respective potential limits of neural retrieval models, we run a reproducibility study on the Touch\'e 2020 data. In our study, we focus on two experiments: (i) a black-box evaluation (i.e., no model retraining), incorporating a theoretical exploration using retrieval axioms, and (ii) a data denoising evaluation involving post-hoc relevance judgments. Our black-box evaluation reveals an inherent bias of neural models towards retrieving short passages from the Touch\'e 2020 data, and we also find that quite a few of the neural models' results are unjudged in the Touch\'e 2020 data. As many of the short Touch\'e passages are not argumentative and thus non-relevant per se, and as the missing judgments complicate fair comparison, we denoise the Touch\'e 2020 data by excluding very short passages (less than 20 words) and by augmenting the unjudged data with post-hoc judgments following the Touch\'e guidelines. On the denoised data, the effectiveness of the neural models improves by up to 0.52 in nDCG@10, but BM25 is still more effective. Our code and the augmented Touch\'e 2020 dataset are available at \url{https://github.com/castorini/touche-error-analysis}.