MultiContrievers: Analysis of Dense Retrieval Representations

TL;DR

This paper analyzes what information dense retrieval models like Contriever preserve or lose compared to underlying language models, revealing insights into their information content, bias, and sensitivity to initial conditions.

Contribution

It provides the first detailed analysis of information retention and bias in dense retrievers, highlighting their high extractability, bias independence, and sensitivity to initializations.

Findings

Contriever models have higher extractability of information.

Extractability correlates poorly with retrieval performance.

Gender bias exists but is not caused by representations.

Abstract

Dense retrievers compress source documents into (possibly lossy) vector representations, yet there is little analysis of what information is lost versus preserved, and how it affects downstream tasks. We conduct the first analysis of the information captured by dense retrievers compared to the language models they are based on (e.g., BERT versus Contriever). We use 25 MultiBert checkpoints as randomized initialisations to train MultiContrievers, a set of 25 contriever models. We test whether specific pieces of information -- such as gender and occupation -- can be extracted from contriever vectors of wikipedia-like documents. We measure this extractability via information theoretic probing. We then examine the relationship of extractability to performance and gender bias, as well as the sensitivity of these results to many random initialisations and data shuffles. We find that (1) contriever models have significantly increased extractability, but extractability usually correlates poorly with benchmark performance 2) gender bias is present, but is not caused by the contriever representations 3) there is high sensitivity to both random initialisation and to data shuffle, suggesting that future retrieval research should test across a wider spread of both.