The Medium Is Not the Message: Deconfounding Document Embeddings via Linear Concept Erasure

TL;DR

This paper introduces a debiasing method that removes confounding attributes from document embeddings, significantly improving similarity and clustering metrics without harming out-of-distribution performance.

Contribution

The paper proposes a linear concept erasure technique to deconfound document embeddings, reducing bias from spurious attributes with minimal computational overhead.

Findings

Bias from confounders is substantially reduced.

Embedding quality improves across multiple tasks.

Out-of-distribution performance remains unaffected.

Abstract

Embedding-based similarity metrics between text sequences can be influenced not just by the content dimensions we most care about, but can also be biased by spurious attributes like the text's source or language. These document confounders cause problems for many applications, but especially those that need to pool texts from different corpora. This paper shows that a debiasing algorithm that removes information about observed confounders from the encoder representations substantially reduces these biases at a minimal computational cost. Document similarity and clustering metrics improve across every embedding variant and task we evaluate -- often dramatically. Interestingly, performance on out-of-distribution benchmarks is not impacted, indicating that the embeddings are not otherwise degraded.