The Medium Is Not the Message: Deconfounding Document Embeddings via Linear Concept Erasure
This addresses bias issues in text similarity for applications that combine diverse corpora, representing an incremental improvement over existing embedding methods.
The paper tackled the problem of document embeddings being biased by spurious attributes like source or language, which affects applications pooling texts from different corpora, and showed that a debiasing algorithm reduces these biases with minimal computational cost, improving similarity and clustering metrics across all evaluated tasks, often dramatically, without degrading out-of-distribution performance.
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.