Top-k sparse autoencoders disentangle sentence embeddings into semantic, syntactic, and pragmatic concepts and enable activation steering to re-rank retrieval results for better user alignment.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.IR 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Aligning Sentence Embeddings to Human Concepts via Sparse Autoencoders
Top-k sparse autoencoders disentangle sentence embeddings into semantic, syntactic, and pragmatic concepts and enable activation steering to re-rank retrieval results for better user alignment.