Complementing Lexical Retrieval with Semantic Residual Embedding

TL;DR

This paper introduces CLEAR, a retrieval model combining lexical and semantic matching using residual embeddings, improving accuracy and efficiency in reranking pipelines.

Contribution

CLEAR's novel residual-based embedding learning explicitly captures semantics beyond lexical matching, enhancing retrieval performance.

Findings

CLEAR outperforms state-of-the-art models

Significant improvements in reranking accuracy

Enhanced efficiency in retrieval pipelines

Abstract

This paper presents CLEAR, a retrieval model that seeks to complement classical lexical exact-match models such as BM25 with semantic matching signals from a neural embedding matching model. CLEAR explicitly trains the neural embedding to encode language structures and semantics that lexical retrieval fails to capture with a novel residual-based embedding learning method. Empirical evaluations demonstrate the advantages of CLEAR over state-of-the-art retrieval models, and that it can substantially improve the end-to-end accuracy and efficiency of reranking pipelines.