LLM-Augmented Retrieval: Enhancing Retrieval Models Through Language Models and Doc-Level Embedding
Keywords: information retrieval, text retrieval, artificial intelligence, large language models, data augmentation
TL;DR: A novel model-agnostic LLM-augmented framework to enhance retriever models' performance to state-of-the-art results
Abstract: Recent advancements in embedding-based retrieval, also known as dense retrieval, have shown state of the art results and demonstrated superior performance over traditional sparse or bag-of-words-based methodologies. This paper presents a model-agnostic document-level embedding framework enhanced by large language model (LLM) augmentation. The implementation of this LLM-augmented retrieval framework has significantly enhanced the efficacy of prevalent retriever models, including Bi-encoders (Contriever, DRAGON) and late-interaction models (ColBERTv2). Consequently, this approach has achieved state-of-the-art results on benchmark datasets such as LoTTE and BEIR, underscoring its potential to refine information retrieval processes.
Primary Area: applications to computer vision, audio, language, and other modalities
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 3320
Loading