DyVo: Dynamic Vocabularies for Learned Sparse Retrieval with Entities

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Nguyen, Thong, Chatterjee, Shubham, MacAvaney, Sean, Mackie, Iain, Dalton, Jeff, Yates, Andrew
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910652087402496
author Nguyen, Thong
Chatterjee, Shubham
MacAvaney, Sean
Mackie, Iain
Dalton, Jeff
Yates, Andrew
author_facet Nguyen, Thong
Chatterjee, Shubham
MacAvaney, Sean
Mackie, Iain
Dalton, Jeff
Yates, Andrew
contents Learned Sparse Retrieval (LSR) models use vocabularies from pre-trained transformers, which often split entities into nonsensical fragments. Splitting entities can reduce retrieval accuracy and limits the model's ability to incorporate up-to-date world knowledge not included in the training data. In this work, we enhance the LSR vocabulary with Wikipedia concepts and entities, enabling the model to resolve ambiguities more effectively and stay current with evolving knowledge. Central to our approach is a Dynamic Vocabulary (DyVo) head, which leverages existing entity embeddings and an entity retrieval component that identifies entities relevant to a query or document. We use the DyVo head to generate entity weights, which are then merged with word piece weights to create joint representations for efficient indexing and retrieval using an inverted index. In experiments across three entity-rich document ranking datasets, the resulting DyVo model substantially outperforms state-of-the-art baselines.
format Preprint
id arxiv_https___arxiv_org_abs_2410_07722
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle DyVo: Dynamic Vocabularies for Learned Sparse Retrieval with Entities
Nguyen, Thong
Chatterjee, Shubham
MacAvaney, Sean
Mackie, Iain
Dalton, Jeff
Yates, Andrew
Information Retrieval
Learned Sparse Retrieval (LSR) models use vocabularies from pre-trained transformers, which often split entities into nonsensical fragments. Splitting entities can reduce retrieval accuracy and limits the model's ability to incorporate up-to-date world knowledge not included in the training data. In this work, we enhance the LSR vocabulary with Wikipedia concepts and entities, enabling the model to resolve ambiguities more effectively and stay current with evolving knowledge. Central to our approach is a Dynamic Vocabulary (DyVo) head, which leverages existing entity embeddings and an entity retrieval component that identifies entities relevant to a query or document. We use the DyVo head to generate entity weights, which are then merged with word piece weights to create joint representations for efficient indexing and retrieval using an inverted index. In experiments across three entity-rich document ranking datasets, the resulting DyVo model substantially outperforms state-of-the-art baselines.
title DyVo: Dynamic Vocabularies for Learned Sparse Retrieval with Entities
topic Information Retrieval
url https://arxiv.org/abs/2410.07722