SpaDE

Choi, Eunseong; Lee, Jimin; Choi, Minijn; Ko, Hyeseon; Song, Young-In; Lee, Jongwuk

doi:10.1145/3511808.3557456

Cited by 11 publications

(5 citation statements)

References 28 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This analysis led to several findings about the components, including that we can remove the query expansion from a SOTA system, leading to a significant latency improvement without compromising the system's effectiveness. While this study covered the most prominent transformer-based LSR methods, several others could not be considered due to time and computing constraints (e.g., [2,4,11,25]). We plan to incorporate them into our implementation as future work.…”

Section: Discussionmentioning

confidence: 99%

See 1 more Smart Citation

A Unified Framework for Learned Sparse Retrieval

Nguyen

MacAvaney

Yates

2023

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Learned sparse retrieval (LSR) is a family of first-stage retrieval methods that are trained to generate sparse lexical representations of queries and documents for use with an inverted index. Many LSR methods have been recently introduced, with Splade models achieving state-of-the-art performance on MSMarco. Despite similarities in their model architectures, many LSR methods show substantial differences in effectiveness and efficiency. Differences in the experimental setups and configurations used make it difficult to compare the methods and derive insights. In this work, we analyze existing LSR methods and identify key components to establish an LSR framework that unifies all LSR methods under the same perspective. We then reproduce all prominent methods using a common codebase and re-train them in the same environment, which allows us to quantify how components of the framework affect effectiveness and efficiency. We find that (1) including document term weighting is most important for a method's effectiveness, (2) including query weighting has a small positive impact, and (3) document expansion and query expansion have a cancellation effect. As a result, we show how removing query expansion from a state-of-the-art model can reduce latency significantly while maintaining effectiveness on MSMarco and TripClick benchmarks. Our code is publicly available. 3

show abstract

Section: Discussionmentioning

confidence: 99%

“…In Table 2, we present a summary of LSR methods fit into our conceptual framework. We cover nearly all transformer-based LSR methods for text ranking in the literature 4 , but omit several due to time and space limitations [2,4,11,25].…”

Section: Surveyed Learned Sparse Retrieval Methodsmentioning

confidence: 99%

A Unified Framework for Learned Sparse Retrieval

Nguyen

MacAvaney

Yates

2023

Lecture Notes in Computer Science

View full text Add to dashboard Cite

show abstract

“…TILDEv2, suggested by Zhuang and Zuccon [75], applies a neural network to learn term-based representations of passages and expand the query. Similarly, SpaDE, proposed by Choi et al [76], uses a neural network to learn a sparse representation of a document and encode the document with two encoders. These models demonstrate the potential of neural networks in enhancing information retrieval by refining term weighting, expanding queries, or learning sparse representations.…”

Section: ) Neural Weighting Schemementioning

confidence: 99%

Information Retrieval: Recent Advances and Beyond

Hambarde

Proença

2023

IEEE Access

View full text Add to dashboard Cite

This paper provides an extensive and thorough overview of the models and techniques utilized in the first and second stages of the typical information retrieval processing chain. Our discussion encompasses the current state-of-the-art models, covering a wide range of methods and approaches in the field of information retrieval. We delve into the historical development of these models, analyze the key advancements and breakthroughs, and address the challenges and limitations faced by researchers and practitioners in the domain. By offering a comprehensive understanding of the field, this survey is a valuable resource for researchers, practitioners, and newcomers to the information retrieval domain, fostering knowledge growth, innovation, and the development of novel ideas and techniques.INDEX TERMS First-stage retrieval, information retrieval, second-stage retrieval.

show abstract

“…Tilde (Zhuang & Zuccon, 2021a) and Tildev2 (Zhuang & Zuccon, 2021b) proposed a deep query and document likelihood based model instead of a query encoder to improve the ranking efficiency. The SpaDE (Choi et al, 2022) model improves the ranking efficiency by using simplified query representations and a dual document encoder containing term weighting and term expansion components. Other approaches also tried to improve the ranking efficiency by compressing document representations (Cohen et al, 2022) and removing unnecessary word representations (COLBERTER) (Hofstätter et al, 2022).…”

Section: Related Workmentioning

confidence: 99%

“…In this paper, we particularly focus on BERT based reranking approaches that jointly model query and document sequences using cross attention approach. Our objective is to explore the trade-off between effective transfer and efficient transfer rather than new architectural improvements as in (MacAvaney et al, 2019;Li et al, 2020;Hofstätter et al, 2020b;Choi et al, 2022;Fan et al, 2023;Leonhardt et al, 2023a). However, we believe our study could be extended to other kinds of reranking models such as dual encoder, hybrid models.…”

Section: Related Workmentioning

confidence: 99%

An in-depth analysis of passage-level label transfer for contextual document ranking

Rudra,

Fernando,

Anand

2023

Inf Retrieval J

View full text Add to dashboard Cite

Pre-trained contextual language models such as BERT, GPT, and XLnet work quite well for document retrieval tasks. Such models are fine-tuned based on the query-document/ query-passage level relevance labels to capture the ranking signals. However, the documents are longer than the passages and such document ranking models suffer from the token limitation (512) of BERT. Researchers proposed ranking strategies that either truncate the documents beyond the token limit or chunk the documents into units that can fit into the BERT. In the later case, the relevance labels are either directly transferred from the original query-document pair or learned through some external model. In this paper, we conduct a detailed study of the design decisions about splitting and label transfer on retrieval effectiveness and efficiency. We find that direct transfer of relevance labels from documents to passages introduces label noise that strongly affects retrieval effectiveness for large training datasets. We also find that query processing times are adversely affected by fine-grained splitting schemes. As a remedy, we propose a careful passage level labelling scheme using weak supervision that delivers improved performance (3-14% in terms of nDCG score) over most of the recently proposed models for ad-hoc retrieval while maintaining manageable computational complexity on four diverse document retrieval datasets.

show abstract

SpaDE

Cited by 11 publications

References 28 publications

A Unified Framework for Learned Sparse Retrieval

A Unified Framework for Learned Sparse Retrieval

Information Retrieval: Recent Advances and Beyond

An in-depth analysis of passage-level label transfer for contextual document ranking

Contact Info

Product

Resources

About