An anatomy for neural search engines

Nakamura, Thiago Akio; Calais, Pedro H.; Reis, Davi; Lemos, André Paim

doi:10.1016/j.ins.2018.12.041

Cited by 22 publications

(5 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In fact, attempts to do away with retrieval based on sparse vector representations date back to latent semantic analysis from the 1990s [Deerwester et al, 1990]. A more recent example is the work of Nakamura et al [2019]: in a standard design where BM25-based first-stage retrieval feeds DRMM for reranking, the authors experimented with replacing first-stage retrieval with approximate nearest-neighbor search based on representations from a deep averaging network [Iyyer et al, 2015]. Unfortunately, the end-to-end effectiveness was much worse, but this was "pre-BERT", prior to the advent of the latest transformer techniques.…”

Section: Open Research Questionsmentioning

confidence: 99%

Pretrained Transformers for Text Ranking: BERT and Beyond

Lin

Nogueira

Yates

2020

Preprint

View full text Add to dashboard Cite

The goal of text ranking is to generate an ordered list of texts retrieved from a corpus in response to a query for a particular task. Although the most common formulation of text ranking is search, instances of the task can also be found in many natural language processing applications. This survey provides an overview of text ranking with neural network architectures known as transformers, of which BERT is the best-known example. The combination of transformers and self-supervised pretraining has, without exaggeration, revolutionized the fields of natural language processing (NLP), information retrieval (IR), and beyond. In the context of text ranking, these models produce high quality results across many domains, tasks, and settings.

show abstract

Section: Open Research Questionsmentioning

confidence: 99%

Pretrained Transformers for Text Ranking: BERT and Beyond

Lin

Nogueira

Yates

2020

Preprint

View full text Add to dashboard Cite

show abstract

“…• Neural models: use shallow or deep neural networks to rank search results in response to a query. Learn language representations from raw text, bridging the gap between the query and document vocabulary [9].…”

Section: Fundamentals Of Information Retrievalmentioning

confidence: 99%

Search and retrieval service for scientic articles on COVID-19

Villalobos,

Mendoza,

Rocha

et al. 2024

Anais Do XVI Congresso Brasileiro De Inteligência Computacional

View full text Add to dashboard Cite

The COVID-19 pandemic was a global health crisis that lasted until May 4, 2023, affecting millions of people and raising many questions about transmission, diagnosis, treatment, vaccine development, and viral pathogens. Unfortunately, misinformation created more socioeconomic damage than the disease itself. To address this problem, we have developed Cognitive Search, a user-friendly application service that uses the latest advances in Natural Language Processing (NLP) to retrieve information from CORD-19, a resource for scholarly articles on COVID-19 and related pathogens. This system uses a combination of Term-Frequency, Semantic Neural Research, and Hybrid Term-Neural algorithms to improve document retrieval performance. The Hybrid Term-Neural approach also considers temporal information in documents to provide more accurate search results. With an intuitive interface, this application can generate valuable insights to help combat outbreaks.

show abstract

“…This can be due to poor-quality content being added and the daily refresh interval of what users view as up-to-date [18]. The relevance scoring function for documents relevant to a query cannot always be ran for all documents in a large scale search engine due to the impact of the corpuses size on computational cost [19]. Both arguments have a huge impact on news relevance marking.…”

Section: Relevance Markingmentioning

confidence: 99%

A Novel Approach to News Archiving from Newswires

Muhammad-Bello

Lukman

Salim

2021

Communications in Computer and Information Science

View full text Add to dashboard Cite

A news archive is the core operational tool a media relations team depends on in order to effectively feed a data-hungry organization. An ingrained approach to news archiving in existence is the use of a relational database. As a consequence, integrating search engines that support full-text search is practically impossible due to the strict data schema that is defined in relational database systems. Therefore, there is a need for news archives that support full-text search with relevance ranking of news. In this paper, an approach that supports full-text search is proposed. The process is started by crawling newswire websites for news that are relevant with respect to some predefined keywords and extracting them. Then, they are stored in a data structure known as an inverted-index which supports full-text search, aggregation, and relevance ranking of search results. Search results are ranked and returned to a user in the order of decreasing relevance to the search term. We were able to provide a software solution written in java, the jsoup library for HTML parsing, and an elasticsearch implementation of a search engine. We tested our solution on nine newswires using ten keywords and were able to retrieve a total of 42 relevant news matching seven keywords. The approach proposed in this paper when compared to the manual approach performed better in terms of retrieval speed and accuracy. We conclude that three main components are important in a good digital archive: relevance, extraction, and search. This work is an integration of a good relevance marking technique, an extraction method, and a search engine.

show abstract

An anatomy for neural search engines

Cited by 22 publications

References 13 publications

Pretrained Transformers for Text Ranking: BERT and Beyond

Pretrained Transformers for Text Ranking: BERT and Beyond

Search and retrieval service for scientic articles on COVID-19

A Novel Approach to News Archiving from Newswires

Contact Info

Product

Resources

About