Pretrained Transformers for Text Ranking
BERT and Beyond
Chapter and Conference Paper
In the information age, health misinformation remains a notable challenge to public welfare. Integral to addressing this issue is the development of search systems adept at identifying and filtering out mislea...
Chapter and Conference Paper
Text retrieval using dense–sparse hybrids has been gaining popularity because of their effectiveness. Improvements to both sparse and dense models have also been noted, in the context of open-domain question a...
Chapter and Conference Paper
Answer retrieval for math questions is a challenging task due to the complex and structured nature of mathematical expressions. In this paper, we combine a structure retriever and a domain-adapted ColBERT retr...
Chapter and Conference Paper
One of the contributions of the landmark Dense Passage Retriever (DPR) work is the curation of a corpus of passages generated from Wikipedia articles that have been segmented into non-overlap** passages of 1...
Chapter and Conference Paper
While much recent work has demonstrated that hard negative mining can be used to train better bi-encoder models, few have considered it in the context of cross-encoders, which are key ingredients in modern re...
Chapter
This section begins by more formally characterizing the text ranking problem, explicitly enumerating our assumptions about characteristics of the input and output, and more precisely circumscribing the scope o...
Chapter
The vocabulary mismatch problem [Furnas et al., 1987]—where searchers and the authors of the texts to be searched use different words to describe the same concepts—was introduced in Section 1.2.2 as a core pro...
Chapter
It is quite remarkable that BERT debuted in October 2018, only around three years ago. Taking a step back and reflecting, the field has seen an incredible amount of progress in a short amount of time. As we ha...
Chapter and Conference Paper
Text retrieval using learned dense representations has recently emerged as a promising alternative to “traditional” text retrieval using sparse bag-of-words representations. One foundational work that has garn...
Chapter
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. The most common formulation of text ranking is search, where the search en...
Chapter
The simplest and most straightforward formulation of text ranking is to convert the task into a text classification problem, and then sort the texts to be ranked based on the probability that each item belongs...
Chapter
Arguably, the single biggest benefit brought about by modern deep learning techniques to text ranking is the move away from sparse signals, mostly limited to exact matches, to continuous dense representations ...
Book
Chapter and Conference Paper
Pseudo-Relevance Feedback (PRF) utilises the relevance signals from the top-k passages from the first round of retrieval to perform a second round of retrieval aiming to improve search effectiveness. A recent res...
Article
This paper introduces the Archives Unleashed Cloud, a web-based interface for working with web archives at scale. Current access paradigms, largely driven by the scope and scale of web archives, generally invo...
Chapter and Conference Paper
While BERT has been shown to be effective for passage retrieval, its maximum input length limitation poses a challenge when applying the model to document retrieval. In this work, we reproduce three passage sc...
Article
Citation recommendation systems for the scientific literature, to help authors find papers that should be cited, have the potential to speed up discoveries and uncover new routes for scientific exploration. We...
Article
Graph processing is becoming increasingly prevalent across many application domains. In spite of this prevalence, there is little research about how graphs are actually used in practice. We performed an extens...
Chapter and Conference Paper
The latest major release of Lucene (version 8) in March 2019 incorporates block-max indexes and exploits the block-max variant of Wand for query evaluation, which are innovations that originated from academia. Th...
Chapter and Conference Paper
When researchers speak of BM25, it is not entirely clear which variant they mean, since many tweaks to Robertson et al.’s original formulation have been proposed. When practitioners speak of BM25, they most li...