English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Neural ranking models (NRMs) and dense retrieval (DR) models have given riseto substantial improvements in overall retrieval performance. In addition totheir effectiveness, and motivated by the proven lack of robustness of deeplearning-based approaches in other areas, there is growing interest in therobustness of deep learning-based approaches to the core retrieval problem.Adversarial attack methods that have so far been developed mainly focus onattacking NRMs, with very little attention being paid to the robustness of DRmodels. In this paper, we introduce the adversarial retrieval attack (AREA)task. The AREA task is meant to trick DR models into retrieving a targetdocument that is outside the initial set of candidate documents retrieved bythe DR model in response to a query. We consider the decision-based black-boxadversarial setting, which is realistic in real-world search engines. Toaddress the AREA task, we first employ existing adversarial attack methodsdesigned for NRMs. We find that the promising results that have previously beenreported on attacking NRMs, do not generalize to DR models: these methodsunderperform a simple term spamming method. We attribute the observed lack ofgeneralizability to the interaction-focused architecture of NRMs, whichemphasizes fine-grained relevance matching. DR models follow a differentrepresentation-focused architecture that prioritizes coarse-grainedrepresentations. We propose to formalize attacks on DR models as a contrastivelearning problem in a multi-view representation space. The core idea is toencourage the consistency between each view representation of the targetdocument and its corresponding viewer via view-wise supervision signals.Experimental results demonstrate that the proposed method can significantlyoutperform existing attack strategies in misleading the DR model with smallindiscernible text perturbations.

Black-box Adversarial Attacks against Dense Retrieval Models: A Multi-view Contrastive Learning Method