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

Ranker and retriever are two important components in dense passage retrieval.The retriever typically adopts a dual-encoder model, where queries anddocuments are separately input into two pre-trained models, and the vectorsgenerated by the models are used for similarity calculation. The ranker oftenuses a cross-encoder model, where the concatenated query-document pairs areinput into a pre-trained model to obtain word similarities. However, thedual-encoder model lacks interaction between queries and documents due to itsindependent encoding, while the cross-encoder model requires substantialcomputational cost for attention calculation, making it difficult to obtainreal-time retrieval results. In this paper, we propose a dense retrieval modelcalled MD2PR based on multi-level distillation. In this model, we distill theknowledge learned from the cross-encoder to the dual-encoder at both thesentence level and word level. Sentence-level distillation enhances thedual-encoder on capturing the themes and emotions of sentences. Word-leveldistillation improves the dual-encoder in analysis of word semantics andrelationships. As a result, the dual-encoder can be used independently forsubsequent encoding and retrieval, avoiding the significant computational costassociated with the participation of the cross-encoder. Furthermore, we proposea simple dynamic filtering method, which updates the threshold during multipletraining iterations to ensure the effective identification of false negativesand thus obtains a more comprehensive semantic representation space. Theexperimental results over two standard datasets show our MD2PR outperforms 11baseline models in terms of MRR and Recall metrics.

A Multi-level Distillation based Dense Passage Retrieval Model