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

In large-scale retrieval, the lexicon-weighting paradigm, learning weightedsparse representations in vocabulary space, has shown promising results withhigh quality and low latency. Despite it deeply exploiting thelexicon-representing capability of pre-trained language models, a crucial gapremains between language modeling and lexicon-weighting retrieval -- the formerpreferring certain or low-entropy words whereas the latter favoring pivot orhigh-entropy words -- becoming the main barrier to lexicon-weightingperformance for large-scale retrieval. To bridge this gap, we propose abrand-new pre-training framework, lexicon-bottlenecked masked autoencoder(LexMAE), to learn importance-aware lexicon representations. Essentially, wepresent a lexicon-bottlenecked module between a normal language modelingencoder and a weakened decoder, where a continuous bag-of-words bottleneck isconstructed to learn a lexicon-importance distribution in an unsupervisedfashion. The pre-trained LexMAE is readily transferred to the lexicon-weightingretrieval via fine-tuning. On the ad-hoc retrieval benchmark, MS-Marco, itachieves 42.6% MRR@10 with 45.8 QPS for the passage dataset and 44.4% MRR@100with 134.8 QPS for the document dataset, by a CPU machine. And LexMAE showsstate-of-the-art zero-shot transfer capability on BEIR benchmark with 12datasets.

LexMAE: Lexicon-Bottlenecked Pretraining for Large-Scale Retrieval