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

Dense retrieval has shown promise in the first-stage retrieval process whentrained on in-domain labeled datasets. However, previous studies have foundthat dense retrieval is hard to generalize to unseen domains due to its weakmodeling of domain-invariant and interpretable feature (i.e., matching signalbetween two texts, which is the essence of information retrieval). In thispaper, we propose a novel method to improve the generalization of denseretrieval via capturing matching signal called BERM. Fully fine-grainedexpression and query-oriented saliency are two properties of the matchingsignal. Thus, in BERM, a single passage is segmented into multiple units andtwo unit-level requirements are proposed for representation as the constraintin training to obtain the effective matching signal. One is semantic unitbalance and the other is essential matching unit extractability. Unit-levelview and balanced semantics make representation express the text in afine-grained manner. Essential matching unit extractability makes passagerepresentation sensitive to the given query to extract the pure matchinginformation from the passage containing complex context. Experiments on BEIRshow that our method can be effectively combined with different dense retrievaltraining methods (vanilla, hard negatives mining and knowledge distillation) toimprove its generalization ability without any additional inference overheadand target domain data.

BERM: Training the Balanced and Extractable Representation for Matching to Improve Generalization Ability of Dense Retrieval