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/

Zendy Plus

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

Zendy Tools

Zendy Open

Recommender system usually suffers from severe popularity bias -- thecollected interaction data usually exhibits quite imbalanced or evenlong-tailed distribution over items. Such skewed distribution may result fromthe users' conformity to the group, which deviates from reflecting users' truepreference. Existing efforts for tackling this issue mainly focus on completelyeliminating popularity bias. However, we argue that not all popularity bias isevil. Popularity bias not only results from conformity but also item quality,which is usually ignored by existing methods. Some items exhibit higherpopularity as they have intrinsic better property. Blindly removing thepopularity bias would lose such important signal, and further deteriorate modelperformance. To sufficiently exploit such important information forrecommendation, it is essential to disentangle the benign popularity biascaused by item quality from the harmful popularity bias caused by conformity.Although important, it is quite challenging as we lack an explicit signal todifferentiate the two factors of popularity bias. In this paper, we propose toleverage temporal information as the two factors exhibit quite differentpatterns along the time: item quality revealing item inherent property isstable and static while conformity that depends on items' recent clicks ishighly time-sensitive. Correspondingly, we further propose a novel Time-awareDisEntangled framework (TIDE), where a click is generated from three componentsnamely the static item quality, the dynamic conformity effect, as well as theuser-item matching score returned by any recommendation model. Lastly, weconduct interventional inference such that the recommendation can benefit fromthe benign popularity bias while circumvent the harmful one. Extensiveexperiments on three real-world datasets demonstrated the effectiveness ofTIDE.

Popularity Bias Is Not Always Evil: Disentangling Benign and Harmful Bias for Recommendation