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

The learning objective plays a fundamental role to build a recommendersystem. Most methods routinely adopt either pointwise or pairwise loss to trainthe model parameters, while rarely pay attention to softmax loss due to itscomputational complexity when scaling up to large datasets or intractabilityfor streaming data. The sampled softmax (SSM) loss emerges as an efficientsubstitute for softmax loss. Its special case, InfoNCE loss, has been widelyused in self-supervised learning and exhibited remarkable performance forcontrastive learning. Nonetheless, limited recommendation work uses the SSMloss as the learning objective. Worse still, none of them explores itsproperties thoroughly and answers ``Does SSM loss suit for itemrecommendation?'' and ``What are the conceptual advantages of SSM loss, ascompared with the prevalent losses?'', to the best of our knowledge.  In this work, we aim to offer a better understanding of SSM for itemrecommendation. Specifically, we first theoretically reveal threemodel-agnostic advantages: (1) mitigating popularity bias; (2) mining hardnegative samples; and (3) maximizing the ranking metric. However, based on ourempirical studies, we recognize that the default choice of cosine similarityfunction in SSM limits its ability in learning the magnitudes of representationvectors. As such, the combinations of SSM with the models that also fall shortin adjusting magnitudes may result in poor representations. One step further,we provide mathematical proof that message passing schemes in graph convolutionnetworks can adjust representation magnitude according to node degree, whichnaturally compensates for the shortcoming of SSM. Extensive experiments on fourbenchmark datasets justify our analyses, demonstrating the superiority of SSMfor item recommendation. Our implementations are available in both TensorFlowand PyTorch.

On the Effectiveness of Sampled Softmax Loss for Item Recommendation