Papers › Scalable Cross-Entropy Loss for Sequential Recommendations with Large Item Catalogs

Scalable Cross-Entropy Loss for Sequential Recommendations with Large Item Catalogs

27 Sep 2024arXiv:2409.18721archive 2025-07-28

Gleb Mezentsev, Danil Gusak, Ivan Oseledets, Evgeny Frolov

Scalability issue plays a crucial role in productionizing modern recommender systems. Even lightweight architectures may suffer from high computational overload due to intermediate calculations, limiting their practicality in real-world applications. Specifically, applying full Cross-Entropy (CE) loss often yields state-of-the-art performance in terms of recommendations quality. Still, it suffers from excessive GPU memory utilization when dealing with large item catalogs. This paper introduces a novel Scalable Cross-Entropy (SCE) loss function in the sequential learning setup. It approximates the CE loss for datasets with large-size catalogs, enhancing both time efficiency and memory usage without compromising recommendations quality. Unlike traditional negative sampling methods, our approach utilizes a selective GPU-efficient computation strategy, focusing on the most informative elements of the catalog, particularly those most likely to be false positives. This is achieved by approximating the softmax distribution over a subset of the model outputs through the maximum inner product search. Experimental results on multiple datasets demonstrate the effectiveness of SCE in reducing peak memory usage by a factor of up to 100 compared to the alternatives, retaining or even exceeding their metrics values. The proposed approach also opens new perspectives for large-scale developments in different domains, such as large language models.

PaperPDFCode

Code

AIRI-Institute/Scalable-SASRec officialmentioned on GitHubpytorch report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

Recommendation SystemsSequential Recommendation

1 archive task tag without a task page not shown.

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Sequential Recommendation Amazon Beauty SASRec-SCE HR@10 0.0935 #1 of 1 Archive leaderboard report
Sequential Recommendation Amazon Beauty SASRec-SCE NDCG@10 0.0544 #1 of 1 Archive leaderboard report
Sequential Recommendation Behance SASRec-SCE COV@1 0.0393 #1 of 1 Archive leaderboard report
Sequential Recommendation Behance SASRec-SCE COV@10 0.25 #1 of 1 Archive leaderboard report
Sequential Recommendation Behance SASRec-SCE COV@5 15.3 #1 of 1 Archive leaderboard report
Sequential Recommendation Behance SASRec-SCE HR@10 0.113 #1 of 1 Archive leaderboard report
Sequential Recommendation Behance SASRec-SCE HR@5 0.0853 #1 of 1 Archive leaderboard report
Sequential Recommendation Behance SASRec-SCE NDCG@1 0.0277 #1 of 1 Archive leaderboard report
Sequential Recommendation Behance SASRec-SCE NDCG@10 0.0663 #1 of 1 Archive leaderboard report
Sequential Recommendation Behance SASRec-SCE NDCG@5 0.0572 #1 of 1 Archive leaderboard report
Sequential Recommendation Gowalla SASRec-SCE COV@1 0.0304 #1 of 1 Archive leaderboard report
Sequential Recommendation Gowalla SASRec-SCE COV@10 0.2190 #1 of 1 Archive leaderboard report
Sequential Recommendation Gowalla SASRec-SCE COV@5 0.126 #1 of 1 Archive leaderboard report
Sequential Recommendation Gowalla SASRec-SCE HR@10 0.0831 #1 of 1 Archive leaderboard report
Sequential Recommendation Gowalla SASRec-SCE HR@5 0.0574 #1 of 1 Archive leaderboard report
Sequential Recommendation Gowalla SASRec-SCE NDCG@1 0.0207 #1 of 1 Archive leaderboard report
Sequential Recommendation Gowalla SASRec-SCE NDCG@10 0.0476 #1 of 1 Archive leaderboard report
Sequential Recommendation Gowalla SASRec-SCE NDCG@5 0.0393 #1 of 1 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Methods

Softmax

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections