Exploiting Preferences in Loss Functions for Sequential Recommendation via Weak Transitivity
Hyunsoo Chung, Jungtaek Kim, Hyungeun Jo, Hyungwon Choi
TL;DR
The paper tackles the limitation of binary labeling in sequential recommender losses by introducing weak transitivity to capture preference-ordered relations among unobserved items. It derives extensions of Pairwise, Pointwise, and Setwise losses (TransBPR, TransBCE, TransSSM) using two-negatives sampling schemes (popularity vs. niche) and a gamma-balanced preference term to enforce $\\hat{s}_{ui} > \\hat{s}_{uj} > \\hat{s}_{uk}$. Empirical results across Beauty, Toys, Sports, and Yelp show consistent gains with TransSSM_pop frequently achieving the best metrics, while weak transitivity provides stronger gradients and robustness than strict alternatives. Overall, the approach improves ranking in implicit feedback settings by leveraging explicit preference structure among unobserved items, with practical implications for more accurate sequential recommendations.
Abstract
A choice of optimization objective is immensely pivotal in the design of a recommender system as it affects the general modeling process of a user's intent from previous interactions. Existing approaches mainly adhere to three categories of loss functions: pairwise, pointwise, and setwise loss functions. Despite their effectiveness, a critical and common drawback of such objectives is viewing the next observed item as a unique positive while considering all remaining items equally negative. Such a binary label assignment is generally limited to assuring a higher recommendation score of the positive item, neglecting potential structures induced by varying preferences between other unobserved items. To alleviate this issue, we propose a novel method that extends original objectives to explicitly leverage the different levels of preferences as relative orders between their scores. Finally, we demonstrate the superior performance of our method compared to baseline objectives.