Skip to content
Book Open access

DART: Residual Trust and Tail-Aware Dual-View Alignment for Sequential Recommendation

Nov 2026 · Proceedings of the 35th ACM International Conference on Information and Knowledge Management · 0 citations · 27 references

Abstract

Sequential recommender systems commonly learn item representations from interaction data. This paradigm works well for frequent items, but becomes unreliable for long-tail items whose ID embeddings receive limited updates. Offline semantic embeddings derived from large language models provide useful item-side knowledge beyond interaction logs. They can therefore compensate for the insufficient interaction supervision of long-tail items. However, directly optimizing such semantic priors for recommendation may distort their original semantic structure. Moreover, semantic information should not be injected uniformly across all items, since head items are already well supported by interactions while long-tail items require semantic compensation more urgently. In this paper, we propose DART, a constrained semantic enhancement framework that couples Residual Trust Semantic Adaptation with Tail-Aware Dual-View Alignment for sequential recommendation. DART treats offline LLM-derived item embeddings as frozen semantic anchors and injects collaborative signals through a trust-region-constrained residual pathway. This design allows item representations to absorb useful collaborative information while limiting semantic drift. To further improve long-tail recommendation, DART introduces a tail-aware dual-view alignment mechanism that concentrates cross-view transfer on long-tail items instead of aligning all items uniformly. Experiments on three public benchmarks show that DART achieves strong overall performance and consistent gains in long-tail settings. Further ablation and representation analyses verify the effectiveness of residual-trust adaptation and tail-aware semantic transfer.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.