Changjie Wang, Bohan Li, Wenlong Wu, Meng Ma, Yuanyang Zhang, Yuhang Dai, Chenhui Feng, Liang Liu, Hua Dai, Haofen Wang · ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 2026
DOI: 10.1145/3799682.3840977
Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).
Session-based Recommendation (SBR) achieves high performance in traditional centralized settings by using global item transition graphs to capture user intent. However, Federated Learning (FL) mandates collaborative training on isolated data, which makes the construction of such global structures unattainable. We identify this critical dilemma as ''Structure Collapse'', where the global graph remains fragmented despite model averaging. This collapse induces a ''Semantic Chasm,'' resulting in incompatible knowledge fusion and imprecise intent prediction. To break this impasse, we propose a novel Cross-Silo Federated Session-based Recommendation framework (CSFSR), which reconstructs global connectivity by harmonizing knowledge priors with local intents. Specifically, our Knowledge-Aware Initialization counteracts Structure Collapse by synthesizing shared attributes into a unified semantic space that anchors isolated clients. Further addressing the Semantic Chasm, the Graph-Guided Aggregation mechanism uses client signatures to direct personalized fusion, effectively filtering noise and ensuring robustness. Extensive experiments on three real-world datasets demonstrate that CSFSR significantly outperforms state-of-the-art federated baselines and exhibits superior noise robustness. Our code is available at https://github.com/Is-Jiejie/CSFSR.
ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 0 citations
ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 0 citations
ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 0 citations
ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 0 citations
No comments yet — start the discussion below.