MEFCL: Reliability-aware multi-view edge-filtered contrastive learning for graph recommendation

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).

Multi-view learning has become an important paradigm for recommender systems by leveraging complementary collaborative semantics from heterogeneous structural perspectives. Existing graph contrastive recommendation methods, however, predominantly rely on random perturbations or heuristic view construction, which may introduce unreliable semantic consistency and noisy collaborative signals. To address these challenges, we propose MEFCL, a reliability-aware multi-view edge-filtered contrastive learning framework for graph recommendation. Specifically, MEFCL formulates recommendation as a multi-view collaborative representation learning problem and constructs complementary local structural and global collaborative views to capture fine-grained neighborhood semantics and high-order structural relationships, respectively. To improve the reliability of multi-view representations, we further develop an edge-level reliability fusion mechanism that jointly evaluates global structural importance and local structural consistency using PageRank centrality and Weisfeiler–Lehman graph kernels. Based on the fused reliability signals, unreliable interactions are filtered before global collaborative modeling to generate noise-resilient contrastive views. In addition, a cross-view contrastive alignment objective is introduced to enhance representation consistency and robustness across different structural views. Extensive experiments on multiple real-world recommendation benchmarks demonstrate that MEFCL consistently outperforms state-of-the-art methods in terms of recommendation accuracy and robustness, particularly under noisy interaction settings and long-tail recommendation scenarios.