Graph　neural　network-based　recommendation　models　have　achieved　remarkable　results　by　performing　information　propagation　on　a　bipartite　graph　consisting　of　user　nodes　and　item　nodes.　Nevertheless,　these　models　typically　focus　on　message　passing　between　nodes　iteratively,　and　pay　insufficient　attention　to　the　intricate　interactions　encapsulated　by　the　graph’s　edges,　which　play　a　pivotal　role　in　recommendation　systems.　Moreover,　the　process　of　node　embedding　learning　often　fails　to　consider　the　potential　similarities　between　users　or　items.　Such　omission　could　curtail　the　representational　learning　efficacy　of　these　models.　To　address　these　challenges,　we　propose　an　edge-based　graph　neighbor　filtering　network　framework　for　recommendation.　This　framework　introduces　an　edge-based　message　passing　mechanism　to　capture　the　interaction　information　between　users　and　items.　Additionally,　a　penalty　term　that　utilizes　neighbor　filtering　is　incorporated　to　coalesce　nodes　exhibiting　analogous　characteristics　more　proximally　within　the　embedding　space.　This　inclusion　serves　as　a　catalyst　for　augmenting　the　prowess　of　representation　learning.　Comprehensive　experiments　and　ablation　studies　are　conducted,　and　the　results　validate　the　effectiveness　of　our　framework.　For　example,　on　the　datasets　book_crossing　and　df_modcloth,　the　proposed　framework　can　achieve　relative　improvements　of　3%　and　over　20%,　respectively,　compared　to　the　strongest　baseline　in　terms　of　NDCG@10　and　Recall@10.　©　The　Author(s),　under　exclusive　licence　to　Springer　Science+Business　Media,　LLC,　part　of　Springer　Nature　2025.