The　metaverse　presents　a　new　Internet　paradigm　where　individuals　can　engage　in　activities　like　play,　work,　and　social　interactions　in　an　immersive　virtual　world.　Empowered　by　edge　computing　technologies,　users　can　seamlessly　delve　into　this　digital　universe　through　ultra-fast　rendering　of　virtual　environments　and　efficient　storage　of　character　data.　However,　the　large-scale　social　interaction　of　the　metaverse　could　strain　network　bandwidth,　exacerbating　congestion　on　the　existing　edge　computing　framework.　To　address　these　challenges,　we　consider　a　human-centric　edge　collaboration　architecture　that　allows　edge　nodes　in　the　same　social　activities　can　collaborate　to　meet　user　needs.　Moreover,　we　introduce　a　socially-aware　traffic　scheduling　algorithm　within　this　architecture　to　address　the　complex　cross-edge　resource　allocation　problem.　By　formulating　the　traffic　scheduling　problem　as　a　Markov　Decision　Process　(MDP)　considering　both　network　and　social　factors,　we　present　a　novel　socially-aware　deep　reinforcement　learning　(DRL)　algorithm　tailored　for　this　MDP.　Extensive　experiments　demonstrate　that　our　approach　satisfies　network　performance　during　extreme　burst　traffic　scenarios,　effectively　managing　the　dynamic　social　demands　of　the　metaverse.　IEEE