The　representation　learning　approach　aims　to　obtain　a　low-dimensional　representation　of　nodes　and　accomplish　community　detection　by　clustering.　Adjacency　matrix　is　the　most　common　form　of　network　representation,　but　it　only　represents　the　direct　connection　relationship　of　network　nodes　and　lacks　more　useful　topological　information.　Existing　approaches,　such　as　jaccard　coefficient　for　topology　extraction,　are　still　limited　to　neighborhoods,　and　the　available　information　is　not　rich　enough.　In　addition,　roles,　another　vital　idea,　lack　a　more　profound　application　to　network　topology.　This　paper　proposes　a　novel　community　detection　algorithm　based　on　enhancing　graph　autoencoder　with　node　structural　role　(CDESR).　On　the　one　hand,　the　structural　role　we　designed　effectively　specifies　the　importance　of　nodes　in　the　network.　Based　on　this　idea,　a　new　strategy　for　computing　node　topological　relations　is　proposed　for　their　information　extraction.　On　the　other　hand,　the　enhancement　matrix　constructed　using　the　extracted　rich　information　efficiently　optimizes　the　graph　autoencoder　to　obtain　a　high-quality　representation.　The　experimental　results　on　real-world　and　synthetic　networks　verify　the　effectiveness　of　our　algorithm.　©　2023,　The　Author(s),　under　exclusive　license　to　Springer　Nature　Singapore　Pte　Ltd.