With　the　rapid　development　of　social　media　technology,　the　enormous　graph　is　difficult　to　be　applied　in　real　applications　such　as　classification,　clustering,　and　link　prediction,　due　to　lack　of　efficient　and　available　node　representation.　Graph　embedding　is　widely　concerned　by　transforming　high-dimensional　sparse　graph　data　into　low-dimensional,　compact　representations.　While　previous　graph　embedding　methods　mainly　focus　on　preserving　the　microscopic　structure,　the　significant　characteristic　of　graph,　namely,　inherent　cluster　structure　of　nodes　in　the　original　graph　is　largely　ignored.　Besides,　embedding　algorithms　on　pure　graphs　have　been　intensively　studied.　However,　nodes　are　often　accompanied　with　abundant　node　attributes,　which　are　highly　correlated　with　node　connections　and　potential　to　improve　the　node　embedding.　How　to　fully　incorporate　attributes　in　node　representation　remains　to　be　further　exploited.　In　this　paper,　we　propose　a　deep　graph　embedding　method　with　clustering　optimization　called　AGEDC(Adaptively　Attribute-enhanced　Graph　Embedding　via　Deep　Clustering　Constraints).　It　first　enriches　the　original　topological　structure　by　an　adaptive　attribute-enhancement　mechanism,　then　it　incorporates　deep　clustering　into　latent　embedding　learning　for　inherent　cluster-structure　constraints.　Further,　the　learnt　low-dimensional　representation　of　node　is　self-corrected　through　a　convolution-like　operation　by　local　and　global　structure　in　the　augmented　graph.　Compared　with　the　previous　graph　embedding　methods,　experimental　results　in　different　application　scenarios　on　real-world　datasets　demonstrate　the　effectiveness　of　our　proposed　algorithm.　©　2021　ACM.