Community　detection　is　a　fundamental　tool　to　uncover　organizational　principles　in　complex　networks.　With　the　proliferation　of　rich　information　available　for　real-world　networks,　it　is　useful　to　detect　communities　in　attributed　networks.　Recently,　many　algorithms　consider　combinating　node　attributes　and　network　topology,　and　the　effect　of　these　methods　is　better　than　using　only　one　information　source.　However,　the　existing　algorithms　still　have　some　shortcomings.　First,　only　a　few　algorithms　can　process　both　categorical　type　and　numerical　type　at　the　same　time.　Second,　the　contribution　between　attributes　and　topology　cannot　be　adjusted　adaptively.　Third,　most　algorithms　do　not　consider　combining　the　high-order　structure　with　the　node　attributes.　Therefore,　we　propose　an　adaptive　seed　expansion　algorithm　based　on　composite　similarity　to　solve　these　problems(ASECS).　We　generate　a　new　weighted　KNN　graph　according　to　the　composite　similarity.　The　composite　similarity　combines　high-order　structure　similarity,　low-order　structure　similarity　and　attributed　similarity　by　weighting　them.　Our　method　can　adaptively　adjust　the　contribution　between　topology　and　node　attributes.　Moreover,　the　designed　attributed　similarity　function　can　process　both　categorical　and　numerical　attributes.　Finally,　we　find　the　seed　nodes　on　the　weighted　KNN　graph　and　expand　the　seed　nodes　to　communities.　The　superiority　of　our　algorithm　is　demonstrated　on　many　networks.