Community　detection　in　complex　networks　can　find　the　community　structure　one　of　the　most　important　properties　of　complex　networks.　Nodes　in　the　same　community　have　more　dense　connections　than　those　in　different　communities,　which　can　be　utilized　to　analyze　the　function　of　complex　networks.　In　addition,　heterogeneous　networks　are　ubiquitous　in　the　real　world.　For　example,　academic　networks　have　different　types　of　nodes　such　as　authors,　papers,　and　conferences.　Network　representation　learning　is　an　important　method　to　discover　the　complex　nonlinear　relationships　between　nodes　in　the　network,　which　is　of　great　help　to　community　detection.　Attention　network　is　a　typical　network　representation　learning　method,　and　it　will　pay　attention　to　the　important　part　in　the　network　for　the　specific　task.　However,　most　existing　heterogeneous　NRL　algorithms　use　metapaths　to　capture　heterogeneous　information,　which　requires　prior　knowledge　to　set　metapaths　in　advance.　This　paper　proposes　a　novel　random-walk-based　heterogeneous　attention　network　(RHAN)　for　community　detection　on　heterogeneous　networks.　Random　walk　is　used　to　generate　the　neighbor　nodes　set　of　nodes,　and　heterogeneous　information　is　considered　by　the　intra-type　attention　and　the　inter-type　attention,　which　is　no　need　for　metapaths.　The　experimental　results　on　four　widely　used　heterogeneous　networks　verify　the　effectiveness　of　RHAN.