Identifying　similar　users　lay　the　foundation　in　many　fields,　such　as　friend　recommendation,　user-based　collaborative　filtering,　and　community　discovery.　It　is　useful　to　analyze　users＇　similarity　based　on　check-in　data,　especially　the　analysis　of　spatiotemporal　and　semantic　information.　The　existing　works　pursue　semantic　similarity　of　user　trajectories　and　cannot　distinguish　the　effects　of　geographical　factors　in　a　fine-grained　way.　This　paper　proposes　a　graph　embedding　approach　to　identify　similar　users　based　on　their　check-in　data.　We　firstly　identify　meaningful　concepts　of　user　check-in　data,　based　on　which　we　design　a　metagraph　for　representing　features　of　similar　user　behaviors.　Then　we　characterize　each　user　with　a　sequence　of　nodes　that　are　derived　through　a　metagraph-guided　random　walk　strategy.　Finally,　the　sequences　are　embedded　to　generate　meaningful　user　vectors　that　are　used　to　the　similarity　among　users　and　thus　identify　similar　users.　We　evaluate　our　proposal　on　two　datasets,　the　results　of　which　show　that　our　proposal　can　outperform　the　baselines.　©　2020　Knowledge　Systems　Institute　Graduate　School.　All　rights　reserved.