Knowledge　graph　is　a　type　of　network　structure　in　which　nodes　represent　entities　and　edges　indicate　relations.　However,　as　the　network　size　explosively　increases,　the　issues　of　data　sparsity　and　computation　inefficiency　on　large-scale　knowledge　graph　become　more　difficult　to　manipulate　and　manage.　Knowledge　graph　embedding,　which　is　a　representation　technique　of　embedding　entities　and　relations　in　the　knowledge　graph　into　continuous,　dense,　and　low-dimensional　semantics　vector　spaces　to　tackle　these　challenges　and　endow　the　model　with　the　abilities　of　knowledge　fusion　and　inference,　has　recently　attracted　much　attention.　In　this　paper,　we　firstly　introduce　the　overall　framework　and　specific　idea　of　embedding　models.　We　then　introduce　two　applications　that　apply　KG　embedding,　compare　the　performance　of　the　methods　in　these　applications.　Finally,　we　summarize　several　challenges　to　overcome,　and　provide　some　prospective　future　research　directions　such　as　deep　learning　network　for　approaches　and　applications.