Patent　similarity　measurement,　as　one　of　the　fundamental　building　blocks　for　patent　analysis,　is　able　to　derive　technical　intelligence　efficiently,　but　also　can　detect　the　risk　of　infringement　and　evaluate　whether　the　invention　meets　the　criteria　of　novelty　and　innovation.　However,　traditional　approaches　make　implicitly　several　assumptions,　such　as　bag　of　words　in　each　component,　semantic　direction　irrelevance　and　so　on.　In　order　to　relax　these　assumptions,　this　study　proposes　an　improved　methodology　on　the　basis　of　entities　and　semantic　relations　(functional　and　non-functional　relations),　which　takes　semantic　direction　of　each　sequence　structure　and　the　word　order　information　of　each　component　into　consideration.　Meanwhile,　an　algorithm　for　calculating　the　global　importance　of　each　sequence　structure　is　put　forward.　Finally,　to　verify　the　effectiveness　and　performance　of　the　improved　semantic　analysis,　a　case　study　is　conducted　on　the　thin　film　head　subfield　in　the　field　of　hard　disk　drive.　Extensive　experimental　results　show　that　our　approach　is　significantly　more　accurate.　©　2021　Elsevier　Ltd.