Analyzing　technical　contradictions　according　to　TRIZ　theory　can　help　us　solve　the　innovative　and　inventive　problems　effectively.　However,　the　existing　analysis　and　extraction　methods　of　technical　contradiction　mainly　rely　on　manual　rule　formulation　and　make　less　use　of　the　semantics　information,　which　limits　the　improvement　of　the　efficiency　and　accuracy　of　the　extraction.　Consequently,　this　paper　proposed　an　extraction　method　based　on　patent　semantic　space　mapping.　It　adopted　the　Doc2Vec　model　to　construct　the　semantic　space　of　the　patent　text　and　trained　extraction　model　by　using　feature　vectors　covering　rich　semantic　relationships,　which　could　make　us　recognize　the　technical　contradictions　from　patents　better.　This　paper　used　patent　data　in　the　automobile　field　as　a　sample　to　perform　experiments.　The　accuracy　of　the　technical　contradiction　recognition　was　improved　compared　with　the　baseline　model　based　on　rule　formulation.　©　2020　ACM.