[Purpose/　significance]　To　facilitate　precise　matching　of　massive　technology　supply　and　demand,　and　to　promote　the　conversion　of　innovative　achievements　into　real　productive　forces.　[Method/　process]　This　paper　proposes　a　personalized　recommendation　method　for　patent　transactions　based　on　enterprise　profiling,　driven　by　Large　Language　Model.　Based　on　the　historical　transaction　information　of　patents　by　enterprises,　the　Large　Language　Model　is　utilized　to　mine　and　analyze　the　fine-grained　demand　preference　characteristics　of　enterprises　in　both　the　short　and　long　term.　By　integrating　demand　preference　characteristics　with　basic　enterprise　information,　an　enterprise　profile　is　constructed.　A　patent　transaction　recommendation　method　based　on　two-stage　matching　is　introduced,　leveraging　the　chain-of-thought　of　Large　Language　Model　to　match　the　enterprise　profile　with　patent　information　in　the　database　through　staged　multi-dimensional　semantic　information　matching,　thereby　achieving　personalized　patent　transaction　recommendations　and　providing　justifications　for　the　recommendations.　[Result/　conclusion]　Experimental　results　indicate　that　the　precision,　recall,　and　F1　score　of　the　proposed　method　are　0.　835,　0.　839,　and　0.　836,　respectively,　significantly　outperforming　traditional　baseline　models.　Utilizing　the　Large　Language　Model’s　extensive　world　knowledge　and　expert-level　reasoning　capabilities　can　accurately　construct　enterprise　profiles,　ensuring　the　interpretability　of　recommendation　results　while　maintaining　recommendation　performance.　[Limitations]　The　construction　of　enterprise　profiles　primarily　relies　on　historical　transaction　information,　and　considering　the　low　frequency　of　patent　transactions,　the　cold　start　recommendation　problem　remains　to　be　further　addressed.　©　2025　Information　studies:　Theory　and　Application.　All　rights　reserved.