With　the　growth　of　the　variety　and　quantity　of　Internet　financial　products,　how　to　effectively　implement　personalized　recommendation　has　become　the　key　issue.　The　particularity　of　Internet　financial　products　lies　in　that　attribute　value　(such　as　7-day　annualized　income)　is　not　fixed.　Users　will　consider　the　attribute　value　of　the　product　at　that　time　when　choosing　the　product　at　different　times.　The　importance　of　timing　factor　in　financial　product　recommendation　cannot　be　ignored.　However,　the　traditional　financial　product　recommendation　algorithm　is　based　on　the　static　attributes　of　users　and　financial　products,　ignoring　the　factors　of　time　in　series　data,　and　the　recommendation　quality　is　low.　With　the　development　of　artificial　intelligence,　deep　learning　technology　has　been　widely　used　in　personalized　recommendation　system.　Therefore,　taking　advantage　of　the　advantages　of　transformer　in　processing　time　series,　an　R-Transformer(Recommendation　system　based　on　transformer)　network　has　proposed.　Two　R-Transformer　networks　are　used　to　mine　users＇　and　financial　products＇　states　based　on　time　series,　and　the　inner　product　of　users＇　and　financial　products＇　states　is　taken　as　the　final　score.　Experimental　results　show　that　compared　with　the　traditional　collaborative　filtering　and　RNN　algorithm,　this　algorithm　can　effectively　reduce　RMSE　and　improve　the　F-measure　of　prediction.