Text　sentiment　analysis　and　its　incorporation　into　recommender　systems　have　been　study　topics　in　the　past　few　years.　Recommendation　accuracy　can　be　increased　by　using　sentiment　elements.　One　of　the　most　popular　is　dichotomous　sentiment　analysis.　Nevertheless,　the　real　emotions　of　users　are　diverse,　and　sentiment　dichotomous　classification　is　not　sufficient　to　fully　express　user　attitudes.　To　address　this　problem,　this　paper　innovatively　proposes　the　Sentiment　Multi-label　Classification　Recommender　System　(SMCRS)　model,　which　establishes　a　six-label　sentiment　classification　module　for　text,　and　extracts　relevant　word　features　for　each　label　by　decomposing　the　sentence　The　label　prediction　is　finally　realized.　Moreover,　the　model　builds　higher-order　interactions　between　various　features　to　realize　the　recommendation　function　by　incorporating　sentiment　as　well　as　user　and　item　information.　Finally,　we　tested　the　performance　of　SMCRS　on　the　JD　dataset,　and　the　accuracy　of　the　sentiment　classification　module　is　as　high　as　78.02%,　while　the　AUC　of　the　whole　model　is　improved　by　6.35%.　This　is　a　great　improvement,　and　it　also　proves　that　sentiment　multi-classification　is　very　helpful　for　the　performance　improvement　of　recommender　systems.　©　2024　SPIE.