Recommendation　systems　play　a　critical　role　in　web　applications　(e.g.e-commerce　sites,　etc.),　so　as　to　alleviate　data　overload　and　suggest　preferred　items　for　users.　Due　to　the　limited　interactions　between　users　and　items　in　real-life　scenarios,　contrastive　learning　has　recently　been　adopted　to　enhance　recommendation　systems.　However,　most　existing　CL(Contrastive　Learning)-based　methods　typically　construct　contrastive　views　through　node　or　edge　perturbation,　which　may　disrupt　the　intrinsic　semantic　structure　in　user-item　interaction　graphs.　And,　these　methods　also　neglect　user　preferences　and　emotional　tendencies　in　reviews.　Therefore,　this　paper　proposes　a　novel　sentiment-aware　graph　contrastive　learning　method　(SentiGCN),　which　can　comprehensively　explore　information　from　both　review　content　and　user-item　graphs.　Firstly,　sentiment　information　is　extracted　from　reviews　using　Natural　Language　Processing　(NLP)　tools.　Then,　the　two　graph　structures　integrating　this　information　are　constructed.　Moreover,　two　contrastive　strategies　are　designed　to　incorporate　self-supervised　signals.　For　the　user-item　interaction　graph　with　sentiment　features,　a　scheme　involving　the　addition　of　random　noise　in　the　embedding　space　is　applied　for　graph　augmentation.　For　the　item-item　neighbor　graph　based　on　sentiment　similarity,　a　cross-layer　contrastive　scheme　is　implemented　to　further　explore　the　potential　associations　between　items.　Finally,　extensive　experiments　on　three　datasets　confirm　the　efficacy　of　SentiGCN.　©　2024　IEEE.