More　and　more　people　become　conscious　of　the　recommendation　system　to　make　good　use　of　the　data　through　their　inherent　advantages　faced　with　the　large　amount　of　data　on　the　Internet.　The　collaborative　filtering　recommendation　algorithm　cannot　avoid　the　bottleneck　of　computing　performance　problems　in　the　recommendation　process.　In　this　paper,　we　propose　a　parallel　collaborative　filtering　recommendation　algorithm　RLPSO-KM-CF　which　is　implemented　based　on　Spark.　Firstly,　the　RLPSO　(reverse-learning　and　local-learning　PSO)　algorithm　is　used　to　find　the　optimal　solution　of　particle　swarm　and　output　the　optimised　clustering　centre.　Then,　the　RLPSO-KM　algorithm　is　used　to　cluster　the　user　information.　Finally,　make　effective　recommendations　to　the　target　user　by　combining　the　traditional　user-based　collaborative　filtering　algorithm　with　the　RLPSO-KM　clustering　algorithm.　The　experimental　results　show　that　the　RLPSO-KM-CF　algorithm　has　a　significant　improvement　in　the　recommendation　accuracy　and　has　a　higher　speed-up　and　stability.　Copyright　©　2018　Inderscience　Enterprises　Ltd.