Community　detection　is　one　of　the　important　methods　for　understanding　the　mechanism　behind　the　function　of　social　networks.　The　recently　developed　label　propagation　algorithm　(LPA)　has　been　gaining　increasing　attention　because　of　its　excellent　characteristics,　such　as　a　succinct　framework,　linear　time　and　space　complexity,　easy　parallelization,　etc.　However,　several　limitations　of　the　LPA　algorithm,　including　random　label　initialization　and　greedy　label　updating,　hinder　its　application　to　complex　networks.　A　new　parallel　LPA　is　proposed　in　this　study.　First,　grey　relational　analysis　is　integrated　into　the　label　updating　process,　which　is　based　on　vertex　similarity.　Second,　parallel　propagation　steps　are　comprehensively　studied　to　utilize　parallel　computation　power　efficiently.　Third,　randomness　in　label　updating　is　significantly　reduced　via　automatic　label　selection　and　label　weight　thresholding.　Experiments　conducted　on　artificial　and　real　social　networks　demonstrate　that　the　proposed　algorithm　is　scalable　and　exhibits　high　clustering　accuracy.　©　2016　Elsevier　B.V.