The　Set　Orienteering　Problem　(SOP)　is　a　variant　of　the　popular　Orienteering　Problem　(OP)　arising　from　a　number　of　real-life　applications.　The　aim　is　to　find　a　tour　across　a　subset　of　customers,　while　maximizing　the　collected　profit　within　a　given　travel　time　limit.　In　SOP,　vertices　(customers)　are　partitioned　into　clusters,　where　a　profit　is　associated　to　each　cluster.　The　profit　of　a　cluster　is　collected　only　if　at　least　one　vertex　belonging　to　the　cluster　is　contained　in　the　tour.　For　NP-hard　problem,　we　present　a　highly　effective　hybrid　evolutionary　algorithm　that　integrates　cluster-based　crossover　operator,　a　randomized　mutation　operator　to　generate　multiple　distinct　promising　offspring　solutions,　and　a　two-phase　local　refinement　procedure　that　explores　feasible　and　infeasible　solutions　in　search　of　high-quality　local　optima.　Extensive　experiments　192　large　benchmark　instances　show　that　the　proposed　algorithm　significantly　outperforms　the　existing　approaches　from　the　SOP　literature.　In　particular,　it　reports　improved　best-known　solutions　(new　lower　bounds)　for　54　instances,　while　matching　the　existing　best-known　for　133　instances.　We　further　investigate　the　contribution　of　the　key　algorithmic　elements　to　success　of　the　proposed　approach.