As　an　effective　tool　to　describe　the　fuzziness　of　qualitative　information,　the　shadowed　sets　have　received　increasing　attention.　By　establishing　reasonable　compromise　between　qualitative　information　and　quantitative　membership　grades,　the　shadowed　sets　provide　a　new　way　to　model　linguistic　variables.　In　this　paper,　we　propose　an　extended　TODIM　method　based　on　shadowed　sets　to　solve　large-scale　group　decision　making　problems　with　linguistic　information.　First,　considering　the　linguistic　terms　cannot　be　directly　computed,　a　codebook　used　to　model　linguistic　terms　is　constructed　with　shadowed　sets　by　the　data-driven　and　the　percentile　method.　Next,　to　improve　the　decision　efficiency,　a　shadowed　set-based　clustering　model　is　proposed　to　cluster　the　decision　makers　on　the　basis　of　new　similarity　measure　and　the　Louvain　algorithm.　Then,　we　propose　a　new　decision-making　method　by　taking　into　account　the　decision　makers’　psychological　behavior　which　has　a　vital　influence　on　the　decision　results.　Finally,　we　utilize　a　case　study　about　assembly　factory　site　selection　to　illustrate　the　feasibility　of　the　proposed　method,　meanwhile　comparison,　parameter　discussion　and　sensitivity　analysis　are　also　conducted　to　verify　the　superiority　and　the　efficiency.　©　2020　Elsevier　Inc.