Large-scale　group　decision-making　(LSGDM)　is　a　complex　process　involving　numerous　decision-makers　(DMs).　However,　considering　such　a　large　number　of　DMs　increases　the　complexity　of　the　process.　it　seems　necessary　to　pay　much　more　attention　to　aspects　such　as　a　proper　dimensionality　reduction　for　scalability,　consensus　processes　with　automatic　feedback,　and　effective　management　of　non-cooperative　DMs.　To　address　such　aspects,　this　paper　presents　a　novel　framework　for　LSGDM,　based　on　Extended　Comparative　Linguistic　Expressions　With　Symbolic　Translation　(ELICIT).　We　first　extend　the　K-means　clustering　algorithm　by　incorporating　individual　assessments　and　trust　relationships　to　classify　DMs　into　subgroups,　enhancing　decision-making　efficiency.　We　then　develop　a　feedback　mechanism　based　on　two　optimization　consensus　models　for　ELICIT　information,　that　automatically　provides　optimal　recommendations.　An　essential　aspect　of　our　proposal　is　the　management　of　non-cooperative　behaviors　by　utilizing　the　normal　distribution　to　detect　and　penalize　misbehaviors.　Furthermore,　we　introduce　a　Data　Envelopment　Analysis　(DEA)　cross-efficiency　method　based　on　ELICIT　values　to　rank　all　alternatives　once　an　acceptable　group　consensus　degree　is　reached.　The　framework＇s　effectiveness　is　demonstrated　through　a　practical　application　case　study,　accompanied　by　a　parametric　analysis.　Comparisons　with　existing　LSGDM　methods　highlight　the　superiority　of　our　proposal　in　terms　of　efficiency.　©　2023　Elsevier　Ltd