Drug　recommendation　is　an　important　task　of　AI　for　healthcare.　To　recommend　proper　drugs,　existing　methods　rely　on　various　clinical　records　(e.g.,　diagnosis　and　procedures),　which　are　commonly　found　in　data　such　as　electronic　health　records　(EHRs).　However,　detailed　records　as　such　are　often　not　available　and　the　inputs　might　merely　include　a　set　of　symptoms　provided　by　doctors.　Moreover,　existing　drug　recommender　systems　usually　treat　drugs　as　individual　items,　ignoring　the　unique　requirements　that　drug　recommendation　has　to　be　done　on　a　set　of　items　(drugs),　which　should　be　as　small　as　possible　and　safe　without　harmful　drug-drug　interactions　(DDIs).　To　deal　with　the　challenges　above,　in　this　paper,　we　propose　a　novel　framework　of　Symptom-based　Set-to-set　Small　and　Safe　drug　recommendation　(4SDrug).　To　enable　set-to-set　comparison,　we　design　set-oriented　representation　and　similarity　measurement　for　both　symptoms　and　drugs.　Further,　towards　the　symptom　sets,　we　devise　importance-based　set　aggregation　to　enhance　the　accuracy　of　symptom　set　representation;　towards　the　drug　sets,　we　devise　intersection-based　set　augmentation　to　ensure　smaller　drug　sets,　and　apply　knowledge-based　and　data-driven　penalties　to　ensure　safer　drug　sets.　Extensive　experiments　on　two　real-world　EHR　datasets,　i.e.,　the　public　benchmark　one　of　MIMIC-III　and　the　industrial　large-scale　one　of　NELL,　show　drastic　performance　gains　brought　by　4SDrug,　which　outperforms　all　baselines　in　most　effectiveness　measures,　while　yielding　the　smallest　sets　of　recommended　drugs　and　26.83%　DDI　rate　reduction　from　the　ground-truth　data.　©　2022　ACM.