In　clinical　EEG　research,　the　construction　of　EEG　datasets　has　an　important　impact　on　the　research　results.　However,　the　needs　of　dataset　construction　are　diverse,　each　of　which　involves　multiple　dimensions,　and　a　single　construction　scheme　cannot　meet　the　needs　of　users.　Meanwhile,　when　extracting　samples,　labels　and　requirements　need　to　be　compared　one　by　one,　and　most　datasets　do　not　provide　negative　examples.　Based　on　the　above　problems,　a　set　of　rules　is　proposed　to　describe　the　requirements　of　constructing　EEG　datasets,　and　a　software　system　framework　supporting　rule-driven　EEG　dataset　generation　is　designed　and　implemented.　It　solves　the　problem　of　too　rigid　rule　making　and　lack　of　negative　example　labels　when　constructing　the　data　set.　It　makes　the　construction　of　data　sets　more　convenient　and　efficient.　Finally,　based　on　the　system,　a　negative　example　construction　scheme　of　active　learning　based　on　uncertain　query　strategy　was　designed.　The　model　trained　using　the　dataset　constructed　by　the　proposed　scheme　has　better　performance　compared　to　the　scheme　where　the　negative　examples　are　randomly　selected.　©　2023　IEEE.