With　the　frequent　use　of　hazardous　chemicals　in　industrial　production　and　experimental　research,　their　recovery　and　hazard　characterization　are　very　important　for　the　protection　of　environmental　health　and　sustainable　development.　However,　it　is　not　efficient　enough　to　analyze　and　judge　the　hazardous　chemicals　in　the　form　of　a　table　query,　so　the　introduction　of　big　data-related　technologies　is　of　great　importance.　To　gain　a　better　understanding　of　the　characteristics　and　degree　of　hazard　associated　with　chemicals,　this　paper　incorporates　the　hazardous　chemicals　domain　knowledge　graph　into　the　process　of　identifying　hazardous　chemical　risks.　A　hazardous　chemical　recovery　text　recognition　model,　KG-TextRCNN,　is　constructed　to　minimize　errors　resulting　from　subjective　evaluations　or　manual　recognition.　The　experimental　results　indicate　that　the　hazardous　chemicals　text　corpus　recognition　has　a　total　average　precision　value　of　99.30%,　with　recall　and　F1　values　above　99%.　During　the　classification　step　of　max　pooling,　the　model　achieved　average　precision,　recall,　and　F1　values　of　85.19%,　86.29%,　and　85.27%,　respectively.　The　model’s　overall　performance　is　stable,　with　high　precision　and　recall　rates　for　most　categories.　This　paper　conducts　ablation　and　comparison　experiments　to　evaluate　and　validate　the　effectiveness　of　the　model.　©　The　Author(s),　under　exclusive　licence　to　Springer　Nature　Switzerland　AG　2025.