The　Priority　Dispatching　Rules　(PDRs)　and　business　solvers　are　widely　employed　for　solving　real-world　scheduling　problems,　such　as　Job　Shop　Scheduling　(JSSP)　and　JSSP　with　maintenance　time　(JSSP-MT)　problems.　However,　the　effective　designs　of　PDRs　and　mathematical　modelings　are　tedious　and　complex　relying　heavily　on　a　myriad　of　specialized　knowledge.　In　this　paper,　we　propose　an　approach　to　automatically　learn　PDRs　based　on　Graph　Node　Embedding　(GNE)　and　Deep　Reinforcement　Learning　(DRL).　We　describe　JSSP　as　a　disjunctive　graph　and　utilize　a　GNE　approach　to　facilitate　better　state　embeddings.　Ablation　studies　demonstrate　the　significant　positive　contribution　of　GNE　both　on　JSSP　and　JSSP-MT.　Experiments　show　that　some　high-quality　combined　PDRs　can　be　learned　with　better　approximate　solutions　against　the　traditional　single　PDRs.　Solutions　produced　by　our　approach　are　much　closer　to　those　from　mathematical　solvers　than　previous　methods.　©　2023　SPIE.