Main　Path　Analysis　(MPA)　is　widely　used　to　trace　the　developmental　trajectory　of　a　technological　field　through　a　citation　network.　The　citation-based　traversal　weight　is　usually　utilized　to　cherrypick　the　most　significant　path.　However,　the　theme　of　documents　along　a　main　path　may　not　be　so　coherent,　and　it　is　very　possible　to　miss　the　main　paths　of　significant　sub-fields　overall　in　a　domain.　Furthermore,　the　global　path　search　algorithm　in　conventional　MPA　also　suffers　from　high　space　complexity　due　to　the　exhaustive　strategy.　To　address　these　limitations,　a　new　method,　named　as　semantic　MPA　(sMPA),　is　proposed　by　leveraging　semantic　information　in　two　steps　of　candidate　path　generation　and　main　path　selection.　In　the　meanwhile,　the　resulting　source　code　can　be　freely　accessed.　To　demonstrate　the　advantages　of　our　method,　extensive　experiments　are　conducted　on　a　patent　dataset　pertaining　to　lithium-ion　battery　in　electric　vehicle.　Experimental　results　show　that　our　sMPA　is　capable　of　discovering　more　knowledge　flows　from　important　subfields,　and　improving　the　topical　coherence　of　candidate　paths　as　well.