The　conventional　keywords　statistics-based　methods　of　Internet　public　opinion　analysis　are　low　accuracy　while　lack　of　semantic　processing　which　is　necessary.　We　propose　a　semantic-based　method　of　Internet　public　opinion　short　text　analysis.　An　extensible　context-free　grammar　is　presented　to　parse　the　IPO　short　text.　We　adopt　ontology　of　IPO　as　a　meta-knowledge　to　help　understanding　IPO　short　text.　Then　we　design　and　implement　a　system　IPOAAS,　which　provides　a　running　platform　for　ECFG　parsing.　Experimental　results　show　high　performance　of　our　method.