With　the　increasingly　rapid　development　of　Internet　applications,　information　is　propagated　in　many　and　diverse　ways.　Real-time　tracking　of　hot　events　presented　in　media　information　streams　involves　core　technologies　in　multiple　fields　to　monitor　and/or　control　information　sources,　and　alert　and　track　new　themes　as　they　appear.　This　has　become　a　popular　field　of　research.　In　this　study　we　used　word　frequency　selection,　chi-square　statistics,　information　gains　and　mutual　information　to　extract　and　calculate　the　impact　of　feature　items,　and　adopted　an　effective　classification　method　to　track　events　with　given　themes.　In　the　test　set,　the　value　of　the　F-Measure　at　optimal　performance　was　86.2%.　The　topic-tracking　technology　has　a　widespread　promise　for　application　in　such　areas　as　information　security.　©　2014　ICIC　International.