The　task　of　topic　detection　becomes　the　hotspot　research　direction　in　the　field　of　natural　language　processing　in　recent　years.　The　cyberspace　public　opinion　has　given　the　significant　impact　on　the　large　numbers　of　internet　users.　This　makes　the　effective　public　opinion　detection　and　tracking　become　very　important.　We　select　the　effective　features　in　the　story　document　and　the　vector　center　model　is　taken　to　represent　the　text　document.　The　algorithm　of　distance　based　clustering　is　carried　out　for　public　opinion　detection.　It　identifies　the　emergence　of　new　events　and　also　merges　the　story　to　the　corresponding　cluster.　Finally,　we　give　the　performance　evaluation　by　the　F-Value　and　entropy　value.　The　system　achieves　the　performance　of　76%　F　value　on　the　test　set.　The　technique　of　topics　detection　can　monitor　the　information　sources　in　various　languages,　and　it　will　provide　the　efficient　guidance　for　judging　the　hot　spots　on　the　web.　©　Springer-Verlag　Berlin　Heidelberg　2011.