The　unique　language　characteristic　of　short　texts　has　made　the　performance　of　traditional　natural　language　processing　methods　degradation,　or　even　unavailable.　Exact　representation　and　calculation　of　the　similarity　between　short　texts　are　great　helpful　to　content　based　clustering.　That　this　paper　treated　each　short　text　as　a　composition　of　characters,　numbers　and　punctuation,　and　a　similarity　measure　based　on　string　similarity　was　proposed.　Then　a　public　opinion　hotspot　detection　and　analysis　system　based　on　short　text　hierarchical　clustering　was　built.　This　method　calculated　the　similarity　directly　which　skipped　the　feature　extraction　and　representation　processing　of　short　text,　to　a　certain　extent,　and　avoided　using　the　sparse　feature　vectors.　Experimental　results　show　the　effectiveness　of　the　proposed　method.