People  are  generating  more  and  more  short texts.  There  is  an  urgent  demand  to  classify short  texts  into  different  domains.  Due  to  the shortness  and  sparseness  of  short  texts,  con-ventional  methods  based  on  Vector  Space Model  (VSM)  have  limitations.  To  tackle  the data scarcity problem, we propose a new mod-

el to directly measure the correlation between a  short  text  instance  and  a  domain  instead  of representing short texts as vectors of weights. We  firstly  draw  domain  knowledge  for  each user-defined  domain  using  an  external  corpus 

of longer documents. Secondly, the correlation is  calculated  by  measuring  the  proportion  of the  overlapping  part  of  the  instance  and  the domain  knowledge.  Finally,  if  the  correlation is greater than a threshold, the instance will be classified  into  the  domain.  Experimental  results show that the classifier based on the proposed  model  outperforms  the  state-of-the-art baselines based on VSM.