Latent semantics for hotspot information clustering

The growing interest about research on hotspot information clustering and discovery can be attributed to our need to harness information on the internet. Currently, there are many similarity calculation methods with their own strength and weakness. In this paper, HowNet, LSA (Latent Semantic Analysis) and LDA (Latent Dirichlet Allocation) are introduced, their strengths and weaknesses evaluated and a revised similarity calculation method is presented. The experimental results show that LSA is advantageous in handling shorter texts when compared with HowNet, which consumes significantly more time to perform the same task. The efficiency of the LSA is over 100 times that of HowNet. It is also found that LDA is more suited to processing longer texts than shorter texts. A prototype is developed to demonstrate the plausibility of our revised similarity method. It presents the evolving of the hot topics within a short period, which will be a great help for hotspot information prediction. © 2014 Binary Information Press