Abstract: We improve the generalized vector space model, and present an original meaning space based text similarity computing method with the “HowNet” sememe. The method converts a feature into a vector in a sememe space with TFIDF weight, and calculates text similarity by the cosine of the angle between different sememe vectors. We eventually perform text clustering and comparative experiments. Experimental results show that the method can well solve the problem of semantic drift of public opinion analysis, and its effect can be significantly improved.

Key words: similarity, HowNet, internet public opinion