Word semantic similarity research based on latent relationships

Word similarity plays an important role on fields of machine translation, semantic disambiguation, information retrieval and others. Singular value decomposition (SVD) is proposed to measure the Chinese words similarity so as to compensate for the data sparseness by vector space model (VSM). Firstly, the thesaurus is used to build the generation templates which represent the relationships between words. Word similarity scores are gotten by calculating the angle cosine between vectors. Experimental results that our accuracy is improved by 5% than traditional VSM.