Information filtration is a kind of classical application of large-scale network content processing. It carries out filtration operation to successively coming information, remaining those information that meets users' demands and deleting those information that falls short of users' demands. Vector space model is one of its common classical models. This thesis made research and improvement on Vector Space Model (VSM), and on this basis put forward to introducing a kind of center space vector to solve the problem of thesaurus. This method has relatively high precision rate and recall rate, and improves the integral capability of the model.