Non-negative Matrix Factorization for Filtering Chinese Document

There are two nasty classical problems of synonymy and polysemy in the filtering systems of Chinese documents. To deal with these two problems, we would ideally like to represent documents not by words, but by the semantic relations between words. Non-negative matrix factorization (NMF) is applied to dimensionality reduction of the words space. NMF is distinguished from the latent semantic indexing (LSI) by its non-negativity constraints. These constraints lead to a parts-based representation because they allow only additive, not subtractive, combinations. Also, NMF computation is based on the simple iterative algorithm; it is therefore advantageous for applications involving large sparse matrices. The experimental results show that, comparing with LSI, NMF method not only improves filtering precision markedly, but also has the merits of fast computing speed and less memory occupancy.