A Content-Based Information Retrieval Model Using Non-Negative Matrix Factorization Method

During the last decade, the demand for content-based retrieval among the mass information continues increasing. Unfortunately, the traditional key words matching method has been useless. Therefore, how to retrieve the expected results in semantic level has become a tough challenge for information retrieval. Theoretically, many methods have been proposed to solve it, among which latent semantic analysis (LSA) model is the most famous one. However, while conventional LSA methods, for example singular value decomposition (SVD), do make advances in this issue, they consume large memory and generate elements with negative value in some dimensions, which is meaningless. With such consideration, in this paper, we propose a content-based retrieval model using non-negative matrix factorization method, which generates a latent semantic space representation with non-negative parts as a low rank approximation to term-document matrix. Experimental results show that our approach gets a good performance in content-based retrieval.