Information Filtering Using Latent Semantics

摘要

We propose an information filtering system using latent semantics obtained by Singular Value Decomposition (SVD) and Independent Component Analysis (ICA). Document vectors usually have too many elements. Thus, we are obliged to spend much time applying the ICA for the document vectors. To solve this problem, the present method combines the SVD which is often used for decreasing dimension and the ICA. Before applying the ICA, we represent documents with singular vectors obtained by the SVD. We measure processing times to carry out the ICA without the SVD and the proposed method for comparison of these methods. In addition, we construct a user profile in space consisting of latent semantics obtained by the present method, and discuss accuracy of recommendation.