Let us mention some problems which this research tries to solve. When discovering the dependencies among features of encoding texts or words into numerical vectors, the Bayesian networks was proposed as the approaches to the text mining tasks, but the complicated analysis is required for using them [1]. The assumption that features are independent of each other causes the requirement of many features for the robustness in encoding so. Since each feature has the very weak coverage, we cannot avoid the sparse distribution of each numerical vector where zero values are dominant with more than 95% [3]. Therefore, this research is intended to solve the problems by considering the feature similarity as well as the feature value similarity.