PURPOSE: A method for adjusting an order is provided to enhance an accuracy of a search by applying bayesian self-organizing feature maps for performing a real time document large group as a target of a related document in accordance with meaning similarity with a user query language combining entropy information. CONSTITUTION: A user records a wanted query language for a search(S10). A user profile constituted by currently searched keywords and a frequently used thereof is designed for reflecting a symbol of the user(S20). Entropy among subjects of the user query language, the user profile, and each web document is calculated(S30). It is judged whether data for a learning of a Kohonen neural network is sufficient(S40). If the data are sufficient, an initial connection weight value of bayesian SOM combined with the Kohonen neural network and a bayesian learning is decided through a decision of prior information being used as each parameter initial value of the network through the bayesian learning(S50). A document large group with respect to a proper document is performed in real time based on the entropy value calculated by the entropy calculation using the bayesian SOM neural network model(S70).
展开▼