Research On Book Weeding Expert System Based On Neural Network Ensembles

With the development of science and technology, the human being's knowledge updates day by day. As an important carrier for transmitting knowledge, books in every library rapidly increase, which push huge pressure on the library's collection. In order to improve the quality of collections and relieve the pressure, every library has to weed books periodically according to the management rules.There are so many disadvantages in the traditional books weeding method that it is essential and urgent to develop an intelligent books weeding method to improve the quality and efficiency of the book's weeding. As a modern tool, the computer is undoubtedly employed in this process.In this dissertation, an intelligent books weeding scheme which combines the expert system technology and neural network ensembles technology is put forward. And the corresponding system model is also built.The work done in this dissertation includes:1. First of all, aiming at the disadvantages of traditional book weeding method, the paper demonstrates the feasibility of applying both the neural network ensembles and expert system to the book weeding. According to the characteristic of book weeding, a neural network ensembles expert system (NNEES) model is designed to weed book intelligently. Based on the traditional expert system, the integrated module of neural network ensembles is applied in this model in order to insure the automatic acquisition of knowledge and reasoning. The main idea in this model is: by training of neural network ensembles to obtain the knowledge needed by the system and finish the primary reasoning work of the system.2. An algorithm that can determines the ensemble structure of neural network automatically is presented. During the training process, the algorithm adjusts the value of connection weights by error function, then the structures of individual neural networks according to their contribution to the whole ensembles, which can gain a better neural network ensemble structure and better knowledge with good generalization ability. The knowledge enhanced the reasoning ability of the system ultimately.3. At last, an intelligent books weeding tool—the book weeding expert system prototype based on neural network ensembles is developed.