Interactive Genetic Algorithm With Behavior-Based Preference Perception And Its Application

Interactive genetic algorithm, by incorporating the human emotion with the evolutionary process of the traditional genetic algorithm, is powerful to solve those optimization problems with qualitative performance indicators. However, the existing IGA usually requires users to assign fitness directly, which severely limits the exploration of IGA in solving complicated practical problems. To overcome such a critical problem, we present an IGA with a new evaluation mode, i.e., an implicit evaluation just based on the interactive behaviors of the user, and apply it to personalized search in E-commerce. The following three main issues are focused.(1) A CP-nets assisted IGA is proposed. The explicit evaluation mode of IGA often brings user fatigue, which greatly limits the performance of IGA in exploration. To deal with such drawbacks, a novel IGA with an implicit evaluation mode is proposed here based on the interactive actions performed by the user and the CP-nets. The model of those possible actions is first built and then the CP-nets adopted to approximate to the preference of the user are constructed according to few interactive actions. Then, the CP-nets model is adopted to estimate the assignments of those individuals not evaluated by the user, and the evolution process is successfully conducted based on the estimated fitness to assist the user finding his/her interested solution as early as possible. The proposed algorithm is applied to a personalized search for psychology books, and the results demonstrate the effectiveness of the algorithm.(2) In IGA, the user’s preference and evaluations are often uncertain and changing along with the evolution. In the evaluation estimation process by using preference model, the estimation will be greatly deviate from the real evaluations assigned by the user, therefore, the evaluation uncertainty is further considered here. First, a Gaussian function is designed to describe the user’s evaluation uncertainty according to his/her interactive actions. Then, the CP-nets with such uncertainties are developed and applied to estimate the fitness of those individuals not evaluated by the user. The proposed algorithm is applied to the personalized search of psychology books system, and compared with the existing typical algorithms. The experimental results show its advantages in reducing the user evaluation uncertainty, alleviating user fatigue and improving the efficiency of search.(3) Based on MFC and SQL Server 2005, we develop a system of personalized search of psychology books. With SQL Server 2005, we first realize the date storage of psychology books and the implementation of preference network. Then, we apply MFC to build system framework, read data and display corresponding psychology books. In order to facilitate the implementation and modification of the algorithm, we design the classes of evolution. The system not only provides the experimental platform for our methods, but also present a new way for the book searching.