Behavior-derived Interactive Genetic Algorithm With Uncertain Preferences And Its Application

Interactive genetic algorithms(IGAs)are powerful to solve those optimization problems with qualitative indicators.However,the existing IGAs usually require users to assign fitness directly,which severely limits the explorations of IGAs in solving complicated practical problems.Accordingly,IGAs with a new evaluation mode is pursued,i.e.,an implicit evaluation just based on the interactive behaviors of the user.In this paper,interactive genetic algorithms based on the user's interactions and similar groups' information are proposed,which emphasize the essential uncertainties of the user's interactions and the preferences by using probabilistic conditional preference nets.The following three main issues are focused:(1)Personal interactions derived interactive genetic algorithm with probabilistic conditional preference nets(PCP-nets)is studied.This paper first analyzes the credibility of the user's interactions and further define a reliability function to reflect the relationships among the uncertain interactions and the preferences.The method of constructing the PCP-nets based on the historical information and the current user's interactions is then followed to quantitatively describe the current user's preference.The fitness of individuals can then be estimated by using the PCP-nets,and the corresponding genetic evolutions are performed.The proposed algorithm is applied to a personalized book search and its superiorities in exploration and feasibility are experimentally demonstrated.(2)Further study an interactive genetic algorithm with preference uncertainties based on group's interactions.According to the collaborative filtering recommendation,which is based on similar users to guide an individual user,this paper first presents the metric for measuring the similarities among the users to identify the similar group based on the user's initial input keywords.Then the PCP-nets of the group are constructed based on the methods presented in(1).The users' preferences of the similar group are incorporated with the PCP-nets of an individual user to refine and update the current user's preference expression.The models of the current user and the social group will be continuously updated in the evolution to timely track the user's changed preference.At last,the genetic operators are operated to generate new individuals and the fitness of the individuals are estimated based on the PCP-nets.The proposed algorithm is verified by being applied to a personalized book search.(3)The prototype platform based on the proposed algorithms for personalized search of psychology books is designed.According to the frameworks of the presented algorithms in(1)and(2),this paper designs a prototype of the application platform.First,the module functions,operation processes and platform involved technics are determined.Then,according to the proposed algorithm,the system is developed and improved.The experimental simulations of the prototype system are briefly introduced.