User profile systems have been studied and used mainly in information filtering. But they have some problems in managing profiles. Some systems need direct human aids and can not deal with various interest domains of a user. All systems have the scope problem when they are applied in information retrieval since queries are more specific in information retrieval than in information filtering.
In this thesis, we propose a new user profile system that solves the problems previous researches have. In the proposed profile system based on multi-population genetic algorithm, a population models an interest domain and many populations cover all interest domains of a user. Each population varies in its size reflecting the strength of an interest domain. Since a population evolves by genetic algorithm, chromosomes that match the interest domain survive. Therefore the population describes the interest domain more precisely (specialization) and adapts itself to the change of the interest domain (adaptation). Genetic operators, crossover and mutation, create and suggest new relevant interest domains from the current interest domain (exploration).
We experiment on query expansion with the proposed user profile system. The expanded queries show better performance than original queries for both cases: the interest domain changes and does not. The results also show that the performance of the queries expanded by the proposed profile system exceeds that of the queries based on the single population genetic algorithm.