Abstract: Projection pursuit method is increasingly used in text categorization to solve the curse of dimensionality. Traditional projection pursuit method considers the projection index optimization as a single-objective problem rather than a multi-objective one, which will reduce the quality of the solution. To solve this problem, this paper proposes a projection pursuit mehod based on multi-objective optimization. Measures are taken like class difference and difference between the classes as two objectives of pursuit index, the projection pursuit method is extended to multi-dimensional projections, and a Chaotic Particle Swarm Optimization(CPSO) is suggested to find the optimal projection direction. Experiment on commonly used text datasets determines the optimal projection direction and dimensions, and then compares the results of different classification models. The results demonstrate that the proposed method can improve the text categorization performance effectively.
curse of dimensionality,
Chaotic Particle Swarm Optimization(CPSO) algorithm