A hybrid evolutionary computation approach with its application for optimizing text document clustering

• We propose a novel hybrid evolutionary computation approach for optimizing text clustering.
• GA improves the initializing strategy of QPSO and yields a preliminary optimization.
• A new position update approach is proposed to normalize the search space of particles.
• This approach enhances the performance evaluated by both fitness and F-measure.

Quantum-behaved particle swarm optimization (QPSO) is a promising global optimization algorithm inspired by concepts of quantum mechanics and particle swarm optimization (PSO). Since the particles are initialized randomly in QPSO, the blindness of initializing particles affects its capacity for complicated optimization. In this paper, we make full use of a hybrid evolutionary computation approach to resolve such an issue. In specific, the robust global search ability of genetic algorithm (GA) improves the initial strategy of particles in QPSO. What is more, the original position update approach of QPSO without the restriction of its upper bound may generate some abrupt features and cause the issue of overstepping boundary, which affects its performance for search of optimum. In this study, a new position update approach is tested to normalize the search range of particles in a proper space. Such an approach enhances its probability to find the optimal solution. Since the clustering problem can be regarded as the centers searching process by using evolutionary optimization approach, the evolutionary process of chromosomes or particles encoded by centers simulates the process of solving clustering problem. In order to testify the clustering performance of our approach, we conduct the experiments on 4 subsets of standard Reuter-21578 and 20Newsgroup datasets. Experimental results show that our method performs better than the state of art clustering algorithms in the light of the evaluations of fitness and F-measure.