关键词: 膜系统, 粒子群优化, 生物医学文本, 特征选择, 条件随机域

Abstract:

In order to delete redundant features in conditional random fields to recognize the gene name from literature, a novel particle swarm optimizer based on the membrane system for feature selection is proposed. In this new algorithm, the particle swarm optimizer is assigned to all sub-regions as sub-algorithms using hierarchy and message mechanism of the membrane system. Based on the structure of the membrane system, the original particle swarm optimizer is disassembled into two parts, the local optimizer and the global optimizer. The local optimizer is assigned to the all outer regions to search for the local best solution and the global optimizer is assigned to the innermost region to search for the global best solution. All outer regions send its best solution to its adjacent inner region and send its worst solution to its adjacent outer region, and the innermost region only sends its worst solution to its adjacent region. When the communication between regions stops in a specific duration or iteration reaches limit times, the iteration is stopped and gets the best solution in the innermost region as the output of the algorithm. We use the maximum log likelihood estimation function of conditional random fields as the objective function, calculate weights of all feature functions by the membrane particle swarm optimizer, and delete those feature functions with a smaller weight than a specific value. Experiment results show that selecting feature functions in conditional random fields by the algorithm we proposed to recognize the gene name from literature can reduce interference produced by redundant features and improve the accuracy of conditional random fields.

Key words: membrane system, particle swarm optimizer, biomedical literature, feature selection, conditional random fields