Generation of simple structured information retrieval functions by genetic algorithm without stagnation

作者：

Highlights：

• The study improves the genetic algorithm, which generates IR ranking functions.

• We solve the problems of evolutionary stagnation and overfitting of the algorithm.

• Several structural metrics are analyzed to detect evolutionary stagnation.

• Several regularizers are analyzed to control structural complexity of functions.

• Generated IR ranking functions outperforms the ones derived theoretically.

摘要

•The study improves the genetic algorithm, which generates IR ranking functions.•We solve the problems of evolutionary stagnation and overfitting of the algorithm.•Several structural metrics are analyzed to detect evolutionary stagnation.•Several regularizers are analyzed to control structural complexity of functions.•Generated IR ranking functions outperforms the ones derived theoretically.

论文关键词：Information retrieval,Genetic programming,Ranking function,Evolutionary stagnation,Overfitting

论文评审过程：Received 15 May 2016, Revised 5 May 2017, Accepted 7 May 2017, Available online 8 May 2017, Version of Record 23 May 2017.

论文官网地址：https://doi.org/10.1016/j.eswa.2017.05.019