Parameter Estimation of Conditional Random Fields Model By Improved Particle Swarm Optimizer

A new parameter estimation algorithm based on improved particle swarm optimizer is proposed to improve the precision and recall rate of conditional random fields model. Aggregation degree of particle swarm is utilized to control particle swarm optimizer’s early local convergence, the relative change ratio of log-likelihood between iterations is employed to end its iterations, and the inertia factor and learning factor are set as linear variables to control the searching scope. We evaluate our method on GENIA, GENETAG and private library. The experiment results prove our method outperforms traditional parameter estimation method on precision and recall.