A hybrid PSO-Viterbi algorithm for HMMs parameters weighting in Part-of-Speech tagging.

Shichang Sun,Hongfei Lin,Hongbo Liu
DOI: https://doi.org/10.1109/SoCPaR.2011.6089149
2011-01-01
Abstract:We propose a new approach to re-optimize Hidden Markov Models (HMMs) using Evolutionary Computation methods. The hybrid algorithm iterates in the neighborhood of original HMMS parameters with a fitness function that evaluates the solution of sequence recognition by knowledge as well as by likelihood. Experiments on POS tagging show that the parameters weighted system outperforms the baseline of the original model. Further improvement is to be achieved by combining the statistical models with more knowledge. © 2011 IEEE.
What problem does this paper attempt to address?