Using smoothing techniques to improve the performance of Hidden Markov's Model
Sweatha Boodidhi · Digital Scholarship - UNLV (University of Nevada Reno) · 2020
The result of training a HMM using supervised training is estimated probabilities for emissions and transitions. There are two difficulties with this approach Firstly, sparse training data causes poor probability estimates. Secondly, unseen probabilities have emission probability of zero. In this thesis, we report on different smoothing techniques and their implementations. We further report on our experimental results using standard precision and recall for various smoothing techniques.