Construction of Khmer entity annotation corpus based on constrained conditional random fields

It is hard to solve the problem of considerable segmentation error caused by out-of-vocabulary words, so for segmentation task based on Constrained Conditional Random Fields Model, the Precision is relatively low. To solve this problem, Constrained Conditional Random Fields Model, which exploits constraints on the Khmer word segmentation and named entity recognition, is proposed in this paper. With a series of segmentation, POS tagging and NER tasks, the Khmer entity annotation corpus which contains more entities is constructed. Several groups of experiments are set up in each link of construction entity annotation corpus. These experimental data have proven that the proposed method is feasible.