An improved-Viterbi Based Chinese Sequence Labeling Enhancement Method
Quan Qi, Bo Wang, Kaihui Mu · 2020
The performance of supervised-learning based Chinese sequence labeling(CSL) is usually limited by the quality of the training corpus. In order to reduce the lack of labeling corpus and make full use of the context information of corpus, For improving the performance of CSL, this paper gives a method that re-decode the sequence using the context labeling results. For re-decoding, we propose an improved-Viterbi algorithm that can utilize the context sequence's labeling results and the order of key information in the results. Experimental results demonstrate that our method can effectively optimize the labeling results for some kinds of Chinese plain text, especially for some domain-specific tasks such as drama scene labeling and resume key information extraction.