Study on Affect of Part-of-Speech on the Performance of Chinese Named Entity Recognition Based on the Conditional Random Fields

Taking advantage of the ability to use arbitrary features as input in CRFs, the action of POS used in the task of Chinese personal name recognition was discussed based on the Conditional Random Fields on the character level. According to the possible expressions as the features in the task and the same feature template of CRFs, multiple experiments of Chinese personal name recognition was token by sequence labeling on common corpus, which were done in similar experiment environment with multiple applications of POS features such as non-POS, POS of first level, POS of second level and POS of every level combined word borders. By comparing and analyzing the results of the experiments, the innovative usage of the combination of second level POS and word borders was obtained from the best effect in the system performance and the recognition of Chinese named entities.