Chinese Chemical Named Entity Recognition and Translation Method Based on Rules

Abstract

Aiming at the problem with the recognition of the Chinese chemical named entity (CCNER), such as a long cost for training time, the result relevant to the corpus, and manual corpus annotation. In addition, the noisy words are introduced by the procedure of the Chinese segmentation, which lower the precision of the CCNER. Therefore, we proposed a novel method for CCNER and translation based on rules. First, we employed the Backus-Naur forms (BNF) to represent the rules. Then, we extracted the constituent parts of the BNFs from existing resource semi-automatically. Moreover, we utilized an accumulation method to generate the candidates. Finally, we applied a post-processing for the CCNER and a translation method by the combination the translated texts of the constituent parts using lexicon resources. Experiments on forty abstracts of Chinese's patents indicated: the precision is 78.8%, and the F1 is 78.5%, verifying the effectiveness of the method.