Improving Chinese Named Entity Recognition with Semantic Information of Character Multi-position Representation

摘要

Named entity recognition is an important basic task for information extraction and construction of knowledge graph, but the recognition rate needs to be further improved, especially in Chinese. There are two main implementations based on the sequence tagging method. Among them, the character-based method lacks the support of word information, and the word-based method is affected by the word segmentation efficiency. In order to comprehensively utilize the information of characters and words and to reflect the semantic information that changes due to different combinations of characters and words in a sentence. We designed a tagging scheme based on word segmentation and dictionaries. Then, neural networks are used for learning multi-position feature vectors and character-based tagging task. Experiments with MSRA datasets show that this method outperforms word-based and character-based baselines and achieves a higher recall rate compared to other methods.