Research On Chinese Entity Linking

As a bridge between information extraction and knowledge base construction, Entity Linking(EL) is becoming increasingly important in the field of natural language processing. As of Chinese entity linking, lack of benchmark corporamakes it difficult to compare the advantage and disadvantage between different EL methods. This paper starts with constructing a Chinese EL corpus, and then generates the candidate set via query expansion, finally proposestwo EL improvements, one is based on similarity computation, and the other is based on supervised learning Our researchcontent includes the next three aspects:1)Building a Chinese entity linking corpus based on ACE 2005 Chinese corpus via automatic annotation and manual annotationas well as a corresponding knowledge base based on Chinese Wikipedia; making statistics and analysis on the corpus; implementinga baseline to pinpoint the difficultiesin Chinese entity linking.2) Designing a method for candidate generation via query expansion on the basis of traditional method, incorporating the characteristics of the corpus to achieve high recall of correct answersin a small scale of candidate set; implementing a method of entity linking based on similarity computation and comparing the impact of different similarity scores.3) Exploring the method of entity linking based on supervised learning and linguistic features on Chinese entity linking. Various mention features, phonetic features, context features are combined via statistical machine learning paradigm to realize the Chinese entity linking.Analysis and experimentation on the Chinese entity linking corpus show that first, we can address the problem of candidate generation via query expansion to achieve a higher recall. Second, in order to overcome the disadvantage that similarity computation fails to fuse various linguistic features, a supervised learning for entity linking is proposed and achieve better performance than similarity computation.