Chinese semantic dependency analysis

Abstract

The first and overwhelmingly major challenge of the Semantic Web is annotating semantic information in text. Semantic analysis is often used to combat this problem by automatically creating the semantic metadata that is needed. However, semantic analysis has been proven difficult to get ideal results, because of two controversial problems; semantic scheme and classification. This chapter presents an answer to these two problems. For semantic scheme, semantic dependency is chosen and for classification a number of machine learning approaches are examined and compared. Semantic dependency is chosen as it gives a deeper structure and better describes the richness of semantics in natural language. The classification approaches encompass standard machine learning algorithms, such as Naive Bayes, Decision Tree and Maximum Entropy, as well as multiple classification and rule-based correction approaches. The best results receive a state-of-the-art accuracy of 85.1%. In addition, an integrated system called SEEN (Semantic dEpendency parsEr for chiNese) is introduced, which combines research presented in this chapter as well as segmentation, part-of-speech, and syntactic parsing modules that are freely available from other researchers.