Research On Semantic Annotation Based On Ontology

The semantic annotation is an important challenge faced with the development of semantic Web. The semantic Web can provide semantic information which could be understood by computers on the Internet, it can also realize intelligent interaction between computer and computer, computer and humans.however, there are still exists some problems in the field of semantic annotation, such as the semantic annotation system can't adopt different annotation methods according to the different characteristics of the data; and the existing methods can't solve all the annotation problems. The rule-based and classification model-based methods can only label mutual independence data information but not interdependent relationship. The first-order linear Conditional Random Fields (CRFs) not only can label linear dependent relationship between data information, but also can model Markov sequence data, that is each labeled state variable only depending on the previous state variable; however, the relationship between labeling information is not a simple linear dependence, and for hierarchical dependence and adjacent dependence relationship, the first-order linear CRFs can't label them.For the annotation problems of hierarchical dependence and adjacent dependence, the paper propose the following solutions:1. Aiming at the first-order linear CRFs couldn't label hierarchical dependence relationship, an improved CRFs model--Tree Conditional Random Fields (TCRFs) is proposed, which can improve the labeling accuracy and recall rate effectively under the condition that the model training (parameter estimation) is not very complexity.2. The TCRFs can improve the labeling accuracy and recall rate significantly when the document is hierarchical dependence relationship; however, for more complex adjacency relationship and long-distance dependency, the TCRFs can't label effectively. Based on the above, Chain Conditional Random Fields (CCRFs) labeling method is proposed. It can label the various and complex long-distance dependence between information effectively by adding Skip-Chain into TCRFs.