基于语义本体的特征降维方法分析-analysis of feature dimensionality reduction method based on semantic ontology.docxVIP

Research of Dimension Reduction Based on Ontology ABSTRACTDimension reduction is an important component in data mining and machine learning,especially automatic text categorization. With the development of semantic web and ontology, semantic-based dimension reduction has received a lot of attention. Many studies have showed that statistic-based dimension reduction methods, like Information Gain (IG), are very effective in document categorization. However, compared with semantic-based dimension reduction algorithms, traditional algorithms suffer from major drawbacks, one major drawback is that they tend to use all the words as features which lead to a huge dimension of the feature space; the other major drawback is that they can’t capture the semantic information that underlies the lexical words, it makes the features selected have no semantic relationship with class label.To use the semantic information between the terms and labels, a novel algorithm is presented to reduce the dimension of feature space based on ontology, named OBDR (Ontology-based Dimension Reduction).First, a good biomedical concept set can be obtained by an ontology-based entity extraction technique as a feature space. Then our original feature space is consisted with semantic terms.Secondly, all the features in the space will be mapped to several clusters based on their semantic sequences in the ontology. If two terms share the same grandfather or father, they will be mapped to the same cluster.At last, each cluster will be represented by only one feature. The feature will be selected by semantic sequence of the term according to the ontology.The main idea of our algorithm is using “Least-Max-Cover” strategy, trying to find the minimum size of subsets of features with most sufficient semantic information. Our experimental results show that, after dimension reduction operating, the classification model based on the new feature space has a good performance on the classification precision, recall a