Towards a SVM-struct Based Active Learning Algorithm for Least Cost Semantic Annotation

The recent growing interests in semantic Web trigger the requirements of annotating various information objects (e.g., documents) on the Web. The main drawback of the existing methods is that they usually require many manually annotated training examples as inputs. This paper proposes a SVM-struct based active learning algorithm for automatic semantic annotation. In particular, the proposed algorithm is underpinned by a novel uncertainty minimization method which can identify the most discriminative examples for re-training so as to reduce the manual annotation cost. Our initial experiments show that the proposed method can achieve comparable annotation performance while requiring a much smaller training set.