Compared to the traditional semi-supervised based anaphora resolution methods

Laplacian SVM(Support Vector Machine) can efficiently explore the similarity and correlations between labeled and unlabeled samples for deriving more accurate classification model.However

traditional Laplacian SVM simply uses Euclidean distance to calculate the distance between two samples

which may result that two samples from different classes may have false high similarity.To address the problem of insufficient Chinese annotated corpus

a data-driven based method is proposed to learn the optimal distance metric.The proposed method takes similarity constraints between sample-pairs into consideration and introduces the Fisher discrimination criterion

so that the similarities of in-class samples are higher than those of between-class samples

and the discriminant features are highlighted in the new metric space.Furthermore

the proposed metric-optimized method is generalized from linear to nonlinear space by the use of kernel

so that it can be used for non-linear classification.Compared with the classical supervised method and other four traditional semi-supervised methods on the ACE2005 Chinese corpus

the proposed method

both the linear form and kernel form

achieves the comparatively better or best performance

with fewer labeled samples.