Cost-sensitive learning addresses the issue of classification in the presence of varying costs associated with different types of misclassification. In this paper, we present a method for solving multi-class cost-sensitive learning problems using any binary classification algorithm. This algorithm is derived using three key ideas: 1) iterative weighting; 2) expanding data space; and 3) gradient boosting with stochastic ensembles. We establish some theoretical guarantees concerning the performance of this method. In particular, we show that a certain variant possesses the boosting property, given a form of weak learning assumption on the component binary classifier. We also empirically evaluate the performance of the proposed method using benchmark data sets and verify that our method generally achieves better results than representative methods for cost-sensitive learning, in terms of predictive performance (cost minimization) and, in many cases, computational efficiency.

Abstract. In this paper, we propose a new method for learning to rank. ‘Ranking SVM ’ is a method for performing the task. It formulizes the problem as that of binary classification on instance pairs and performs the classification by means of Support Vector Machines (SVM). In Ranking SVM, the losses for incorrect classifications of instance pairs between different rank pairs are defined as the same. We note that in many applications such as information retrieval the negative effects of making errors between higher ranks and lower ranks are larger than making errors among lower ranks. Therefore, it is natural to bring in the idea of cost-sensitive learning to learning to rank, or more precisely, to set up different losses for misclassification of instance pairs between different rank pairs. Given a cost-sensitive loss function we can construct a Ranking SVM model on the basis of the loss function. Simulation results show that our method works better than Ranking SVM in practical settings of ranking. Experimental results also indicate that our method can outperform existing methods including Ranking SVM on real information retrieval tasks such as document search and definition search. 1

In this paper, we perform an empirical study of the impact of noise on cost-sensitive (CS) learning, through observations on how a CS learner reacts to the mislabeled training examples in terms of misclassification cost and classification accuracy. Our empirical results and theoretical analysis indicate that mislabeled training examples can raise serious concerns for cost-sensitive classification, especially when misclassifying some classes becomes extremely expensive. Compared to general inductive learning, the problem of noise handling and data cleansing is more crucial, and should be carefully investigated to ensure the success of CS learning. 1

1 Introduction Social Network Analysis has long been an important field ofresearch in the social sciences. Recent developments such as the proliferation of the online communities and communi-cation networks has shown the need for scalable techniques for extracting, analyzing and mining large real-world socialnetworks. These networks consist of entities linked by various relations. Predictive models which exploit both the at-tributes of entities and relations and their relational patterns are important for identifying key actors and important (oranomalous) links.