An Embedded Co-AdaBoost and Its Application in Classification of Software Document Relation

To enhance classification performance by making use of easily available unlabelled data to overcome the scarcity of labelled data, this paper proposes an Embedded Co-Adaboost algorithm that integrates multi-view learning into the Adaboost learning framework and at the same time leverages the advantages of Co-training algorithm for performance enhancement. Experimental results demonstrate the effectiveness of the proposed algorithm in terms of the convergence rate, the accuracy, and the steady performance as compared to the original AdaBoost algorithm, without relying on redundant and sufficient feature sets. As a algorithm application in software engineering, the Embedded Co-AdaBoost has been applied to the classification of software document relations to improve the quality of the architecture design documents and the reusability of design knowledge.