Search

Standard supervised learning approaches apply the decision model to the new document to be classified in a “context-free” manner. To exploit the potential information in the relationship between the items, there are link-based methods which model the relationship as a graph and exploit further information for classification from the neighbourhood. These link-based methods learn from the collective information of the whole neighbourhood. Rather than taking the whole neighbourhood into consideration, the author proposes a model to calculate link certainty to evaluate the influence of individual link into the classification process. The link certainty is later combined with the content-only model to form a complex model named individual-link model. This new approach reduces the effects of loss of information. The author systematically evaluates new model on standard data sets to compare with traditional content-only methods and other link-based method to see the improvement of the new model. Besides the promising results when working on the graph of documents, the individual link model can be applied on the general graph of objects in the future.

Association Link Network (ALN) is proposed to establish association relations among Web resources, aiming at extending the hyperlink-based Web to an association-rich network and effectively supporting Web intelligence activities, such as Web-based learning. However, it is difficult to build the ALN one-off by direct computing since the huge number and quickly increasing learning resources on the Web. Thus, how to rapidly and accurately acquire the association relations between the new coming and existing learning resources has become a challenge in the incrementally building process of ALN. In this paper, a new algorithm is developed for incrementally updating ALN to cater for the dynamic management of learning resources increasing with time.

We propose a novel technique to track the motion of the left ventricular myocardium on a sequence of transthoracic two-dimensional echocardiogram. Although the proposed method is generally based on the traditional correlation based motion tracking, the tracking accuracy and stability are improved by the introduction of new objective function. In our method, the objective function is defined from the model where multiple Regions of Interest (ROIs) are connected to each other with elastic links. The elastic links reflect both elastic and shape properties of myocardium. The proposed method is implemented and evaluated using ten clinical subjects. The experimental results show that the tracking accuracy of the proposed method is higher than that of the conventional method.