Research On Algorithms For Link-Based Classification

From single relation to multiple relations, it becomes a great step for data mining approaches. Simultaneously, a key challenge for data mining is tackling the problem of mining richly structured heterogeneous datasets. Especially, the objects of richly structured datasets always have some relationship between each other, the relationships are called link. Links among the objects may show certain patterns. Link mining is a newly emerging multiple relation research area. It concentrates on mining links between objects.This thesis concentrates on classification of link mining, in other words link-based classification.First of all, the thesis proposes an approach about link-based classification, which combines the link and the content information. This approach uses the information from the links and the attributes of the objects in the dataset. Combine the two parts information by probability, the approach is as same as Bayesian approach.The next step, the thesis proposes an approach to refine the initial dataset. The generic link-based dataset we consider is essentially a directed graph, which the nodes are objects and edges are links between objects. Web pages are very special. Multiply web pages'PR and frequency as a weight of the edges and refine the edges according to their weight.According to the approaches above, techniques are evaluated on real-world database and the anticipated results are realized.