Abstract:

The analysis of hyperlink structure on the Web has been employed for detecting high quality documents. In approaches such as PageRank, the Web graph is modelled as a Markov chain and the quality of a document corresponds to the probability of visiting it during a random walk. However, it is not always straightforward to consider the Web graph as a Markov chain. For example, PageRank introduces a universal document, in order to transform the Web graph to a Markov chain. In this paper, we present the Absorbing Model, a hyperlink analysis model based on absorbing Markov chains, where the Web graph is transformed by adding one absorbing state for each document. We provide an authority-oriented and a utility-oriented interpretation of the Absorbing Model, and show that the latter is more effective than the authority-oriented model. Thus, we believe that it is quite important to make this distinction between the two types of hyperlink analysis. In addition, we provide evidence that support the investigation of more elaborate hyperlink analysis methods on a query-by-query basis.