Web Mining and Social Networking: Techniques and ... - tud.ttu.ee

106 5 Web Linkage Miningbetween users make the pages that are constrained by Q alone to be more heterogeneous incontent than the pages that are constrained by both U and Q.Based on a query log, two kinds of implicit link relations are defined according to twodifferent constraints:1. L I 1: d i and d j appear in the same entries constrained by U and Q in the query log. Thatis d i and d j are clicked by the same user issuing the same query;2. L I 2: d i and d j appear in the same entries constrained by the query Q only. That is d i andd j are clicked according to the same query, but the query may be issued by different users.It is clear that the constraint for L I 2 is not as strict as that for L I 1. Thus, more links of L I 2can be found than L I 1, however they may be more noisy.Similar to the implicit links, three kinds of explicit links are defined based on the hyperlinksamong the Web pages according to the following three different conditions:L E (i, j)={ 1 CondE0 Other(5.13)1. Cond E 1 : there exist hyperlinks from d j to d i (In-Link to d i from d j );2. Cond E 2 : there exist hyperlinks from d i to d j (Out- Link from d i to d j );3. Cond E 3 : either Cond E 1orCond E 2 holds.These three types of explicit links under the above conditions are denoted as LE1, LE2,LE3 respectively. In the above definitions, the inlink and out-link are distinguished because,given the target Web page, the in-link is the hyperlink created by other Web page editors whorefer to the target Web page. In contrast, the out-link is created by the editor of the sourceWeb page. They may be different when used to describe the target Web page. From the abovedefinitions, D.Shen et al. [80]said that the implicit links give the relationships between Webpages from the view of Web users. However, the explicit links reflect the relationships amongWeb pages from the view of Web-page editors.To utilize these link information for classification, D.Shen et al. [80] propose two approach.One is to predict the label of the target Web page by the labels of its neighborsthrough majority voting.This algorithm is similar to k-Nearst Neighbor (KNN). However, kis not a constant as in KNN and it is decided by the set of the neighbors of the target page.The other enhances classification performance through the links by constructing virtualdocuments. Given a document, the virtual document is constructed by borrowing some extratext from its neighbors. Although originally the concept of the virtual document is pioneeredby [101], the notion are extended by including different links. After constructing the virtualdocument through links, any traditional text classification algorithm could be employed toclassify the web pages. In [80], they take Naive Bayesian classifier and Support Vector Machineas the classifiers. The experimental results show that implicit links can improve theclassification performance obviously as compared to the explicit links based methods. Moredetails are in [80].SummaryIn this chapter, we have reported some interesting research topics in the context of web linkagemining. The principal technologies used in linkage analysis have been substantially addressedand the real applications have been covered as well. We start with the theory of co-citation andbibliographic couple, followed by two most famous linkage mining algorithms, i.e. PageRank