DRANK+: A Directory Based Pagerank Prediction Method for Fast Pagerank Convergence

As the increasing of importance in search engines, Internet users change their behavior browsing the Internet little by little. In recent years, most part of search engines use link analysis algorithms to measure the importance of web pages. They employ the conventional flat web graph constructed by web pages and link relation of web pages to measure the relative importance of web pages. The most famous link analysis algorithm is PageRank algorithm. However, previous researches in recent years have found that there exists an inherent bias against newly created pages in PageRank. For this issue, some researchers have proposed a new ranking algorithm called Page Quality to solve it. Page Quality utilizes the difference of PageRank at continuous time stages to predict a reasonable importance score of pages at next time stage. We also have proposed a new ranking algorithm called DRank to solve the same issue last year. It utilizes the intrinsic characteristic of hierarchical structure embedded in URL and the cluster phenomenon of PageRank in a directory to predict the possible importance of pages in the future and to diminish the inherent bias of search engines to new pages. In this paper, we modify the original DRank algorithm and propose a new ranking algorithm called MDRank to complement the weaker part of DRank which could fail while the number of pages in directory is not enough. The integrated algorithm is called DRank+, which combines DRank and MDRank. In our experiments, the modified DRank algorithm obtains more accuracy in predicting the importance score of pages at next time stage than the original DRank algorithm. Furthermore, MDRank can also obtain more accuracy in predicting the future importance score of pages while the number of pages in directories is few. It also interprets that DRank+ not only alleviates the bias of newly created pages successfully but also reaches more accuracy than Page Quality and original DRank in predicting the importance of newly created pages.