Link Analysis Ranking -- Help Page


Welcome! This is a Web site with the experimental results for my Ph.D. thesis "Link Analysis Ranking". The purpose of the site is to present the rankings of  10 different Link Analysis Ranking algorithms on 34 different queries, and various statistics about their behavior.  This help page will help you navigate through the site. Navigation assumes a prior familiarity with the material presented in my thesis.

There are four types of pages in the site. All pages have a common layout. The top part of the page consists of a header that has the title of the page, and, underneath, links for each algorithm. On the left-hand area of the page there is a list with links for the 34 different queries. The middle area is where the main information is presented, while the right-hand area contains additional, or side information. We will now describe each page type.

Entry Page: This is the first page you encounter. You can always return to this page by clicking on "Queries" on the left-hand area of the page.  When entering the Entry Page, the middle area displays a Distance Table with the average distances between all algorithms, where the average is taken over all queries. The right-hand area contains a Performance Table with the average relevance ratios for all algorithms, where the average ratios are taken over all queries.

Clicking on an algorithm link will take you to an Algorithm Page for the corresponding algorithm. Clicking on a query link will take you to a Query Page for the corresponding query.

Algorithm Page: This page displays aggregate statistics about an algorithm. Algorithm Pages can only be reached from the Entry Page. In an algorithm page, the middle area contains a Distance Table,  with the distances of the specific algorithm to all other algorithms for each query.  The right-hand area contains a Performance Table, with the relevance ratios of the algorithm for each query. Distance and Performance Tables were supposed to be aligned, but unfortunately it does not always work.

Back to top

Query Page: This page is the entry point for accessing the results of the algorithms for a specific query. You can access  this page by clicking to the query name on the left-hand area. The main area contains a Distance Table with the distances between the algorithms for this query. The left-hand area contains a Performance Table, with the relevance ratios for all algorithms for this query. At the bottom of the table, the number of users that rated the documents is shown. It also contains a Matrix Plot of the graph for this query.

Clicking on an algorithm link from this page will take you to a Results Page with the top 100 results of this algorithm for this query.

Back to top

Results Page: This page presents the results of a specific algorithm, for a specific query. The middle area contains the top-100 results of the algorithm. The title and URL of the pages have been truncated when necessary, so that they fit into the  allocated space. Clicking on the title of each result opens a new browser window with the corresponding page.  Beneath each result there is some relevance information is presented for each page. Namely, we present the fraction of users that have voted the page as Highly Relevant (HR), Relevant (R), Non-Relevant (NR), or Dont Know (DN). Depending on the relation of these fractions, we label the page as Highly Relevant (HR), Relevant (R) or Non-Relevant (NR) as described in the thesis. The label is shown in boldface. The pid and qid numbers are for our own debugging purposes.

On the right-hand area we present Performance Table for this query. The relevance ratios of the specific algorithm are printed in red. The number of users that rated the documents are shown at the bottom. There is also a Matrix Plot with the plot  for this query, that shows the position of the top-10 results of the algorithm. Below that, there is Distance Table, with the distance of the algorithm to the remaining algorithms.

When at this page, clicking on an algorithm link will take you to a Results Page for this algorithm, on this query.

Back to top

We now describe the various tables that we present.

Distance Table: A distance table presents the pairwise distances between all algorithms. We consider four different distance/similarity measures described in the thesis: Intersection over top-10 (I), Weighted Intersection over top-10 (WI), Manhattan geometric distance (d), rank distance (R). The measure presented in printed at the bottom of the table. When a page that contains a distance table is loaded, the default is the Intersection measure. In the top-right cell of the table, there are links to tables with the pairwise distances for all the remaining measures.

Back to top

Performance Table: A performance table contains relevance ratios for each algorithm. There are two types of relevance ratios, Labeled and Non-Labeled, as described in the thesis. The default is to present the Non-Labeled statistics. In the top-right cell of the table there is a link to the other option (L is for Labeled, NL for Non-Labeled). Note, that when the table presents Non-Labeled data, the link will be L, and when the table contains Labeled data the link will be NL.

Back to top

Matrix Plot: This is a gif image with the matrix plot of the graph, as described in the thesis. When viewing the results of a specific algorithm, the  rows of the top-10 results of the algorithm are highlighted (again as described in the thesis). Clicking on the image will open a new window with the image in bigger size.

Back to top