New Algorithm Ranks Sports Teams like Google’s PageRank
https://phys.org/news/2009-12-algorithm-sports-teams-google-pagerank.html
After learning about how PageRank works, I was curious about how ideas originating from the algorithm could be applied in other contexts. The above link describes a variant of the PageRank algorithm, which the article calls the “Power Rank” algorithm. The purpose of the algorithm is to rank sports teams (in this case, NFL and college football teams) using concepts derived from the PageRank algorithm.
PageRank uses a network in which websites are the nodes of the graph, and edges between the nodes represent links from one webpage to another. Similarly, the Power Rank algorithm is based on a network in which each node represents a team in the league, and edges exist in between teams that have played each other. However, unlike the PageRank algorithm in which a link represents 1 directional edge between two nodes, nodes in the Power Rank algorithm are linked by 2 bidirectional edges (if the teams have played each other). The value of each edge is the probability that the team at the head of the edge beats the team at the tail of the edge. Based on all of the games a team has played, every team/node will receive a base-value which is based on the value of its edges.
Similar to how PageRank ranks websites which are linked to by more authoritative websites, Power Rank ranks teams based on the quality of the other teams they have beaten (which is based on the values of those team’s nodes). Let’s look at the case in which team 1 achieves a certain base-value because it has beaten a number of low-value teams, and team 2 achieves an equal base-value but after beating an equal number of high-value teams. The Power Rank algorithm would rank team 2 higher because it has beaten more high-quality teams. This is similar to how PageRank would behave in the case of ranking two websites which are linked to an equal number of times, but the webpages linking to website 2 are much more authoritative than those linking to website 1 (website 2 would be ranked higher).
From this, we can see that the concepts associated with the PageRank algorithm can be cleverly applied in contexts outside of a web-search engine, particularly in the case of determining rankings.