Social Network Analysis is a technique that enables us to map and analyze the relations between entities. Louvain and Page Rank algorithms are two key techniques used to do this. In the earlier article, we introduced the Louvain algorithm. Let’s now talk about the page rank algorithm.
While clusters expose regions with highly localized interactions, dominant nodes in which clusters form may govern such interactions. We use the ranking to identify these dominant nodes.
PageRank is an algorithm that calculates a node's transitive impact or connectivity. PageRank is a web page ranking system developed at Stanford University by Google founders Larry Page and Sergey Brin. This is done by measuring the consistency and quantity of its connections, to assign each page a numerical score of value and authority.
Though we define PageRank here in its most popular website ranking sense, it can be used to rate nodes of any kind. PageRank of a web site is determined by three factors:
Number of links, Strength of links and Source of links
An incoming heavier-weight hyperlink means a larger influx of traffic to the website. To learn which website draws the most people, we might model the actions of many people surfing the web and analyze which website they finally land on. The total number of visitors to each website will then correlate to its PageRank – the higher the rank, the more visitors it draws. For example, A high PageRank country would be one in an oil trade network that engages in many high-value trade deals with other high-ranking countries — making it an influential player with a high degree of centrality in the global oil trade network.
The PageRank algorithm has one drawback, given its ease of use: it is skewed against older nodes. To stop that, PageRank rankings should be periodically changed to allow new websites a chance to boost ranks as they build up their reputations. The PageRank algorithm ranks nodes within a network based on their number of connections, as well as their intensity and source.
Although this allows us to classify dominant nodes within a network, it is also biased towards newer nodes, which would have had less time to establish significant connections.
It has recently started being used for fraud detection in many companies. Page Rank is appropriate for detecting fraud but may also be extended to many other business cases I will post soon about its implementation.
If you found this Article interesting, why not review the other Articles in our archive.