Applying Link Analysis beyond the Web
In class, we discussed the application of link analysis using hubs and authorities in PageRank. What about its application in other fields? An interesting article, “Hidden Relationships and Networks: Financial Institutions at Risk” by Oracle Financial Services, explore the application of link analysis in identifying hidden links between accounts and introduces its usefulness in combating money laundering or other financial crime.
Unlike the Web, links of financial accounts do not point to each other. There is no link you can click on to direct you to a new account like you can do with webpages. Instead, financial institutions approach the problem by examining shared links or information. Consider there exist 3 accounts that are owned by people under different names. account A and account B are linked by the same address, and account B and account C are linked by the same phone number. All three accounts are thus considered linked together.Note that unlike PageRank where hubs and authorities have scores and scores are updated in each update, it is not possible to assign scores to each account as the accounts’ transactions do not always appear suspicious in a standalone basis. Here, link analysis is used to find hidden links between accounts and uses the information to form a set of interrelated accounts that may appear to be unrelated accounts at first glance. Once a set of accounts is determined, financial institutions will examine if there is a pattern of transactions that take place within the set of accounts that constitute no apparent economic or lawful purposes.
It is really interesting to learn how the link analysis can be applied in other fields, specifically the financial field in this case. It also appears that the way that link analysis is used may vary depending on the purpose of the application. While a pattern instead of a weighted score was examined in this case, a similar logic applies. Just like an authority is likely to be highly endorsed if a reputable hub points to it, if an account that is already identified to be a money laundering account have systematic transactions with another account. It is likely that the other account is related to the money laundering activities. In a sense, I think we can view the accounts as hubs and authorities. If there is an identified account that has complex and unusually large transactions with many different other accounts, we can see it as a hub with a high score. Whichever account it has a transaction with, the score of the other account can be updated once. This will allow us to have a numerical sense of how all the accounts are linked together and provide a more quantitative observation of the patterns in the situation.
http://www.oracle.com/us/industries/financial-services/045950.pdf