Implementation of Oskarsdottir et al (2020) - Social network analytics for supervised fraud detection
cookbook
show how I extended BrianAronson's BiRank, implementing support for prior vectors of known fraudulent claims. I also provide explanation on how should one check for correctness of the birank score calculated, through two methods presented in the original He et al (2017) BiRank paper. Providing support for known fraudulent parties could also be done in a similar fashion, but it doesn't make sense for this specific use-case and has not been implemented.
Modelling was done on the Sample Network in Figure 1 and verified by comparing Claim C1's features values generated from the algorithm against data from Figure 5, Table 7 and Table 8. birank
notebook shows provides the extended Bipartite Class in birank.py
for implementation in your own work, with additional generate_prior
and check_birank
methods as discussed in cookbook
.