Bin hopping optimizes the virtual-to-physical map-ping for accesses that target repeated sequences of vir- tual memory locations. In practice, consecutively al- located physical pages that support consecutively allo- cated virtual pages are selected not to collide in the. Bin hopping therefore exploits temporal locality of applications.
Have you tried the reference  cited in the paper?
Kessler, R.E., Hill, M.D.: Page placement algorithms for large real-indexed caches. ACM Trans. Comput. Syst. 10 (1992) 338359.
From "An Empirical Study of the Effects of Careful Page Placement in Linux" by Bahadur, et al (http://portal.acm.org/ft_gateway.cfm?id=275365&type=pdf), I found this:
In dynamic coloring, bin selection is based upon some dynamic estimate of each bins activity level. Without special hardware such as that proposed in [J. Baker. The Effects of Virtual Page Placement on Cache Miss Ratios. M.S. Thesis, Department of Computer Science, Clemson University, Clemson SC, 1992], accurate estimates may be impossible to obtain. The principle of temporal locality indicates that a bin that has been recently assigned a new page is likely to experience relatively high activity. This observation motivates the round robin assignment of bins which is commonly known as bin hopping. Two variations of bin hopping have been studied and reported upon: global bin hopping in which the page fault manager simply maintains a single next-bin identifier and local bin hopping in which a next-bin identifier is maintained for each process.
I would also suggest trying texts on virtural memory systems.