The Big-M method can be seen as a special case of the Two-Phase Simplex, now instead of $y_1,y_2,...,y_n$ dummy vars to reformate the base, we use a large value $M$. It is calculated pretty similarly to the Two-Phase Simplex as you can see here for the problem. Again the fundamental idea is this reduced cost formula $\bar{\bf{c'}_j}=c_j'-\bf{c'}_B B^{-1}A_j$. The Bertsimas has something on the page 117 but not as explicitly calculated example as here. I hope it helps!
Related information
Example

where the $15M$ should be $-15M$, a typo, ie $\bar{c}_0:=-\bf{c}_B'\bf{B}^{-1}b=-15M$. The rest is pretty much the traditional Simplex. The key is the differences with M -- you consider this very large $M$ as a number, not as a limit for the positive infinite.
Hard part
I haven't yet solved this but I got infeasible base with $[x_2,x_1,x_4]$ and not optimum situation here. Then I zwitched $x_4$ and $x_3$ but the new base $[x_2,x_1,x_2]$ is infeasible and not optimum as well. I may have done some mistake somewhere -- calculating the big M terms errorsome unless some thinking mistake in my simplex.