I encountered in a text the unsupported assertion that the series expansion of
$$(1-w)^{y_1/y_2+1} + \left(\frac{x}{y_2z_2}\right)(1-w)^{y_1/y_2} = 1$$
is
$$w = \left(\frac{1}{y}\right)\left(\frac{1}{z_2}\right)x - \left(\frac{1}{2y^2z}\right)x^2 - \left(\frac{1}{3y^3z^2}\right)\left(z_1-z_2\right)x^3 - \left(\frac{1}{8y^4z^3}\right)\left(2z_1^2 - 5z_1z_2 + 2z_2^2\right)x^4 + \ldots $$
for $0 \le \frac{x}{y_2z_2} \le 1$ and given that $\{ y_1, y_2, z_1, z_2 \} \in \mathbb{N}$, $y=y_1+y_2$, $z=z_1\cdot{}z_2$, $y_1z_1=y_2z_2$, and $1-w \ge 0$.
The series appears to have the basic form
$$w = -\sum_{i=1}^\infty \left(\frac{x^i}{i!! \cdot{} y^i z^{i-1}}\right) \left(??\right)$$
where the term indicated by question marks lacks a recognizable pattern ($\{-1/z_2, 1, z_1-z_2, (2z_1-z_2)(z_1-2z_2), \ldots \}$).
By what approach can this series expansion be derived?