I am wondering if we can show in a simple way that $$ I=\int_0^1 \frac{\ln^2(1+x)}{x}dx = \int_1^2 \frac{\ln^2(t)}{t-1}dt = \frac{\zeta(3)}{4}. $$
Because the end result is very simple, I suspect that there might be a fast way to prove this. Can you prove it without using polylog identities? Complex analysis is allowed. It may be easier to show the equivalent identity $$ \sum_{k=1}^\infty \frac{(-1)^k H_k}{k^2} = -\frac{5 \zeta (3)}{8} $$ I know you can do that one with the generating function of the harmonic numbers, but that gives a nasty expression with polylogs which I would like to avoid.
Let us denote $I_{\pm}=\displaystyle \int_{0}^1\frac{\ln^2(1\pm x)}{x}dx$. We will express $I_+$ in terms of $I_-$, which is itself related to the standard integral representation of $\zeta(z)$ by the change of variables $x=1-e^{-t}$: $$I_-=\int_0^{\infty}\frac{t^2dt}{e^{t}-1}=2\zeta(3).$$ Indeed, we have \begin{align} \int_0^1\frac{\ln^2\frac{1+x}{1-x}}{x}dx=\int_0^{\infty}\frac{16t^2}{2\sinh 2t}dt&=\int_0^{\infty}16t^2\left(\frac{1}{e^{2t}-1}-\frac{1}{e^{4t}-1}\right)dt=\frac74 I_- \tag{1} \end{align} where the first equality is obtained by setting $x=\tanh t$. Also, it is easy to show ($x^2\to x$) that $$\int_{0}^1\frac{\ln^2(1-x^2)}{x}dx=\frac12I_-. \tag{2}$$ Summing (1) and (2), one finds that $ 2I_+ +2I_-=\left(\frac74+\frac12\right)I_-$, and hence $\displaystyle I_+=\frac{I_-}{8}=\frac{\zeta(3)}4$.