To decrypt with this system, the decryptor first computes $g^{ab}$ (which he can do because he knows one of the two private exponents); then, he computes the modular inverse of $g^{ab}$; that is written as $(g^{ab})^{-1}$.
The modular inverse is defined the same way that the regular multiplicative inverse is defined in the reals (although there it is commonly referred to as the reciprocal): $b = a^{-1}$ is true iff $a\times b = 1$.
So, $(g^{ab})^{-1}$ is that value $x$ where $x \times g^{ab} \equiv 1 \bmod n$.
Modular inverses can be computed using the Extended Euclidean method; another method (which is usually more expensive, but not drastically so) that works if $n$ is prime is using the identity (for $n$ prime and $a \neq 0$):
$$a^{-1} \equiv a^{n-2} \bmod n$$
Once you have that, then the decryption can then be computed by a single modular multiplication:
$$m = y \times (g^{ab})^{-1} \bmod n$$
In addition: When using DH for encryption using the method you are using, it is more normally called ElGamal (after the person who first invented it); you might want to use that as the search term to look up more information on it.
In addition, there is a distinct (but generally better) alternative method for using the DH primitive to perform encryption; that is known as the Integrated Encryption Scheme; you might want to study that as well. One advantage that has is that it doesn't require a modular inverse to decrypt.