The RW algorithm has been proposed recently to solve the exact graph matching problem. This algorithm exploits Random Walk theory to compute a topological signature which can be used to match the nodes in two isomorphic graphs. However, the algorithm may suffer from the presence of colliding signatures in the same graph, which may prevent the procedure from finding the complete mapping between the matching nodes. In this paper we propose an improved version of the original algorithm, the RW2 algorithm, which progressively expands the node signatures by a recursive visit of the node descendants and ancestors to disambiguate the colliding signatures. The experimental results, performed on a benchmark dataset, show that the new algorithm attains a better matching rate with almost the same computational cost as the original one.