Compute similarity functions using graph traversal and computation along the path, eliminating the need for complex matrix extraction and computation. The computed similarity can be stored back in the graph as additional knowledge.