Cytoscape will read in pretty much any text file that defines a network. Essentially, you want to construct a file such that each row represents an edge:

Term1 Term2
Term1 Term3
Term2 Term3
...

What's not clear to me from your description is how you want those terms connected. Typically, a gene will have multiple GO terms associated with it, so you are more likely to do something like:

Gene1 Term1
Gene1 Term2
Gene1 Term3
...

and then you would get a bipartite network between genes and terms. Another way to do the visualization is to construct a network of the GO hierarchy and for each term show how many genes have that annotation.

So, the question becomes -- what are you hoping to see with your network?

-- scooter