First, sit down with a stubby pencil and a sheet of paper. Work out the steps you'd need to take and the criteria by which you'll decide which step to take for any given instance of your data or, at least, for a representative sample of the data.

Then turn the result into pseudo-code -- or directly to code if you have the required knowledge and skills.

When you run into problems with particular pieces of the code you're trying to produce, try Tutorials, Super Search, the CB and if all that fails, post your question showing what you've done along with a clear explanation of what's happening that you didn't expect or not happening that you did expect.

Add a small sample of output, error messages and warnings, and -- of course -- a small sample of your data (enclosed in <code>...</code> tags). For more on formatting (your question above would surely be easier to read were it divided into several paragraphs or structured as a list), see Markup in the Monastery.

Actually, of course, since you didn't just get here yesterday, you probably know most of this but on the other hand, it appears to be your first post, so perhaps some or all the above will be useful.