What is appropriate number of keywords for conference paper?

Honestly, I don't know if there is much reason other than history and tradition to continue to put keywords on conference papers. Most people will find your paper via search engines, rather than by searching for keywords in some specialized interface, and search engine technology is good enough these days that keywords are not usually particularly necessary for improving the chance of your paper being found.

The places where I know that keywords still actually matter, at least marginally, are:

• Putting in a keyword can be a good way to include a term that isn't otherwise explicitly stated in the paper (and thus might not show up on search engines).
• Some reviewing systems use an author's history of keywords to help figure out which papers they are likely to be a good reviewer for in the future.

So, given all of this, how many keywords should you choose? My own inclinations are as follows:

• When required to use the ACM classification system: two, because it's a pain to look up something appropriate in their ontology, but I feel like I'm shirking too obviously if I try to get away with only one.
• When I get to pick my own keywords: 3-6, plucked from the air over the course of a few minutes by free association.

From my experience in CS, I would put about four or five keywords. I would sort them from more specific to more general. But what you should actually do is to check other papers in the conference/journals where you will submit to know what other people have been doing in that specific community because it can vary depending on the community, the publisher, etc.