Data-driven programming can probably have different meanings, but here 
is the one I use it for: it is a style of programming in which 
specialization is done through data structures and not boilerplate code.

Examples are probably a better way to understand the concept.

Instead of writing (pseudocode):

  DoSomething(a,b)
  DoSomething(c,d)
  DoSomething(e,f)

You write:

  things_to_do = [ [a,b], [c,d], [e,f] ]
  for each thing in things_to_do
    DoSomething(thing)

If you have first-class functions, instead of writing:

  switch x
    case a
      f(y)
    case b
      g(y)

You write:

switch = { a: f, b: g}
switch[x](y)

Now, more realistic examples. This one is extracted from a Ruby library 
I wrote, and it makes some methods defined on Arrays work on Strings. I 
think it illustrates the power of data-driven programming coupled with 
metaprogramming quite well.

  methods = %w{first first= head head= init last last= tail}
  methods.each{|n| define_method(n){|*a,&b| chars.to_a.send(n,*a,&b).join}}

If you prefer C, an excellent example is the way Redis commands are 
defined: see the "Globals" section of 
https://github.com/antirez/redis/blob/master/src/redis.c.