You seem to've answered your own question, rhetorically speaking.
Is that a question of how we define words? If so, then what is the difference between data and knowledge?
The simple answer to the former is "yes". To the latter... you've already fitted them to the same definition by the way you framed your question.
You began by proposing two definitions of knowledge which are generally intertwined with belief in the semantic, intentional sense. "Familiarity" presupposes a host of emotional responses in sentient creatures, unless you want to define it in some other way from common use. Plato's "Justified True Belief" is the same, and you even say so.
But then you try to draw a contrast with terms like "knowledge management" and "knowledge representation", which is pointless. The fact that a term exists doesn't imply that it has a meaning. By using it here, you seem to be trying to propose that people use the terms in a way to refer to knowledge that objectively exists with no subjective stance - or, if I'm being charitable, at least that it exists if there's merely the possibility that a subjective stance could be held.
However, if you simply define "knowledge" to not include intentional beliefs, such as in regards to "knowledge representation", or if you define it in such a way that you've included data sets like RDF files, JPEGs, etc., then all you've done is use the word "knowledge" in the wrong way. You characterize knowledge as data sets, so obviously no one can draw a difference between them that's meaningful without providing a different meaning for knowledge - specifically, a definition which includes beliefs, which you've already excluded by drawing this contrast.
So is it a question of how we define words? Obviously. What's then then difference between knowledge and data? None, by that definition.
On the other hand, if you want to define knowledge a little more traditionally, then we can show that if you define data as any fixed string of bits, then data sets are obviously not considered knowledge. A person can take any random string of digits and then claim it's meaningful, given a particular context or encoding. This string still couldn't be considered "knowledge" until it maps to reality in some valid respect, but it would still be "data".
But - that very context or encoding that has to be added in order to provide meaning to the data, is exactly what defines our difference. Data by itself is opaque; it's just gibberish until someone supplies it with semantics. Knowledge must be transparent; if it doesn't map to the way the world is in some respect, it's not knowledge.