It would be very nice
I've experienced the same problem. It's relatively easy to extend your knowledge to a field close to your current knowledge. When you try to start understanding a field where you base knowledge is next to zero, the start can be pretty exhausting (usually not because the field is so hard, but because there is just such an information overflow, that it's next to impossible to get a good 'big picture' of the basis).
I would see the ideal solution as a combination of meta data, and search engines. Meta data needed for such a thing (as mentioned already by many others), would prerequisites (and skill level on them), skills 'gained' from the text (and an indepthness those are gone into), etc.. To not be 'total chaos' of key words, the pre/post classifications should be codes refering to a standardized tree of 'concepts' (which would surely be much easier task to create than a full tree trying to teach those concepts).
A search engine would build graphs based on the pre and post elements, and allow the person to then choose learning routes him/herself. Some of the reasons for such a meta-data/search engine approach, versus some static tree/graph approach: More scalable. Anyone can participate. Experts often totally disagree with each other, which I think would make creating a static moderated structure hard, without choosing sides.
Some of the problems would include: Abuse by web-site advertisers. Prevention of the system getting over flooded with average texts.
Perhaps some kind of a centralized rating system might help here, where a centralized database would keep average score given by readers to the articles for several different viewpoints, such as qualify of writing, relevance, how well it fits prerequisites, etc.. Of course the rating system could be abused too, but I could see ways on how this might be made statistically insignificant using some heuristics and weights.