Are there any good social bookmarking tools out there?

What do social bookmarking and curation tools look like in  the semantic web?

• Is metadata user entered plaintext, suggested by its frequency? Is metadata scrupulously written out by content makers? Is it natural language processing keywords which extract concepts which can then be used as tags? Machine learning to learn what data is relevant, and recommending new links to you? How should ideas be linked, and how can these relationships be visualized? Lists? Tag clouds? Trees? Graphs?

• Answer:

  There's a structure continuum with NLP of less structured data on one end, folksonomies (easy tagging) near the middle, and ontologies with more structured data on the other end. Your choice depends on the amount of structure you need and how important it is to do the structuring. All three methods are relevant and helpful, some more than others. NLP looks at the structure that's implicit in natural language. Folksonomies add a bit more. Ontologies are machine readable, human assisted, explicitly stated, logical relationships.For example, if I state in a triple that Alice-->friendOf-->Carol, that's an assertion and a relationship the machine can read and use to infer more relationships, none of which might be implicit in the content on its own. We wouldn't need to add structure if the whole context of the data were implicit in it. It's not. When you ask Google a question and it doesn't answer it directly and precisely, the gap that's apparent between the answer you want and the often vague results list you get could be closed with better "tagging" or "structuring". The better the structure you add, the more machines can infer additional structure. So if you're going to invest to add structure because the data is important, ontologies are the the most powerful mechanism, because they enable that additional structuring to scale, and the newly machine-inferred connections to be made. We're building logic into the data layer to be able to serve up the content and data of most need to the points that content and data are aggregated and consumed. The object is to help users get to relevant insights and the things they're interested in seeing, reading and listening to faster. What the BBC and Ontotext have done with the World Cup site is a good example, and John O'Donovan explains it well. http://www.bbc.co.uk/blogs/bbcinternet/2010/07/bbc_world_cup_2010_dynamic_sem.html NLP gets at a fraction of the structure that would be helpful to make the content Web truly analyzable. It helps us get to a bit of that data layer logic. Let's say for the sake of argument that NLP and brute-force statistical analysis methods get us 30 percent there. Maybe it's more, maybe less. At some point, we will have to invest more to get more valuable linkages out of the data. Linkages and the integration they imply are where the power is, how you obtain more relevance. Otherwise, you're just left with the same old silos. As for how to represent these linkages, a visual approach is best. Half the population learns visually. Graphs with both nodes and edges labeled is a standard approach.The challenge is how to depict graphs that can get complicated quickly.Some sort of zoom in/out capability, or the ability to switch "altitudes" quickly is necessary. No need to reinvent the wheel. The RDF/RDFS/OWL stack is a rich set of standards. Those standards are being used a lot these days, to good effect. It's essential to get solid, practical guidance from those who understand the tools so you can cut to the chase. The Sunday workshops at SemTech are a good way to get an introduction to the standard methods: http://semtech2011.semanticweb.com/agenda.cfm?pgid=1. Here's a good overview my colleagues and I put together on the whys, hows and wherefores of the topic: http://www.pwc.com/us/en/technology-forecast/spring2009/index.jhtml