It may not be efficient but:

• Schemaless (easy updates, easy to combine data)
• Use an RDF processor that has reasoning
• Use various ontologies that will give different insights
• Take about data on the web ( not local )
• Well defined standard