brotherhood_of_LAN - 10:17 pm on Oct 7, 2012 (gmt 0)
>which G themselves say is part of their "knowledge graph"
tbh isn't it a stretch to say the images are part of the 'knowledge graph'? At the most in your example of "things to do in paris", Google gets the intent, and the proper noun and most surely understands it's a location.... and can infer to display images it thinks are in the area of Paris, but the last part is more likely to do with plain old keyword matching rather than 'knowing' each individual image's relationship with paris. It's a bit like saying a moments glance at something is knowledge, without the thoughts behind it.
Though, this is perhaps getting off topic :o)