I have scripts that are crawling the web in places like twitter, reddit and facebook to generate keyword ideas.

Right now, I'm filtering out the noise and crud manually! So it's a combination of automation and curation by hand.

additional question, maybe related to the question on noise : what is the Y axis? number of queries I suppose but are they normalized?

Yeah they're normalized relative to a scale of 100, as it's Google trends data. In the future I want to include actual search volume numbers as a line on the chart too though. Just need to figure out where to get that data first!