http://neuvc.com/labs/vcdelta or https://twitter.com/vcdelta, for daily feed
It wasn't in practice that hard to set up, it's a scraper in python with a DSL to tell it how each VC site set up their portfolio page.
The pain is maintaining it. Every week a few of the scrapers break because the VC changes their site. And recently, with so many seed funds starting up, it's missing a lot of the interesting new VCs.
[I once scraped Crunchbase for each VC firm, and the descriptions of every company they invested in, ran them all through a naive bayes model, and set it up so that you could type in your company description and it would tell you which firm was most likely to invest in you. It didn't work at all, giving me new appreciation for people who actually know how to make ML models work.]
In theory, not hard. It would be fun to run a LDA model and see what topics come up.
In practice, hard. Given that every VC website probably presents this info in slightly different ways, not to mention the variety in the portfolio company websites, not to mention the survivorhip bias (where is the info on the portfolio companies which failed)...
But if we hired a person to gather and structure all of this data... seriously, mechanical turks have their uses.