Just a thought about how the product recommendation works on MS. Is the recommendation "score" just an average all the reviews on a product?
If it is just an average wouldn’t the recommendation score become outdated over long periods of time? Lets say a company started off in 2005 and they weren’t doing pretty well till 2009 and received quite a lot of bad reviews for 4 years on MS. But somehow the company pulled itself together and fixed all it’s issues and started getting good reviews till date. If the score was a plain average, wouldn’t the company’s recommendation score be a false negative? It has to get more number of good reviews within the 3 years after it turned around to be good for it’s customers than all the bad reviews it accumulated for 4 long years.
Personally I think the averaging approach wouldn’t work very well for this scenario. Besides people tend to rather complain more than appreciate online and this can’t be fixed very well mathematically.
I know scoring is one of the hardest problems for a community based website and even play store and iTunes are struggling with this. But there is one very good approach to solve this issue of false negatives and false positives. An exponential decay function for the score would be almost perfect. It may not be a perfect solution, but it is better than averaging. Ubuntu’s launchpad community uses this approach and call their user point system "karma". I actually like it. Karma starts "decaying" over time if you are not involved in the community anymore.
The idea in my mind would be the value of a review decays over time gradually and say between 6 months to 2 years (depending on the popularity of the product, number of people who found it useful, frequency of updates and any other parameters) it hits 0 and at this point, the user who wrote the initial review can get a chance to rewrite it or let the review disappear into the vastness of your server hard disks ( I am sure you have lots and they replicate on a raid :) ).
It’s just a personal thought and who knows if it might work or not ;-). If the scoring system is something more robust, my apologies to the MS tech team for doubting their prowess and honestly I am not!